neighbors as negatives: relative earnings ...luttmer/relative.pdfneighbors as negatives: relative...
TRANSCRIPT
NEIGHBORS AS NEGATIVES RELATIVE EARNINGS ANDWELL-BEING
ERZO F P LUTTMER
This paper investigates whether individuals feel worse off when othersaround them earn more In other words do people care about relative position anddoes ldquolagging behind the Jonesesrdquo diminish well-being To answer this questionI match individual-level data containing various indicators of well-being to infor-mation about local average earnings I find that controlling for an individualrsquosown income higher earnings of neighbors are associated with lower levels ofself-reported happiness The datarsquos panel nature and rich set of measures ofwell-being and behavior indicate that this association is not driven by selection orby changes in the way people define happiness There is suggestive evidence thatthe negative effect of increases in neighborsrsquo earnings on own well-being is mostlikely caused by interpersonal preferences that is people having utility functionsthat depend on relative consumption in addition to absolute consumption
I INTRODUCTION
Classical economists understood that individuals are moti-vated at least partly by concerns about relative position AdamSmith [1759] for example wrote ldquoNothing is so mortifying as tobe obliged to expose our distress to the view of the public and tofeel that though our situation is open to the eyes of all mankindno mortal conceives for us the half of what we suffer Nay it ischiefly from this regard to the sentiments of mankind that wepursue riches and avoid povertyrdquo Arthur Pigou [1920] approv-ingly quotes John Stuart Millrsquos observation that ldquomen do notdesire to be rich but richer than other menrdquo 1 Of course thebelief that people compare themselves with others around themgoes back much further After all the framer of the Ten Com-mandments apparently judged it necessary to forbid humansfrom coveting their neighborrsquos possessions Not all humans how-ever appear to abide by this Commandment and possible effects
I would like to thank Iris Bohnet Suzanne Cooper David Cutler AngusDeaton Thomas DeLeire Joseph Doyle Susan Dynarski David Ellwood AmyFinkelstein Roland Fryer Edward Glaeser Caroline Hoxby Brian Jacob Chris-topher Jencks Matthew Kahn Lawrence Katz Elizabeth Keating Asim KhwajaBrian Knight Jeffrey Liebman Helen Levy Ellen Meara Benjamin Olken Mar-tin Sandbu Jesse Shapiro Cybele Raver Mark Rosenzweig Bas ter Weel Rich-ard Zeckhauser three anonymous referees and seminar participants at CarnegieMellon University Columbia University the NBER Labor Studies Summer In-stitute Princeton University the Wharton School and the University of Mary-land for helpful discussions and useful comments All errors are my own
1 This is quoted in Graham and Pettinato [2002]
copy 2005 by the President and Fellows of Harvard College and the Massachusetts Institute ofTechnologyThe Quarterly Journal of Economics August 2005
963
of social comparisons on consumption and savings behavior areanalyzed in the classic works of Veblen [1899] and Duesenberry[1949]
Though contemporary economists are aware that individualsmay care about relative position the accepted mainstream modelstates that individuals derive utility from their own consumptionU(C) rather than from a combination of own and relative con-sumption U(CCC ) where C denotes some measure of the con-sumption of relevant others2 For many applications it does notmatter whether utility has a relative component whenever C isfixed or given U(C) and U(CCC ) are isomorphic Indeed unlessan individualrsquos behavior can affect C U(C) and U(CCC ) cannotbe distinguished by individual behavior without placing addi-tional structure on the utility function3 In light of this it isperhaps not surprising that most economists tend to rely on anabsolute formulation of utility U(C)
Whereas individuals may in many cases take C as givenpolicy decisions often affect C Hence the distinction betweenabsolute and relative formulations of utility has important impli-cations for tax and expenditure policy as analyzed by Boskin andSheshinski [1978] Layard [1980] Oswald [1983] Ng [1987]Seidman [1987] Ireland [1998] Ljungqvist and Uhlig [2000] andAbel [2005] In particular if utility depends on relative consump-tion one personrsquos increase in consumption has a negative exter-nality on others because it lowers the relative consumption ofothers In this case taxes that discourage consumption are not asdistortionary as previously thought because they also serve tointernalize the negative externality of consumption on othersThe distinction between relative and absolute formulations ofutility is also pertinent to the welfare implications of residential
2 Becker [1974] introduces a more general framework for incorporatingsocial considerations into a utility function Samuelson [2004] and Rayo andBecker [2004] offer evolutionary explanations of relative consumption effectsPostlewaite [1998] discusses the advantages and drawbacks of modeling relativeposition as an argument of the utility function rather than as an instrument forgetting greater consumption in the future
3 A structure that specifies that relative concerns are more important forsome goods (eg present consumption or luxury consumption items) than forother goods (eg leisure or future consumption) has behavioral implications Seefor example Pollak [1976] or Frank [1985 1999] As Dupor and Liu [2003] makeclear if the consumption of others affects own marginal utility rather the level ofown utility the consumption of others can influence asset pricing risk takingsavings intensity of job search work effort economic growth and income inequal-ity See for example Abel [1990] Robson [1992] Galı [1994] Carroll Overlandand Weil [1997] Campbell and Cochrane [1999] Stutzer and Lalive [2004] andBecker Murphy and Werning [2005]
964 QUARTERLY JOURNAL OF ECONOMICS
sorting by income and to the debate about whether the povertyline should be absolute (a fixed consumption basket) or relative (afraction of mean or median income)
This paper provides evidence that suggests that utility de-pends in part on relative position I use panel data on individualsrsquoself-reported happiness other measures of well-being and othercharacteristics from the 1987ndash1988 and the 1992ndash1994 waves ofthe National Survey of Families and Households (NSFH) I matchthese data to information on local earnings where localities areso-called Public Use Microdata Areas (ldquoPUMAsrdquo) which haveabout 150000 inhabitants on average Average annual earningsin each PUMA are estimated by applying national earnings byindustry occupation and year from the Current Population Sur-vey to the industry and occupation mix of that PUMA from the1990 Census I find that higher PUMA-level earnings are associ-ated with lower levels of happiness controlling for a host ofindividual characteristics including income4 This is robust tochanges in specification and highly statistically significant Anincrease in neighborsrsquo earnings and a similarly sized decrease inown income each lead to a reduction in happiness of about thesame order of magnitude
This paper builds on previous papers that have empiricallyexamined the relationship between relative position and well-being5 In a series of papers Easterlin [1974 1995 2001] notesthat income and self-reported happiness are positively correlatedacross individuals within a country but that average happinesswithin countries does not seem to rise over time as countriesbecome richer Easterlin interprets these findings as evidencethat relative income rather than absolute income matters forwell-being but other studies have found that happiness is notpurely a relative concept [Veenhoven 1991 Diener et al 1993]Using European micro data Van de Stadt Kapteyn and Van deGeer [1985] Clark and Oswald [1996] Senik [2004] and Ferrer-i-Carbonell [2005] find that well-being is partly driven by relativeposition where reference groups are defined by demographic
4 Though at a conceptual level relative consumption rather than relativeincome or earnings affects well-being I use measures of earnings and income asproxies for consumption in the empirical section because of data availability
5 Frey and Stutzer [2002] provide an excellent review of this literatureLayard [2003] also discusses much of this literature as part of his engaging LionelRobbins Memorial Lectures on happiness
965NEIGHBORS AS NEGATIVES
characteristics6 Using U S data McBride [2001] and Blanch-flower and Oswald [2004] both find tantalizing evidence thatrelative income affects subjective well-being but they cautionabout the statistical reliability of their findings7
This paper contributes to this literature in three ways Firstit takes seriously the concern that living in an affluent area mightaffect onersquos definition of happiness even if it does not affect onersquostrue or experienced well-being I use other outcome measuresthat are less prone to definition shifts in response to neighborsrsquoearnings in order to investigate this concern and conclude that itis unlikely that this concern is driving the results
Second the paper examines whether the inverse relationshipbetween happiness and neighborsrsquo earnings might be spuriousdue to omitted individual or local characteristics The panel na-ture of the NSFH data enables me to run specifications withindividual fixed effects its detailed geographical information al-lows for the inclusion of local housing prices and state fixedeffects and the use of a predicted measure of local earnings filtersout many local earnings shocks caused by unobserved local fac-tors that might simultaneously influence happiness The resultshold up under these specifications reducing the concern that theyare due to omitted variable bias
Third the paper offers suggestive evidence concerning themechanism mediating the negative relationship between neigh-borsrsquo earnings and happiness I find evidence that the results arestronger for people who socialize more with neighbors but not forthose who socialize more with friends outside the neighborhoodThe paperrsquos findings indicate that interpersonal preferences that
6 In addition Clark [2003] finds effects of local unemployment on happinessthat may be explained by concerns about relative position
7 Using a sample of 324 individuals from the General Social SurveyMcBride [2001] finds that controlling for own income self-reported happinessdepends negatively on the average income in onersquos age cohort though this effectis only just significant at the 5 percent level in one of his two specifications and notsignificant in the other Blanchflower and Oswald [2004] find that controlling forown income there is a sizable but statistically insignificant negative effect of percapita state income on self-reported happiness providing suggestive evidence thatindividuals care about relative position They also find that relative income enterssignificantly if entered as household income per capitastate income per capitaHowever because this regression controls for the log of household income percapita rather than the level of household income per capita the relative incometerm may be significant because it offers an alternative functional form for ownhousehold income and not because of variation in state income per capita UsingCanadian data Tomes [1986] relates self-reported happiness to own income andincome in the local community He concludes that his results ldquodefy any simplecharacterization in terms of inequality aversion or relative economic statusrdquo
966 QUARTERLY JOURNAL OF ECONOMICS
incorporate relative income concerns drive the negative associa-tion between neighborsrsquo earnings and own well-being
II EMPIRICAL STRATEGY
Can data on behavior reveal whether peoplersquos well-being isaffected by the incomes of others around them Unless one as-sumes that neighborsrsquo incomes affect an individualrsquos marginalutility of a subset of goods the only behavior affected is theindividualrsquos choice of reference group implicit in the decisionabout where to locate8 Individualsrsquo concerns about relative posi-tion might then be capitalized in house prices with houses inhigh-income neighborhoods costing relatively less than similarhouses in low-income neighborhoods because a homeowner in arich neighborhood needs to be compensated for being relativelypoor9 This prediction of course only holds if individuals are both(i) aware that their utility depends on relative position and (ii)correctly forecast the utility effect of the change in referencegroup associated with moving Loewenstein OrsquoDonoghue andRabin [2003] describe a number of experiments that show sys-tematic biases in individualsrsquo predictions of their future utilityWith respect to endogenous reference groups they note that
when people make decisions that cause their comparison groups to changemdashsuch as switching jobs or buying a house in a new neighborhoodmdashprojectionbias predicts that people will underappreciate the effects of a change incomparison groups and hence consistent with Smithrsquos assertion overesti-mate the long-term satisfaction that would accompany such a change As aresult people may be prone to make reference-group-changing decisions thatgive them a sensation of status relative to their current reference group If aperson buys a small house in a wealthy neighborhood in part because it hasa certain status value in her apartment building she may not fully appreci-ate that her frame of references may quickly become the larger houses andbigger cars that her new neighbors have
These considerations make credible identification of relative in-come concerns from mobility decisions or housing price informa-tion very challenging
8 See Falk and Knell [2004] for evidence on reference group choices from aquestionnaire study
9 This prediction is derived from Frankrsquos [1984] model in which he analyzesthe effects of relative income concerns on wage distributions He assumes that thereference group consists of coworkers and deduces that a worker in a firm withhighly paid workers must be paid more than a similar worker in a low productivityfirm because the worker surrounded by highly paid workers needs to be compen-sated for the utility loss of being a relatively low earner
967NEIGHBORS AS NEGATIVES
The identification of relative income concerns therefore prob-ably falls in the limited set of research questions for which oneneeds to turn to a proxy for utility to answer it (see Di TellaMacCulloch and Oswald [2001] Gruber and Mullainathan[2002] or Frey Luechinger and Stutzer [2004] for other exam-ples of such questions)10 Though some skepticism toward self-reported measures of well-being is warranted (see eg Bertrandand Mullainathan [2001] and Ravallion and Lokshin [2001])Frey and Stutzer [2002] cite ample psychological evidence thatconfirms the validity and reliability of self-reported happiness asa measure of well-being and conclude that ldquothe existing researchsuggests that for many purposes happiness or reported subjec-tive well-being is a satisfactory empirical approximation to indi-vidual utilityrdquo
To determine whether well-being depends partly on relativeincome concerns one might then estimate an equation of theform11
1 well-being fown income average income in locality controls
To make the discussion of the empirical strategy more concrete avery basic linear OLS regression with self-reported happiness (ona 1ndash7 scale) as the dependent variable yields a coefficient of 020(se of 0014) on log own household income and a coefficient of017 (se of 004) on average log household income in onersquoslocality (PUMA)12 This regression previews the general findingsof the more elaborate regressions presented in detail in the re-sults section below Can this finding of a negative coefficient onaverage income in locality (and a positive one on own income) beinterpreted as evidence that utility is at least partly determinedby relative income Ideally we would have objective measures of
10 Zeckhauser [1991] Solnick and Hemenway [1998] and Johansson-Sten-man Carlsson and Daruvala [2002] find evidence for positional concerns byasking subjects to make choices over hypothetical scenarios with different levelsof absolute and relative income Van Praag and various coauthors pioneered theuse of subjective well-being questions in economics starting in the late sixtiesMuch of this work is nicely reviewed in Van Praag and Ferrer-i-Carbonell [2004]
11 I enter average income in locality separately in this equation rather thanin the form of the ratio of own income to average income in locality I do thisbecause in practice I have a number of proxies for own income instead of a singlemeasure
12 For simplicity no other controls are included in this illustrative regres-sion This is a pooled cross-section regression with the sample consisting of allNSFH main respondents from the balanced panel with nonmissing own income(N 16280) Average log household income in the PUMA is calculated from the1990 Census Public Use Micro Sample and refers to income in 1989 Standarderrors are adjusted for clustering at the PUMA level
968 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
of social comparisons on consumption and savings behavior areanalyzed in the classic works of Veblen [1899] and Duesenberry[1949]
Though contemporary economists are aware that individualsmay care about relative position the accepted mainstream modelstates that individuals derive utility from their own consumptionU(C) rather than from a combination of own and relative con-sumption U(CCC ) where C denotes some measure of the con-sumption of relevant others2 For many applications it does notmatter whether utility has a relative component whenever C isfixed or given U(C) and U(CCC ) are isomorphic Indeed unlessan individualrsquos behavior can affect C U(C) and U(CCC ) cannotbe distinguished by individual behavior without placing addi-tional structure on the utility function3 In light of this it isperhaps not surprising that most economists tend to rely on anabsolute formulation of utility U(C)
Whereas individuals may in many cases take C as givenpolicy decisions often affect C Hence the distinction betweenabsolute and relative formulations of utility has important impli-cations for tax and expenditure policy as analyzed by Boskin andSheshinski [1978] Layard [1980] Oswald [1983] Ng [1987]Seidman [1987] Ireland [1998] Ljungqvist and Uhlig [2000] andAbel [2005] In particular if utility depends on relative consump-tion one personrsquos increase in consumption has a negative exter-nality on others because it lowers the relative consumption ofothers In this case taxes that discourage consumption are not asdistortionary as previously thought because they also serve tointernalize the negative externality of consumption on othersThe distinction between relative and absolute formulations ofutility is also pertinent to the welfare implications of residential
2 Becker [1974] introduces a more general framework for incorporatingsocial considerations into a utility function Samuelson [2004] and Rayo andBecker [2004] offer evolutionary explanations of relative consumption effectsPostlewaite [1998] discusses the advantages and drawbacks of modeling relativeposition as an argument of the utility function rather than as an instrument forgetting greater consumption in the future
3 A structure that specifies that relative concerns are more important forsome goods (eg present consumption or luxury consumption items) than forother goods (eg leisure or future consumption) has behavioral implications Seefor example Pollak [1976] or Frank [1985 1999] As Dupor and Liu [2003] makeclear if the consumption of others affects own marginal utility rather the level ofown utility the consumption of others can influence asset pricing risk takingsavings intensity of job search work effort economic growth and income inequal-ity See for example Abel [1990] Robson [1992] Galı [1994] Carroll Overlandand Weil [1997] Campbell and Cochrane [1999] Stutzer and Lalive [2004] andBecker Murphy and Werning [2005]
964 QUARTERLY JOURNAL OF ECONOMICS
sorting by income and to the debate about whether the povertyline should be absolute (a fixed consumption basket) or relative (afraction of mean or median income)
This paper provides evidence that suggests that utility de-pends in part on relative position I use panel data on individualsrsquoself-reported happiness other measures of well-being and othercharacteristics from the 1987ndash1988 and the 1992ndash1994 waves ofthe National Survey of Families and Households (NSFH) I matchthese data to information on local earnings where localities areso-called Public Use Microdata Areas (ldquoPUMAsrdquo) which haveabout 150000 inhabitants on average Average annual earningsin each PUMA are estimated by applying national earnings byindustry occupation and year from the Current Population Sur-vey to the industry and occupation mix of that PUMA from the1990 Census I find that higher PUMA-level earnings are associ-ated with lower levels of happiness controlling for a host ofindividual characteristics including income4 This is robust tochanges in specification and highly statistically significant Anincrease in neighborsrsquo earnings and a similarly sized decrease inown income each lead to a reduction in happiness of about thesame order of magnitude
This paper builds on previous papers that have empiricallyexamined the relationship between relative position and well-being5 In a series of papers Easterlin [1974 1995 2001] notesthat income and self-reported happiness are positively correlatedacross individuals within a country but that average happinesswithin countries does not seem to rise over time as countriesbecome richer Easterlin interprets these findings as evidencethat relative income rather than absolute income matters forwell-being but other studies have found that happiness is notpurely a relative concept [Veenhoven 1991 Diener et al 1993]Using European micro data Van de Stadt Kapteyn and Van deGeer [1985] Clark and Oswald [1996] Senik [2004] and Ferrer-i-Carbonell [2005] find that well-being is partly driven by relativeposition where reference groups are defined by demographic
4 Though at a conceptual level relative consumption rather than relativeincome or earnings affects well-being I use measures of earnings and income asproxies for consumption in the empirical section because of data availability
5 Frey and Stutzer [2002] provide an excellent review of this literatureLayard [2003] also discusses much of this literature as part of his engaging LionelRobbins Memorial Lectures on happiness
965NEIGHBORS AS NEGATIVES
characteristics6 Using U S data McBride [2001] and Blanch-flower and Oswald [2004] both find tantalizing evidence thatrelative income affects subjective well-being but they cautionabout the statistical reliability of their findings7
This paper contributes to this literature in three ways Firstit takes seriously the concern that living in an affluent area mightaffect onersquos definition of happiness even if it does not affect onersquostrue or experienced well-being I use other outcome measuresthat are less prone to definition shifts in response to neighborsrsquoearnings in order to investigate this concern and conclude that itis unlikely that this concern is driving the results
Second the paper examines whether the inverse relationshipbetween happiness and neighborsrsquo earnings might be spuriousdue to omitted individual or local characteristics The panel na-ture of the NSFH data enables me to run specifications withindividual fixed effects its detailed geographical information al-lows for the inclusion of local housing prices and state fixedeffects and the use of a predicted measure of local earnings filtersout many local earnings shocks caused by unobserved local fac-tors that might simultaneously influence happiness The resultshold up under these specifications reducing the concern that theyare due to omitted variable bias
Third the paper offers suggestive evidence concerning themechanism mediating the negative relationship between neigh-borsrsquo earnings and happiness I find evidence that the results arestronger for people who socialize more with neighbors but not forthose who socialize more with friends outside the neighborhoodThe paperrsquos findings indicate that interpersonal preferences that
6 In addition Clark [2003] finds effects of local unemployment on happinessthat may be explained by concerns about relative position
7 Using a sample of 324 individuals from the General Social SurveyMcBride [2001] finds that controlling for own income self-reported happinessdepends negatively on the average income in onersquos age cohort though this effectis only just significant at the 5 percent level in one of his two specifications and notsignificant in the other Blanchflower and Oswald [2004] find that controlling forown income there is a sizable but statistically insignificant negative effect of percapita state income on self-reported happiness providing suggestive evidence thatindividuals care about relative position They also find that relative income enterssignificantly if entered as household income per capitastate income per capitaHowever because this regression controls for the log of household income percapita rather than the level of household income per capita the relative incometerm may be significant because it offers an alternative functional form for ownhousehold income and not because of variation in state income per capita UsingCanadian data Tomes [1986] relates self-reported happiness to own income andincome in the local community He concludes that his results ldquodefy any simplecharacterization in terms of inequality aversion or relative economic statusrdquo
966 QUARTERLY JOURNAL OF ECONOMICS
incorporate relative income concerns drive the negative associa-tion between neighborsrsquo earnings and own well-being
II EMPIRICAL STRATEGY
Can data on behavior reveal whether peoplersquos well-being isaffected by the incomes of others around them Unless one as-sumes that neighborsrsquo incomes affect an individualrsquos marginalutility of a subset of goods the only behavior affected is theindividualrsquos choice of reference group implicit in the decisionabout where to locate8 Individualsrsquo concerns about relative posi-tion might then be capitalized in house prices with houses inhigh-income neighborhoods costing relatively less than similarhouses in low-income neighborhoods because a homeowner in arich neighborhood needs to be compensated for being relativelypoor9 This prediction of course only holds if individuals are both(i) aware that their utility depends on relative position and (ii)correctly forecast the utility effect of the change in referencegroup associated with moving Loewenstein OrsquoDonoghue andRabin [2003] describe a number of experiments that show sys-tematic biases in individualsrsquo predictions of their future utilityWith respect to endogenous reference groups they note that
when people make decisions that cause their comparison groups to changemdashsuch as switching jobs or buying a house in a new neighborhoodmdashprojectionbias predicts that people will underappreciate the effects of a change incomparison groups and hence consistent with Smithrsquos assertion overesti-mate the long-term satisfaction that would accompany such a change As aresult people may be prone to make reference-group-changing decisions thatgive them a sensation of status relative to their current reference group If aperson buys a small house in a wealthy neighborhood in part because it hasa certain status value in her apartment building she may not fully appreci-ate that her frame of references may quickly become the larger houses andbigger cars that her new neighbors have
These considerations make credible identification of relative in-come concerns from mobility decisions or housing price informa-tion very challenging
8 See Falk and Knell [2004] for evidence on reference group choices from aquestionnaire study
9 This prediction is derived from Frankrsquos [1984] model in which he analyzesthe effects of relative income concerns on wage distributions He assumes that thereference group consists of coworkers and deduces that a worker in a firm withhighly paid workers must be paid more than a similar worker in a low productivityfirm because the worker surrounded by highly paid workers needs to be compen-sated for the utility loss of being a relatively low earner
967NEIGHBORS AS NEGATIVES
The identification of relative income concerns therefore prob-ably falls in the limited set of research questions for which oneneeds to turn to a proxy for utility to answer it (see Di TellaMacCulloch and Oswald [2001] Gruber and Mullainathan[2002] or Frey Luechinger and Stutzer [2004] for other exam-ples of such questions)10 Though some skepticism toward self-reported measures of well-being is warranted (see eg Bertrandand Mullainathan [2001] and Ravallion and Lokshin [2001])Frey and Stutzer [2002] cite ample psychological evidence thatconfirms the validity and reliability of self-reported happiness asa measure of well-being and conclude that ldquothe existing researchsuggests that for many purposes happiness or reported subjec-tive well-being is a satisfactory empirical approximation to indi-vidual utilityrdquo
To determine whether well-being depends partly on relativeincome concerns one might then estimate an equation of theform11
1 well-being fown income average income in locality controls
To make the discussion of the empirical strategy more concrete avery basic linear OLS regression with self-reported happiness (ona 1ndash7 scale) as the dependent variable yields a coefficient of 020(se of 0014) on log own household income and a coefficient of017 (se of 004) on average log household income in onersquoslocality (PUMA)12 This regression previews the general findingsof the more elaborate regressions presented in detail in the re-sults section below Can this finding of a negative coefficient onaverage income in locality (and a positive one on own income) beinterpreted as evidence that utility is at least partly determinedby relative income Ideally we would have objective measures of
10 Zeckhauser [1991] Solnick and Hemenway [1998] and Johansson-Sten-man Carlsson and Daruvala [2002] find evidence for positional concerns byasking subjects to make choices over hypothetical scenarios with different levelsof absolute and relative income Van Praag and various coauthors pioneered theuse of subjective well-being questions in economics starting in the late sixtiesMuch of this work is nicely reviewed in Van Praag and Ferrer-i-Carbonell [2004]
11 I enter average income in locality separately in this equation rather thanin the form of the ratio of own income to average income in locality I do thisbecause in practice I have a number of proxies for own income instead of a singlemeasure
12 For simplicity no other controls are included in this illustrative regres-sion This is a pooled cross-section regression with the sample consisting of allNSFH main respondents from the balanced panel with nonmissing own income(N 16280) Average log household income in the PUMA is calculated from the1990 Census Public Use Micro Sample and refers to income in 1989 Standarderrors are adjusted for clustering at the PUMA level
968 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
sorting by income and to the debate about whether the povertyline should be absolute (a fixed consumption basket) or relative (afraction of mean or median income)
This paper provides evidence that suggests that utility de-pends in part on relative position I use panel data on individualsrsquoself-reported happiness other measures of well-being and othercharacteristics from the 1987ndash1988 and the 1992ndash1994 waves ofthe National Survey of Families and Households (NSFH) I matchthese data to information on local earnings where localities areso-called Public Use Microdata Areas (ldquoPUMAsrdquo) which haveabout 150000 inhabitants on average Average annual earningsin each PUMA are estimated by applying national earnings byindustry occupation and year from the Current Population Sur-vey to the industry and occupation mix of that PUMA from the1990 Census I find that higher PUMA-level earnings are associ-ated with lower levels of happiness controlling for a host ofindividual characteristics including income4 This is robust tochanges in specification and highly statistically significant Anincrease in neighborsrsquo earnings and a similarly sized decrease inown income each lead to a reduction in happiness of about thesame order of magnitude
This paper builds on previous papers that have empiricallyexamined the relationship between relative position and well-being5 In a series of papers Easterlin [1974 1995 2001] notesthat income and self-reported happiness are positively correlatedacross individuals within a country but that average happinesswithin countries does not seem to rise over time as countriesbecome richer Easterlin interprets these findings as evidencethat relative income rather than absolute income matters forwell-being but other studies have found that happiness is notpurely a relative concept [Veenhoven 1991 Diener et al 1993]Using European micro data Van de Stadt Kapteyn and Van deGeer [1985] Clark and Oswald [1996] Senik [2004] and Ferrer-i-Carbonell [2005] find that well-being is partly driven by relativeposition where reference groups are defined by demographic
4 Though at a conceptual level relative consumption rather than relativeincome or earnings affects well-being I use measures of earnings and income asproxies for consumption in the empirical section because of data availability
5 Frey and Stutzer [2002] provide an excellent review of this literatureLayard [2003] also discusses much of this literature as part of his engaging LionelRobbins Memorial Lectures on happiness
965NEIGHBORS AS NEGATIVES
characteristics6 Using U S data McBride [2001] and Blanch-flower and Oswald [2004] both find tantalizing evidence thatrelative income affects subjective well-being but they cautionabout the statistical reliability of their findings7
This paper contributes to this literature in three ways Firstit takes seriously the concern that living in an affluent area mightaffect onersquos definition of happiness even if it does not affect onersquostrue or experienced well-being I use other outcome measuresthat are less prone to definition shifts in response to neighborsrsquoearnings in order to investigate this concern and conclude that itis unlikely that this concern is driving the results
Second the paper examines whether the inverse relationshipbetween happiness and neighborsrsquo earnings might be spuriousdue to omitted individual or local characteristics The panel na-ture of the NSFH data enables me to run specifications withindividual fixed effects its detailed geographical information al-lows for the inclusion of local housing prices and state fixedeffects and the use of a predicted measure of local earnings filtersout many local earnings shocks caused by unobserved local fac-tors that might simultaneously influence happiness The resultshold up under these specifications reducing the concern that theyare due to omitted variable bias
Third the paper offers suggestive evidence concerning themechanism mediating the negative relationship between neigh-borsrsquo earnings and happiness I find evidence that the results arestronger for people who socialize more with neighbors but not forthose who socialize more with friends outside the neighborhoodThe paperrsquos findings indicate that interpersonal preferences that
6 In addition Clark [2003] finds effects of local unemployment on happinessthat may be explained by concerns about relative position
7 Using a sample of 324 individuals from the General Social SurveyMcBride [2001] finds that controlling for own income self-reported happinessdepends negatively on the average income in onersquos age cohort though this effectis only just significant at the 5 percent level in one of his two specifications and notsignificant in the other Blanchflower and Oswald [2004] find that controlling forown income there is a sizable but statistically insignificant negative effect of percapita state income on self-reported happiness providing suggestive evidence thatindividuals care about relative position They also find that relative income enterssignificantly if entered as household income per capitastate income per capitaHowever because this regression controls for the log of household income percapita rather than the level of household income per capita the relative incometerm may be significant because it offers an alternative functional form for ownhousehold income and not because of variation in state income per capita UsingCanadian data Tomes [1986] relates self-reported happiness to own income andincome in the local community He concludes that his results ldquodefy any simplecharacterization in terms of inequality aversion or relative economic statusrdquo
966 QUARTERLY JOURNAL OF ECONOMICS
incorporate relative income concerns drive the negative associa-tion between neighborsrsquo earnings and own well-being
II EMPIRICAL STRATEGY
Can data on behavior reveal whether peoplersquos well-being isaffected by the incomes of others around them Unless one as-sumes that neighborsrsquo incomes affect an individualrsquos marginalutility of a subset of goods the only behavior affected is theindividualrsquos choice of reference group implicit in the decisionabout where to locate8 Individualsrsquo concerns about relative posi-tion might then be capitalized in house prices with houses inhigh-income neighborhoods costing relatively less than similarhouses in low-income neighborhoods because a homeowner in arich neighborhood needs to be compensated for being relativelypoor9 This prediction of course only holds if individuals are both(i) aware that their utility depends on relative position and (ii)correctly forecast the utility effect of the change in referencegroup associated with moving Loewenstein OrsquoDonoghue andRabin [2003] describe a number of experiments that show sys-tematic biases in individualsrsquo predictions of their future utilityWith respect to endogenous reference groups they note that
when people make decisions that cause their comparison groups to changemdashsuch as switching jobs or buying a house in a new neighborhoodmdashprojectionbias predicts that people will underappreciate the effects of a change incomparison groups and hence consistent with Smithrsquos assertion overesti-mate the long-term satisfaction that would accompany such a change As aresult people may be prone to make reference-group-changing decisions thatgive them a sensation of status relative to their current reference group If aperson buys a small house in a wealthy neighborhood in part because it hasa certain status value in her apartment building she may not fully appreci-ate that her frame of references may quickly become the larger houses andbigger cars that her new neighbors have
These considerations make credible identification of relative in-come concerns from mobility decisions or housing price informa-tion very challenging
8 See Falk and Knell [2004] for evidence on reference group choices from aquestionnaire study
9 This prediction is derived from Frankrsquos [1984] model in which he analyzesthe effects of relative income concerns on wage distributions He assumes that thereference group consists of coworkers and deduces that a worker in a firm withhighly paid workers must be paid more than a similar worker in a low productivityfirm because the worker surrounded by highly paid workers needs to be compen-sated for the utility loss of being a relatively low earner
967NEIGHBORS AS NEGATIVES
The identification of relative income concerns therefore prob-ably falls in the limited set of research questions for which oneneeds to turn to a proxy for utility to answer it (see Di TellaMacCulloch and Oswald [2001] Gruber and Mullainathan[2002] or Frey Luechinger and Stutzer [2004] for other exam-ples of such questions)10 Though some skepticism toward self-reported measures of well-being is warranted (see eg Bertrandand Mullainathan [2001] and Ravallion and Lokshin [2001])Frey and Stutzer [2002] cite ample psychological evidence thatconfirms the validity and reliability of self-reported happiness asa measure of well-being and conclude that ldquothe existing researchsuggests that for many purposes happiness or reported subjec-tive well-being is a satisfactory empirical approximation to indi-vidual utilityrdquo
To determine whether well-being depends partly on relativeincome concerns one might then estimate an equation of theform11
1 well-being fown income average income in locality controls
To make the discussion of the empirical strategy more concrete avery basic linear OLS regression with self-reported happiness (ona 1ndash7 scale) as the dependent variable yields a coefficient of 020(se of 0014) on log own household income and a coefficient of017 (se of 004) on average log household income in onersquoslocality (PUMA)12 This regression previews the general findingsof the more elaborate regressions presented in detail in the re-sults section below Can this finding of a negative coefficient onaverage income in locality (and a positive one on own income) beinterpreted as evidence that utility is at least partly determinedby relative income Ideally we would have objective measures of
10 Zeckhauser [1991] Solnick and Hemenway [1998] and Johansson-Sten-man Carlsson and Daruvala [2002] find evidence for positional concerns byasking subjects to make choices over hypothetical scenarios with different levelsof absolute and relative income Van Praag and various coauthors pioneered theuse of subjective well-being questions in economics starting in the late sixtiesMuch of this work is nicely reviewed in Van Praag and Ferrer-i-Carbonell [2004]
11 I enter average income in locality separately in this equation rather thanin the form of the ratio of own income to average income in locality I do thisbecause in practice I have a number of proxies for own income instead of a singlemeasure
12 For simplicity no other controls are included in this illustrative regres-sion This is a pooled cross-section regression with the sample consisting of allNSFH main respondents from the balanced panel with nonmissing own income(N 16280) Average log household income in the PUMA is calculated from the1990 Census Public Use Micro Sample and refers to income in 1989 Standarderrors are adjusted for clustering at the PUMA level
968 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
characteristics6 Using U S data McBride [2001] and Blanch-flower and Oswald [2004] both find tantalizing evidence thatrelative income affects subjective well-being but they cautionabout the statistical reliability of their findings7
This paper contributes to this literature in three ways Firstit takes seriously the concern that living in an affluent area mightaffect onersquos definition of happiness even if it does not affect onersquostrue or experienced well-being I use other outcome measuresthat are less prone to definition shifts in response to neighborsrsquoearnings in order to investigate this concern and conclude that itis unlikely that this concern is driving the results
Second the paper examines whether the inverse relationshipbetween happiness and neighborsrsquo earnings might be spuriousdue to omitted individual or local characteristics The panel na-ture of the NSFH data enables me to run specifications withindividual fixed effects its detailed geographical information al-lows for the inclusion of local housing prices and state fixedeffects and the use of a predicted measure of local earnings filtersout many local earnings shocks caused by unobserved local fac-tors that might simultaneously influence happiness The resultshold up under these specifications reducing the concern that theyare due to omitted variable bias
Third the paper offers suggestive evidence concerning themechanism mediating the negative relationship between neigh-borsrsquo earnings and happiness I find evidence that the results arestronger for people who socialize more with neighbors but not forthose who socialize more with friends outside the neighborhoodThe paperrsquos findings indicate that interpersonal preferences that
6 In addition Clark [2003] finds effects of local unemployment on happinessthat may be explained by concerns about relative position
7 Using a sample of 324 individuals from the General Social SurveyMcBride [2001] finds that controlling for own income self-reported happinessdepends negatively on the average income in onersquos age cohort though this effectis only just significant at the 5 percent level in one of his two specifications and notsignificant in the other Blanchflower and Oswald [2004] find that controlling forown income there is a sizable but statistically insignificant negative effect of percapita state income on self-reported happiness providing suggestive evidence thatindividuals care about relative position They also find that relative income enterssignificantly if entered as household income per capitastate income per capitaHowever because this regression controls for the log of household income percapita rather than the level of household income per capita the relative incometerm may be significant because it offers an alternative functional form for ownhousehold income and not because of variation in state income per capita UsingCanadian data Tomes [1986] relates self-reported happiness to own income andincome in the local community He concludes that his results ldquodefy any simplecharacterization in terms of inequality aversion or relative economic statusrdquo
966 QUARTERLY JOURNAL OF ECONOMICS
incorporate relative income concerns drive the negative associa-tion between neighborsrsquo earnings and own well-being
II EMPIRICAL STRATEGY
Can data on behavior reveal whether peoplersquos well-being isaffected by the incomes of others around them Unless one as-sumes that neighborsrsquo incomes affect an individualrsquos marginalutility of a subset of goods the only behavior affected is theindividualrsquos choice of reference group implicit in the decisionabout where to locate8 Individualsrsquo concerns about relative posi-tion might then be capitalized in house prices with houses inhigh-income neighborhoods costing relatively less than similarhouses in low-income neighborhoods because a homeowner in arich neighborhood needs to be compensated for being relativelypoor9 This prediction of course only holds if individuals are both(i) aware that their utility depends on relative position and (ii)correctly forecast the utility effect of the change in referencegroup associated with moving Loewenstein OrsquoDonoghue andRabin [2003] describe a number of experiments that show sys-tematic biases in individualsrsquo predictions of their future utilityWith respect to endogenous reference groups they note that
when people make decisions that cause their comparison groups to changemdashsuch as switching jobs or buying a house in a new neighborhoodmdashprojectionbias predicts that people will underappreciate the effects of a change incomparison groups and hence consistent with Smithrsquos assertion overesti-mate the long-term satisfaction that would accompany such a change As aresult people may be prone to make reference-group-changing decisions thatgive them a sensation of status relative to their current reference group If aperson buys a small house in a wealthy neighborhood in part because it hasa certain status value in her apartment building she may not fully appreci-ate that her frame of references may quickly become the larger houses andbigger cars that her new neighbors have
These considerations make credible identification of relative in-come concerns from mobility decisions or housing price informa-tion very challenging
8 See Falk and Knell [2004] for evidence on reference group choices from aquestionnaire study
9 This prediction is derived from Frankrsquos [1984] model in which he analyzesthe effects of relative income concerns on wage distributions He assumes that thereference group consists of coworkers and deduces that a worker in a firm withhighly paid workers must be paid more than a similar worker in a low productivityfirm because the worker surrounded by highly paid workers needs to be compen-sated for the utility loss of being a relatively low earner
967NEIGHBORS AS NEGATIVES
The identification of relative income concerns therefore prob-ably falls in the limited set of research questions for which oneneeds to turn to a proxy for utility to answer it (see Di TellaMacCulloch and Oswald [2001] Gruber and Mullainathan[2002] or Frey Luechinger and Stutzer [2004] for other exam-ples of such questions)10 Though some skepticism toward self-reported measures of well-being is warranted (see eg Bertrandand Mullainathan [2001] and Ravallion and Lokshin [2001])Frey and Stutzer [2002] cite ample psychological evidence thatconfirms the validity and reliability of self-reported happiness asa measure of well-being and conclude that ldquothe existing researchsuggests that for many purposes happiness or reported subjec-tive well-being is a satisfactory empirical approximation to indi-vidual utilityrdquo
To determine whether well-being depends partly on relativeincome concerns one might then estimate an equation of theform11
1 well-being fown income average income in locality controls
To make the discussion of the empirical strategy more concrete avery basic linear OLS regression with self-reported happiness (ona 1ndash7 scale) as the dependent variable yields a coefficient of 020(se of 0014) on log own household income and a coefficient of017 (se of 004) on average log household income in onersquoslocality (PUMA)12 This regression previews the general findingsof the more elaborate regressions presented in detail in the re-sults section below Can this finding of a negative coefficient onaverage income in locality (and a positive one on own income) beinterpreted as evidence that utility is at least partly determinedby relative income Ideally we would have objective measures of
10 Zeckhauser [1991] Solnick and Hemenway [1998] and Johansson-Sten-man Carlsson and Daruvala [2002] find evidence for positional concerns byasking subjects to make choices over hypothetical scenarios with different levelsof absolute and relative income Van Praag and various coauthors pioneered theuse of subjective well-being questions in economics starting in the late sixtiesMuch of this work is nicely reviewed in Van Praag and Ferrer-i-Carbonell [2004]
11 I enter average income in locality separately in this equation rather thanin the form of the ratio of own income to average income in locality I do thisbecause in practice I have a number of proxies for own income instead of a singlemeasure
12 For simplicity no other controls are included in this illustrative regres-sion This is a pooled cross-section regression with the sample consisting of allNSFH main respondents from the balanced panel with nonmissing own income(N 16280) Average log household income in the PUMA is calculated from the1990 Census Public Use Micro Sample and refers to income in 1989 Standarderrors are adjusted for clustering at the PUMA level
968 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
incorporate relative income concerns drive the negative associa-tion between neighborsrsquo earnings and own well-being
II EMPIRICAL STRATEGY
Can data on behavior reveal whether peoplersquos well-being isaffected by the incomes of others around them Unless one as-sumes that neighborsrsquo incomes affect an individualrsquos marginalutility of a subset of goods the only behavior affected is theindividualrsquos choice of reference group implicit in the decisionabout where to locate8 Individualsrsquo concerns about relative posi-tion might then be capitalized in house prices with houses inhigh-income neighborhoods costing relatively less than similarhouses in low-income neighborhoods because a homeowner in arich neighborhood needs to be compensated for being relativelypoor9 This prediction of course only holds if individuals are both(i) aware that their utility depends on relative position and (ii)correctly forecast the utility effect of the change in referencegroup associated with moving Loewenstein OrsquoDonoghue andRabin [2003] describe a number of experiments that show sys-tematic biases in individualsrsquo predictions of their future utilityWith respect to endogenous reference groups they note that
when people make decisions that cause their comparison groups to changemdashsuch as switching jobs or buying a house in a new neighborhoodmdashprojectionbias predicts that people will underappreciate the effects of a change incomparison groups and hence consistent with Smithrsquos assertion overesti-mate the long-term satisfaction that would accompany such a change As aresult people may be prone to make reference-group-changing decisions thatgive them a sensation of status relative to their current reference group If aperson buys a small house in a wealthy neighborhood in part because it hasa certain status value in her apartment building she may not fully appreci-ate that her frame of references may quickly become the larger houses andbigger cars that her new neighbors have
These considerations make credible identification of relative in-come concerns from mobility decisions or housing price informa-tion very challenging
8 See Falk and Knell [2004] for evidence on reference group choices from aquestionnaire study
9 This prediction is derived from Frankrsquos [1984] model in which he analyzesthe effects of relative income concerns on wage distributions He assumes that thereference group consists of coworkers and deduces that a worker in a firm withhighly paid workers must be paid more than a similar worker in a low productivityfirm because the worker surrounded by highly paid workers needs to be compen-sated for the utility loss of being a relatively low earner
967NEIGHBORS AS NEGATIVES
The identification of relative income concerns therefore prob-ably falls in the limited set of research questions for which oneneeds to turn to a proxy for utility to answer it (see Di TellaMacCulloch and Oswald [2001] Gruber and Mullainathan[2002] or Frey Luechinger and Stutzer [2004] for other exam-ples of such questions)10 Though some skepticism toward self-reported measures of well-being is warranted (see eg Bertrandand Mullainathan [2001] and Ravallion and Lokshin [2001])Frey and Stutzer [2002] cite ample psychological evidence thatconfirms the validity and reliability of self-reported happiness asa measure of well-being and conclude that ldquothe existing researchsuggests that for many purposes happiness or reported subjec-tive well-being is a satisfactory empirical approximation to indi-vidual utilityrdquo
To determine whether well-being depends partly on relativeincome concerns one might then estimate an equation of theform11
1 well-being fown income average income in locality controls
To make the discussion of the empirical strategy more concrete avery basic linear OLS regression with self-reported happiness (ona 1ndash7 scale) as the dependent variable yields a coefficient of 020(se of 0014) on log own household income and a coefficient of017 (se of 004) on average log household income in onersquoslocality (PUMA)12 This regression previews the general findingsof the more elaborate regressions presented in detail in the re-sults section below Can this finding of a negative coefficient onaverage income in locality (and a positive one on own income) beinterpreted as evidence that utility is at least partly determinedby relative income Ideally we would have objective measures of
10 Zeckhauser [1991] Solnick and Hemenway [1998] and Johansson-Sten-man Carlsson and Daruvala [2002] find evidence for positional concerns byasking subjects to make choices over hypothetical scenarios with different levelsof absolute and relative income Van Praag and various coauthors pioneered theuse of subjective well-being questions in economics starting in the late sixtiesMuch of this work is nicely reviewed in Van Praag and Ferrer-i-Carbonell [2004]
11 I enter average income in locality separately in this equation rather thanin the form of the ratio of own income to average income in locality I do thisbecause in practice I have a number of proxies for own income instead of a singlemeasure
12 For simplicity no other controls are included in this illustrative regres-sion This is a pooled cross-section regression with the sample consisting of allNSFH main respondents from the balanced panel with nonmissing own income(N 16280) Average log household income in the PUMA is calculated from the1990 Census Public Use Micro Sample and refers to income in 1989 Standarderrors are adjusted for clustering at the PUMA level
968 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
The identification of relative income concerns therefore prob-ably falls in the limited set of research questions for which oneneeds to turn to a proxy for utility to answer it (see Di TellaMacCulloch and Oswald [2001] Gruber and Mullainathan[2002] or Frey Luechinger and Stutzer [2004] for other exam-ples of such questions)10 Though some skepticism toward self-reported measures of well-being is warranted (see eg Bertrandand Mullainathan [2001] and Ravallion and Lokshin [2001])Frey and Stutzer [2002] cite ample psychological evidence thatconfirms the validity and reliability of self-reported happiness asa measure of well-being and conclude that ldquothe existing researchsuggests that for many purposes happiness or reported subjec-tive well-being is a satisfactory empirical approximation to indi-vidual utilityrdquo
To determine whether well-being depends partly on relativeincome concerns one might then estimate an equation of theform11
1 well-being fown income average income in locality controls
To make the discussion of the empirical strategy more concrete avery basic linear OLS regression with self-reported happiness (ona 1ndash7 scale) as the dependent variable yields a coefficient of 020(se of 0014) on log own household income and a coefficient of017 (se of 004) on average log household income in onersquoslocality (PUMA)12 This regression previews the general findingsof the more elaborate regressions presented in detail in the re-sults section below Can this finding of a negative coefficient onaverage income in locality (and a positive one on own income) beinterpreted as evidence that utility is at least partly determinedby relative income Ideally we would have objective measures of
10 Zeckhauser [1991] Solnick and Hemenway [1998] and Johansson-Sten-man Carlsson and Daruvala [2002] find evidence for positional concerns byasking subjects to make choices over hypothetical scenarios with different levelsof absolute and relative income Van Praag and various coauthors pioneered theuse of subjective well-being questions in economics starting in the late sixtiesMuch of this work is nicely reviewed in Van Praag and Ferrer-i-Carbonell [2004]
11 I enter average income in locality separately in this equation rather thanin the form of the ratio of own income to average income in locality I do thisbecause in practice I have a number of proxies for own income instead of a singlemeasure
12 For simplicity no other controls are included in this illustrative regres-sion This is a pooled cross-section regression with the sample consisting of allNSFH main respondents from the balanced panel with nonmissing own income(N 16280) Average log household income in the PUMA is calculated from the1990 Census Public Use Micro Sample and refers to income in 1989 Standarderrors are adjusted for clustering at the PUMA level
968 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
utility and all variation in neighborsrsquo earnings would be due toexogenous shocks such as national demand shocks to industriesoverrepresented in onersquos area Lacking such an ideal experimentI discuss below the three most serious threats to a causal inter-pretation of the coefficient on average income in locality andconsider ways of testing them
The first alternative story is that the definition of happinessshifts people answer the question about their happiness in rela-tive rather than absolute terms [Tversky and Griffin 1991 Fred-erick and Loewenstein 1999]13 In this case self-reported happi-ness would be a proxy for relative experienced well-being ratherthan absolute experienced well-being Suppose for example thateach individualrsquos experienced well-being Ui is equal to her incomeYi and that individuals are asked whether they are happy or notIndividuals now face the task of translating experienced well-being into an answer to this question If people respond that theyare happy whenever Ui exceeds some fixed (but possibly individ-ual-specific) cutoff value then they answer the question in abso-lute terms In this case an increase in everyonersquos income by thesame factor would increase the fraction of people answering thatthey are happy However if people respond that they are happywhenever their Ui exceeds some cutoff value that is a function ofthe population distribution of Ui (such as the mean or medianUi) they are answering the question in relative terms In thiscase an increase in everyonersquos income by the same factor may notaffect the proportion of individuals answering that they arehappy even though every individualrsquos experienced utility ishigher14 I address this concern by using alternative outcomemeasures that have a relatively objective definition such as thefrequency of marital disagreements about various topics
The second alternative story is that the results are driven byunobserved local area characteristics that are correlated withboth average local income and self-reported happiness Onemight expect most of this type of omitted variable bias to go in theother direction eg one would expect higher-income areas tohave less crime better local schools and other amenities that
13 King et al [2004] explain how vignettes can be used to anchor the scaleof subjective questions thereby making the answers comparable across people Nosuch vignettes however were used in the NSFH
14 This is also a potential explanation for the findings in a number of studiesthat levels of self-reported happiness or life satisfaction remain remarkably con-stant in a country over time even as incomes rise
969NEIGHBORS AS NEGATIVES
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
raise happiness The concern about local omitted variables driv-ing the result is addressed in three ways First if the results holdup after inclusion of state fixed effects they cannot be driven byunobservables that operate at that level such as climate statepolicies or regional shocks Second instead of using actual localincome I use a predicted measure of local earnings The predictoris based on the industry occupation composition of the localityat one point in time (1990) and national industry occupationearnings trends (excluding data from onersquos own state)15 Thuspredicted local earnings vary across areas for two reasons (i) theindustry occupation mix at a point in time and (ii) nationalearnings trends by industry and occupation which we have noreason to believe to be correlated with unobserved local shocksThis predictor therefore filters out any local shocks (such asquality of local government) that both affect happiness and localincomes conditional on industry occupation mix I refer to thismeasure as PUMA earnings or more informally as neighborsrsquoearnings and use it as the measure of local earnings throughoutthis paper unless otherwise noted The use of predicted localearnings however does not rule out the possibility that areaswith an overrepresentation of high-paying industries and occu-pations may tend to have unobserved characteristics such ashigher housing prices that reduce happiness Third to addressthe concern that neighborsrsquo earnings simply proxy for local hous-ing prices I examine whether the results are robust to includinglocal housing price measures as controls and whether the resultshold up when we control for an individualrsquos predicted real income(based on education age and average national earnings for some-one in the same industry and occupation as the respondent)instead of the personrsquos actual income
The third alternative story is that the results are driven byomitted individual characteristics that influence both the deci-sion where to live and self-reported happiness In particularselection of individuals with unobservables that make them rela-tively happy (or relatively likely to report being happy) intolocalities with relatively low incomes would also result in a nega-tive coefficient on average income in locality Though one mightexpect that most selection would go in the opposite direction (high
15 This predictor follows similar predictors used by Bartik [1991] Blan-chard and Katz [1992] Bound and Holzer [2000] and Autor and Duggan [2003]See Appendix 3 for details on the construction of this predictor
970 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
income in onersquos locality proxying for higher unobserved own in-come) there may be selection effects that lead to a spuriousnegative effect of average income in locality For example intrin-sically happy people might be better able to deal with the rougheraspects of low-income areas and thus choose to live there Thispaper exploits the panel aspect of the NSFH data to deal with thisconcern If after inclusion of individual fixed effects averageincome in the locality still matters then we know that time-invariant unobserved individual characteristics cannot be drivingthe results
To preserve statistical power the baseline specification totest for relative income concerns is a pooled cross-section OLSregression of the form
(2) Happinessipst PUMA earningspt1 Xit2
Xp3 wavet4 s εipst
where i indexes individuals p indexes PUMAs s indexes statesand t indexes the wave of the survey Happiness is self-reportedhappiness while PUMA earnings are average predicted log earn-ings in the PUMA of the respondent where the prediction is basedon the PUMArsquos industry occupation composition and nationalearnings trends The vector Xit is a set of individual-specificcontrols that include a number of proxies for income as well asbasic demographics while the vector Xp contains other PUMAcharacteristics such as its racial composition Finally wavet is adummy for the wave of the survey s is a full set of statedummies and εipst is an error term that may be clustered withinPUMAs If individuals derive utility in part from relative posi-tion we would expect 1 to be negative
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH I limit the sample tomarried or cohabiting individuals for two reasons First for theseobservations we have information about spouses or interactionswith onersquos spouse which are useful in a number of further tests ofthe baseline results Second it turns out that married individualsdrive the baseline results though neighborsrsquo earnings still have anegative and significant effect on happiness in the full samplethat includes nonmarried individuals
Since most survey questions are asked of both the mainrespondent and his or her spouse it makes sense to exploitspousal information Adding this information as a separate ob-
971NEIGHBORS AS NEGATIVES
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
servation to the regression may bias standard errors downwardsbecause the error term of the respondent and the error term of hisor her spouse are likely to be correlated16 Instead I average thevalues of the individual-level variables for the main respondentand his or her spouse and enter those as a single observation inthe regression I thus do not exploit intrahousehold-level varia-tion but this does not matter since the main explanatory vari-ables of interest (neighborsrsquo earnings and own household income)do not vary within households17
In the results section I will present various modifications ofthis baseline regression to investigate whether the baseline re-sults are spurious whether they are robust and what mecha-nisms drive them These modifications will be explained in detaillater and include exploring other outcome or control variablesadding individual fixed effects and adding interactions betweenPUMA earnings and other variables
III DATA
IIIA National Survey of Families and Households
The data on subjective well-being as well as the individual-level control variables come from the National Survey of Familiesand Households18 The NSFH consists of a nationally represen-tative sample of individuals aged nineteen or older (unless mar-ried or living in a household with no one aged nineteen or older)living in households and able to speak English or Spanish Thefirst wave of interviews took place in 1987ndash1988 and a secondwave of interviews took place in 1992ndash1994 Though the ques-tionnaires are not identical in both waves many questions wereasked twice making it possible to treat the data as a panel ofabout 10000 individuals This data set is particularly well-suitedfor this paper because it can be merged with detailed geographicinformation The respondents of the baseline sample live in 580
16 I already cluster at the PUMA level (the level at which neighborsrsquo earn-ings vary) and therefore cannot easily cluster at the household level at the sametime In unreported regressions I have entered the information of the mainrespondent and the spouse as separate observations This yields basically thesame point estimates with somewhat smaller standard errors
17 See Clark [1996] for a study of comparison effects within households18 The NSFH is a survey that was primarily designed for demographers
interested in family and household issues More information on the NSFH can befound in Sweet Bumpass and Call [1988] in Sweet and Bumpass [1996] or at theNSFH website httpwwwsscwiscedunsfhhomehtm
972 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
separate Public Use Microdata Areas in the first wave in 965PUMAs in the second wave while 555 PUMAs have respondentsliving there in both waves (more about the definition of PUMAslater)
The main outcome variable is self-reported happiness whichis the answer to the question ldquoNext are some questions abouthow you see yourself and your life First taking things all to-gether how would you say things are these daysrdquo Respondentsanswer on a seven-point scale where 1 is defined as ldquovery un-happyrdquo 7 is defined as ldquovery happyrdquo and intermediate values arenot explicitly defined Figure I shows the distribution of responsesto this question from the main respondents and spouses in thebaseline sample Other interesting outcome measures include thefrequency of open disagreements with onersquos spouse on a numberof topics items from Radloffrsquos [1977] depression scale and onlyin the second wave self-reported satisfaction with various as-pects of onersquos life and the frequency of financial worries Appen-dices 1 and 2 contain detailed definitions and summary statisticsfor these outcome variables as well as for the control variablesfor income labor market participation and other demographiccharacteristics
IIIB Census and Current Population Survey
The smallest geographical area in the 1990 Census 5 percentPublic Use Micro Sample (PUMS) is the so-called Public UseMicrodata Area PUMAs consist of neighborhoods towns or
FIGURE IDistribution of Self-Reported Happiness
973NEIGHBORS AS NEGATIVES
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
counties aggregated up or subdivided until they contain at least100000 inhabitants In 1990 there were 1726 PUMAs in theUnited States and the median and mean size of a PUMA was127000 and 144000 inhabitants respectively The 1990 Censusmicrodata are used to estimate the three-digit industry three-digit occupation composition of each PUMA which is later used topredict PUMA earnings In addition I use the Census to estimateaverage earnings and income in 1989 for each PUMA which willserve as a check on the predictor
I use the Merged Outgoing Rotation Groups (MORG) fromthe Current Population Survey (CPS) in the years 1987ndash1988 and1992ndash1994 to estimate average earnings by three-digit indus-try three-digit occupation cell in each of the two time periodswhen NSFH interviews took place For each PUMA I calculatethese average national earnings by time-period industry occupation cell excluding data from the state in which the PUMAlies
Predicted PUMA earnings for each time period are aweighted average of national earnings by industry occupationcell during that time period where the weights are the employ-ment shares of each industry occupation cell in the PUMA in1990 The resulting predictor of PUMA earnings thus variesacross PUMAs because of variation in the industry occupationmix of PUMAs in 1990 As the industry occupation employmentshares are held constant within each PUMA the predictor variesover time solely due to changes in national earnings by indus-try occupation cell Details of this procedure are found inAppendix 3
IV RESULTS
IVA Basic Results
The first column of Table I shows the baseline specification infull This is a pooled cross-section OLS regression of self-reportedhappiness on predicted PUMA log earnings individual controlsstate fixed effects and controls for metropolitan area size andracial composition in the PUMA Individual level variables in thisregression are averages of the main respondent and his or herspouse Robust standard errors are corrected for clustered errorterms at the PUMA level and the sample includes all NSFHrespondents who are married or cohabiting in both waves The
974 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
TABLE IBASELINE REGRESSION
Dependent variable (1) (2) (3)
Self-reported happinessBaseline
Only mainrespondent
IV for ownincome
Coeff SE Coeff SE Coeff SE
PUMA ln earnings (predicted) 0239 0066 0248 0083 0296 0076ln Household income 0123 0020 0111 0024 0361 0102ln Value of home 0068 0021 0073 0025Renter 0172 0032 0209 0038ln Usual working hours 0072 0044 0113 0036 0138 0052Unemployed 0431 0126 0254 0115 0355 0150Not in the labor force 0141 0043 0059 0043 0240 0065Female 0044 0034Age 0031 0005 0029 0007 0038 0008Age2100 0033 0006 0031 0007 0040 0008White (omitted)Black 0044 0060 0019 0068 0025 0065Hispanic 0277 0069 0197 0080 0297 0071Asian 0047 0117 0125 0131 0011 0117Other raceethnicity 0115 0212 0153 0396 0072 0206Years of education 0010 0007 0010 0007 0006 0012ln Household size 0180 0036 0199 0042 0188 0037Catholic (omitted)No religion 0168 0060 0161 0062 0162 0060Jewish 0289 0098 0272 0111 0280 0098Baptist 0112 0042 0050 0047 0118 0042Episcopalian 0090 0091 0027 0099 0089 0093Lutheran 0014 0057 0106 0066 0009 0057Methodist 0035 0046 0051 0052 0032 0047Mormon 0041 0108 0091 0128 0039 0110Presbyterian 0018 0074 0109 0079 0016 0074Congregational 0018 0097 0097 0102 0041 0100Protestant no denomination 0052 0106 0038 0078 0071 0106Other Christian 0096 0048 0017 0054 0107 0050Other religionsmissing 0049 0104 0066 0120 0076 0107ln Metropolitan area population 0005 0014 0000 0018 0007 0014Nonmetropolitan area 0036 0040 0058 0050 0047 0040Fraction Black in PUMA 0200 0123 0213 0154 0252 0125State fixed effects Yes Yes YesAdjusted R2 00388 00247 Number of observations 8944 8023 8944
Significance levels 10 percent 5 percent Robust standard errors adjusted for clustering on PUMAs(1000 clusters in specifications (1) and (3) 974 clusters in specification (2)) ln usual hours ln home value andln Metropolitan area population are demeaned All regressions also include dummy variables for independentvariables with missing values and for logarithms of dollar values smaller than $100year Self-reportedhappiness is measured on a scale of 1 to 7 with 7 representing ldquovery happyrdquo The sample consists ofrespondents of NSFH waves 1 and 2 who are married or cohabiting in both waves In specifications (1) and(3) the variables are the average of the respondentrsquos value and that of his or her spouse In specification (3)ln household income is instrumented by predicted in household earnings where predicted earnings are basedon the industry occupation of the respondent and his or her spouse
975NEIGHBORS AS NEGATIVES
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
first row shows that predicted PUMA earnings have a signifi-cantly negative effect on self-reported happiness In other wordscontrolling for other factors individuals living in richer areasreport being less happy As expected own household income hasa positive effect on happiness but its coefficient may be relativelysmall because the regression includes other proxies for incomesuch as the value of onersquos home and a dummy variable for rent-ing19 Usual working hours has an insignificant negative effectunemployment status has a large and significant negative effectwhile a dummy for being out of the labor force has a significantpositive effect on happiness20 The other demographic controlsyield few surprising insights
Column 2 shows the same regression using only data fromthe main respondent rather than averaging the respondentrsquos datawith that of the spouse This regression confirms that the resultsare not sensitive to the averaging procedure used in the baselineregression though power increases moderately when the data ofthe main respondent and the spouse are averaged
A one-standard deviation (027) increase in neighborsrsquo earn-ings reduces self-reported happiness by 0065 or 6 percent of astandard deviation This effect may seem small though oneshould keep in mind that there is a lot of idiosyncratic variationin self-reported happiness It is instructive to compare the size ofthe effect of neighborsrsquo earnings to that of onersquos own income Inthe baseline specification this comparison may be misleadingbecause the coefficient on own household income is likely biaseddownwards due to measurement error in income and the inclu-sion of other income proxies In specification (3) I try to get a moreaccurate estimate of the effect of own income by eliminating twoother income proxies (home value and the dummy for renter) andby instrumenting log household income by the predicted house-hold earnings where the prediction is based on the industry occupation information of the respondent and the spouse andnational earnings information (excluding the own state) by indus-try occupation and time period from the CPS MORG21 Instru-menting results in an estimated effect of own household income
19 ln Value of home has been demeaned Hence the dummy for renter isrelative to a homeowner with a home of average value
20 ln Usual working hours has been demeaned Hence the dummies forunemployment and out of the labor force are relative to an employed personworking the average number of hours
21 See Appendix 3 for details on the construction of this predictor
976 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
on happiness that is about three times as large as the estimate inthe baseline specification22 Moreover the estimate is now largerthan the absolute value of the estimate of neighborsrsquo earningsthough this difference is not statistically significant Thus thepoint estimates imply that if both own income and neighborsrsquoearnings rise by the same percentage a person would feel betteroff indicating that onersquos absolute economic situation matters forhappiness in addition to onersquos relative position though I cannotreject the hypothesis that only relative position matters
IVB Could the Results Be Spurious
Table II investigates whether omitted area characteristics orselection could be driving the results The first row of the tablereproduces the baseline results
The second row includes individual-specific fixed effects thuscontrolling for all time-invariant individual characteristics Thecoefficient on neighborsrsquo earnings remains negative and statisti-cally significant This finding discounts the possibility that thecross-section results are driven by selection of people who arehappier by nature into areas that are relatively poor Of coursethis specification does not rule out selection based on unobservedtime-variant characteristics One might worry that movers mayhave had something unobserved happen to them and that per-haps this unobserved factor caused their happiness to be in-versely related to average neighborsrsquo earnings The third rowtests this by showing the baseline regression excluding all re-spondents who moved to a different PUMA Again the coefficientson neighborsrsquo earnings and own income are hardly affected show-ing that the baseline results are not just driven by movers
The regression in the fourth row is identical to the one in thethird row except that it includes individual fixed effects Becausethe sample is limited to nonmovers the individual fixed effectsalso serve as PUMA fixed effects (ie any PUMA fixed effectswould be absorbed by the individual fixed effects) Thus thecoefficient on neighborsrsquo income in this regression is purely iden-tified from changes in neighborsrsquo earnings that are solely due todifferent national trends in earnings in different industry
22 The coefficient on own income rises to 016 if the controls for home valueand the renter dummy are dropped Thus most of the rise of the coefficient on ownincome is due to the instrument In the remainder of the paper I use specification(1) rather than specification (3) as the baseline specification because it allows fora richer set of controls for own income
977NEIGHBORS AS NEGATIVES
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
occupation cells The individual fixed effects absorb any time-invariant individual characteristics as well as any effect corre-lated with the industry occupation composition of each PUMATime-varying unobserved characteristics cannot affect onersquosneighborsrsquo earnings because the sample is limited to nonmoversUnfortunately the standard error in this specification is too largefor this regression to provide meaningful evidence on selection
Because the baseline regression already includes state fixedeffects log metropolitan population size a dummy for nonmetro-politan areas and the fraction of the PUMA population that isBlack the results are driven by variation in neighborsrsquo earningsthat occurs within states and that is unrelated to metropolitanarea size or the racial composition of the PUMA Still one mayworry that neighborsrsquo earnings might proxy for local omittedvariables in particular local price levels In this case the nega-tive coefficient on neighborsrsquo earnings would simply reflect that
TABLE IITESTING FOR SELECTION
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
PUMA lnhousing price
AdjR2 NSpecification Coeff SE Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Individual fixed
effects0369 0165 0054 0053 03133 8944
(3) Observationsremaining in samePUMA
0224 0084 0129 0024 00423 6894
(4) Observationsremaining in samePUMA and individualfixed effects
0414 1460 0046 0065 03172 6894
(5) Using predictedhousehold incomeinstead of householdincome and homevalue
0164 0066 00333 8944
(6) Controlling for PUMAln housing price(adjusted for housingcharacteristics)
0225 0079 0123 0020 0018 0057 00387 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 Whenever individual fixed effects are included spousal variables are only used if the identity of the spouseremains the same in both waves
978 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
happiness is lower in areas where high price levels depress realincome23 Because neighborsrsquo earnings are predicted based on thelocal industry occupation mix and national earnings data (ex-cluding data from onersquos own state) local price variation wouldonly be picked up by my measure of neighborsrsquo earnings to theextent that it is correlated with the local industry occupationmix I examine this possibility in two ways
First in order for neighborsrsquo earnings to proxy for local pricelevels it must be the case that earnings for the average respon-dent are higher in nominal terms for respondents living in moreexpensive areas Thus if we do not measure a respondentrsquos in-come using monetary variables (such as household income andhome value) but instead use proxies that do not respond to localwage levels (national earnings in the respondentrsquos industry occupation cell) then there would be no role for neighborsrsquo earn-ings to serve as a control for local prices The fifth row of Table IIestimates the baseline regression purged of any controls thatproxy for the respondentrsquos nominal income The coefficient onPUMA earnings drops in size but remains negative and signifi-cant thus ruling out that PUMA earnings is just picking upvariation in local price levels
Second I try to control directly for housing prices which isprobably the most important component of local prices Howeverhousing values vary because of variation in both the price ofhousing services and the quantity of housing services provided bythe property Ideally only the former component should be in-cluded as a control because the latter component will also proxyfor neighborsrsquo incomes since housing services are a normal goodTo isolate the price component from the quantity component Iuse the 1990 Census 5 percent PUMS to run a hedonic regressionof log home value on a set of PUMA fixed effects and all availablehousing characteristics To the extent that the measured housingcharacteristics adequately capture consumption of housing ser-vices the PUMA fixed effects capture the price component In row6 I control for local housing prices (as measured by these PUMA
23 Note however that if higher local prices reflect positive local amenitiesthey do not reduce real income in effect the individual is purchasing the localamenity by locating in an expensive area This means that only variation in pricesdue to transportation cost or local production costs could possibly explain thefindings Because of the state fixed effects the transportation or production costdifferences should be within states to explain away the results The scope for suchvariation is considerably less than the scope of transportation and production costdifferences in the nation as a whole
979NEIGHBORS AS NEGATIVES
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
fixed effects) but the coefficient on PUMA earnings remainssimilar in size and statistically significant confirming that localprice variation does not drive the effect of neighborsrsquo earnings24
Table III investigates the robustness of the baseline resultsThe first row again reproduces the baseline regression One
24 The finding that local housing prices (adjusted for housing characteris-tics) do not enter significantly in the regression may seem surprising This is trueeven if I use unadjusted average log home value as the measure of local housingprices ie this result is not driven by some peculiarity of the hedonic home valueregression Instead it appears that the state fixed effects and the control formetropolitan area size already absorb most of the housing price variation With-out the state fixed effects and the control for metropolitan area size both theunadjusted and the adjusted measure of local housing prices enter negatively andsignificantly In that case the coefficient on neighborsrsquo earnings remains negativeand significant with adjusted local housing prices as a control but it is no longersignificant if unadjusted local housing prices are used as a control This latterfinding is not surprising since my predicted measure of neighborsrsquo earnings andunadjusted local housing values are highly correlated (r 069) and both areimperfect measures of neighborsrsquo true incomes
TABLE IIIROBUSTNESS CHECKS
Dependent variableSelf-reported happiness
PUMA lnearnings ln HH income
AdjR2 NSpecification Coeff SE Coeff SE
(1) Baseline 0239 0066 0123 0020 00388 8944(2) Actual PUMA ln earnings
in 1989 as control(instead of predictedPUMA ln earnings)
0333 0085 0123 0020 00390 8944
(3) PUMA ln income in 1989as control (instead ofpredicted PUMA lnearnings)
0241 0069 0122 0020 00387 8944
(4) ln earnings of R andspouse as control (insteadof ln HH income)
0221 0066 0080 0017 00387 8944
(5) Full balanced panelsample (includingnonmarried individuals)
0145 0051 0127 0017 00655 15568
(6) Controlling for 5th-orderpolynomial in lnhousehold income
0243 0066 0161 0030 00391 8944
(7) Ordered probit 0228 0065 0112 0020 00126 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 The regression in specification (5) also includes four marital status dummies as controls The terms in thepolynomial in specification (6) are demeaned Hence the coefficient on the first term (reported in the table)is the slope of ln household income for someone with mean ln household income
980 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
might be concerned that the results are driven by the somewhatcomplicated procedure used to predict PUMA earnings The sec-ond row alleviates this concern if anything the estimate onPUMA earnings becomes more negative and more significant ifwe replace the predicted value by the actual value in 1989Similarly the third row shows that using PUMA log incomeinstead of earnings yields similar results while the fourth rowshows that the results are insensitive to replacing log own house-hold income with the couplersquos log total earnings The fifth rowruns the baseline regression on all the observations in the bal-anced panel (rather than only the ones married in both waves)The coefficient on neighborsrsquo earnings remains highly significantbut drops somewhat in magnitude giving a first indication thatthe estimates are primarily driven by the married subsampleThis issue will be explored further in Table V discussed below
Could neighborsrsquo earnings proxy for nonlinearities in theeffect of own income This concern is ruled out by the sixth rowwhich shows that the estimate on neighborsrsquo earnings hardlychanges after the inclusion of a fifth-order polynomial in loghousehold income Since the outcome variable self-reported hap-piness is ordinal rather than cardinal an OLS regression maynot be appropriate Specification (7) estimates the baseline re-gression using an ordered Probit and finds that the coefficient onneighborsrsquo earnings remains negative and highly significantMoreover the ratio of the coefficient on neighborsrsquo earnings to thecoefficient on own income remains roughly constant25 An alter-native way to deal with the ordinal nature of the happinessquestion is to create five dummy variables corresponding to areported level of happiness of at least 3 4 5 6 and 7 respec-tively26 Unreported regressions of these dummy variables onneighborsrsquo earnings and the remaining controls of the baselineregression yield a significantly negative coefficient on neighborsrsquoearnings though the effect is only marginally significant in theregressions for attaining a happiness level of at least 3 or at least7 These regressions show that the effect of neighborsrsquo earningsoperates throughout the happiness distribution
25 The coefficients in the ordered Probit turn out to be similar in magnitudeto those in the baseline regression partly because the root mean squared error ofthe baseline regression is 106 and thus close to unity to which the error term inthe latent model of the ordered Probit is normalized
26 I do not partition the happiness distribution at 2 because only a fewpercent of the observations report such low levels of happiness
981NEIGHBORS AS NEGATIVES
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
IVC Do Neighborsrsquo Earnings Affect Other Outcomes
One might be concerned that an increase in neighborsrsquo earningsmerely changes how individuals define happiness rather than theirtrue underlying well-being This concern is hard to rule out defini-tively but using another outcome measure that is arguably lessprone to shifts in definition yields some insights One might expectthat a couple surrounded by neighbors earning more would havemore disagreements about material issues as their aspirationsmight be shaped by the spending patterns of those around them[Stutzer 2004] The regressions in specification (1) of Table IV showthat higher neighborsrsquo earnings are significantly associated withmore frequent open disagreements about money but not signifi-cantly with the frequency of disagreements about household tasksthe children sex in-laws or spending time together Because thequestions about open disagreements are about behavior they seemless prone to a shift in definition in response to neighborsrsquo earn-ings27 This finding therefore offers suggestive evidence that theestimated effect of neighborsrsquo earnings on self-reported happiness isnot simply due to a shift in the definition of happiness
The second specification in Table IV considers a measure ofdepression which is the sum of the twelve items from the Radloffdepression scale that are included in both waves of the surveyThough many of the items also have subjective definitions (egldquofeeling lonelyrdquo ldquosleeping restlesslyrdquo) to the extent that it is harderto use neighborsrsquo behavior as a reference the depression scale maybe less prone to shifting definitions On the other hand depressionand well-being though correlated are two distinct concepts and it isvery well possible that increases in neighborsrsquo earnings reduce truewell-being without increasing depression As the regression showsneighborsrsquo earnings have no significant impact on the depressionindex Because the depression index is quite skewed I also looked atthe effect of neighborsrsquo earnings on the probability of being in eachpart of the distribution of depression index and find that higherneighborsrsquo earnings significantly increase the probability of being in
27 The response categories for the frequency of disagreements (ldquoNever lessthan once a month several times a month about once a week several times aweek almost every dayrdquo) are defined by the survey instrument and thus lesssusceptible to subjective interpretation However what counts as ldquoan open dis-agreementrdquo remains somewhat subjective This question is asked after the self-reported happiness question so it is possible that answers to this question arecorrelated with self-reported happiness because of a bias toward giving answersconsistent with previous answers However it is not clear why this bias would bemore severe for disagreements about money than other disagreements
982 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
the top four quintiles of the depression index but do not significantlyaffect the probability of being in the top three top two or topquintile of the depression index Thus the effect of neighborsrsquo earn-ings seems to be limited to the bottom of the depression index
TABLE IVOTHER OUTCOME MEASURES
Dependent variable
PUMA lnearnings ln HH income
AdjR2 NCoeff SE Coeff SE
(1) Frequency of opendisagreements abouta money 0189 0057 0116 0019 01478 9125b household tasks 0084 0057 0060 0016 01184 9122c the children 0068 0076 0045 0030 00533 8522d sex 0046 0062 0053 0017 00863 9044e spending time together 0027 0065 0058 0021 00925 9120f in-laws 0008 0057 0053 0025 00677 9096
(2) Depression index (sum of 12Radloff items) 0570 0856 1345 0255 00681 8782
(3) Health status relative to agegroup 00002 0041 0097 0013 01503 9173
(4) Satisfaction with (wave 2only)a amount of leisure time 0274 0106 0126 0037 02373 5010b friendships 0180 0085 0060 0026 00595 5008c sex life 0171 0104 0141 0037 00300 4965d financial situation 0158 0100 0580 0044 02120 5006e health 0155 0090 0179 0031 00515 5011f home 0151 0100 0078 0030 01478 5012g family life 0119 0079 0057 0027 00410 5010h present job 0097 0117 0259 0044 00606 4323i physical appearance 0093 0083 0063 0030 00444 5009j neighborhood 0152 0125 0058 0035 00966 5010k city or town 0366 0119 0061 0036 00768 5008
(5) Financial worries (wave 2only) 0082 0069 0389 0034 02262 5001
(6) Labor supply (hoursweek) 2224 1007 03041 8944
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at thePUMA level All regressions include the same controls as the baseline regression reported in Table I column1 except for specification (6) which does not include controls for own income hours worked unemploymentand labor force participation The frequency of open disagreements is measured on a scale of 1 (ldquoneverrdquo) to 6(ldquoalmost every dayrdquo) The depression index is the sum of the twelve Radloff items that appear in both wavesof the NSFH Each item is the number of days in the past week that the respondent felt or experienced asymptom related to depression Examples of such symptoms are ldquosleeping restlesslyrdquo ldquotalking less thanusualrdquo and ldquofeeling sadrdquo Specification (2) uses only data from the main respondent because the depressionquestions were only asked of the main respondent in all waves Self-reported health status is the answer tothe question ldquoCompared with other people your age how would you describe your healthrdquo where 1 corre-sponds to ldquovery poorrdquo and 5 to ldquoexcellentrdquo Self-reported satisfaction is measured on a seven-point scale where1 denotes ldquovery dissatisfiedrdquo and 7 denotes ldquovery satisfiedrdquo The variable Financial worries exists only in wave2 and is the answer to the question ldquoHow often do you worry that your total family income will not be enoughto meet your familyrsquos expenses and billsrdquo where 1 corresponds to ldquoneverrdquo and 5 to ldquoalmost all the timerdquo
983NEIGHBORS AS NEGATIVES
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
distribution (ie only those farthest from being depressed comesomewhat closer to being depressed) Overall the findings usingdepression as an outcome variable only slightly alleviate concernsabout shifting definitions of happiness though depression may be asufficiently different concept from well-being to pick up relativeposition effects28
Since a large literature examines the effect of relative posi-tion on health outcomes the third specification uses self-reportedhealth status (relative to onersquos age group) as an outcome mea-sure29 I find no significant relation between average neighborsrsquoearnings and self-reported health This finding of course doesnot rule out that such a relationship might exist but it does notshow up using my baseline specification30
IVD Mechanisms Behind the Association between NeighborsrsquoEarnings and Happiness
One can think of overall self-reported happiness as being com-posed of onersquos satisfaction with various domains of life such asfamily life financial situation or friendships [Van Praag Frijtersand Ferrer-i-Carbonell 2003] In wave 2 the NSFH asks respon-dents to rate their satisfaction with eleven such domains on a7-point scale Specification (4) uses these satisfaction measures asoutcome variables in order to understand which components ofhappiness drive the relationship between happiness and neighborsrsquoearnings though of course there may be drivers of happiness notcovered by these satisfaction questions As shown in specification(4) each measure of satisfaction increases with own income andexcept for satisfaction with neighborhood and city or town all sat-isfaction measures decrease with neighborsrsquo earnings However in
28 In their study of the randomized Moving to Opportunity experimentKling et al [2004] find that individuals moving to lower-poverty census tractsreport lower levels of psychological distress Thus for mental health measuressuch as depression or psychological distress the benefits associated with richerareas (such as lower crime) apparently outweigh any relative position effect
29 Self-reported health is measured by the question ldquoCompared with otherpeople your age how would you describe your healthrdquo with possible answersbeing ldquovery poorrdquo ldquopoorrdquo ldquofairrdquo ldquogoodrdquo and ldquoexcellentrdquo Since the question abouthealth explicitly asks respondents to compare themselves with other people oftheir age this outcome cannot be used to address any concerns about a shiftingdefinition of happiness
30 In his survey of this literature Deaton [2003] concludes that the evidenceon the relation between income inequality and health needs to be treated skep-tically though he believes there is convincing biological evidence that increases insocial rank can be protective of health Eibner and Evans [forthcoming] findevidence that in the United States relative deprivation (which is a measure ofrank and the income gap with those who are richer) increases mortality
984 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
only three cases is the relationship with neighborsrsquo earnings statis-tically significant at the 5 percent level Neighborsrsquo earnings signifi-cantly increase satisfaction with onersquos city or town which indicatesthat respondents are aware of the tangible benefits of living in anarea with richer people Neighborsrsquo earnings significantly reducesatisfaction with the amount of leisure time (even though the re-gression controls for labor force participation and hours worked) andsatisfaction with onersquos friendships
If neighborsrsquo consumption patterns shape onersquos aspira-tions one might have expected that higher neighborsrsquo earningswould significantly reduce onersquos satisfaction with material out-comes such as onersquos financial situation or onersquos home Thisseems not to be the case and is confirmed by specification (5)which shows that the frequency of financial worries does notincrease significantly with neighborsrsquo earnings Instead peopleappear to be giving up leisure to allow their friendships tosuffer and to work more perhaps in an attempt to mimic thematerial living standards of their neighbors31 This is consis-tent with Frey and Stutzer [2004] who argue that people sub-stitute goods yielding extrinsic satisfaction (material posses-sions status) for those yielding intrinsic satisfaction (time forfamily friendships hobbies) because predictions of futureutility from the extrinsic attributes of consumption are system-atically biased upwards Furthermore in unreported results Irun the baseline regression with all the satisfaction measuresadded as additional controls and find that the negative effect ofneighborsrsquo earnings remains significant at the 5 percent levelHence apparently a significant part of the negative effect ofneighborsrsquo earnings on happiness runs through drivers of hap-piness not captured by the satisfaction questions
IVE Interaction Effects
Does the relationship between neighborsrsquo earnings and happi-ness operate across a range of demographic subgroups The firstspecification of Table V interacts both neighborsrsquo earnings and ownincome with a set of marital status dummies Though the hypothesisthat the coefficients on neighborsrsquo earnings are all equal to eachother cannot be rejected at the 10 percent level there seems to be
31 The labor supply effect shown in specification (6) should be treated withcaution since one would expect neighborsrsquo earnings to be correlated with own earn-ings of which labor supply is one component Neumark and Postlewaite [1998] andBowles and Park [forthcoming] find effects of reference groups on labor supply
985NEIGHBORS AS NEGATIVES
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
TABLE VINTERACTION EFFECTS
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(1) Marital status transitions 00537 14500Remains married or cohabiting 0186 0076 0122 0023 [8023]Remains divorced or separated 0363 0155 0157 0049 [1524]Remains widowed 0103 0182 0087 0058 [1032]Remains never married 0076 0171 0122 0051 [995]Marital status change 0035 0095 0123 0034 [2926]
P-value on test of equal coefficients [0153] [0924]
(2) Household income is 00385 8944below the PUMA median 0238 0095 0119 0035 [3401]above the PUMA median 0237 0074 0124 0036 [5543]
P-value on test of equal coefficients [0993] [0915]
(3a) Socialize with a neighbor 00416 8944Less than once a month or missing 0161 0080 0132 0026 [5076]Once a month or more frequently 0335 0075 0108 0027 [3868]
P-value on test of equal coefficients [0038] [0502]
(3b) Socialize with relatives 00399 8944Less than once a month or missing 0176 0086 0145 0034 [2927]Once a month or more frequently 0267 0074 0109 0023 [6017]
P-value on test of equal coefficients [0300] [0353]
986Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
TABLE V(CONTINUED)
Dependent variableSelf-reported happiness PUMA ln earnings ln HH income
Adj R2 NSpecification CoeffSE
[p-value] CoeffSE
[p-value]
(3c) Socialize with friends who live outsidethe neighborhood 00393 8944
Less than once a month or missing 0263 0090 0157 0031 [4119]Once a month or more frequently 0217 0075 0088 0026 [4825]
P-value on test of equal coefficients [0644] [0092]
(3d) Socialize with people one works with 00392 8944Less than once a month or missing 0221 0077 0140 0023 [6325]Once a month or more frequently 0272 0089 0065 0036 [2619]
P-value on test of equal coefficients [0607] [0077]
Significance levels 10 percent 5 percent Robust standard errors are adjusted for clustering at the PUMA level Each specification is a single OLS regression in whichPUMA ln earnings and ln household income are interacted with an exhaustive set of dummies All regressions also include as controls the uninteracted set of dummy variables aswell as the same controls as the baseline regression reported in Table I column 1 All specifications use the baseline sample consisting of individuals married or cohabiting in bothwaves except for specification (1) where only information from the main respondent is used (in order allow for marital status transitions) The number of observations in each categoryis denoted between square brackets Frequency of social contacts is the frequency of social contacts of the main respondent because this variable was not collected of spouses in bothwaves In wave 1 respondents were asked how often they ldquospend a social eveningrdquo with various types of people while in wave 2 they were asked how often they ldquoget together sociallyrdquowith these types of people
987N
EIG
HB
OR
SA
SN
EG
AT
IVE
S
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
little or no effect of neighborsrsquo earnings on happiness for never-married individuals or those experiencing marital status transi-tions This is not surprising because these individuals are likely tobe less settled and thus less inclined to consider their neighbors astheir reference group In similar but unreported regressions I testedwhether the effect of neighborsrsquo earnings on happiness varies bygender age group educational attainment homeownership pres-ence of children in the household or length of time lived in onersquoscurrent home In all cases the point estimates on neighborsrsquo earn-ings are negative for each demographic subgroup and in no case canthe hypothesis that the effect is the same across subgroups berejected at the 10 percent level
One might wonder whether people predominantly comparethemselves with neighbors from their own demographic sub-group I explore this hypothesis for subgroups defined by collegeattendance The point estimates of an unreported regression inwhich earnings of neighbors by educational attainment is inter-acted with own education indeed show that the happiness of thosewithout a college degree declines with the earnings of neighborswithout a college degree but is relatively insensitive to the earn-ings of neighbors with a college degree Similarly the happinessof those with a college degree declines with the earnings of neigh-bors with a college degree but is relatively insensitive to theearnings of neighbors without one However these point esti-mates are not significantly different from each other thus thefinding that individuals mostly compare themselves with neigh-bors with the same educational attainment is suggestive at best
For two reasons one might have expected individuals to beinfluenced more by neighborsrsquo earnings if they are below themedian PUMA income than if they are above the median FirstRizzo and Zeckhauser [2003] show that individuals respond morestrongly to shortfalls below their reference income Thus if neigh-borsrsquo earnings serve as a reference income one would expect astronger response for those earning less than that Second to theextent that people also derive some happiness from ldquofitting inwith the Jonesesrdquo an individual below the median is furthernegatively affected by an increase in neighborsrsquo earnings whilethe negative effect is attenuated for a person above the medianSpecification (2) however shows that the effect of neighborsrsquoearnings is almost identical for those above and below the medianincome in their PUMA In additional unreported regressions Iexamine other nonlinearities but find no evidence of such effects
988 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
The point estimates on a quadratic term in neighborsrsquo earnings orthe interaction between own household income and neighborsrsquoearnings are both small and statistically insignificant In addi-tion adding various measures of inequality in onersquos PUMA to thebaseline regression did not reveal any significant effects of in-equality on happiness in contrast to Alesina Di Tella and Mac-Culloch [2004] who find that inequality in onersquos state affectshappiness
If neighborsrsquo earnings reduce self-reported happiness be-cause people engage in social comparisons we would expect astronger effect for those with more contacts with their neighborsThe NSFH asks all main respondents about the frequency ofsocial interactions with neighbors relatives friends living out-side the neighborhood and people they work with Specifications(3a-d) of Table V compare the effect of neighborsrsquo earnings forthose who have infrequent social contacts (less than once amonth) with those with frequent social contacts with the type ofperson indicated controlling for the direct effect of social inter-actions32 The regressions show that the effect of neighborsrsquo earn-ings is significantly stronger for those who socialize more fre-quently with neighbors but not for those who socialize morefrequently with relatives friends outside the neighborhood orpeople they work with These findings are consistent with whatone would expect if social comparisons with neighbors partlydetermine peoplersquos happiness
V CONCLUSION
This paper shows that individualsrsquo self-reported happiness isnegatively affected by the earnings of others in their area Bylooking at alternative outcome measures such as frequency ofmarital disagreements I provide suggestive evidence that thisfinding is not simply an artifact of the way people report happi-ness I investigate the concern that the finding could be driven byomitted variables but find no evidence of selection in a number ofspecification tests Though the mechanism by which increases inneighborsrsquo earnings reduce happiness is hard to identify pre-cisely I find that increased neighborsrsquo earnings have the stron-
32 The direct effect is positive in all cases but only significant for socializingwith neighbors or relatives Of course the direct effect should not be interpretedcausally
989NEIGHBORS AS NEGATIVES
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
gest negative effect on happiness for those who socialize more intheir neighborhood I conclude that the negative effect of neigh-borsrsquo earnings on well-being is real and that it is most likelycaused by a psychological externality that is people having util-ity functions that depend on relative consumption in addition toabsolute consumption
The size of the effect is economically meaningful An increasein neighborsrsquo earnings and a similarly sized decrease in ownincome each have roughly about the same negative effect onwell-being This suggests that an increase in own income leads toa negative externality on neighborsrsquo well-being that is of the sameorder of magnitude as the positive effects on own well-beingUnless one chooses to disallow these negative externalities on theground that they appear to stem from an interpersonal preferencecomponent that may be morally questionable33 externalities ofthis size can in principle substantially affect optimal incometaxation consumption taxation and residential sorting policies
APPENDIX 1 VARIABLE DEFINITIONS
The baseline sample consists of individuals who are marriedor cohabiting in both waves of the NSFH and have nonmissinginformation on own household income Unless otherwise notedthe variables in the baseline sample consist of the average of thenonmissing values of that variable for the main respondent andhis or her spouse All dollar amounts are converted to 1982ndash1984real dollars using the CPI-U from Bureau of Labor Statistics
Variable name Description
Self-reported happiness The answer to the question ldquoNext are some questionsabout how you see yourself and your life Takingthings all together how would you say things arethese daysrdquo where 1 denotes ldquovery unhappyrdquo and 7denotes ldquovery happyrdquo Values in between did nothave explicit labels
33 See Frank [2005] for arguments why these externalities should not bedisallowed
990 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
PUMA ln earnings(predicted)
Average predicted pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSee Appendix 3 for details
PUMA ln earnings in1989 (actual)
Average actual pretax ln earnings of allnoninstitutionalized working-age persons (16 age 65) with nonmissing industry and occupationcodes living in the PUMA of the NSFH respondentSource 1990 Census PUMS
PUMA ln income in1989 (actual)
Average actual pretax ln household income of allnoninstitutionalized persons living in the PUMA ofthe NSFH respondent Source 1990 Census PUMS
PUMA ln unadjustedhouse price
Average ln property value of owner-occupied units inthe PUMA of the NSFH respondent Source 1990Census PUMS
PUMA ln adjustedhouse price
The PUMA fixed effects of an OLS regression of lnproperty value on all housing characteristicsavailable in the 1990 Census PUMS
ln Metropolitan areapopulation
ln Census population counts for the metropolitan areaof which the PUMA is part Source 1990 CensusGeographic Equivalency file
Fraction Black inPUMA
Fraction of the noninstitutionalized population in thePUMA that is non-Hispanic Black Source 1990Census PUMS
ln Household income ln Household total pretax income constructed by theNSFH using information from both the mainrespondent and his or her spouse NSFH variablenames ldquoIHTOT2rdquo in wave 1 and ldquoMUHHTOTrdquo inwave 2
ln Predicted earningsof R and spouse
ln Pretax earnings of the main respondent and hisher spouse based on the industry occupationcodes of the respondent and the spouse and theearnings by industry occupation time-periodfrom the CPS MORG See Appendix 3 for details
ln Actual earnings of Rand spouse
ln Pretax earnings of the main respondent and hisher spouse
ln Value of home ln Current value of the respondentrsquos home if therespondent is a homeowner Answer to the questionldquoHow much do you think your home would sell fornowrdquo
Renter Dummy for renting onersquos homeln Usual working
hoursln Total usual working hours in onersquos main and
secondary job Nonworking individuals are assignedthe mean log working hours to make the dummiesfor unemployment and not-in-the-labor force moreeasily interpretable
991NEIGHBORS AS NEGATIVES
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
Variable name Description
Unemployed Dummy for being unemployed defined as notcurrently employed and having looked for workduring the past 4 weeks
Not in the labor force Dummy for those neither currently employed norunemployed
Non-Hispanic WhiteBlack HispanicAsian Other race orrace na
Answer to the question ldquoWhich of the groups on thiscard best describes you 01-Black 02-White-not ofHispanic origin 03-Mexican American ChicanoMexicano 04-Puerto Rican 05-Cuban 06-OtherHispanic 07-American Indian 08-Asian 09-OtherrdquoCategories 3 4 5 and 6 are combined intoldquoHispanicrdquo and categories 7 9 and no answer arecombined into ldquoOther race or race nardquo
Married or cohabitingSeparated DivorcedWidowed Nevermarried
Cohabiting individuals (even if separated divorced orwidowed) are coded into the Married or cohabitingcategory
Educationalattainment (years)
Years of completed education at the time of the wave1 interview as defined by the NSFH constructedvariable ldquoEDUCATrdquo
Religious affiliationdummies (13)
Answer to the question ldquoWhat is your religiouspreference (IF PROTESTANT ASK) What specificdenomination is thatrdquo I grouped the 65 possibleresponses to this question into 13 major categories
Wave The wave of the NSFH data Wave 1 was fieldedbetween March 1988 and May 1989 with most ofthe interviews conducted in the summer of 1988Wave 2 was fielded between July 1992 and July1994 with most of the interviews conducted in1993
Frequency of openmaritaldisagreements abouthousehold tasksmoney spendingtime together sexin-laws the children
The answer to the question ldquoThe following is a list ofsubjects on which couples often havedisagreements How often if at all in the last yearhave you had open disagreements about each of thefollowing [household tasks money spending timetogether sex in-laws the children] 1-Never 2-Less than once a month 3-Several times a month4-About once a week 5-Several times a week 6-Almost everydayrdquo
992 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
Depression Index(based on Radloff)
Twelve items from the Radloff depression index wereasked in both waves ldquoNext is a list of the ways youmight have felt or behaved during the past weekOn how many days during the past week did you[Feel bothered by things that usually donrsquot botheryou Not feel like eating your appetite was poorFeel that you could not shake off the blues evenwith help from your family or friends Havetrouble keeping your mind on what you weredoing Feel depressed Feel that everything youdid was an effort Feel fearful Sleep restlesslyTalk less than usual Feel lonely Feel sad Feelyou could not get going] (number of days)rdquo Theanswer to each question is the number of days (0ndash7) on which the condition applied The depressionindex is the sum of the answers to these 12questions This variable is only based on theresponses by the main respondent because thedepression items were not asked of the spouse inthe first wave
Self-reported healthstatus
The answer to the question ldquoCompared with otherpeople your age how would you describe yourhealth 1-Very poor 2-Poor 3-Fair 4-Good 5-Excellentrdquo
Satisfaction with homeneighborhood city ortown financialsituation amount ofleisure healthphysical appearancefriendships sex lifefamily life presentjob
The answer to the question ldquoOverall how satisfiedare you with [Your home Your neighborhoodYour city or town Your financial situation Theamount of leisure time that you have Yourhealth Your physical appearance Yourfriendships Your sex life Your family life Yourpresent job] where 1 denotes ldquovery dissatisfiedrdquoand 7 denotes ldquovery satisfiedrdquo Values in betweendid not have explicit labels These questions wereonly asked in wave 2
Frequency of financialworries
The answer to the question ldquoHow often do you worrythat your total family income will not be enough tomeet your familyrsquos expenses and bills Would yousay 1-Almost all the time 2-Often 3-Once in awhile 4-Hardly ever 5-Neverrdquo This question wasonly asked in wave 2 The variable is reverse codedin the regressions
993NEIGHBORS AS NEGATIVES
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
APPENDIX 2 SUMMARY STATISTICS
Variable
Whole sampleBaseline
sample onlyN 23010 N 8944
MeanStddev N Mean
Stddev N
Self-reported happiness 5346 138 19795 5553 108 8944PUMA ln earnings (predicted) 9395 028 22979 9422 027 8944PUMA ln earnings in 1989 (actual) 9390 024 22979 9399 024 8944PUMA ln income in 1989 (actual) 10056 032 22979 10091 031 8944PUMA ln unadjusted house price 4202 063 22979 4198 062 8944PUMA ln adjusted house price 0111 059 22979 0130 056 8944ln Metropolitan area population 13802 125 18664 13728 122 7021Fraction Black in PUMA 0141 020 22979 0106 016 8944ln Household income 9886 110 19037 10346 075 8944ln Predicted earnings of R and
spouse9825 112 16984 10197 092 8125
ln Actual earnings of R and spouse 9801 074 19676 10125 062 8091ln Value of home 10871 088 12408 10979 082 6788Renter 0409 049 23010 0217 041 8944ln Usual working hours 3692 044 13416 3694 039 7255Unemployed 0029 017 23010 0019 010 8944Not in the labor force 0342 047 23010 0317 036 8944Female 0605 049 23010 na na naAge 44669 1691 23010 43282 1403 8944Non-Hispanic White 0734 044 23010 0828 036 8944Black 0179 038 23010 0091 028 8944Hispanic 0073 026 23010 0059 022 8944Asian 0009 010 23010 0010 009 8944Other race or race na 0005 007 23010 0012 008 8944Married or cohabiting 0601 049 23010 1000 000 8944Separated 0041 020 23010 0000 000 8944Divorced 0121 033 23010 0000 000 8944Widowed 0104 031 23010 0000 000 8944Never married 0133 034 23010 0000 000 8944Educational attainment (years) 12518 311 22924 12997 262 8940ln Household size 0914 056 23010 1139 039 8944No religion 0084 028 23010 0084 023 8944Catholic 0240 043 23010 0242 040 8944Jewish 0020 014 23010 0023 014 8944Baptist 0235 042 23010 0201 037 8944Episcopalian 0020 014 23010 0021 013 8944Lutheran 0054 023 23010 0063 022 8944Methodist 0097 030 23010 0106 028 8944Mormon 0022 015 23010 0028 016 8944Presbyterian 0033 018 23010 0041 018 8944Congregational 0017 013 23010 0021 012 8944
994 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
Protestant no denomination 0043 020 23010 0030 013 8944Other Christian 0114 032 23010 0118 029 8944Other religionsmissing 0022 015 23010 0023 012 8944Wave 1435 050 23010 1538 050 8944Additional variables for Table IVFrequency of open marital
disagreements aboutmoney 2069 115 13019 2079 099 9125the children 2097 127 10277 2366 167 8664household tasks 1992 107 13026 2022 093 9122sex 1748 111 12700 1782 095 9044spending time together 1997 129 12974 1998 107 9120in-laws 1494 088 12765 1568 150 9156
Depression Index (based onRadloff)
14913 1668 21511 12215 1449 8782
Self-reported health status 3949 086 21992 4021 065 9173Variables only available in wave 2 Observations in
wave 2 10004Baseline
observations inwave 2 5067
Satisfaction with family life 5749 142 9682 5881 106 4806Satisfaction with financial
situation4525 171 9658 4780 140 4800
Satisfaction with sex life 5011 187 9232 5276 139 4766Satisfaction with home 5491 152 9742 5622 118 4806Satisfaction with health 5384 154 9694 5436 116 4805Satisfaction with present job 5161 168 7036 5301 136 4150Satisfaction with amount of leisure 4558 180 9685 4514 147 4805Satisfaction with friendships 5697 135 9688 5676 103 4803Satisfaction with city or town 5317 154 9643 5383 123 4803Satisfaction with physical
appearance5144 147 9692 5118 112 4805
Satisfaction with neighborhood 5420 160 9719 5532 128 4805Frequency of financial worries 3093 121 9713 2997 102 5001
APPENDIX 3 CONSTRUCTION OF PREDICTOR FOR PUMA- AND
INDUSTRY OCCUPATION EARNINGS
This appendix provides a detailed description of the construc-tion of both the PUMA-level earnings predictor and the predictorof average earnings at the industry occupation level The con-struction of these predictors is described in five steps
Step 1 Calculation of average industry occupation time-period earnings from CPS MORG
I use the NBER CPS labor abstracts to obtain usual weeklyearnings by industry occupation for the years 1987ndash1988 and
995NEIGHBORS AS NEGATIVES
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
1992ndash1994 for all employees (self-employed individuals are ex-cluded because their earnings data are not available) To ensurecompatibility over time all earnings are deflated by the 1982ndash1984 CPI-U and all topcoded earnings are replaced by twice thevalue of the topcode34 This yields the variables ln Earn1 and lnEarn2 ln average annual earnings by industry occupation cellin 1987ndash1988 and in 1992ndash1994 These variables differ by statebecause data from the own state is excluded in calculating theaverage
Step 2 Prediction of average earnings by industry occupation time-period cell
Because many industry occupation cells have very fewobservations or are empty in the CPS data but nonempty in theNSFH or Census PUMS I use a linear regression to predictearnings in each industry occupation cell in each time period35
I regress ln Earn1 on a full set of three-digit industry dummiesand a full set of three-digit occupation dummies where eachindustry occupation cell is weighted by the weighted number ofobservation in that cell in the MORG I run this regressionseparately for each state (because of the exclusion of the ownstate information) On average these regressions have about20000 observations The regressions yield the earnings predictorln EarnPred1 for 1987ndash1988 and ln EarnPred2 for 1992ndash1994
Step 3 Taking a weighted average between actual and predictedearnings
The benefit of the predictor described above is that it yieldspredicted earnings for industry occupation time-period cellsthat are empty in the CPS MORG and that it increases theprecision of the earnings estimate for industry occupation time-period cells with very few observations However by notallowing industry occupation interactions the regression maynot be the best predictor of earnings for cells with a large number
34 This is the expectation of a topcoded earnings variable if the tail of theearnings distribution follows a Pareto distribution with a parameter of 2 whichseems to be the case empirically [Saez 2001]
35 The industry and occupation codes in the first wave of the NSFH as wellas in the 1987 and 1988 CPS MORG files are three-digit 1980 Census codes whilethree-digit 1990 Census codes were used in the second NSFH wave the 1992ndash1994 CPS MORG files and (obviously) the 1990 Census PUMS I create a cross-walk to make these two sets of generally very similar codes compatible I resolveany differences by merging several three-digit occupations or industries togetherThis results in 225 unique and comparable industry codes and 496 unique andcomparable occupation codes
996 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
of observations To balance these two concerns I take a weightedaverage ln EarnHatt between predicted earnings and actualearnings where the weight depends on n the number of obser-vations in that cell Specifically for wave 1
ln EarnHat1 n1eff
n1eff n ln EarnPred1 n
n1eff n ln Earn1
and similarly for wave 2 The parameter nteff or the relative
weight on predicted earnings is estimated by a nonlinear leastsquares regression of ln average earnings from the PUMS on theright-hand side of the equation above where industry occupa-tion cells are weighted by the weighted number of observations inthem in the PUMS This yields neff of 229 and 251 for wave 1 andwave 2 respectively The R2 of these regressions are 074 and073 respectively The variable ln EarnHatt is merged with theNSFH data (based on the wave of the survey and the respondentrsquosindustry occupation and state) and is the instrument for ownhousehold income used in specification (3) of Table I
Step 4 Taking averages by PUMA
Using data from the 1990 Census 5 percent PUMS I assignto each noninstitutionalized person aged 16ndash65 the predictedearnings ln EarnHatt based on that personrsquos industry occupa-tion and state I do this separately for predicted earnings in1987ndash1988 and in 1992ndash1994 Next I calculate the average ofthis variable for each PUMA for each time period yieldingPumaPredt
Step 5 Rescaling of PumaPredt
To test the predictive power of PumaPredt I regress bothPumaPred1 and PumaPred2 on PUMA-level ln earnings in 1989(from the 1990 Census) for all the PUMAs that occur in the NSFHdata Though both regressions have a reasonably high R2 (075and 076 respectively) the coefficient on 1989 PUMA ln earningsis only around a half Thus while the predictor apparently does agood job ranking PUMAs by earnings it underestimates earningsdifferences To correct for this underprediction I scale both pre-dictors up by a factor (241) such that a regression of the averageof PumaPred1 and PumaPred2 on actual PUMA ln earnings in1989 yields a coefficient of exactly one This rescaling reduces thecoefficient on PUMA-level earnings as an independent variableby a factor of about two but it of course does not affect the
997NEIGHBORS AS NEGATIVES
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
statistical significance of any result in the paper In the rest of thepaper I refer to the rescaled predictor of PUMA ln earnings asPUMA earnings (predicted) or when there is no risk of confusionPUMA earnings for short
Validation using NSFH Data
As a check on both the PUMA-level and industry occupa-tion-level predictors of earnings I regress ln earnings of NSFHrespondents (with earnings of at least $100year) on these pre-dictors The first row shows that the predictor of PUMA levelearnings is highly significant has a coefficient of about 075 andexplains about 37 percent of the individual-level earnings varia-tion The second row shows that results are similar if we use theactual average ln earnings in the PUMA of the respondent in1989 This shows that the use of predicted rather than actualearnings does not entail a huge loss of predictive power Theresults are similar for each wave separately and also hold up ifstate fixed effects are included (not reported)
Row 3 repeats this check for the measure of industry occupation earnings constructed in step 3 above This measure ishighly significant and has a coefficient reasonably close to oneRow 4 shows that results remain similar if we instead use lnaverage earnings in 1989 by industry occupation cell from the1990 Census This confirms that results are not driven by anypeculiarities of my procedure of creating the measure for indus-try occupation earnings
VALIDATION OF EARNINGS PREDICTOR BY PUMA AND BY INDUSTRY OCCUPATION
Dependent variable ln Respondentrsquos Real Earnings
Coefficient SE Adj R2 NIndependent variable Comment
(1) PUMA ln earnings(predicted)
Baseline 0753 004 00369 14845
(2) PUMA ln earnings(actual 1989)
Actual 1989 notpredicted
0941 005 00425 14845
(3) Industry occupationln earnings
Baseline 0859 002 02732 13980
(4) Census Industry occupation lnearnings in 1989
Average 1989 ln earningsby ind occ from1990 Census
0721 002 02513 12541
In specifications (1) and (2) standard errors are corrected for clustering at the PUMA level while inspecifications (3) and (4) they are adjusted for clustering at the industry occupation level
JOHN F KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY
998 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Abel Andrew B ldquoAsset Prices under Habit Formation and Catching Up with theJonesesrdquo American Economic Review LXXX (1990) 38ndash42
mdashmdash ldquoOptimal Taxation When Consumers Have Endogenous Benchmark Levelsof Consumptionrdquo Review of Economic Studies LXXII (2005) 1ndash19
Alesina Alberto Rafael Di Tella and Robert J MacCulloch ldquoInequality andHappiness Are Europeans and Americans Differentrdquo Journal of PublicEconomics LXXXVIII (2004) 2009ndash2042
Autor David H and Mark G Duggan ldquoThe Rise in the Disability Rolls and theDecline in Unemploymentrdquo Quarterly Journal of Economics CXVIII (2003)157ndash205
Bartik Timothy Who Benefits from State and Local Economic Development Pol-icies (Kalamazoo MI Upjohn Institute 1991)
Becker Gary S ldquoA Theory of Social Interactionsrdquo Journal of Political EconomyLXXXII (1974) 1063ndash1093
Becker Gary S Kevin M Murphy and Ivan Werning ldquoThe Equilibrium Distri-bution of Income and the Market for Statusrdquo Journal of Political EconomyCXIII (2005) 242ndash310
Bertrand Marianne and Sendhil Mullainathan ldquoDo People Mean What TheySay Implications for Subjective Survey Datardquo American Economic ReviewXCI (2001) 67ndash72
Blanchard Olivier Jean and Lawrence F Katz ldquoRegional Evolutionsrdquo BrookingsPapers on Economic Activity (1992) 1ndash61
Blanchflower David G and Andrew J Oswald ldquoWell-Being over Time in Britainand the USArdquo Journal of Public Economics LXXXVIII (2004) 1359ndash1386
Boskin Michael and Eytan Sheshinski ldquoOptimal Redistributive Taxation WhenIndividual Welfare Depends on Relative Incomerdquo Quarterly Journal of Eco-nomics XCII (1978) 589ndash601
Bound John and Harry J Holzer ldquoDemand Shifts Population Adjustments andLabor Market Outcomes during the 1980srdquo Journal of Labor EconomicsXVIII (2000) 20ndash54
Bowles Samuel and Yongjin Park ldquoEmulation Inequality and Work Hours WasThorstein Veblen Rightrdquo Economic Journal forthcoming
Campbell John Y and John H Cochrane ldquoBy Force of Habit A Consumption-Based Explanation of Aggregate Stock Market Behaviorrdquo Journal of PoliticalEconomy CVII (1999) 205ndash251
Carroll Christopher D Jody Overland and David N Weil ldquoComparison Utilityin a Growth Modelrdquo Journal of Economic Growth II (1997) 339ndash367
Clark Andrew E ldquoLrsquoUtilite Est-Elle Relative Analyse a lrsquoAide de Donnees sur lesMenagesrdquo Economie et Prevision CXXI (1996) 151ndash164
mdashmdash ldquoUnemployment as a Social Norm Psychological Evidence from Panel DatardquoJournal of Labor Economics XXII (2003) 323ndash351
Clark Andrew E and Andrew J Oswald ldquoSatisfaction and Comparison IncomerdquoJournal of Public Economics LXI (1996) 359ndash381
Deaton Angus ldquoHealth Inequality and Economic Developmentrdquo Journal of Eco-nomic Literature XLI (2003) 112ndash158
Di Tella Rafael Robert J MacCulloch and Andrew J Oswald ldquoPreferences overInflation and Unemployment Evidence from Surveys of Happinessrdquo Ameri-can Economic Review XCI (2001) 335ndash341
Diener Ed Ed Sandvik Larry Seidlitz and Marissa Diener ldquoThe Relationshipbetween Income and Subjective Well-Being Relative or Absoluterdquo SocialIndicators Research XXVIII (1993) 195ndash223
Duesenberry James S Income Savings and the Theory of Consumer Behavior(Cambridge MA Harvard University Press 1949)
Dupor Bill and Wen-Fang Liu ldquoJealousy and Equilibrium OverconsumptionrdquoAmerican Economic Review XCIII (2003) 423ndash428
Easterlin Richard A ldquoDoes Economic Growth Improve the Human Lot SomeEmpirical Evidencerdquo in Nations and Households in Economic Growth Essaysin Honor of Moses Abramowitz Paul A David and Melvin W Reder eds(New York NY Academic Press 1974)
999NEIGHBORS AS NEGATIVES
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
mdashmdash ldquoWill Raising the Incomes of All Increase the Happiness of Allrdquo Journal ofEconomic Behavior and Organization XXVII (1995) 35ndash48
mdashmdash ldquoIncome and Happiness Towards a Unified Theoryrdquo Economic Journal CXI(2001) 465ndash484
Eibner Christine E and William N Evans ldquoRelative Deprivation Poor HealthHabits and Mortalityrdquo Journal of Human Resources forthcoming
Falk Armin and Markus Knell ldquoChoosing the Joneses On the Endogeneity ofReference Groupsrdquo Scandinavian Journal of Economics CVI (2004) 417ndash435
Ferrer-i-Carbonell Ada ldquoIncome and Well-Being An Empirical Analysis of theComparison Income Effectrdquo Journal of Public Economics LXXXIX (2005)997ndash1019
Frank Robert H ldquoAre Workers Paid Their Marginal Productsrdquo American Eco-nomic Review LXXIV (1984) 549ndash571
mdashmdash Choosing the Right Pond (New York NY Oxford University Press 1985)mdashmdash Luxury Fever (New York NY Free Press 1999)mdashmdash ldquoPositional Externalities Cause Large and Preventable Welfare Lossesrdquo
American Economic Review XCV (2005) forthcomingFrederick Shane and George Loewenstein ldquoHedonic Adaptationrdquo in Well-Being
The Foundations of Hedonic Psychology Daniel Kahneman Ed Diener andNorbert Schwarz eds (New York NY Russell Sage Foundation 1999)
Frey Bruno S and Alois Stutzer ldquoWhat Can Economists Learn from HappinessResearchrdquo Journal of Economic Literature XL (2002) 402ndash435
Frey Bruno S and Alois Stutzer ldquoEconomic Consequences of MispredictingUtilityrdquo manuscript University of Zurich 2004
Frey Bruno S Simon Luechinger and Alois Stutzer ldquoValuing Public Goods TheLife Satisfaction Approachrdquo CESifo Working Paper No 1158 2004
Galı Jordi ldquoKeeping up with the Joneses Consumption Externalities PortfolioChoice and Asset Pricesrdquo Journal of Money Credit and Banking XXVI(1994) 1ndash8
Graham Carol and Stefano Pettinato ldquoFrustrated Achievers Winners Losersand Subjective Well-Being in New Market Economiesrdquo Journal of Develop-ment Studies XXXVIII (2002) 100ndash140
Gruber Jonathan and Sendhil Mullainathan ldquoDo Cigarette Taxes Make Smok-ers Happierrdquo The BE Journals in Economic Analysis amp Policy forthcoming
Ireland Norman J ldquoStatus-Seeking Income Taxation and Efficiencyrdquo Journal ofPublic Economics LXX (1998) 99ndash113
Johansson-Stenman Olof Fredrik Carlsson and Dinky Daruvala ldquoMeasuringFuture Grandparentsrsquo Preferences for Equality and Relative Standingrdquo Eco-nomic Journal CXII (2002) 362ndash383
King Gary Christopher J L Murray Joshua A Salomon and Ajay TandonldquoEnhancing the Validity and Cross-Cultural Comparability of Measurementin Survey Researchrdquo American Political Science Review XCVIII (2004) 191ndash207
Kling Jeffrey R Jeffrey B Liebman Lawrence F Katz and Lisa SanbonmatsuldquoMoving to Opportunity and Tranquility Neighborhood Effects on AdultEconomic Self-Sufficiency and Health from a Randomized Housing VoucherExperimentrdquo manuscript Princeton University 2004
Layard Richard ldquoHuman Satisfactions and Public Policyrdquo Economic Journal XC(1980) 737ndash750
mdashmdash ldquoHappiness Has Social Science A Cluerdquo Lionel Robbins Memorial LecturesLondon School of Economics (httpceplseacuklayard) 2003
Ljungqvist Lars and Harald Uhlig ldquoTax Policy and Aggregate Demand Manage-ment under Catching Up with the Jonesesrdquo American Economic Review XC(2000) 356ndash366
Loewenstein George Ted OrsquoDonoghue and Matthew Rabin ldquoProjection Bias inPredicting Future Utilityrdquo Quarterly Journal of Economics CXVIII (2003)1209ndash1248
McBride Michael ldquoRelative-Income Effects on Subjective Well-Being in theCross-Sectionrdquo Journal of Economic Behavior and Organization XLV (2001)251ndash278
Neumark David and Andrew Postlewaite ldquoRelative Income Concerns and the
1000 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
Rise in Married Womenrsquos Employmentrdquo Journal of Public Economics LXX(1998) 157ndash183
Ng Yew-Kwang ldquoRelative-Income Effects and the Appropriate Level of PublicExpenditurerdquo Oxford Economic Papers XXXIX (1987) 293ndash300
Oswald Andrew J ldquoAltruism Jealousy and the Theory of Optimal Non-LinearTaxationrdquo Journal of Public Economics XX (1983) 77ndash87
Pigou Arthur C The Economics of Welfare (London Macmillan 1920)Pollak Robert A ldquoInterdependent Preferencesrdquo American Economic Review
LXVI (1976) 309ndash320Postlewaite Andrew ldquoThe Social Basis of Interdependent Preferencesrdquo European
Economic Review XLII (1998) 779ndash800Radloff Lenore S ldquoThe CES-D Scale A Self-Report Depression Scale for Re-
search in the General Populationrdquo Applied Psychological Measurement I(1977) 385ndash401
Ravallion Martin and Michael Lokshin ldquoIdentifying Welfare Effects from Sub-jective Questionsrdquo Economica LXVIII (2001) 335ndash357
Rayo Luis and Gary S Becker ldquoEvolutionary Efficiency and Mean Reversion inHappinessrdquo manuscript University of Chicago 2004
Rizzo John A and Richard J Zeckhauser ldquoReference Incomes Loss Aversionand Physician Behaviorrdquo Review of Economics and Statistics LXXXV (2003)909ndash922
Robson Arthur J ldquoStatus the Distribution of Wealth Private and Social Atti-tudes to Riskrdquo Econometrica LX (1992) 837ndash857
Saez Emmanuel ldquoUsing Elasticities to Derive Optimal Income Tax Ratesrdquo Re-view of Economic Studies LXVIII (2001) 205ndash229
Samuelson Larry ldquoInformation-Based Relative Consumption Effectsrdquo Economet-rica LXXII (2004) 93ndash118
Seidman Laurence S ldquoRelativity and Efficient Taxationrdquo Southern EconomicJournal LIV (1987) 463ndash474
Senik Claudia ldquoRelativizing Relative Incomerdquo manuscript DELTA and Univer-sity Paris-IV Sorbonne 2004
Smith Adam The Theory of Moral Sentiments (London A Millar Edinburgh AKincaid and J Bell 1759)
Solnick Sara J and David Hemenway ldquoIs More Always Better A Survey onPositional Concernsrdquo Journal of Economic Behavior and Organization XXX-VII (1998) 373ndash383
Stutzer Alois ldquoThe Role of Income Aspirations in Individual Happinessrdquo Journalof Economic Behavior and Organization LIV (2004) 89ndash110
Stutzer Alois and Rafael Lalive ldquoThe Role of Social Work Norms in Job Search-ing and Subjective Well-Beingrdquo Journal of the European Economic Associa-tion II (2004) 696ndash719
Sweet James A and Larry L Bumpass ldquoThe National Survey of Families andHouseholdsmdashWaves 1 and 2 Data Description and Documentationrdquo Centerfor Demography and Ecology University of Wisconsin-Madison (httpwwwsscwiscedunsfhhomehtm) 1996
Sweet James A Larry L Bumpass and Vaughn Call ldquoThe Design and Contentof The National Survey of Families and Householdsrdquo Center for Demographyand Ecology University of Wisconsin-Madison NSFH Working Paper No 11988
Tomes Nigel ldquoIncome Distribution Happiness and Satisfaction A Direct Test ofthe Interdependent Preferences Modelrdquo Journal of Economic Psychology VII(1986) 425ndash446
Tversky Amos and Dale Griffin ldquoEndowment and Contrast in Judgements ofWell-Beingrdquo in Strategy and Choice Richard J Zeckhauser ed (CambridgeMA MIT Press 1991)
Van de Stadt Huib Arie Kapteyn and Sara van de Geer ldquoThe Relativity ofUtility Evidence from Panel Datardquo Review of Economics and StatisticsLXVII (1985) 179ndash187
Van Praag Bernard M S and Ada Ferrer-i-Carbonell Happiness Quantified ASatisfaction Calculus Approach (Oxford UK Oxford University Press 2004)
1001NEIGHBORS AS NEGATIVES
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS
Van Praag Bernard M S Paul Frijters and Ada Ferrer-i-Carbonell ldquoThe Anat-omy of Subjective Well-Beingrdquo Journal of Economic Behavior and Organiza-tion LI (2003) 23ndash49
Veblen Thorstein The Theory of the Leisure Class (New York NY Macmillan1899)
Veenhoven Ruut ldquoIs Happiness Relativerdquo Social Indicators Research XXIV(1991) 1ndash24
Zeckhauser Richard J ldquoThe Strategy of Choicerdquo in Strategy and Choice RichardJ Zeckhauser ed (Cambridge MA MIT Press 1991)
1002 QUARTERLY JOURNAL OF ECONOMICS