health, inequality, and economic...

Journal of Economic Literature Vol. XLI March 2003) pp. 113–158

Health, Inequality, and EconomicDevelopment

ANGUS DEATON1

113

1. Introduction

SUPPOSE THAT INCOME causes good health.People live longer and are healthier in

rich countries than in poor countries, peoplelive longer and are healthier than theirgrandparents and great-grandparents wholived in poorer times and, within a country ata moment in time, rich people live longerand are healthier than poor people. Supposetoo that this relationship is concave, so thatincome has a larger effect on health andlongevity among the poor than among therich. Then income redistribution from rich topoor, within countries, or between countries,will improve population health (SamuelPreston 1975). Yet there may be more to itthan that. Income inequality, or other relatedsocial inequalities with which it is correlated,may be directly hazardous to individualhealth. According to a recent body of litera-ture, equal societies have more social cohe-sion, more solidarity, and less stress; they

offer their citizens more public goods, moresocial support, and more social capital; andthey satisfy humans’ evolved preference forfairness. That equal societies are healthier isan argument particularly associated withRichard Wilkinson (1992, 1996, 2000), aswell as the collection of papers edited byIchiro Kawachi, Bruce Kennedy, andWilkinson (1999). One study (John Lynch etal. 1998) goes so far as to claim that, in theUnited States in 1990, the loss of life from in-come inequality “is comparable to the com-bined loss of life from lung cancer, diabetes,motor vehicle crashes, HIV infection, sui-cide, and homicide in 1995.” If even a frac-tion of this effect were real or if, morebroadly, income distribution affects popula-tion health even indirectly, economic and fis-cal policy has effects on well-being that aretypically ignored by economists or policymakers. And if economists are skeptical ofsuch mechanisms, many policy makers arenot; shortly after being elected, British PrimeMinister Tony Blair stated in Parliament that“there is no doubt that the published statis-tics show a link between income, inequality,and poor health.”2

This paper explores the theoretical basisand empirical evidence for a connection be-tween inequality and health, among poor aswell as rich countries. The proposition that

1 Research Program in Development Studies andCenter for Health and Wellbeing, Princeton Univer-sity. Based on a paper prepared for Working Group 1 ofthe WHO Commission on Macroeconomics andHealth. I am grateful to Anne Case for many helpfuldiscussions during the preparation of this paper and toGeorge Alleyne, David Cutler, Sandy Jencks, MichaelMarmot, Jim Smith, Duncan Thomas, RichardWilkinson, and three referees for comments on earlierversions. I am grateful for financial support from theJohn D. and Catherine T. MacArthur Foundation andfrom the National Institute of Aging through grants toPrinceton and to the NBER.

2 Quoted in Michaela Benzeval, Ken Judge, and SueShouls (2001).

Article 3 2/21/03 4:20 PM 13

income inequality is an individual health riskwas first proposed for wealthy countries thathave passed through the epidemiologicaltransition, where chronic diseases have re-placed infectious diseases as the main causeof mortality. But as we shall see, many of thearguments that income inequality is a healthrisk are as plausible for poor as for richcountries and, in some cases, more so.

I devote a substantial portion of the paperto theory, albeit interspersed with evidence.With a few exceptions, the literature doesnot specify the precise mechanisms throughwhich income inequality is supposed to af-fect health. In consequence, there is littleguidance on exactly what evidence weshould be examining, or whether the propo-sitions are refutable at all. Section 2 lays outsome of the possible stories, starting withthe simple case in which health is affectedby income, and there is no direct effect ofinequality. This is sometimes referred to usthe “absolute income hypothesis,” to empha-size that it is income that matters for health,not income relative to other peoples’ in-comes, nor income inequality. A name thatwould be at least as good is the “poverty” hy-pothesis, that ill health is a consequence oflow income, in the sense that more incomeimproves health by more among those withlow incomes than among those with high in-comes. It is important to start by establish-ing the full range of implications for thissimple case, and in particular what role isplayed by income inequality. I also discusswhat happens when we make health de-pend, not on absolute income, but on rela-tive income, and what this version of the rel-ative income hypothesis implies for therelationship between health and inequality.

Much of the health-economics literaturedoes not accept the existence of any causaleffect running from income to health, ex-cept possibly through the purchase of healthcare, arguing that the correlation betweenthem is driven in part by a causality runningfrom health to income, and in part by thirdfactors, such as education, or rates of time

preference. For much of this paper, I shalltake it as given that a causal link from in-come to health is at least part of the story,and follow the consequences. Even so, I in-clude in section 2 a brief discussion of thetopic, recognizing that an adequate discus-sion of the relationship between income andhealth would require a survey much longerthan the present one.

I discuss a number of other theoreticallinks between inequality and health, includ-ing possible effects of income on invest-ments in health and education, the two-waylink between nutrition and earnings at lowlevels of income, and the arguable negativeeffects of inequality on the ability of the po-litical process to deliver public goods. I alsoconsider arguments that our evolutionaryhistory predisposes us toward fairness, andsickens us when we live in unequal environ-ments. Such an account can be made consis-tent with a story in which relative depriva-tion is a cause of ill health and according towhich, within groups, inequality has no ef-fect on health conditional on income, whileacross groups, both average income and in-come inequality determine health. Finally, Iconsider the important case in which in-come inequality is in part a consequence ofill health, so that policies that reduce thelikelihood of sickness, shorten its duration,or ameliorate its effects on earnings can alsonarrow income inequalities. Some incomeinequality is a consequence of the fact thatearnings cannot be completely insuredagainst ill health, so that better health insur-ance is likely to help reduce inequalities inincome. Better insurance will typically re-duce inequalities, both in health and inincome.

Section 3 turns to the evidence, most ofwhich comes from developed economies. Ireview cross-country studies on adult mor-tality for rich countries and on child mortal-ity for both rich and poor countries. A majorquestion is whether the international data onincome distribution is of sufficiently highquality to support the inferences that are

114 Journal of Economic Literature, Vol. XLI (March 2003)

Article 3 2/21/03 4:20 PM Page 114

being made. Better data, or, at least, moreconsistent data are available within coun-tries, and some of the most interesting evi-dence on inequality and health comes fromstudies looking across areas and over timewithin developed countries, such as Britain,Canada, and the United States, and in a fewpoor countries. I also review the evidencefrom studies that link individual mortalityand morbidity to the ambient level of in-come inequality, as well as those that studythe links over time between income inequal-ity and patterns of mortality.

My conclusion is that there is no directlink to ill health from income inequality perse; all else equal, individuals are no morelikely to be sick or to die if they live in placesor in periods where income inequality ishigher. The raw correlations that exist in(some of the) data are most likely the resultof factors other than income inequality,some of which are intimately linked tobroader notions of inequality or unfairness.The fact that income inequality itself is not ahealth risk does not mean that inequalitiesmore generally are not important, let alonethat the social environment in which peoplelive is irrelevant for their health. Indeed, Ishall argue precisely the opposite. But wemust narrow and focus our search if we areto make the leap from correlation to policy.

The literature that is reviewed in this pa-per comes from a number of fields otherthan economics, particularly epidemiology,public health, sociology, psychology, and his-tory. Different fields have different styles ofpresenting theory and evidence, as well asdifferent standards for what counts as credi-ble evidence. Nevertheless, the ideas are of-ten important and should not be dismissedby economists, if only because they arewidely accepted by many policy makers andby scholars in other disciplines, and econo-mists need to confront, not ignore, them. Inthis spirit, it is worth presenting materialthat is not fully worked out theoretically norconvincingly demonstrated empirically.Economists can make important contribu-

tions to this work, and in turn are likely tobenefit from other social scientists’ some-times well-argued skepticism of economists’methodologies and prejudices.

More broadly, ideas about income, in-come inequality, and health are importantfor welfare economics. Health is a compo-nent of well-being, so that if health were af-fected by income inequality, tax and transferpolicies that affect the distribution of in-come would have effects that work, not onlythrough the usual mechanisms—for exam-ple, through an equity-preferring social-welfare function defined over levels of in-come or consumption—but also throughindividual levels of health. If income has anonlinear effect on health, and even if thereis no direct effect of income inequality onhealth, redistribution of income toward thepoor will improve their average health bymore than the loss of health among the rich.People who are income-poor are also health-poor, so that seeing well-being as dependenton both income and health reveals widerdisparities between rich and poor thanare recognized by standard, income-basedapproaches.

2. Theoretical Accounts of Income Inequality and Health

2.1 Individual, Group, and National Health

2.1.1 Health, Income, and Poverty

Figure 1 shows a recent version of thePreston (1975) curve, the international rela-tionship between life expectancy and na-tional income in current purchasing-powerparity dollars. Among the poorest countries,increases in average income are strongly as-sociated with increases in life expectancy,but as income per head rises, the relation-ship flattens out, and is weaker or even ab-sent among the richest countries. As Prestonoriginally noted, if this relationship werealso to characterize the relationship betweenmortality and income within countries,

Deaton: Health, Inequality, and Ecomomic Development 115

Article 3 2/21/03 4:20 PM Page 115

there would be a negative relationshipacross countries between income inequalityand life expectancy. Redistribution of in-come from rich to poor, whether within orbetween countries, will increase the healthof the poor more than it hurts the health ofthe rich, and thus improve average nationalor world health. Countries with particularlyhigh income inequality, such as the UnitedStates, will have lower life expectancy thantheir average income would warrant.

As Preston surmised, there is indeed arelationship between income and mortalitywithin rich countries, even those on the flatpart of the international curve. Figure 2was calculated from the NationalLongitudinal Mortality Survey in theUnited States. This is a national follow-upstudy of about 1.3 million people (withabout three-quarters of a million people in

the public release data) who were inter-viewed in a current population survey or ina census-related sample around 1980, andwhose deaths over a follow-up period of3,288 days were ascertained by matching tothe National Death Index (see EugeneRogot et al. 1992 for details). The surveycollects each person’s family income(within one of seven ranges) so that it ispossible to link the probability of deathduring the follow-up to family income andother variables, the most important ofwhich are sex and age. For adults betweenthe ages of 18 and 85, the log odds of mor-tality are approximately linear in age, so aconvenient way to summarize the data is toestimate a logit model for the probability ofdeath in which age is entered linearly, to-gether with a series of dummy variables forthe income categories.


Source: World Development Indicators CD-ROM, World Bank (2002)Note: Circles are proportional to population and some of the largest (or most interesting) countries are labeled. The solid line is a plot of a population-weighted nonparametric regression. Luxembourg, with per capita GDP of $50,061 and life expectancy of 77.04 years, is excluded.

80

70

60

50

40

0 10,000

China

MexicoSpain

FranceItaly

USA

Equatorial Guinea

Botswana

India

South AfricaNamibia

Nigeria

Gabon

BangladeshPakistan

IndonesiaRussia

Argentina

Korea

Brazil

GermanyUK

Japan

20,000 30,000 40,000

GDP per capita, 2000, current PPP $

Figure 1. The Preston Curve: Life Expectancy versus GDP Per Capita

life

expe

ctan

cy, 2

000

Article 3 2/21/03 4:20 PM Page 116

As it turns out, the log odds of mortality areclose to being linear in the logarithm of fam-ily income, so that the probability of death isas illustrated in the figure, which shows, formen and women separately, the probabilitiesof a fifty-year old dying during the follow-upperiod. The protective effects of income aresubstantial; Rogot et al. calculate that peoplewhose family income was more than $50,000in 1980 have a life expectancy that is about25 percent longer than people whose familyincome was less than $5,000. This differenceis overestimated by the assumption that peo-ple stay in the same income class throughouttheir lives, but underestimated by the poormeasurement of income in much of theNLMS, as well as by the fact that current in-come is likely a poor proxy for the longer-term measures of income that are presum-

ably more relevant. For both men andwomen, the effect of income on mortality isgreater among the poor than among therich; that the curve for women appears lesscurved than the curve for men is an insignif-icant consequence of the choice of scale. Ifthese curves come from a causal link be-tween income and health, income redistri-bution will improve population health.

It is often useful to think of the redistribu-tive story, not in terms of inequality, but interms of poverty. If a country with a high av-erage income has a great deal of income in-equality, then there is a relatively large num-ber of people with low income whose healthis poor. Although figure 2 shows no povertyline below which income matters and abovewhich it does not, it is at the bottom of theincome distribution that the relationship


Figure 2. Age Adjusted Probability of Death and Family Income, National Longitudinal Mortality Study

Source: Author's calculations from public use sample of the NLMS.Note: Probability of death is the probability that a 50-year-old man or woman at the time of interview (around 1980) died during the follow-up period of 3,288 days. Restricted to people aged 15 to 85 at the time of interview.

0.15

0.10

0.05

0

0 20,000 40,000 60,000

family income in 1980, 1980 $

white males

white femalesprob

abili

ty o

f dea

th a

t age

50

Article 3 2/21/03 4:20 PM Page 117

between income and health really matters.And if a rich country has a lot of poor peo-ple, it will have low average health relative toits per capita income. To illustrate,Wilkinson (1989) looks at mortality differ-ences by social class in Britain from 1921 to1981, and argues that mortality fell mostrapidly at times when income differentialswere narrowed, particularly at times whenincomes of the poor rose more rapidly thanthose of the rich, such as during World WarII. Similarly, Amartya Sen (1999, figures 2.2and 2.3) shows how life expectancy inEngland and Wales from 1901 to 1960 grewmost rapidly in the decades 1911–21 (by 6.5years) and 1940–51 (by 6.8 years), and moreslowly at other times, 4 years in 1901–11, 2.4years in 1921–31, 1.4 years in 1931–41, and2.8 years in 1951–60. Sen shows that thedecadal rate of growth of GDP per capita isstrongly negatively correlated with decadalincreases in life expectancy and, likeWilkinson, focuses on the degree of sharingduring the two wars, as well as on the directnutritional and health interventions thattook place during and immediately after thesecond war. Both wars brought well-payingemployment opportunities to many peoplein Britain for the first time, including manywomen. Richard Hammond (1950) dis-cusses how wartime food policy in the 1940snot only protected essential food supplies,but brought fresh milk and vitamins to work-ing people to the extent that, unlike most ofthe citizens of Europe, their nutritional sta-tus improved during the war. Reductions inincome inequality during the wars, if indeedthey took place, marked an improvement inthe conditions of the working people, amongwhom better incomes and better nutritionwould have had the largest effects onmortality.

In contrast to these arguments, I shall ar-gue in section 3.5 that it is not possible tolink recent increases in income inequality inthe United States and Britain to mortalitychanges. It should be noted, however, thatrecent increases in income inequality,

though large enough by postwar standards,are probably not large relative to earliercompressions, particularly those associatedwith the world wars (see Peter Lindert 2000,and Claudia Goldin and Lawrence Katz2001). They are also characterized more byincreases in the incomes of the rich than bydecreases in the incomes of the poor.

The fact that a concave relationship be-tween health and income implies that aver-age health is negatively associated with in-come inequality has come to be known inthe literature as a “statistical artefact” (HughGravelle 1998). The usage is designed to dis-tinguish it from mechanisms in which in-come inequality has a direct effect on indi-vidual health, but it is unfortunate insuggesting that there is no real link betweenincome inequality and health, and that redis-tributive policy cannot improve averagepopulation health. This is far from the case;if income causes health, and if there are di-minishing returns, redistribution from richto poor will improve average populationhealth.

2.1.2 Does Income Cause Health?

The nonlinearity argument for income re-distribution, in common with almost all ofthe literature in epidemiology and publichealth, takes it for granted that increases inincome cause improvements in health andreductions in mortality. Yet Preston’s originalpaper emphasized that income is only partof the story, and he shows that the curveshave shifted upward over time, a movementthat he attributes to improved public healthmeasures, particularly in middle-incomecountries. And the 2000 Preston curve infigure 1 has developed a non-monotonicityaround $5,000, driven by the collapse of lifeexpectancy in Russia, and by AIDS in Africa,particularly in South Africa; neither of thesedevelopments is income driven. Beyondthis, much of the economics literature hasbeen skeptical about any causal link from in-come to health, and instead tends to empha-size causality in the opposite direction, from


Article 3 2/21/03 4:20 PM Page 118

good health to higher earnings, and frompoor health to the inability to work and, insome countries, to running down assets topay for care. Although these arguments areobviously correct in some cases, and no oneseriously challenges that ill health is often areason for early retirement from the work-force, they have been strongly resisted bymany noneconomists, particularly in Britain;see Sally MacIntyre (1997) for an insightfuldiscussion of the political background ofthese debates. Economists have also madethe argument that the correlation reflectsthe operation of some third factor, particu-larly education (Michael Grossman 1972,1975, 2000), or preference factors such asdiscount rates. Those who are impatient orhave poor self-control are likely both to failto make investments to protect their healthand to fail to acquire the education and skillsthat generate higher earnings (Victor Fuchs1993). Such arguments also focus on the roleof health-related behaviors, such as smok-ing, binge-drinking, or lack of exercise,which are correlated both with low incomeand poor health.

In private or partially private systems,money can also pay for more expensivehealth care, but this mechanism is often dis-counted, sometimes because of a generalskepticism about the degree to which deathcan be postponed by medical care or, morespecifically, because of doubts about the rel-ative effectiveness of expensive over basichealth care. It is also true that the relation-ship between income and mortality, the“gradient,” exists in countries with and with-out public provision of health care; seeAnton Kunst, Feikje Groenhof, and JohanMackenbach (1998) for a review ofEuropean countries. Thomas McKeown(1976, 1979) famously demonstrated that,for disease after disease, the introduction ofeffective prophylaxis was not the primaryfactor driving historical declines in mortality,and the persuasiveness of his arguments hasleft a presumption, at least among medicalsociologists and historical demographers,

against the argument that differences inmortality across population groups are pri-marily driven by differences in access to orquality of health care. Yet McKeown’s argu-ments, even if true historically, need not betrue now, and the case against health care isfar from well-established. Indeed, manystudies today take it for granted that differ-ences in health care are the primary and ob-vious reason for disparities in outcomes, forexample “unequal treatment” of blacks andwhites in the United States (Brian Smedley,Adrienne Stith, and Alan Nelson 2002).

For the purpose of the current discussion,I need only that there be some effect run-ning from income to health, a position thathas increasingly gained currency amongeconomists in recent years; see for examplethe discussions in Jonathan Feinstein(1993), Preston and Paul Taubman (1994),and more recently, James Smith (1999). InDeaton (2002) I have tried to summarize myown position. An earlier, usefully skepticalreview is by Alan Garber (1989). A full dis-cussion of the theory and evidence relatingincome and health would occupy another re-view, at least the length of this one. Even so,it is worth emphasizing a few key points.

The literature outside of economics hasdocumented many hundreds of cases of apositive correlation between health andsome measure of socioeconomic status; the“modern” literature goes back to the eigh-teenth century, with precursors in ancientRome and China; see Nancy Krieger (2001).One of the most influential of the currentstudies is of civil servants in Britain, theWhitehall study, which has documentedsubstantial morbidity and mortality differ-ences by administrative rank; see MichaelMarmot, Martin Shipley, and Geoffrey Rose(1984), Marmot et al. (1991), and the moregeneral discussions in Marmot (1994, 2002).This literature tends to see income as a“marker” for an underlying concept of so-cioeconomic status, which is treated as theunderlying cause of health discrepancies.Because income is only one such marker, it


Article 3 2/21/03 4:20 PM Page 119

is often treated interchangeably with wealth(or parts of wealth, such as vehicle owner-ship or housing), or with consumption, orearnings, or even with rank, power, social oroccupational class, education, or even raceand geography. Such a framework is not veryuseful for thinking about social policy ashealth policy—for example, in a simple taxsystem, a higher marginal rate of income taxwill change disposable income, but will oftenhave no effect on rank in the income distri-bution—nor does it take us very far in think-ing about whether redistribution of income,as opposed to wealth, or even power, is likelyto improve public health.

That said, there is an increasing numberof studies that try to separate out differentinfluences and confront economists’ argu-ments about confounding by third factors.There are studies that look at the separateeffects of both education and income. Inthe NLMS (and figure 2) the protective ef-fects of income are not much affected bycontrolling for education (Irma Elo andPreston 1996; Deaton and Christina Paxson2001a; and E. Backlund et al. 1999), andsimilar findings have come from other datasets (Evelyn Kitigawa and Philip Hauser1973; Paula Lantz et al. 1998; RichardRogers, Robert Hummer, and Charles Nam2000). Income and education are also sepa-rately protective for self-reported healthstatus (Jacob Nielsen 2002). Economists’skepticism about a direct role for income issupported by relationships between mortal-ity, income, and education at the state ormetropolitan statistical area level, where av-erage education drives out average income,or even turns it into a risk factor (seeRichard Austen, Irving Leveson, andDeborah Sarachek 1969; Joseph Newhouseand Lindy Friedlander 1980), a finding thatcontinues to hold in more recent data (seeChristopher Ruhm 2000; and Deaton andDarren Lubotsky 2003). The conflict be-tween the individual and aggregate data re-mains unresolved. The relationship be-tween income and health is reduced, but far

from eliminated, by controlling for behav-ioral risk factors (in the Whitehall studies,which have full medical and behavioraldata, only about a quarter of the associationbetween rank and mortality is eliminated bycontrolling for risk factors), or for employ-ment status.

The most difficult issue of all is sortingout the dual causality between income andhealth. In poor countries, where malnutri-tion remains a major issue, there is wideagreement that income has a direct causaleffect; for excellent empirical evidence onthe effects of South Africa’s generous socialpension, see Anne Case (2001, 2002). Manyhave argued the same about rich countries,emphasizing the negative effects of povertyand of poor housing, and the positive effectsof not spending one’s life worrying aboutmoney. Even malnutrition may still have ef-fects in rich countries today; GabrieleDoblhammer and James Vaupel (2001) andDoblhammer (2002) have established a re-lationship between month of birth and re-maining life-expectancy at age fifty (Austria,Denmark, Australia) and mean age at death(those who died between 1989 and 1997 inthe United States). Those born in Octoberand November do better than those born inMarch and April, although the seasonal pat-tern is reversed in the southern hemisphere(except for those who died in the south butwere born in the north.) These effects areplausibly attributed to the effects of in-trauterine nutrition more than half a cen-tury ago, particularly to the seasonal avail-ability of fresh fruit and vegetables, and areindeed wearing off for more recently borncohorts. Theoretical mechanisms for sucheffects are developed in the work of DavidBarker (1994, 1995). The fetal origins(“womb with a view”) hypothesis “statesthat fetal undernutrition in mid to late ges-tation, which leads to disproportionate fetalgrowth, programs later coronary heart dis-ease” (Barker 1995, p.171). There is nowextensive empirical evidence supportingthese views, including evidence on the very


Article 3 2/21/03 4:20 PM Page 120

long-lasting (or long-postponed) effects onhealth.

The studies of birth cohorts, particularlyin Britain, have thrown up a wide range ofdynamic effects between health and earn-ings, often operating with long lags. Paneldata in Britain (British Household PanelSurvey/BHPS) and the United States(Health and Retirement Survey/HRS) arecurrently being used to examine health andincome dynamics. For example, in BHPS,Benzeval and Ken Judge (2001) find thatcontrolling for initial (self-reported) healthstatus reduces but does not eliminate the ef-fect of current income or current health sta-tus. A more sophisticated approach, closelyrelated to Granger causality testing, is em-ployed by Peter Adams et al. (2003) usingdata from HRS; they find no causal linkfrom wealth to health. This analysis, in par-ticular, and the ability of “post hoc, propterhoc” inference to capture causality in gen-eral, has been seriously challenged byFabrizia Mealli and Donald Rubin (2002). Itis clear that there are influences between in-come and health that run in both directions,and that, in some cases, the lags can be aslong as a human lifetime. In such circum-stances, establishing clear causal patterns isa matter of great difficulty.

2.1.3 Poverty, Health and the Direct andIndirect Effects of Inequality

The absolute income (poverty) hypothe-sis tells us a good deal about how we can ex-pect average income and income inequalityto affect population health at different lev-els of development. At the most obvious,the effects of per capita income mirrorthose of individual income, and become lessimportant the richer is the country.Eventually, we would expect income in-equality to lose its effect too, but it is notenough that average income be highenough, we also need everyone’s income tobe high enough. The bottom tail of the in-come distribution has to be pulled up be-yond the point at which income has much

effect on health. Before that, there will stillbe health-diminishing poverty even in richcountries, so that income inequality will stillmatter as well as average income. In conse-quence, the absolute income or poverty hy-pothesis implies that, among the poorestcountries, average income is what mattersfor population health, and income inequal-ity is relatively less important. Among richcountries, average income is less important,and income inequality relatively more im-portant. Eventually, neither will mattermuch for population health but, under plau-sible assumptions, the effect of income in-equality relative to that of average incomecontinues to grow as countries becomericher. These implications of the absoluteincome hypothesis are important because itis often claimed that the differential effectsof average income and income inequality onhealth change with economic developmentas described, and that the observation helpsestablish the case for a direct effect of in-equality on health in rich countries. Whilethe finding is certainly consistent with suchan effect, it is also consistent with a simplemodel in which income has a larger effecton health among the poor.

Suppose that individual i lives in countrys, and that her health his is a quadratic func-tion of her family income yis so that if, forconvenience, I write everything relative toworldwide means, signified by overbars, wehave

his − h = α+ β(yis − y)−γ(yis − y)2 − θvs. (1)

Income promotes health, but by less for therich, so that both β and γ are positive.Income inequality, here represented by thevariance vs of income around the countrymean, ys , is allowed to have a direct effect onhealth; I shall be particularly concerned withthe case where θ is zero, so that the only ef-fects of nonlinearity are the aggregation ef-fects associated with the nonlinearity.Equation (1) is assumed to hold true every-where, for all individuals, whether they live


Article 3 2/21/03 4:20 PM Page 121

in poor or rich countries, and wherever theyare in the epidemiological transition.Averaging (1) over everyone within eachcountry, we get the cross-country relation-ship

hs − h = α+ β(ys − y)−γ(ys − y)2 − (γ + θ)vs (2)

where variables subscripted by s are popula-tion means. Even if θ is zero, the varianceappears in (2) through the aggregation.

Equation (2) can be used to illustrate anumber of points. First, consider the effectof economic development in the form ofhigher levels of ys on the derivatives ofhealth with respect to average income andincome inequality. The derivative with re-spect to ys is β − 2γ(ys − y) , which, since γis positive, decreases with average income.By contrast, the derivative with respect to vsis constant. As a result, as a country becomesricher and average income rises, the effecton population health of income inequalitybecomes more important relative to the ef-fect on population health of average income.This is true whether or not θ , the direct ef-fect of inequality, is zero.

The second point illustrated by (2), due toDouglas Miller (2000), is that the parame-ters γ and θ are both identified in the equa-tion so that it is possible, at least in principle,to test for a direct effect of inequality usingonly aggregate data. Intuitively, this worksbecause the curvature in the individual rela-tionship (1) is transmitted to the aggregaterelationship (2), so that it is possible, usingonly aggregate data, to eliminate the part ofthe inequality effect that works through theaggregation. Miller applies this technique tomortality differences across the U.S. states,and finds that the relationship between mor-tality and income inequality is entirely ex-plained by the curvature in the aggregate re-lationship between average income andmortality, leaving no room for a direct effectof inequality. This result, however, is quitecontrary to that of Michael Wolfson et al.(1999), who use the NLMS data to estimate

the relationship between income and mor-tality at the individual level (as in figure 2),and then show that the implied relationshipbetween income inequality and mortalityacross the states accounts for only a fractionof the actual relationship. I shall return tothe state data in section 3.3 below, but thecontradiction between these two sets of re-sults most likely reflects a deeper differencein the relationship between income andmortality when estimated from individualand aggregate data.

The quadratic model is useful for illustra-tive purposes, but not very plausible, if onlybecause it implies that increases in incomewill eventually reduce health. A more realis-tic model links individual health to a latentvariable, itself a function of income, and as-sumes that death takes place when this la-tent variable falls below some critical value.Under suitable assumptions, such a formula-tion yields explicit predictions for the proba-bility of death for individuals, as well as thefraction dying in communities, and the re-sulting equations provide a better way oflinking population health to population in-come and income inequality. As before, Iwrite his for the health of individual i in pop-ulation s, and assume that health is linear inthe logarithm of individual income,

his = a+ blnyis + dωs + εis (3)

where the random term εis is assumed to benormally distributed with mean zero andvariance σ2, and ωs is the within-countryvariance of log income, so that, once again Iallow for a possible direct effect of inequal-ity on health. The individual dies when hisfalls below the critical level c. The probabil-ity that this happens, or the probability ofdeath, is written pis which is

pis = p(death | yisωs) =

Φ(c− a− blnyis − dωs

σ

)(4)

where Φ is the distribution function of thestandard normal. Suppose that, within each


ln

ln

Article 3 2/21/03 4:20 PM Page 122

country s, the logarithm of income is nor-mally distributed with mean µs variance ωs.We can then rewrite (3) as

his = a+ bµs + dωs

+b(lnyis − µs) + εis (5)

so that ps, the fraction of people dying in thestate, or the probability of death conditionalon the state mean and variance of log in-come, can be written

ps = p(death | µs, ωs)

= Φ

(c− a− bµs − dωsσ√

1 + b2ωs/σ2

). (6)

Equation (6) links population mortality ratesto the population mean of log income, µs,and the variance of log income, ωs.

The main points of (2) carry through to themore realistic (6). The parameter d is identi-fied in (6) from the aggregate data; eventhough the log variance appears in both thenumerator and denominator, its coefficientin the latter is the same as the coefficient ofmean income in the numerator. Second, sup-pose that d is zero, and we differentiate, firstwith respect to µs, and then with respect toωs, it is readily shown that (a) both deriva-tives tend to zero as µs goes to infinity, and(b) that the ratio of the inequality derivativeto the mean log income derivative grows lin-early in the latter. As countries become suffi-ciently rich, and in the absence of a direct ef-fect of income inequality, neither populationmean income nor income inequality haveany effect on population mortality rates. Butthe effect of income inequality relative to theeffect of income is larger among richer coun-tries. Once again, this has nothing to do witha direct effect of inequality on health.

2.1.4 Relative Income, Absolute Income,and Inequality

If income causes health, it is possible thathealth is determined, not by absolute in-come, but by income relative to some aspira-tion level, or relative to the incomes of oth-

ers. Richard Easterlin (1975) long ago foundevidence that happiness is independent ofincome in the long run, and health may fol-low the same pattern. That health dependson income relative to average incomes ofone or more reference groups is what wemight call a relative income hypothesis. Thiscould happen in a number of different ways.One case is where relative income deter-mines access to material goods, for examplewhen the people who live in a town are themarket for local land for housing, with therichest getting the hilltop plots with fineviews, and the poorest getting the plotsdownwind of the smokestacks. The localhousing case is an example where it is notmoney itself that is important, but rank,here determined by money. More generally,rank at work is important in determininghow much control people have over theirworking lives and, as demonstrated in theWhitehall studies, the degree of control atwork accounts for much of the relationshipbetween occupational status and health(Marmot et al. 1997).

The relative income theory is consistentwith an effect of income inequality, althoughit does not imply it. The distinction is ofsome importance, if only because a direct in-fluence of inequality on health, as proposedby Wilkinson, is frequently labeled “the rela-tive income hypothesis.” The argument ap-pears to be that, if health is lower for thosewhose income is relatively low, then higherinequality makes the poor even poorer inrelative terms, and so worsens populationhealth. In section 2.1.6 below, I shall de-velop a model in which just this happens,but the effect is not an automatic conse-quence of a relative income model.

Suppose that, as before, individuals are la-beled i, but that we use s to index the rele-vant comparison or reference group. An in-dividual’s health might then be above orbelow the population mean depending onwhether her income is above or below theaverage income in the reference group.Hence, with β positive, we might write:


ln

Article 3 2/21/03 4:20 PM Page 123

his = h+ β(yis − ys). (7)

For the group as a whole, group averagehealth hs is just average health h, which isthe same for all groups, so that neither groupincome nor group income inequality has anyeffect on group health. This annihilation ofthe effect of income on health as we movefrom the individual to the group will alwayshappen if the reference group is within theunit over which we are aggregating. For ex-ample, if reference groups are geographicallylocal, average income will be unrelated to av-erage health at the region, province, or statelevel. Such an account offers one explanationfor why income might be related to healthwithin groups, but not between them.

The relative income story is also consis-tent with a role for income inequality at thegroup level, and the mechanism is the sameas for the absolute income hypothesis, thenonlinearity of the relationship between in-come, now relative income, and health.Again, an obvious example is provided bythe quadratic version of equation (7), inwhich individual health depends on incomerelative to group income and the square ofincome relative to group income:

his = h+ β(yis − ys)− γ(yis − ys)2 (8)

When we average health over individualswithin each group, the first term vanishes, asbefore, while the second term becomes thevariance of income,

hs = h− γvs (9)

This provides us with one possible accountof the oft-cited phenomenon that was previ-ously discussed, that within states or coun-tries, individual health is related to individ-ual income, while between them, averagehealth is only weakly dependent on averageincome but is negatively related to incomeinequality.

Yet another possibility is that health is de-termined by rank in an income distribution.In the individual data, this hypothesis gener-

ates a monotone increasing nonlinear rela-tionship between income and health, andthere will be associated inequality effectsover aggregated groups. If rank is all thatmatters about income, higher income foreverybody will have no effect on anyone’shealth. This rank hypothesis is worth notingfor two reasons. First, rank is a plausible de-terminant of power, of social status, or ofcontrol over others, all of which have beenassociated with health. Second, rank is notnecessarily affected by the usual redistribu-tive policies. For example, consider a redis-tributive policy that raises the marginal taxrate and uses the proceeds to pay lump-sumbenefits to everyone. Provided the elasticityof labor supply is greater than –1, there willbe no effect on rank; if A has more incomethan B before the change, she will have moreafter the change (Kevin Roberts 1981).

2.1.5 The Impossibility of IdentifyingReference Groups

An immediate problem with the imple-mentation of any model of relative income isthe identification of the relevant referencegroup. In a few cases, such as the Whitehallcivil servants, the reference group (or atleast what is likely people’s most importantreference group) is a ready-made part of thedesign. More usually, reference groups arenot clearly defined and people will oftenhave multiple such groups, comparingthemselves to their neighbors, to their co-workers, to those they meet in social and re-ligious organizations, and to those they seeon television or read about in newspapers.One way of dealing with this is to recognizethat reference group incomes cannot be ob-served, and to work out the effects of theomission on the relationship between thetwo things that we can observe, health andincome. As shown in Deaton (2001a), thisprocedure brings inequality back into thestory even when it has no direct role.

Figure 3 illustrates the simple case wherethere are two groups, labeled “economists”


Article 3 2/21/03 4:20 PM 4

(on the left) and “doctors” on the right.Income is measured on the horizontal axis,health on the vertical axis. The two ellipsesshow where economists and doctors are lo-cated in terms of their health and incomes;members of each group are scattered withinthe two elliptical areas. Doctors have higherincomes than economists, and within eachgroup, individual health depends on individ-ual income relative to other members of thegroup. The two parallel steep lines show therelationships between income and health foreach of the two groups. Although doctorshave higher incomes than economists, theirindividual health is no better on average be-cause their absolute income does not matter,only their income relative to other membersof their group. Suppose that an epidemiolo-gist analyzes the data on economists’ anddoctors’ health, but without knowing whichis which. When the data are pooled, the rela-tionship between health and income is the

flatter, broken line. By mixing the twogroups, omitting the relevant information ongroup, the relationship is flattened out orattenuated.

Inequality comes into this story becausethe degree of flattening depends on the ratioof within-group income inequality to be-tween-group income inequality. If doctorsand economists are moved further apart, bymoving the two ellipses horizontally awayfrom each other, the broken line will be-come more attenuated. If within-group in-equality is increased, holding the between-group difference fixed, so that the ellipsesare stretched out along their individual in-come to health lines, the broken line will be-come steeper, increasing the gradient be-tween income and health. The steepness ofthe gradient depends positively on the ratioof within- to between-group inequality. Forexample, in Whitehall, if there is only onereference group, that of British civil servants,


Figure 3. Health and Income in Two Reference Groups: the Effects of Within- and Between-Group Inequality on the Gradient.

heal

th

Group 1“Ecomonists”

Group 2“Doctors”

income

Article 3 2/21/03 4:20 PM Page 125

the gradient of health with respect to in-come (or rank) is likely to be steeper than ina study containing a mixture of indistin-guishable reference groups. More generally,if health depends on relative, not absolute,income, and there is an increase in incomeinequality that increases inequality withingroups more than it increases it betweengroups, the slope of the gradient will in-crease.

In Deaton (2001a), I show that if healthdepends on relative income as in equation(7), so that

his = h+ β(yis − ys) + εis (10)

where εis is a random term that ensures thatthere is a scatter around the line, then theexpectation of health conditional on individ-ual income, but unconditional on referencegroup income, takes the form:

E(his | yis) = h+βσ2

w

σ2w + σ2

b

ys (11)

where σ2w and σ2

b are the within- andbetween-group variances of income. Theformula is the same as the attenuation biasformula for measurement error in regressionanalysis.

2.1.6 Evolution, Equality, Deprivation, and Fairness

It might easily be supposed that hierar-chic, unequal societies are an inevitable partof the human condition. Yet for the vast ma-jority of our evolutionary history, humanslived in hunter-gatherer groups that werenot only not hierarchic, but aggressivelyegalitarian; see the survey of the literatureby David Erdal and Andrew Whiten (1996).As has long been argued, perhaps first byMcKeown (1976, 1979), human health ismaximized when we live under the condi-tions under which we evolved, pursuing reg-ular exercise (walking ten to fifteen miles aday, as foragers and hunters did), and eatinglow-fat, low-salt, low-meat, low-sugar, high-fibre, largely vegetarian diets. By the sametoken, given that hierarchies and social in-

equalities were unknown for most of our his-tory, modern inequalities are likely to be ahazard to our health. This argument is force-fully and eloquently put by Wilkinson(2000).

That foraging groups were egalitarian ap-pears to be widely agreed. Such arrange-ments could perhaps have come from lack ofa technology for storing food. When a killhas been made, and the meat is too much tobe consumed at once, sharing and subse-quent reciprocity are the only mechanismsthat can turn meat today into meat tomor-row. Humans, unlike other predatoryspecies, lack a long-term ability to store glu-cose, and so cannot summon the energy tohunt after long periods without food, whichmay be why hunter-gatherer bands evolvedin the first place (Arthur Robson 2002).Members of groups that enforced strictsharing would therefore have a survival ad-vantage over members of those who did not,so that a preference for sharing, fairness,and reciprocation may be evolved attributes(see also Samuel Bowles and Herbert Gintis1998). With the invention of settled agricul-ture, with the associated ability to fill andhold food in granaries, as well as to buildherds of animals, egalitarian and reciprocalsharing was less efficient, and could give wayto hierarchies within which rich and power-ful individuals could dominate others.Although such systems and their industrialsuccessors are vastly more productive than isforaging, the benefits come at the price of anagging and health-compromising outrageover the loss of equality. And while humanswill perhaps evolve to suit this new environ-ment, we have only given up foraging for avery short time, only 10,000 to 20,000 yearsof our one- to two-million-year history.Adaptation to the new environment has itsbenefits, in terms of production, longevity,health, and population size, but it has a lin-gering cost that prevents our health fromreaching its full potential.

Wilkinson and others have begun toweave together a plausible story of the


Article 3 2/21/03 4:20 PM Page 126

processes that support such an account.Psychosocial stress is the main pathwaythrough which inequality affects health.Wilkinson draws a contrast between soci-eties in which relationships “are structuredby low-stress affiliative strategies which fos-ter social solidarity” on the one hand, andsocieties characterized by “much morestressful strategies of dominance, conflictand submission. Which social strategy pre-dominates is mainly determined by howequal or unequal a society is” (Wilkinson2000, p. 4). Equality is seen as a precondi-tion for the existence of stress-reducingnetworks of friendships, while inequalityand relative deprivation are seen as com-promising individual dignity, and promot-ing shame and violence. At the same timethe biological mechanisms through whichchronic stress compromises health are be-ginning to be understood; excellent surveyscan be found in Robert Sapolsky (1998),Eric Brunner and Marmot (1999), andWilkinson (2000, ch. 2.)

The story outlined above is persuasive inmany respects. That the social environmentin which we live helps determine our healthis surely right, and the effects of psychoso-cial stress are now well-documented inboth human (particularly Whitehall civilservant) and animal populations, wherecontrolled experiments are possible. Yet itis less clear why income inequality is theonly villain, or even one of the villains, inthe piece. One attempt to explain is givenin Deaton (2001b), which is extended inthe following account. I treat income rela-tive to other members of a reference groupas the key variable, and hypothesize thatstress on each individual depends on thedifferences between that individual’s in-come and those of others in the group. Theexistence of incomes above you poses athreat, while incomes below you can be ei-ther costly, if you value fairness, or benefi-cial, if you derive pleasure from the statusassociated with having more income thanothers. A utility function with these proper-

ties has been proposed by Ernst Fehr andKlaus Schmidt (FS) (1999), who note thatpeople with such preferences, unlikepurely selfish actors, are willing to give uptheir own resources to affect other people’soutcomes. They show that the assumptionthat a fraction of the population has suchpreferences can account for a wide range ofexperimental and actual social behaviors.

The FS utility function is written

u(y) = y − β1

∫ ∞y

(x− y)dF (x)

+ β2

∫ y

0

(y − x)dF (x) (12)

where β1 > 0 captures the extent to which itis harmful to have people with incomesabove you, while β2, which can be positive,negative, or zero, but in all cases greaterthan –1, captures the benefits or costs thatarise from the existence of people with in-comes less than yours. FS assume that β2 isnegative, so that people with (12) are averseto people being either below or above them,caring about “fairness,” or more precisely,“self-centered inequity aversion,” but theynote that almost all of their results are robustto allowing β2 to be positive. β2 is less thanβ1 which guarantees that people prefer thattheir own incomes be higher.

If we note that the integral of y – x overthe range of x is y – µ, for mean income µ,we can rewrite the FS utility function as

u(y) = y + β2(y − µ)− (β1 − β2)×∫ ∞y

(x− y)dF (x) = y + β2(y − µ)

−(β1 − β2)µR(y) (13)

where Ry is the measure of relative depriva-tion, proposed by Shlomo Yitzhaki (1979).Relative deprivation is a term frequentlyused by Wilkinson in his discussions of socialstress, and whose implications for health,also using the Yitzhaki measure, have beenexplored by Christine Eibner (2001).Although the meaning here is more special-ized (and more precise) than that intended


Article 3 2/21/03 4:20 PM Page 127

by Wilkinson, the formalization appears tobe well-chosen, in that it generates many ofthe effects that he proposes. The derivativeof health (utility) with respect to income is1 + β2 + (β1 − β2)(1 − F(y)), which is positive,while the second derivative is −(β1 − β2)f(y)which is always negative. So if we think ofhealth as proportional to utility, FS utilityimplies a concave relationship between in-come and health within groups. Becausehealth is entirely determined by income, in-equality has no effect on health conditionalon income.

If we look at group health, and computethe average by integrating (13) over y, weobtain

u(y) =∫ ∞

0

u(y)dF (y)

= µ[1− (β1 − β2)g] (14)

where µ is mean income, and g is the ginicoefficient of income inequality. Acrossgroups, utility—or by supposition health—depends positively on mean income but neg-atively on income inequality. (Note also thatutility evaluated at mean income µ has thesame form as (14) with the gini coefficientreplaced by the relative mean deviation,sometimes called the Pietra or Robin Hoodindex, the last frequently used in the epi-demiological literature. Equation (14) con-forms almost exactly to Wilkinson’s (1997)statement that “income inequality summa-rizes the health burden of individual relativedeprivation.”

In summary, this theory of mortality riskhas three important implications: (i) withingroups, health is a concave increasing func-tion of income; (ii) conditional on an indi-vidual’s income within the group, inequalitydoes not matter for individual health; and(iii) for groups, average health dependspositively on group income and negativelyon group income inequality. The results oftesting this theory are postponed to section3, where they can be presented in the con-text of other, related work on geographicalmortality differences in the United States.

2.2 Other Mechanisms, Other Inequalities

2.2.1 Credit Constraints, Health, andIncome Inequality

Income inequality may affect investment,including investments in health and educa-tion. If poor people are more likely to becredit constrained than rich people, if onlybecause they have less collateral, redistribu-tion from those without credit constraints tothose with them may increase investment, amechanism that has been explored in the re-cent literature on economic growth; see forexample Roland Benabou (1996); PhillipeAghion, Eve Caroli, and Cecilia García-Peñalosa (1999); Pranab Bardhan, Bowles,and Gintis (2000). Relevant health invest-ments may include taking a sick child to ahospital, paying for vaccinations, or evenproviding adequate nutrition; even if thesehave high rates of return over the long run,poor parents may lack the money to pay forthem. Education is also relevant, in its ownright, as an investment in human capital andhigher earnings, but also because of its linkswith health. For example, Elo and Preston(1996) estimate that, around the world, ayear of extra education reduces mortalityrates by about 8 percent, half of which worksthrough the effects of additional earnings,and half directly.

Figure 4 shows some relevant data on ed-ucation for rural India. The 52nd round ofthe National Sample Survey collected datain 1995–96 on education enrolment and sta-tus of respondents. The graph shows thepercentage of boys and girls aged fromseven to twelve inclusive that are currentlyenrolled in school (as reported by thehousehold in the survey) as a function of thelogarithm of total household expenditureper head, a close proxy for income per head.The graphs are calculated using nonpara-metric regression using 20,307 boys and17,321 girls aged seven to twelve. Boys aremore likely to be in school than girls, bothare more likely to be in school when theylive in better-off households, and the effects


Article 3 2/21/03 4:20 PM Page 128

of additional resources are larger for girlsthan for boys. Although the slopes of thecurve vary somewhat with per capita expen-diture, they are flatter among better offhouseholds. In consequence, and with theusual caveats about omitted variables andcausality, redistribution from richer topoorer households will increase the totalpercentage of children in school, and willincrease girls’ enrolment more than boys’.Education, in turn, is likely to improvehealth, not only of those receiving it, but oftheir children, particularly in the case ofwomen.

Figure 5 shows some contrasting data onchild vaccinations from the same Indian sur-vey. As was the case for school enrolment,the fractions of children under five who haveall three of the most important vaccinations(BCG, DPT, and OPV) is lower for poorthan for rich households. However, unlikeeducation, the slope is only slightly flatteramong better-off than among poorer house-

holds, so that this graph indicated no majoreffect of redistributing income on raisingchild vaccination rates. Nor is there anydifference between vaccination rates of boysand girls.

These examples are only suggestive. Thecurves in figures 4 and 5 make no attempt tocontrol for other factors—such as parentaleducation, school quality, or health serviceprovision—that are likely to be positively cor-related with both household income and theoutcome, so that the effects of income are al-most certainly overstated. Nevertheless, thegraphs illustrate that income redistributionmay (or may not) increase health and educa-tional investments in children, and may evendifferentially favor girls.

2.2.2 Nutritional Wages, Health, and Inequality of Land

The nutritional wage model provides anaccount of how inequality affects both healthand earnings while explicitly recognizing that


Figure 4. School Enrolment of Boys and Girls Aged 7 to 12 in Rural India, 1995–96.

Source: Author's calculations (locally weighted regressions) based on National Sample Survey data.

perc

enta

ge r

epor

ting

enro

lmen

t

girls

boys

100

80

60

40

100 300

household per capita total expenditure, rupees per head

500 700

Article 3 2/21/03 4:20 PM Page 129

health and earnings are simultaneously de-termined. It pinpoints inequality in landholdings, not inequality in income, as thedeterminant of poor health, and providesthe first of several examples of an importanttheme of this survey, that inequality in di-mensions other than income may have an ef-fect on health. The ideas go back to HarveyLeibenstein (1957), with the fundamentalwork by James Mirrlees (1975) and JosephStiglitz (1976); see Christopher Bliss andNicholas Stern (1978) for a survey. That nu-tritional wage models can account for per-sistent poverty and destitution in poor coun-tries is eloquently argued in ParthaDasgupta and Debraj Ray (1986, 1987) andDasgupta (1993); an excellent textbook sum-mary is provided by Ray (1998, ch. 13).According to this story, unemployed anddestitute workers, even though they are will-ing to work for less than the wage being cur-rently paid to employed workers, cannot un-

derbid them because their undernourish-ment and poor health lowers their marginalproduct to make them unattractive hireseven at the lower wage. More formally, theremay be no combination of work and calorieconsumption that is feasible for the workerand acceptable to the employer. Under suchconditions, unemployment, and the chronicmalnourishment and sickness associatedwith it, cannot be eliminated by decreasingthe wage.

What is the role of inequality in this story?The root of the problem is seen in the distri-bution of land. If the destitute had sufficientland of their own, they would be able tomeet their basic nutritional needs by grow-ing food on small plots without having towork for wages. Seen this way, the problemof unemployment and malnutrition is aproblem of landlessness, and can be re-solved by redistributing land so as to give toevery family a plot sufficient to guarantee


Figure 5. Vaccination Rates of Children Under 5 in Rural India, 1995–96.

Source: Author's calculations (locally weighted regressions) based on National Sample Survey data.

perc

enta

ge r

epor

ting

full

vacc

inat

ions

girls

girls

boys

boys

70

60

50

40

30

100 300

household per capita total expenditure, rupees per head

500 700

Article 3 2/21/03 4:20 PM Page 130

their basic needs, and to enable them to par-ticipate gainfully in the labor market.Destitution is therefore seen as a problemthat can be dealt with by a more equitabledistribution of land.

The nutritional explanation of destitutionand chronic malnutrition in India byDasgupta and Ray is echoed by RobertFogel’s (1994) account of the economic andhealth history of Europe. He argues that,even at the end of the eighteenth century inEngland and France, food production wasso low that, given its likely distribution overpeople, perhaps a fifth of the population wascapable of no more than a few hours of lightwork each day. These people were chroni-cally malnourished, were short in stature,and died young. Only the increase in agricul-tural productivity in the nineteenth centurypermitted an escape from this nutritionaltrap, and meant the beginning of the transi-tion to better health, lower mortality, andlonger life.

This story of nutrition and wages hasmuch to commend it. It directly incorpo-rates the two-way causality between healthand earnings and provides a general equilib-rium explanation of unemployment andpoor health that has an obvious relevance topoor countries now as well as to the histori-cal record in now-rich countries. It also pin-points land reform as the appropriate redis-tributive policy prescription. Yet thepredictions of the model have been chal-lenged; see for example Hans Binswangerand Mark Rosenzweig (1984), Rosenzweig(1988, pp. 720–28), and John Strauss andDuncan Thomas (1998). Workers who aretrapped by their low nutrition and inabilityto work would devote all their energies tofinding food, and would have no energy forconsuming anything other than food, forsaving, or even for procreation (MarkGersovitz 1983). And even in the pooresteconomies, food is typically too cheap rela-tive to the wage rate to make the trap plausi-ble. Shankar Subramanian and Deaton(1996) calculate that in rural Maharashtra in

1983, 2,000 calories (in the form of standardcoarse cereals) could be purchased for lessthan 5 percent of the day wage, a findingthat is consistent with the observation thatpoor agricultural workers in India typicallyeat their fill of cheap calories at the end ofthe work day; see also Anand Swamy (1997).With food so cheap, the nutritional wagetrap seems too easy to escape. More broadly,the model does not always draw a suffi-ciently clear distinction between nutrition,which comes from the food that can bebought for money, and nutritional status,which depends on disease as well as on nu-tritional inputs. A plentiful supply of foodwill not nourish someone whose drinkingwater and food are contaminated, andchronic malnutrition typically needs to beaddressed through public health measuresas well as by increasing the supply of food.This criticism applies, not only to the nutri-tional wage theory as an account of destitu-tion in poor countries, but also to McKeown(1976) and Fogel’s (1994) arguments aboutthe historical importance of nutrition; seeparticularly Preston (1996).

2.2.3 Public Goods and Political Inequality

While the nutritional wage story pinpointsinequality in land holdings as a cause of poorhealth, another line of literature sees the po-litical process as a route through which in-come and other inequalities can affecthealth. One story is that heterogeneity inpreferences makes it more difficult for peo-ple to agree on the provision of publicgoods, such as health, water supply, wastedisposal, education and police. Such mecha-nisms have long been recognized in the lit-erature on political economy, and a simplebut suggestive account has been provided byAlberto Alesina, Reza Baqir, and WilliamEasterly (1999). They point out that the av-erage value of public good to members of acommunity diminishes with the heterogene-ity of their preferences, a heterogeneity thatcould be located in racial, income, or geo-graphical differences. Provision is therefore


Article 3 2/21/03 4:20 PM Page 131

likely to be lower where heterogeneity isgreater. Although they note the potential ap-plication to income inequality, Alesina,Baqir, and Easterly think of their model asapplying to racial divisions in the UnitedStates, and to ethnic fractionalization morewidely, a variable that often appears to havenegative consequences in cross-countrygrowth studies. In the context of the citiesand counties of the United States examinedby Alesina et al., ethnic fractionalization isclosely related to the fraction of the popula-tion that is black, which is positively relatedto total spending, but negatively related tothe shares in spending of “productive” pub-lic goods, such as health, roads, and educa-tion. For health, the total effect offsets theshare effect, so that the absolute amount ofhealth spending is positively associated withfractionalization. Income inequality is in-cluded in the models, although the resultsare not presented in the paper. However, Iunderstand from one of the authors that theeffects vary from model to model and thatthere are no robust negative effects of in-equality on either total spending or its distri-bution. However, it might also be noted thatRobert Putnam (1983), in his study of socialcapital in Italy, also sees equality as an im-portant element in a well-functioning civiccommunity.

Simon Szreter (1988) has provided a per-suasive account of politics and sanitation inBritain in the mid-nineteenth century; seealso Easterlin (1996). The urbanization ofpopulation associated with the industrialrevolution led to a sharp reduction in publichealth, with mortality higher in cities than inthe countryside, and a decline in overall lifeexpectancy. Urban populations often had noaccess to clean water, and no facilities fordisposal of human and other wastes, whichwere allowed to accumulate as a perpetualhazard to health. Crowding aided the trans-mission of infectious diseases, some ofwhich can only be sustained in populationsabove a critical size. Pollution from smokeand other factory discharges contaminated

the atmosphere and the environment. Yetmany cities, in Britain and in Europe, wereslow to address these problems, even whenthe necessary policies were well understood.Although the germ theory of disease wasgenerally accepted only after 1870, earlierexplanations, such as the “miasma” theory,also emphasized the importance of cleaningup the environment. And while money is al-ways a factor limiting public construction,these were periods of relatively rapid eco-nomic growth. Indeed, the coexistence ofrapid economic growth and mortality in-crease (as well as a decrease in stature) dur-ing this period is regarded as something of apuzzle by economic historians—such asMichael Haines and Hallie Kintner (2000),Roger Schofield and David Reher (1991),and Fogel (1997)—who are typically so con-fident of the link between health and in-comes that they often use measures of theformer—such as stature—as reliable indica-tors of the latter (Richard Steckel 1995).

Szreter (1988) argues that the key tounderstanding the mortality transition inEngland lies in local politics. Although theindustrializing cities were in fact well sup-plied with fresh water, it was used for com-mercial purposes, not supplied to homes.And the new entrepreneurial class, richthough it was, saw no point in spending eachothers’ money for public sanitation whichhad no obvious commercial benefit. It wasonly after political reform, and particularlythe limited political emancipation of work-ing men, that new political coalitions coulddevelop that made sanitation and publichealth a priority.

This is a story of nineteenth-centuryEngland, not of the world today, and it isabout political, not income, inequality. Yetthere is a marked parallel with the effects ofcivil rights legislation in the American Southin the late 1960s. Douglas Almond, KenChay, and Michael Greenstone (2001), usingdetailed data from Mississippi, documentthat, prior to 1965, hospitals were strictlysegregated by race and that, especially in


Article 3 2/21/03 4:20 PM Page 132

rural areas, there was limited access forblack mothers and their infants. Federal ac-tions, i.e., the 1963 court decision outlawing“separate but equal” clauses in federal hos-pital construction, and the 1964 Civil RightsAct, made segregation illegal, but it was thepassage of Medicare in 1965, which prohib-ited Medicare funding to segregated hospi-tals, that effectively enforced desegregation.Almond, Chay, and Greenstone show that,from 1965 to 1971, there was a large reduc-tion in black post-neo-natal infant mortalityrates, particularly in conditions amenable totreatment, such as diarrhea and pneumonia,and that this period saw the only sustainednarrowing of the gap in black-white infantmortality rates after 1950.

In the story of Britain in the nineteenthcentury, the central actors in the plot are theReform Acts and their (limited) extension ofdemocracy which may or may not have (di-rectly) reduced income inequality, but cer-tainly reduced political inequality. In thestory of the South, the civil rights legislationnot only improved the health of AfricanAmericans, it also markedly improved theirincomes relative to those of whites (JohnDonohue and James Heckman 1991; andDavid Card and Alan Krueger 1993).Reducing political inequality—that some canvote and others cannot, or some can get ac-cess to schools, hospitals or jobs while otherscannot—is not only valuable in its own right,it also helps reduce other inequalities, inhealth, incomes, and education, in accordwith the general thesis of Amartya Sen(1999). In a contemporary context, it is hardnot to see political action, or at least the lackof it, as one of the reasons behind the lowlevel of provision of schools and clinics inIndian villages. Rhagobendra Chattopadhyayand Esther Duflo’s (2001) work in Indiashows how the mandated representation ofwomen as leaders of village councils (in arandomly selected third of all Indian villages)has led to small but perceptible gains in pub-lic goods important to women and children,particularly water, fuel, and roads (whose

construction provides employment opportu-nities for women) though, perhaps surpris-ingly, not in the limited amount of preschooleducation that is under panchayat control.

The work on villages in India, on nine-teenth-century Britain, and on cities in theUnited States all point to the possible im-portance for health of political arrange-ments, and in particular to the consequencesof inequalities in political power. Onceagain, the story is not about income inequal-ity per se, though income inequality canhardly be unrelated to political inequality.But by taking a narrow focus on income in-equality, we may be missing the inequalitiesthat have the largest effect on health.

2.3 Income Inequality as a Consequence of Ill Health

Most of this section has started from theposition that income causes health, develop-ing the consequences for the relationshipbetween income inequality and health. Buthealth also causes income, and policies thatdirectly affect health are likely to have ef-fects on income inequality.

When interactions between income andhealth are important, the distribution of in-come will depend on the level and distribu-tion of health. Any measure that reduces thespread of health conditions across the popu-lation, or improves the health environment,will narrow the distribution of income.Health shocks are important determinantsof earnings and consumption in developingcountries; see for example Paul Gertler andJonathan Gruber (2002). In some countrieshealth care expenses can be large relative toincome or wealth (Adam Wagstaff 2002). Inagricultural villages around the world, thepoorest people are often those who cannotwork as a result of some long-term disabilityor injury. Anything that helps people recovermore rapidly from an illness will reduce thepersistence of ill health, which reduces thelong-term variance of health across the pop-ulation. Better insurance arrangements, orbetter and more widespread clinics are


Article 3 2/21/03 4:20 PM Page 133

obvious candidates to reduce such persist-ence, and so will not only improve popula-tion health (if they work at all), but also im-prove the distribution of income. Cleanwater, whose lack affects the poor more thanthe rich, will improve the incomes of the poorrelative to the rich, and reduce income in-equality. Malaria eradication campaigns andvaccination drives will have the same effect,not only improving population health, butalso narrowing the distribution of income.This would be true even in a Lewis world inwhich there is an unlimited supply of labor atthe subsistence wage. Better health may notimprove the wage rate, and cannot do so un-der the Lewis assumption, but it can enablemore people to work at that wage.

Once again, a simple formal model illus-trates these points. Suppose that health andincome satisfy the following equations:

ht = α1 + β1ht−1 + γ1yt + ε1t

yt = α2 + β2ht + ε2t (15)

where ε1t and ε2t are stochastic shocks tohealth, each with zero mean, variances σ11and σ22 and covariance σ12. For the processto be stationary, I assume that 0 ≤ β1 < 1.The conceptual experiment here of a groupof identical people, who differ only in therandom shocks that they receive to their in-comes and health. These shocks are drawnfrom the same joint distribution, but inde-pendently across people. In such circum-stances, the joint distribution of health andincome across people is the same as thelong-run time-series distribution of healthand income from (15). It would be straight-forward to introduce heterogeneity bymaking the parameters individual specificand, if the health and income shocks aredistributed independently of the individualeffects, the inequality from the dynamicscan be added to the inequality from theheterogeneity. While stationarity is not re-alistic for finitely-lived people, it helps usisolate the contribution of random shocksto inequality.

The parameters in (16) can be thought ofin terms of the various policies that might af-fect them. The variance of the health shock,σ11, captures the health environment, andwould be reduced by vector eradicationcampaigns, cleaning up the environment, ormore comprehensive vaccinations. The pa-rameter β1 which is non-negative, controlsthe speed of the healing process once healthhas been compromised so that, for example,better health insurance and better health-care would reduce β1. The parameter β2which is positive, controls the effect ofhealth on income. Disability insurance,which limits the effects of health on income,would decrease β2, towards zero. If we takeincome to be net of health care expenses, β2will also reflect the costs of health care, andbetter health insurance, like disability insur-ance, will decrease it towards zero. The pa-rameter γ1 controls the direct causal linkfrom income to health and, for the purposesof the argument, I shall assume it is zero soas to examine the effects of health on in-come inequality in the absence of any linkfrom income to health.

We can use (15) to calculate the variancesand covariance of income and health. In par-ticular

var(y) = σ22 +β2

2σ11

1− β21

(16)

so that the variance of income is the varianceof the income innovation plus a term thatdepends on the health environment and theinsurance system. A reduction in the vari-ance of health shocks—vector control, forexample—reduces income inequality. So doimprovements in health care, in health in-surance, and in disability insurance. Placesthat have more of these policies in place willhave better health and a more equal distri-bution of income. Even without any causallink from income to health, the regressioncoefficient of health on income (the gradi-ent) is positive, and can be written

b =cov(hy)var(y)

=β2σ11

β22σ11 + σ22(1− β2

1)(17)


Article 3 2/21/03 4:20 PM Page 134

This gradient is larger the larger is the vari-ance of the health shock relative to that ofthe income shock, and is smaller the moreeffective is the healthcare and health insur-ance system.

Income differences between countries arealso affected by their differences in popula-tion health. In consequence, the speed atwhich new health technology is transmittedfrom the industrialized to developing coun-tries affects their relative population healthsas well as their relative incomes; see JeffreySachs (2001). Current debates about interna-tional pricing of drugs and patent protectionare important in this context. Whether thefaster transmission of drugs and vaccines willimprove the distribution of income withinpoor countries depends on the prevalence ofdisease within the income distribution. ForAIDS, prevalence in poor countries is (still)higher among the rich, while for malaria ortuberculosis, prevalence is higher among thepoor. In the past, health innovations have of-ten widened health inequalities when first in-troduced; examples are better sanitation acentury or more ago (Preston and Haines1991); tobacco and lung cancer more re-cently; and child health improvements incontemporary Brazil (Cesar Victora et al.2000). Faster transmission of best-practicehealth care may therefore widen the incomedistribution within the receiving countries, atleast in the first instance.

3. Empirical Evidence on Inequality and Health

3.1 Measuring Income Inequality

The measurement of income inequalityinvolves conceptual and practical issues.Comprehensive treatments of the theory ofinequality measurement were developed inthe 1970s by Anthony Atkinson (1970) andSen (1973), the latter updated in JamesFoster and Sen (1999). Although there are anumber of axioms on the nature of inequal-ity that are broadly accepted, these are in-

sufficient to permit us to make unambiguousinequality rankings between any two distri-butions of income. Instead, the axioms in-duce a partial ordering whereby we cansometimes rank one distribution as more un-equal while, in other cases, we can judge theinequality of distributions only by choosing aspecific inequality measure, with differentmeasures giving different results. In particu-lar, different inequality measures give a dif-ferent emphasis to different parts of the dis-tribution. For example, the gini coefficient ismore sensitive to inequality (or to measure-ment error) at the top of the income distri-bution, whereas measures that work withthe logarithms of income, such as the Theilmeasures, or the variance of logarithms, arequite sensitive to inequality at the bottom.Although neither the conceptual issues northe choice of inequality indicator are themost important issues for health, it shouldbe noted that many of the indexes that areused in the public health literature do notsatisfy even the generally accepted axioms.For example, the Robin Hood index (moreusually known as the relative mean devia-tion; Kennedy, Kawachi, and DeborahProthrow-Stith 1996a) is unaffected bytransfers between individuals on the sameside of the mean. If a transfer program wereto transfer incomes from those just belowthe median to those near the bottom, theRobin Hood index would not change, eventhough there would have been a real reduc-tion in inequality (and very likely a decreasein mortality risk too.) Perhaps most of thepublic health work uses as its inequalitymeasure the share of income accruing to thebottom x (often 50) percent of the popula-tion. Once again, transfers within the bot-tom x percent, or within the top 1 − x per-cent, will leave the measure unaffected,even though such transfers are capable ofhaving a substantial effect on income in-equality more broadly.

Some authors, including Lynch et al.(1998), Luiza Franzini, John Ribble, andWilliam Spears (2001), Paul Brodish, Mark


Article 3 2/21/03 4:20 PM Page 135

Massing, and Herman Tyroler (2000), andDiane McLaughlin and Shannon Stokes(2001) also use the 90:10 ratio defined, notas is usually the case, as the ratio of the 90thto the 10th percentile of the income distri-bution, but the ratio of the share of incomeaccruing to the bottom 90 percent to theshare accruing to the bottom 10 percent.This “inequality” measure has the unfortu-nate property (for an inequality measure)that it is reduced by transferring incomefrom the middle of the distribution to thetop tail; not surprisingly, this measure issometimes strongly negatively correlatedwith more usual measures such as the gini;see for example Franzini, Ribble, andSpears. Those authors who use it to demon-strate that inequality affects mortality maywell have shown the opposite!

There has also been a good deal of discus-sion about the appropriate definition ofhousehold income, and in particular thetreatment of household size. The standardprocedure in economics would be to “equiv-alize” household income by dividing incomeby some measure of household size, eitherhousehold size itself, or the number ofequivalent adults, for example the numberof adults plus half the number of children, orthe square root of the total number of peo-ple in the household. Such per-equivalentmeasures attempt to capture the resourcesavailable to each person in the household,and recognize that, at the same level of in-come, members of a larger household areworse off than members of a smaller house-hold. Indeed, public health statisticians,studying pellagra, were among the first todevelop equivalent measures; see EdgarSydenstricker and Willford King (1921).(Even so, it should be noted that there is ev-idence in Elo and Preston 1996 that, condi-tional on family income, larger family sizemay not increase mortality. Nor do equiva-lized measures give any purchase onwhether income is allocated equally withinthe household) When income inequality iscalculated, it is also important that equivalized

income be assigned to individuals, and thatinequality be calculated over persons, nothouseholds. These apparently technical de-tails can sometimes have serious effects onthe measurement of income inequality, andtheir treatment in the public health litera-ture has often been cavalier. There are alsoobvious constraints when authors restrictthemselves to the inequality measures pre-calculated by the Census Bureau; these re-strictions may sometimes be necessary in or-der to avoid having to deal with other dataproblems with the public-release microdata.

Conceptual problems are dwarfed bymeasurement problems. Income itself ishard to measure, and the difficulties multi-ply when measuring income inequality.Measurement error in income, even if it haslittle effect on the measurement of mean in-come, will inflate the measured variance andmeasured income inequality. The measure-ment of income is sensitive to survey design,particularly to the choice of the referenceperiods for income (longer reference peri-ods give lower measured inequality), and toexactly how the income question is asked.The degree of disaggregation of income cat-egories is important, as is whether incomesare reported as a number, or in a set of pre-defined ranges, typically ending with anopen-ended top category. The choice of cut-off points for the ranges is important, partic-ularly the top band which effectively limitsthe highest income that can be reported.Some surveys permit people to report nega-tive incomes (losses from business activities)and some do not. Some surveys collect dataon income, and some on consumption; thelatter is almost always less unequally distrib-uted than the formal. The response ratefrom surveys varies over space and time, andricher households are typically less likely toagree to participate, in many cases becausethey live in communities where the enumer-ators cannot reach them; see Robert Grovesand Nick Couper (1998). In rich countries,and in some not-so-rich countries, response


Article 3 2/21/03 4:20 PM Page 136

rates have been falling over time, perhaps inresponse to the increasing competition frommarket researchers. There are also differ-ences across countries in the degree towhich people are prepared to cooperatewith government surveys. Response rates inthe United States are typically much higherthan in Western Europe.

Because different countries use differentsurvey instruments, and have different sur-vey protocols, useful cross-country compar-isons of income inequality require detailedknowledge of the specific surveys. Similar is-sues sometimes arise with comparisons overtime within one country, even where the sta-tistical service is of the highest quality.Specifically, the U.S. Census Bureau, in thesummer of 2000, decided that the large in-crease in household income inequality be-tween 1992 and 1993, which it previouslypresented as real, was in some unknowablepart due to changes in survey methodology,particularly changes in the highest level ofincome permitted in the questionnaires, aswell as the introduction of computer-aidedinterview technology; see Arthur Jones andDaniel Weinberg (2000). In consequence,the United States no longer has a consistentcontinuous time series of household incomeinequality. That the United States is worse(as opposed to more transparent) than othercountries seems unlikely.

International data on income inequalitycome from a number of standard sources.Perhaps the most reliable is theLuxembourg Income Study (LIS) whichcontains information on the distributions ofdisposable income for 25 (wealthy) coun-tries over a period of twenty years, althoughnot all countries have data for all years; seePeter Gottschalk and Timothy Smeeding(2000, pp. 273–74). The LIS permits accessto the micro data from broadly comparableincome data for the covered countries. Notethat the underlying surveys do not use thesame questionnaire, so that the comparabil-ity is not perfect, nor are response rates thesame for all countries. Nevertheless, the

data are well-documented and have beenwidely analyzed, so that their properties arewell understood. Some authors have takenincome inequality data for industrializedcountries from other, non-LIS sources, suchas Malcolm Sawyer (1976); these are nowsuperceded by the LIS.

Matters are a good deal more difficult forincome distribution data from the largenumbers of developing countries that arenot covered by the LIS. For many years,popular sources of income distribution datawere Shail Jain (1975) and Felix Paukert(1973), which are essentially compendia ofinequality estimates then available in theWorld Bank and International Labor Office,respectively. In more recent years, researchon international patterns of income inequal-ity has been transformed by the availabilityon the World Bank web site of the inequalitydata assembled by Klaus Deininger and LynSquire (DS) (1996). These data, which haveseen widespread use, contain more than2,600 observations on gini coefficients (andmany quintile shares) for more than 100 de-veloped and developing countries for datesbetween 1947 and 1994. To be included inthe DS data set, estimates have to comefrom an identifiable source, be national incoverage, and be based on either consump-tion or income. (Which comes from which isidentified.) A subset of the observations arelabeled “high-quality” and these have beenwidely (and mostly uncritically) used in alarge number of papers, including papers onincome inequality and health. Much of thehigh-quality data comes from industrializedcountries, so that many researchers inter-ested in poor countries have used at leastsome data not so labeled.

While DS’s data and documentation are agreat improvement over what was previouslyavailable, they do not support the uncriticaluse that has been made of them, as shown inan important study by Atkinson and AndreaBrandolini (2001). Atkinson and Brandolinifocus their attention on the subset of the DSdata for the OECD countries, for which


Article 3 2/21/03 4:20 PM Page 137

there are good, well-documented surveys(including the LIS) which can be used forcomparison. DS’s “high-quality” estimatesdo not do well in this comparison, eitheracross countries, or in some cases over timewithin countries. For example, DS showsSweden as one of the more unequal coun-tries in the OECD, with more income in-equality than the United Kingdom, whereasin the LIS (as reported in Gottschalk andSmeeding 2000), Sweden has the lowest in-come inequality and the United Kingdomthe highest apart from the United States. Insome cases, such as Germany, the DS timeseries of inequality is quite different fromthat computed directly from the surveys.Although the DS data may be more reliablefor poor countries than for rich, it is unlikely,especially since the poor countries contain amuch larger fraction of the data that are notendorsed by DS themselves.

We are currently in the position of nothaving any consistently reliable set of dataon income inequality outside the countriescovered by the LIS. This is in spite of the ex-istence of the fifty or so surveys that havebeen collected under the aegis of the WorldBank’s Living Standard MeasurementSurvey (LSMS), which was set up in 1980with the original purpose of generating com-parable data on income distribution for awide range of countries. While the LSMSsurveys are broadly comparable, the ques-tionnaires are not identical across surveys,and some have differed a great deal.

Research on inequality and health has alsoused data on measures of inequality forareas within countries. In principle, suchmeasures are less problematic, if only be-cause they are usually calculated from na-tional surveys using a uniform survey instru-ment so that, even if there are errors, thepatterns across areas may not be much af-fected. One problem is sample size, espe-cially for small areas. Inequality measuresare usually less precisely estimated thanmeans so that, for example, the U.S. Bureauof the Census publishes estimates of mean

income by state using the CurrentPopulation Survey, which has a sample sizeof around 50,000 households each year.However, it publishes inequality measuresby state only for three-year moving averages.Several developing countries also have regu-lar, national household surveys that are largeenough to support considerable disaggrega-tion; India, Indonesia, and Pakistan are ex-amples. The Indian National Sample Survey(NSS) collects detailed consumption datafrom more than 120,000 households everyfive years or so, and these surveys are de-signed to be representative for more thanseventy regions of the country. Yet even thissurvey does not support the measurement ofincome inequality for districts, the level atwhich the Indian census publishes much ofits data on child mortality.

Most household surveys have a two-stagestratified design in which, at the first stage,primary sampling units (PSUs) are randomlyselected with probability proportional topopulation size. These PSUs are typicallysmall geographical units, such as villages orcensus tracts. Within each PSU, the samenumber of households are selected so that,over the two stages together, each householdin the population has an equal chance of be-ing selected into the survey. In some studies,investigators have calculated measures of lo-cal income inequality based on the house-holds in each PSU. This procedure is obvi-ously dangerous when there are only a fewhouseholds in each PSU and, even when thisis not the case, respondents within PSUs aresometimes not randomly selected, so thatthe relationship between the sample esti-mate and its population counterpart cannotbe assessed. Note too that PSUs are selectedfor statistical convenience, not analyticalmeaning, and frequently do not correspondto any sensible definition of a community.

3.2 Cross-Country Studies of Income Inequality and Health

Cross-country studies have played an im-portant part in the literature on income


Article 3 2/21/03 4:20 PM Page 138

inequality and health. Preston’s (1975) semi-nal analysis looked at international patternsof GDP and life expectancy, and it was onthe basis of his findings that Preston sug-gested that there should be a negative rela-tionship between income inequality andhealth. G. B. Rodgers (1979) and A. T. Flegg(1982) were early studies that followedPreston’s lead, explicitly looking for (andfinding) effects of income inequality onmortality. Rodgers used the Paukert (1973)data for 56 (unnamed) countries and, con-trolling for income and other variables,found hazardous effects of inequality on life-expectancy at birth, life-expectancy at agefive, and on the rate of infant mortality, withthe last only significant in the developedcountries in the sample. Flegg (1982) lookedonly at child mortality, and found significanteffects of income inequality on child mortal-ity in developing countries using the Jain(1975) data. These authors, like Preston,thought of the nonlinearity in the relation-ship between income and health as the basisfor their results, and did not propose any di-rect effect of income inequality at the micro-economic level. A direct effect was found byRobert Waldmann (1992) who used U.N.and World Bank sources, supplemented byincome inequality data from Jain (1975), toinvestigate infant mortality on a cross sec-tion of up to 57 developing and developedcountries. As expected, he found that, condi-tional on mean income, the share of incomegoing to the poorest 20 percent of the popu-lation decreased infant mortality, and moresurprisingly, that the share of income goingto the top 5 percent increased infant mortal-ity. This is a direct effect of inequality; theinfant mortality rates among the poor in-crease when the rich get richer, even whentheir own incomes do not suffer.

Perhaps the single most cited finding inthe literature is Wilkinson’s (1992, 1994,1996) demonstration of a relationship be-tween income inequality and life expectancyacross a number of industrialized countries,not only in levels but, more impressively, in

changes over time. Countries, such asFrance and Greece, that narrowed their in-come distributions by reducing relativepoverty, increased their life-expectancies,while those, such as the United Kingdomand Ireland, whose income distributionswidened, fell behind (Wilkinson 1996, figure5.4). Wilkinson interprets these results asshowing that, as countries become wealthierand move through the epidemiological tran-sition, the leading cause of differences inmortality moves from material deprivationto social disadvantage. Material deprivationprovokes poverty and infectious disease,while social disadvantage provokes stressand chronic disease.

Later research has cast considerabledoubt on the robustness and reliability ofmany of these findings. As expected, one ofthe main difficulties lies in the unreliabilityof the data on income inequality. For exam-ple, using the Deininger and Squire data,Gravelle, John Wildman, and MatthewSutton (2002) fail to replicate Rodgers(1979) results for developed and developingcountries. Jeffrey Mellor and Jennifer Milyo(2001) use a sample of 47 developing anddeveloped countries in 1990 and find thatthe positive correlation between the gini co-efficient and infant mortality vanishes oncesecondary school enrolment is controlledfor, while the negative correlation betweenincome inequality and life expectancy iseliminated by controlling for income perhead. (Commenting on Mellor and Milyo,Kawachi and Tony Blakely 2001 argue thatpoor education may be a consequence ofhigh income inequality, and it is thereforeinvalid to include it as a control.) Mellor andMilyo also fail to replicate Wilkinson’s re-sults for developed economies. Although theDS data have their own problems, the origi-nal results are clearly not robust; see alsoLynch et al. (2001) who use the LIS data todemonstrate how the choice of “sample”points (i.e. countries) affects the link be-tween income inequality and adult mortality.The same is true of Waldmann’s findings;


Article 3 2/21/03 4:20 PM Page 139

Andrew Baumbusch (1995) replicatedWaldmann’s analysis using data of the samevintage, but found that income accruing tothe top 5 percent reduced infant mortalityonce the data were updated from the 1993edition of the World Bank’s World Devel-opment Report.

The single most convincing study of theLIS countries is by Judge, Jo-Ann Mulligan,and Benzeval (1997), who emphasize thepoor quality of the data in previous work,and use the LIS data in their own examina-tion of life expectancy and infant mortality inAustralia, Belgium, Canada, Finland,France, Germany, Ireland, Italy, Luxem-bourg, The Netherlands, Norway, Sweden,Switzerland, the United Kingdom, and theUnited States. In these data, which are thebest international data currently available,the correlation between the gini coefficientand life expectancy is –0.17, insignificantlydifferent from zero, and neither the gini norother measures of income inequality areclose to significance in any of the regressionsexplaining life expectancy. The situation issomewhat different for infant mortalityrates, where there is a significant positive(i.e. harmful) effect of the ratio of the 90thto the 10th percentile. This measure of in-equality exerts a significant effect in severalof the regressions, though it becomes in-significant when controls are added for thenegative effects on mortality of female laborforce participation, see also Lynch et al.(2001) for a similar finding. In these data,the raw correlation between infant mortalityand inequality is driven largely by theUnited States, which is very unequal and hasrelatively high infant mortality. Inequality inthe United States may indeed have some-thing to do with its high infant mortalityrate; however, as I shall argue below, any sat-isfactory discussion of income inequality andmortality in the United States must take ex-plicit account of the effects of race.

If these results are not entirely definitive,it is because the LIS data, although betterthan any other, are neither fully comparable

nor fully accurate. The debate betweenWilkinson (1998) and Judge, Mulligan, andBenzeval (1998) has focused on differencesin response rates across the LIS surveys, andtheir possible effect on the results. It is alsopossible that, as with the difference betweeninfant mortality and life-expectancy, therewill be links between specific causes at spe-cific ages and the plausibly associated mea-sures of inequality, see for example SandraMcIsaac and Wilkinson (1997). Yet, it issurely time to agree that there is currentlyno evidence that income inequality driveslife expectancy and all-cause adult mortalitywithin the industrialized countries. It re-mains to be seen whether this means there isno relationship, or whether there is a rela-tionship that is being obscured by still inade-quate data. Judgement on that depends agood deal on whether there exists a relation-ship between income inequality and healthin other contexts, on which more below.

That, conditional on income, there shouldbe a cross-country relationship between in-fant mortality and income inequality, at leastin poor countries and possibly in rich coun-tries, is both theoretically plausible, andrather better supported by the (admittedlyinadequate) data that are available. Theplausibility comes from recent work fromthe World Bank which, following themethodology pioneered by Deon Filmerand Lant Pritchett (1999b), has used demo-graphic and health surveys around the worldto construct a synthetic measure of wealth,which is then used to explain infant mortal-ity rates; see Davidson Gwatkin (2000) foran overview. The measure of wealth is anindex based on the ownership of variousdurable goods, and while the measure isundoubtedly correlated both with actualwealth and income, we have no way of cali-brating the transformation and thus of usingthe results to relate income to child health.(The transformation is also different in dif-ferent countries, which effectively precludesmeaningful cross-country comparisons.)Nevertheless, the results show very strong


Article 3 2/21/03 4:20 PM Page 140

gradients in child health, with infant mortal-ity rates heavily concentrated at the bottomof the distribution. Wagstaff (2000) usesnine (mostly) LSMS surveys from develop-ing countries to calculate child mortalityrates by quintile of equivalent consumption,and shows that child and infant mortalityrates typically decline most rapidly betweenthe bottom and second quintile. Whetherthese results imply that infant mortality ratesare convex in income depends on the degreeof convexity of the relationship between in-come and the asset index, and on the densityfunction of equivalent consumption, but thatinfant mortality is concentrated at the bot-tom of the income distribution seems likely.Yet there is also some evidence on the otherside. In particular, Mamta Murthi, Anne-Catherine Guio, and Jean Drèze (1995) findvery little effect of poverty on child mortalityacross districts in India once they control forother factors, most importantly female liter-acy and urbanization.

To the extent that the DS data are ac-cepted, there is a good deal of empirical evi-dence from developing countries linking in-fant and child mortality to the DS measuresof income inequality conditional on the levelof GDP per head and a range of other vari-ables (for example in Pritchett andLawrence Summers 1996; Filmer andPritchett 1999a; and Simon Hales et al.1999). Whether this evidence extends toadult mortality and life expectancy is diffi-cult to know, not only because of the datadifficulties with income inequality, but be-cause of the quality of the data on adult mor-tality. Few poor countries have completeregistration systems for deaths, so that goodevidence on adult mortality (or life ex-pectancy at age five, for example) is hard tocome by. In a few cases, such as India, thereare sample registration surveys, and somedata on adult deaths can be gleaned fromthe demographic and health surveys. But formany countries, data on life expectancy areextrapolated from the data on infant mortal-ity rates, and contain little additional infor-

mation. An exception to this generalizationcomes from the countries of Eastern Europeand the former Soviet Union, where life ex-pectancy has been falling as income inequal-ity has increased (see Marmot and MartinBobak 2000, fig. 3, which shows a twelve-country correlation coefficient of –0.63).Peter Walberg et al. (1998) also show thatlife-expectancy fell fastest in the initiallymost unequal regions of Russia, a findingthat is not obviously related to the inequalityhypothesis. Furthermore, as is widely recog-nized, the Eastern European experience isdifficult to interpret because so much elsehas been going on, so that it is hard to isolatethe effect of income inequality.

Finally, there are a number of cross-coun-try studies that link other health outcomes toincome inequality. Steckel (1995) finds a re-lationship between human stature (a mea-sure of cumulative nutritional status) and in-come inequality on a sample of developedand developing countries using the incomedistribution from Jain (1975). Mead Over(1998) looks across cities in the developingworld and finds that the U.S. CensusBureau’s estimates of HIV infection ratesare positively related to the DS measures ofcountrywide income inequality. He inter-prets his findings in terms of upper incomemen demanding the services supplied bylower income women, and these results per-haps come closest to providing substance toPaul Farmer’s (1999) contention that diseaseoccurs along the “fault lines” in the incomedistribution. Marie Gaarder (2001) arguesthat income inequality is likely to worsen thehealth consequences of pollution becausethe poor have lower baseline health and aretherefore more susceptible. She includesthe gini coefficient in a meta-analysis of pre-vious estimated effects of particulate con-centration on mortality at various sitesaround the world and finds significant posi-tive effects. Pablo Fajnzylber, DanielLederman, and Norman Loayza (2000) finda significant relationship between DS ginicoefficients and both homicide and robbery


Article 3 2/21/03 4:20 PM Page 141

rates for a group of 45 (for homicides) and34 (for robberies) developed and developingcountries. A good deal of this is driven byLatin American countries, where both crimeand inequality are very high. Using data onseventeen countries in the Americas, andgini coefficients from DS, Oscar Mujica etal. (2000) confirm the positive correlation(0.55) between homicide and income in-equality, but find a negative correlation(–0.78) between suicide rates and incomeinequality.

3.3 Within-Country Area Studies of Income Inequality and Health

As skepticism has grown about the inter-national relationship between income in-equality and health, attention has switchedto studies within countries, particularly ofmortality and income inequality across thestates of the United States. Two studies, byGeorge Kaplan et al (1996), and byKennedy, Kawachi, and Prothrow-Stith(1996a,b), both published in the BritishMedical Journal, and inspired byWilkinson’s (1992) cross-country work,found a relationship across the states be-tween various measures of income inequal-ity and age-adjusted all-cause mortality, aswell as a number of other measures, includ-ing infant mortality rates, deaths from can-cer, coronary heart disease, and homicide,as well as disability, low birth weight, andcrime. Kawachi and Kennedy (1997) estab-lished that the results were robust to thechoice of inequality indicator, while Lynchet al. (1998) extended the results to 282metropolitan statistical areas (MSAs) in1990, finding that the loss of life from in-come inequality “is comparable to the com-bined loss of life from lung cancer, diabetes,motor vehicle crashes, HIV infection, sui-cide, and homicide in 1995.” Kawachi,Kennedy, and Prothrow-Stith (1997) arguethat income inequality works by reducingsocial capital, in particular the degree oftrust between people, a state-level measureof which is constructed from the General

Social Survey. Such an account is very muchin the spirit of stories of psychosocial stresswithin unequal social structures. In supportof this explanation, Wolfson et al. (1999) es-timate the degree of nonlinearity in the in-come to mortality curve using the NLMS(as in figure 2) and show that the effects ofincome inequality on state mortality ratesare too large to be explained by the nonlin-earity argument alone, so that there must besome direct effect of income inequality onindividual mortality. The implications ofthese results for economic policy have notgone unnoticed. See for example Kaplanand Lynch’s (2001) editorial in theAmerican Journal of Public Health entitled“Is Economic Policy Health Policy?”

These within-nation results do not sufferfrom the same data problems as do the in-ternational comparisons. Income inequal-ity is usually measured from incomes col-lected in the census, which is administeredin the same form to all households in allstates. Nor is there any question about theexistence of the correlation. Figure 6shows a typical scatter plot between the logodds of age-adjusted mortality (the log ofthe ratio of the fraction dying to the frac-tion not dying) on the vertical axis, and thegini coefficient of household income perequivalent, with equivalents defined as 1for adults and 0.5 for children aged eigh-teen and less. The District of Columbia isincluded and, although it is an outlier inthe sense of having higher income inequal-ity and higher mortality than any state, itlies along the regression line defined bythe other observations. Across the states ofthe United States, income inequality isstrongly negatively correlated with income;poor states, many of which are in theSouth, also have the weakest safety nets forthe poor. Even so, the correlation betweenincome and mortality is much weaker thanthe correlation between income inequalityand mortality, and adding income to a re-gression does not eliminate the effect ofincome inequality.


Article 3 2/21/03 4:20 PM Page 142

Nevertheless, there are serious questionsabout whether the correlation between in-come inequality and mortality is robustthrough time, and whether it comes fromthe effects of income inequality or someother factor that is correlated with it. Mellorand Milyo (2001) use data for the 48 conti-nental states from five census years—1950,1960, 1970, 1980, and 1990—and reproducethe strong hazardous effect of the gini coef-ficient on all-cause mortality when only yeardummies, the age composition of the state,and median income are included as controls.The inclusion of controls for the averagelevel of education in each state eliminatesthe significance of the gini coefficient and,once the authors include controls for thefractions of people in each state who are ur-banized and who are black, the gini coeffi-cient attracts a negative sign, though onethat is not significantly different from zero.

Similar reversals are found for the fractionof births that are low birth weight while,over the five decades, there is no relation-ship across states between deaths from car-diovascular disease, from malignant neo-plasms, or from liver disease. Indeed, for thefirst two, income inequality has a negativeand significant relationship with deaths oncecontrols are entered for income, education,race, and urbanization. Only for homicidesand, to a lesser extent, infant mortality anddeaths from accidents, is the gini coefficienta risk factor conditional on the other con-trols. Mellor and Milyo also subject the hy-pothesis to a much more stringent test, look-ing at the relationship between ten- andtwenty-year changes in mortality and thecorresponding changes in income inequality.This is almost certainly too severe a test be-cause it places a great deal of weight on thetiming of the link between income inequality


Figure 6. Income Inequality and Age-Adjusted Mortality, U.S. States, 1990

Source: CDC for state (all race, both sex) mortality data, author's calculations from 1990 census for income inequality.Note: Circles are proportional to population size.

12

10

8

6

NH

NM

OK

AZ

TXNY

CAFL

HI

MNUT

IAWI

ME IN

LA

DC

MSALGASC

MDDE

VT

.33 .35 .37 .39 .41 .43 .45 .47 .49

gini of per equivalent income

age

adju

sted

mor

talit

y pe

r 10

00

Article 3 2/21/03 4:20 PM Page 143

and mortality. Even so, it is worth notingthat, with one exception, none of the incomeinequality to mortality relationships survivesthe test. The exception is homicide, wherethe relationship with income inequality iswell-determined and holds over time as wellas in the cross-section.

Another concern is the pooling of dataacross racial or ethnic groups with differentincomes and different mortality rates. In theUnited States, African Americans havehigher mortality rates and lower incomesthan whites, so that states with a high frac-tion of blacks tend to have higher mortalityrates as well as higher income inequality. Ascan be seen from figure 6, such states tendto be predominately in the South wheremany other special factors are likely to oper-ate; by contrast, the states with low mortalityare states where there are few or no AfricanAmericans. If data are pooled for 1980 and1990, the log odds of age-adjusted mortalityresponds to the gini coefficient with a coeffi-cient of 1.7 for males, and 1.1 for females. Inthe same regression, the coefficient on themean of the logarithm of equivalized incomeis negative for men, but barely significant,and not at all for women (Deaton 2001b). Toillustrate the size of the effect of inequality,the 1990 gini coefficients for Louisiana andNew Hampshire were 0.45 and 0.33 respec-tively, which would account for a 12-percentdifference in mortality rates, more than halfof the difference shown in figure 6.

If we now confine the calculations towhite mortality alone, so that we no longerhave the mechanical effects describedabove, the coefficient on the gini drops to1.1 for men and to 0.6 for women, about athird lower than for all-race mortality.Nevertheless, these effects remain signifi-cantly different from zero, and still showthat inequality is a health hazard for thewhite population alone. If we recalculate thegini coefficients so as to measure only in-equality among whites, the effects are fur-ther reduced, to 0.6 for men, and 0.4 forwomen, and only the former is (marginally)

significantly different from zero. This resultmeans that the effect of inequality on whitescomes, not from the inequality of white in-comes, but from the inequality betweenwhites and blacks, raising the suspicion thatthe effect has more to do with race than withincome inequality. Such a suspicion is borneout by controlling for the fraction of thepopulation that is black in each state. Itturns out that a high fraction of blacks raisesmortality rates among both males and fe-males (note that these are whites) and thatconditional on race, income inequality hasno effect on mortality. It is unclear why thefraction black should exert such a strong ef-fect on white mortality (black mortality isalso higher in states where there are rela-tively many blacks), though it might be ar-gued that the fraction black is itself somesort of marker for the inequality that charac-terizes race relations in the United States.Even so, the effect is not one that worksthrough income inequality; once the fractionblack is included in the regression, the ginicoefficient has no effect.

There is an obvious concern here that Ihave simply replaced one invalid variable—income inequality—with another—racialcomposition—and that both stand proxy forsomething else. This is particularly the casewith the state data, where there are at most51 observations (or 102 observations if wepool data from 1980 and 1990), and where itwould be easy to confound racial composi-tion (or income inequality) with geographi-cal factors, especially given the peculiar roleof the South. Nevertheless, Deaton andLubotsky (2003) show that the results carrythrough to the 287 MSAs that can be consis-tently identified between 1980 and 1990.These data can be used to replicate the find-ings of Lynch et al. (1998), and to show that,once again, the inclusion of racial composi-tion eliminates (and sometimes even re-verses) the effect of income inequality. Andbecause there are more MSAs than states, itis possible to work within regions, and toshow that whether we look at cities in the


Article 3 2/21/03 4:20 PM Page 144

South, or cities in other regions, and condi-tioning on city average income, white mor-tality is higher in cities where the fraction ofblacks is higher. The results also holds acrossnearly all age and sex groups. In apparentcontradiction to these results, McLaughlinand Stokes (2002) find that, with countydata, controlling for race does not eliminatethe effects of income inequality on mortality.Perhaps these findings come from the un-fortunate choice of inequality measure (theratio of the share of the bottom 90 percentto the bottom 10 percent—see above);Deaton and Lubotsky’s findings are repli-cated on the county data.

That the fraction black eliminates the ef-fects of income inequality or mortality iseasy enough to understand mechanically;blacks are poorer and whites richer in citieswith a large fraction black, so that incomeinequality is higher where the fraction blackis higher, but it is the fraction black, not in-come inequality, that matters for both whiteand black mortality. But the finding tells usnothing about what it is about largely blackcities that makes them less healthy. One ar-gument is similar to the case against incomeinequality, that racism generates stress andlack of social capital, sickening both the per-petrators and the victims. But there areother more mundane stories. While there isconsiderable evidence that the quality ofhealth care is lower for African Americansthan for whites (Brian Smedley, AdrienneStith, and Alan Nelson 2002), there is muchless evidence that treatment is different be-tween blacks and whites within the sametreatment area, technically within a hospitalreferral region (Amitabh Chandra andJonathan Skinner 2002). There is also evi-dence in Fuchs, Mark McClellan, andSkinner (2002) that the association betweenthe fraction black and white mortalityamong the elderly is strongest for cardiovas-cular disease. So one possibility is that, in ar-eas where African Americans make up alarge fraction of the population, hospitals areless well-funded or otherwise of lower qual-

ity and that, in those areas, both blacks andwhites have higher mortality rates. In theevent of an acute myocardial infarction(heart attack), people are taken to the near-est facility whatever their income, health in-surance status, or race, so that whites, likeblacks, will receive the health care providedin their community. As a result, both blacksand whites suffer from the lower-qualityhealth care provided to blacks. Furthermore,in the many metropolitan areas that havemultiple hospital referral regions, the effectsof the fraction black on white mortality willbe attenuated, and on black mortality exag-gerated by residential segregation. Segre-gation effectively transfers the burden ofpoor treatment from whites to blacks; seeDouglas Massey (1991) for this argument ina more general context.

If the apparent effects of inequality ormortality are nothing to do with income in-equality per se, but are instead tied to racein America, then there should be no rela-tionship between income inequality andmortality across regions in countries whererace does not have the same salience. Thisseems to be the case. Nancy Ross et al.(2000) find that there is no relation betweenincome inequality and mortality for the tenprovinces and 53 metropolitan areas ofCanada, and Ross has similar, unpublishedfindings for Australia. For Britain, there ap-pears to be no area study on income inequal-ity and health, though Yoav Ben-Shlomo,Ian White, and Marmot (1996) find thatmortality in the 8,464 wards of England isaffected not only by an index of deprivationbased on household characteristics, but alsoby the within-area dispersion of the depriva-tion index. Again, this is what is to be ex-pected if the deprivation measure is moreclosely linked to mortality among high-deprivation people. Likewise, DebbieStanistreet, Alex Scott-Samuel, and MarkBellis (1999) find a link between mortalityand imputed income and income inequalityacross 366 English local authorities. Buttheir income measures are imputed from


Article 3 2/21/03 4:20 PM Page 145

the occupational structure of each area, as-suming that wage rates are identical every-where, so that they are measures of local oc-cupational patterns, not of income norincome inequality.

Tung-liang Chiang (1999) looks at mortal-ity rates in the 21 counties and cities ofTaiwan in 1976, 1985, and 1995 using house-hold survey data to calculate measures of in-come and income inequality. He finds strongprotective effects of income in 1976 and1985, and little effect of income inequality,but finds that the situation is reversed in1995, at which date income inequality is ahazard, and income has no effect. Chiang in-terprets his findings as support forWilkinson’s idea that income is important atlow levels of income, and income inequalityat high levels of income, which, as we haveseen, is also consistent with a nonlinear ef-fect of income and no direct influence of in-equality. Enrique Regidor et al. (1997) findno relationship between (a nonstandard)measure of income inequality and the preva-lence of long-term disability across the sev-enteen regions of Spain.

It is widely believed that there is a link be-tween income inequality and crime (includ-ing homicide) in the United States. I have al-ready noted Mellor and Milyo’s (2001)finding that homicide was the only negativehealth outcome that was robustly linked toincome inequality in their tests, and suchfindings have consistently appeared in theliterature since Isaac Ehrlich (1973); seeC. C. Hsieh and M. D. Pugh (1992) for anoft-quoted review and meta-analysis.

There appear to be few relevant studiesfrom developing countries, even where itwould be possible to do so, for example inthe work on fertility and child mortality inIndia by Murthi, Guio, and Drèze (1995)and Drèze and Murthi (2001). However,Drèze and Reetika Khera (2000) find thathomicide rates across India are unrelated tomeasures of consumption inequality, but arepositively associated with the fraction of“missing” women. Although the authors do

not make the point, a link between homicideamong men and the shortage of women in-vites a socio-biological explanation in termsof mating behavior. Rodolfo Peña, Stig Wall,and Lars-Åke Persson (2000) find that infantmortality risks are higher among the poor inNicaragua, and higher still when the poorlive in relatively wealthy neighborhoods,which is consistent with a negative role forinequality.

That there is no interregional relationshipbetween income inequality and mortality issomewhat surprising, given the nonlinear re-lationships between income and health withwhich I began. Perhaps the curvature is toomild to generate any strong relationship—and Wolfson et al. (1999) provide such evi-dence—or perhaps income is not causal. Ihave already referred to the regional esti-mates for the United States where educationdrives out education from aggregated mor-tality relationships. Another place to look isthe time-series data and the patterns of in-come, income inequality, and mortality overtime, and I shall examine that evidence insection 3.5 below.

3.4 Studies of Income Inequality andHealth Using Individual Data

Studies using individual-level data facedifferent data problems from either the na-tional or the area studies. They have the ad-vantage of being able to look for a direct ef-fect of income inequality without having toallow for the effects of inequality that workthrough aggregation. But there are compen-sating difficulties. Because mortality is a rareevent, large sample sizes (or long follow-ups) are required to give enough deaths toestimate mortality rates reliably. At the sametime, those few health-related surveys thatfollow people from interview to death aretypically very poorly endowed with eco-nomic information, including incomes.Nevertheless, there are several surveys inthe United States that have been used tolook at the determinants of mortality at theindividual level. The National Longitudinal


Article 3 2/21/03 4:20 PM Page 146

Mortality Study starts from data collected inthe Current Population Survey, mostlyaround 1980, and then uses the NationalDeath Index to check whether members ofresponding households are dead by eachfollow-up date. Currently, around 1.3 mil-lion people have been tracked for up to adecade. In principle, the CPS provides ex-cellent and detailed economic information,but many of the rounds used for the NLMSwere not the March surveys, when incomedata are collected, and so contain only rudi-mentary information on household incomes.

There are two other U.S. health surveyswith later merges of death-certificate data:the National Health Interview Survey(NHIS), which interviews around 50,000households every year, and the NationalHealth and Nutritional Examination Survey(NHANES) the first round of which sur-veyed more than 14,000 people between1971 and 1975, for whom information ondeaths has been merged up to 1987. A finalsource of mortality data is the Panel Study ofIncome Dynamics (PSID), which has fol-lowed around 5,000 households (and theirchildren and split-offs) since 1968. Becausethis is a panel survey returning regularly toeach household, deaths are reported by sur-viving family members. All four of these sur-veys have been used to look at the relation-ship between mortality, income, and incomeinequality.

Sweden also has nationally representativedata that works in the same way as theNLMS in the United States, though witheven more comprehensive data. Since 1975,Statistics Sweden has interviewed around7,000 individuals each year in its Survey ofLiving Conditions, and these people havebeen linked, not only to the national deathstatistics, but to income information fromthe national income tax statistics; see UlfGerdtham and Magnus Johannesson (2001).

A second line of work has used, not mor-tality, but self-reported measures of overallhealth status. The questions are included ina large number of surveys, if only because

they are asked easily and quickly. They askrespondents to rate their health on a five-point scale from “poor,” to “fair,” to “good”to “very good” and “excellent.” Many investi-gators convert this to a binary indicator ofpoor health, corresponding to the “poor”and “fair” categories; such an indicator hasbeen validated as a powerful predictor ofsubsequent morbidity and mortality, evenconditional on a physician’s examination;(Ellen Idler and Yeal Benyamini 1997).Even so, there is no automatic link betweenmorbidity and mortality; indeed there arereasons why they might move in opposite di-rections; see Christopher Murray (1996) fora good discussion. Self-reported generalhealth questions are included in the NHIS,in the Behavioral and Risk Factor Sur-veillance Study (BRFSS) and, from 1995 on-wards, in the Current Population Survey.

The interpretation of the individual stud-ies, and of the extent to which they support alink from inequality to health depends agreat deal on who is doing the interpreting.Nevertheless, there is general agreementthat the results from these studies areweaker and more ambiguous than the areastudies. For example, Kimberley Lochner etal. (2001), using the NHIS and merged mor-tality data, find only a small effect of state in-come inequality on mortality (relative risk ofliving in the top five most unequal statescompared with the ten most equal states of1.12). This effect, which is estimated withcontrols for family income and the state-level poverty rate, is only statistically signifi-cant for near-poor whites. Kevin Fiscellaand Peter Franks (1997) find that, once theycontrol for individual income, there is no ef-fect of PSU-level inequality on the probabil-ity of dying in the NHANES follow-up, butthere are real questions about whether theirmeasure of inequality—the share of incomeaccruing to the bottom 50 percent of thepopulation—can be adequately measuredfrom the PSU data in the NHANES itself,within which respondents are not randomlydrawn. Mary Daly et al. (1998) find no effect


Article 3 2/21/03 4:20 PM Page 147

of state-level income inequality on individualfive-year mortality rates using data from thePSID. As I shall explain in more detail be-low, state-level income inequality also has nodetectable effect on individual mortality inthe NLMS.

Several studies find a range of results,with at least some studies finding an effectof income inequality on self-reported mor-bidity. See Mah-Jabeen Soobader andFelicia LeClere (1999, using the NHIS withmerged census county-level inequality data),Kennedy et al. (1998, using the BRFSSmerged with state-level inequality data),Fiscella and Franks (2000, using the NHIS),Mellor and Milyo (2002) and Blakely,Lochner, and Kawachi (2002) using the CPSwith inequality data from the CPS at variousgeographical levels; and Roland Sturm andCarole Roan Gresenz (2002) and RobertKahn et al. (2000) using other health sur-veys. The estimated effects are typicallymodest, and in several cases are eliminatedwhen controls are added for individual in-come. For example, Mellor and Milyo showthat their effects are removed once controlsare introduced for income and its square, aswell as for fixed state effects. (Note againthat this last is a severe test; Mellor andMilyo only have three years of CPS data, sothey are effectively demanding a link be-tween changes in morbidity and changes inincome distribution between 1995 and1997). But LeClere and Soobader (2001)also demonstrate considerable fragility inthe results, showing that the effects seem towork only for whites aged 18–44 in high-inequality counties, and middle-aged whitesin very-high-inequality counties. There areno effects for other whites, nor for non-Hispanic blacks. Blakely, Lochner, andKawachi (2002) is the most recent and oneof the most careful studies; it concludes that,after controlling for income, and apart fromthe possible effects of income inequality thatwork through aggregation, there is little as-sociation between income inequality and be-ing in poor health in the CPS data from 1996

and 1998. This conclusion is echoed forJapan, where the raw correlation betweenpoor health and the prefecture gini coeffi-cient is eliminated once individual income iscontrolled for (Kenji Shibuya, HidekiHashimoto, and Eiji Yano 2002).

As noted, Wolfson et al. (1999) showthat the degree of curvature in the NLMS(figure 2) is insufficient to explain the largeeffects of income inequality on mortality atthe state level. In Deaton (2001b), I ad-dress more directly the role of state in-come inequality on mortality in the NLMS.The NLMS distinguishes seven incomegroups, so that at the first stage of theanalysis, I use a logit model to estimate thelog odds of dying during the ten-year fol-low up as a linear function of age includingdummy variables for each of the seven in-come groups. These logits are estimatedfor white males and females separately, us-ing data only for those aged 18 to 75 at thetime of first interview. (The log odds ofmortality is approximately linear in ageover this range.) In order to conduct astate-level analysis, each of these models isfitted to data for a single state, thus allow-ing inequality—or any other state-level ef-fect—an unrestricted effect on the rela-tionship between mortality and income.The first stage produces numbers for eachstate like the points shown as circles in fig-ure 1, so that, at the second stage, it is pos-sible to examine whether these points arehigher in states where income inequality ishigher. Note that this two-stage procedureis as general as a single-stage model inwhich individual mortality is linked tostate-level data on income inequality.

My original concern was to test the modelof relative deprivation presented in section2.2.5 above. This was done by comparing theeffects of each income group in each statewith the predicted values from computingrelative deprivation. While states are by nostretch of the imagination plausible refer-ence groups, it is at the state level that theinequality effects were first discovered, so it


Article 3 2/21/03 4:20 PM Page 148

is tempting to try to account for the effectusing the relative deprivation story. Withinstates, the relative deprivation story doeswell, outperforming a simple model inwhich income itself accounts for the differ-ences across income groups; Eibner (2000)finds similar results using a range of defini-tions for reference groups. However, the rel-ative deprivation model accounts for essen-tially none of the variation in mortalityacross states which, given the theory, meansthat the gini coefficient does not predict in-terstate mortality differences in the NLMSdata. This finding is supported (even with-out controls for income) using the 1.3 mil-lion observations in the full NLMS. Table 13in Rogot et al. (1992) shows no correlationacross the states between age-adjusted mor-tality and income inequality, a finding that isin direct contradiction with figure 6 and withthe findings listed at the outset of section 3.4above. This contradiction is resolved (atleast in part) by the demonstration that, inthese individual-level data as in the aggre-gate state-level data, the fraction black ineach state is a powerful predictor of whitemortality.

Once again, there is no direct effect of in-come inequality. Because the racial compo-sition of areas is such a strong predictor ofmortality in the aggregate state, MSA, andcounty-level data, as well as in the individualdata from the NLMS, it would be interestingto discover whether the Lochner et al.(2001) findings on mortality in the NHISfollow-up can also be attributed to racialcomposition; given their partition of statesinto inequality groups, it seems likely.

Taking income and mortality together, theSwedish data used by Gerdtham andJohanesson (2001) are probably of higherquality than anything currently availablein the United States. Gerdtham andJohanesson used the 284 municipalities ofSweden as their communities, and examineindividual mortality for 41,006 individualsaged between 20 and 84 who were inter-viewed between 1980 and 1986 and whose

mortality was followed until the end of 1996.Mortality was assessed relative to individualincome, community income, and commu-nity income inequality, with the latter twomeasured from the survey data itself, a pro-cedure subject to the reservations raisedabove. As in all similar studies, individual in-come was strongly protective, even allowingfor education and a host of other variables,including initial health status, but neither in-equality nor mean community income ap-peared to have any effect. The last result isevidence against the relative income hypoth-esis so that, once again, we are led back tothe original model in which health is an in-creasing nonlinear function of absolute in-come. Finally, in another impressive Nordicstudy, Merete Osler et al. (2002) use data onnearly 26,000 people from Copenhagen whowere followed for an average of 12.8 years.These survey data, which include data onbehavioral risk patterns, were merged withadministrative data on housing, income, ed-ucation, and occupation, as well as incomeequality (share accruing to the bottom 50percent) at the parish level, calculated frompopulation, not survey, data. Without con-trols, income inequality is positively relatedto mortality, but the effect is eliminated byadding the other variables; indeed, addingindividual income is sufficient to eliminatethe effect of inequality on mortality.

3.5 Inequality and Mortality Decline in the United States and Britain

A final source of evidence comes from ex-amining whether there is a link betweenmortality and the increase in income in-equality in the 1980s in both Britain and theUnited States. Wilkinson (1996) argues thatfor Britain, mortality rates for infants and foryoung adults fell less rapidly after 1985 thanwould have been the case had income in-equality remained constant. Wilkinson(1996, figure 5.10, p. 97) plots a time seriesof mortality, not only of infants, but also ofchildren and young adults, and shows thatthe sum of age-adjusted mortality rates fell


Article 3 2/21/03 4:20 PM Page 149

less rapidly after 1985 than it did in thedecade from 1975 to 1985. These findings,together with the corresponding evidencefor the United States were recently exam-ined in Deaton and Paxson (2001b). Theirresults are as follows.

There were large increases in income in-equality in both Britain and the UnitedStates in the 1970s and 1980s. In both coun-tries, inequality in family and household in-come increased from the early or mid-1970suntil around 1990, with (arguably) little in-crease but certainly no decline since. By theearly to mid-1980s, inequality had risen tonew postwar highs and continued to in-crease, at least until 1990. As pointed out byWilkinson, the rate of decline of infant mor-tality was particularly rapid in the decadefrom 1975 to 1985, and less rapid thereafter.The same is true in the United States,though the period of rapid decline startssomewhat earlier, in the late 1960s, and fin-ishes earlier, around 1980. In both countries,the rate of decline of mortality rates amongyoung adults has slowed steadily, and by1985 mortality rates are either flat or actu-ally rising. For infants and young adultstaken together, the rate of mortality declinehas therefore been a good deal slower inrecent years than in the period before theincrease in income inequality.

Even so, it is unlikely that income in-equality has much to do with these mortalitytrends. First, the episodes of rapid decline inthe infant mortality rate are episodes, nottrends. Prior to the periods of rapid declinein each country, progress was slower, with arate of decline comparable to that after theend of the episode. Yet income inequalitywas not high prior to the onset of the rapiddecline, so that in neither country since1950 has there been a consistent relation-ship between income inequality and the rateof decline of infant mortality. Second, weknow a good deal about the causes of the de-cline in infant mortality, much of which canbe attributed to declines in perinatal mortal-ity through new techniques for preventing

the deaths of low birth-weight babies. Thesetechniques diffused more quickly in theUnited States than in the United Kingdom,so that the rapid decline in mortality startedfirst in the United States, and its possibilitieswere more rapidly exhausted there. Thereseems no reason other than coincidence tolink the timing of this exhaustion to the risein income inequality. Third, among youngadults, much of the increase in mortality isattributable to HIV/AIDS, for which the risein income inequality in the mid-1980s is notthe cause. Finally, if we look at mortalityrates of adults aged 45 and above, there is aperiod of unusually rapid mortality decline(particularly although not exclusively formen) that began around 1970 (again a littleearlier in the United States), and continuesto the present. So if income inequality ishazardous for the young, it is protective fortheir elders! Once again, a more convincingexplanation lies in the increased use, first inthe United States and later in Britain, of life-saving technologies for dealing with cardio-vascular disease, angioplasty, coronary arterybypass grafts, and the use of clot-bustingdrugs and even aspirin.

The time-series results also cast doubt onthe role of income in promoting health. Formost age groups, the rate of mortality declinewas slow during the period of rapid growth inmedian incomes from 1950 until the early1970s, and much more rapid thereafter dur-ing the period of the productivity slowdown.In Britain, there is no slowdown in the rate ofgrowth of real disposable personal incomesafter the mid-1970s, so that the pattern of in-come growth was very different from that inthe United States. Yet the patterns of mortal-ity decline were very similar. Neither incomenor income inequality help us explain Britishand American mortality patterns in the twoquarter centuries after 1950.

4. Summary and Conclusions

The stories about income inequality af-fecting health are stronger than the evidence.


Article 3 2/21/03 4:20 PM Page 150

Judging by the explosion of interest and ofcitations, there is a strong appeal to the ideathat before the epidemiological transition,income determines mortality, while after it,income inequality determines mortality; thatin poor countries, income protects againstpoor sanitation, unhealthy working and liv-ing environments, poor nutrition, and aplethora of infectious diseases; that in richcountries, where these evils are but distantmemories, income inequality is an indicatorof the quality of social arrangements, ofstress, and of mortality. Yet, as we have seen,even if it is true that, at higher income levels,income inequality becomes relatively moreimportant as a cause of death, there is noneed to assume that the relationship be-tween income and mortality changes witheconomic development. If it is poverty, notinequality, that drives mortality, so that in-come has a much bigger effect on health atlow than high incomes, average income willeventually cease to be associated with poorhealth, while the effects of inequality willendure for much longer because, even inrich economies, there are some who are notso rich. Income inequality will continue toaffect mortality until everyone ceases to bepoor, which happens long after average in-come has risen out of the range of poverty.

But it is not true that income inequality it-self is a major determinant of populationhealth. There is no robust correlation be-tween life expectancy and income inequalityamong the rich countries, and the correla-tion across the states and cities of the UnitedStates is almost certainly the result of some-thing that is correlated with income inequal-ity, but that is not income inequality itself.The rapid increases in income inequality inthe 1980s have not been associated with anyslowdown in the rate of mortality decline.Studies of individual mortality and incomeinequality show no link, except for one sur-vey where the estimated effects are smalland are confined to one population group.Infant and child mortality in developingcountries is primarily a consequence of

poverty so that, conditional on average in-come, income inequality is important onlybecause it is effectively a measure ofpoverty. But it is low incomes that are im-portant, not inequality, and there is no evi-dence that making the rich richer, howeverundesirable that may be on other grounds, ishazardous to the health of the poor or theirchildren, provided that their own incomesare maintained. The only exception to thesegeneralizations is perhaps the case of homi-cide, where income inequality itself appearsto play a genuine role.

These conclusions are not different fromthose of earlier commentators, particularlyJudge (1995); Judge, Mulligan, andBenzeval (1997); Wagstaff and Eddy vanDoorslaer (1999); and Judge and IainPatterson (2001). Yet these results must notbe misinterpreted. While they do imply thatincome inequality is not important per se,other than its effects through poverty, theydo not imply that the social environment isnot important for individual health, let alonethat individual health is determined by indi-vidual characteristics and the provision ofpersonal medical care. We know fromWhitehall and from other studies that posi-tions in hierarchies matter, perhaps throughan ability to control one’s life, but in any casethrough some mechanism that worksthrough relationships with other human be-ings. In the United States in 1999, blackmales had a life expectancy at birth 6.8 yearsshorter than white males; for females, thediscrepancy was 5.2 years (CDC 2002).There is a twenty-year gap in life expectancybetween white men in the healthiest coun-ties and black men in the unhealthiest coun-ties (Murray 1998 quoted by Marmot 2001).Hispanic Americans have longer life-expectancies than whites. And there are dif-ferences within racial and ethnic groups byeducation and income. As I have arguedabove, across cities, states, and counties ofthe United States, places that have a largerAfrican-American population are unhealth-ier, with both whites and African Americans


Article 3 2/21/03 4:20 PM Page 151

dying younger than would otherwise be thecase. We do not know why this should be thecase, whether the nature of race relations inAmerica is so difficult that it is actually ahealth hazard, or whether what we see is theresult of lower-quality health care in placeswhere there are large black populations. Ifthe latter, residential segregation is likely toimprove the health of whites at the expenseof blacks, and there is some evidence in theliterature that, indeed, segregation harmsthe health of African Americans (ThomasLaveist 1993 and Ingrid Gould Ellen 2000).All of these cases involve inequality of onekind or another, residential or racial, but notincome inequality per se.

I have also emphasized several other caseswhere reductions in deprivation in one di-mension—whether it be land ownership (thenutritional wage story), democratic rights(England in the 1870s or the AmericanSouth in the 1960s), women’s agency (healthand fertility in India), or income (throughmalnutrition or freedom from stress)—willbring benefits not only in and of themselves,but also to the relief from other deprivations,in this case particularly the deprivation of illhealth. This is of course Sen’s (1999) themein Development and Freedom, that relieffrom any one of a number of interlinked dep-rivations, each of which is an important “un-freedom” in its own right, helps promote re-lief from the others. This is quite differentfrom a story in which income inequality isthe principal actor and main villain.

These results are entirely consistent withwhat is known of about the biology of stressand disease. That rank can be protective ofhealth, including enhancing the resistance toinfection, has been demonstrated in bothanimal and human experiments (SheldonCohen 1997, 1999). That repeated stress as-sociated with the insults and lack of controlthat come from low rank has a well-devel-oped biochemical basis (see for exampleRobert Sapolsky 1993, and Bruce McEwen1998). Yet there is nothing that necessarilyimplicates income, let alone income in-

equality, as the main correlate of the degreeto which people experience disease-induc-ing insults. Some of the institutions whererank is likely most protective—prisons or themilitary—are not characterized by wide in-come inequality, nor even by important dif-ferences in income. Inequality may be im-portant, but there is little that suggests it isincome inequality.

My conclusions carry a number of implica-tions for the direction of future research.The most obvious is that attention should bedirected away from income inequality per se.Instead, the urgent need is to investigate therole that income plays in promoting health—whether the effects of income come from in-come itself, or from correlates such as educa-tion, wealth, control, or rank. We need toknow why income is so important in the indi-vidual level studies, and so apparently unim-portant at the aggregate level over space orover time. If income is indeed directly pro-tective, we need to know whether the effectis really nonlinear and by how much, becauseit is this, and not any direct effect of incomeinequality on health, that determineswhether and by how much income redistri-bution can improve population health.

REFERENCES

Adams, Peter; Michael D. Hurd, Daniel McFadden,Angela Merrill and Tiago Ribeiro. 2003. “Healthy,Wealthy, and Wise?” J. Econometrics, 112, pp. 3–56.

Aghion, Philippe; Eve Caroli and Cecilia García-Peñalosa. 1999. “Inequality and Economic Growth:The Perspective of the New Growth Theories,” J.Econ. Lit. 37:4, pp. 1615–60.

Alesina, Alberto; Reza Baqir and William Easterly.1999. “Public Goods and Ethnic Divisions,” Quart.J. Econ. 114, pp. 1243–84.

Almond, Douglas V.; Kenneth Y. Chay and MichaelGreenstone. 2001. “Civil Rights, the War on Poverty,and Black-White Convergence in Infant Mortality inMississippi,” econ. dept., U. Cal. Berkeley.

Atkinson, Anthony B. 1970. “On the Measurement ofInequality,” J. Econ. Theory 2, pp. 244–63.

Atkinson, Anthony B. and Andrea Brandolini. 2001.“Promise and Pitfalls in the Use of ‘Secondary’ DataSets: Income Inequality in OECD Countries,” J.Econ. Lit. 39, pp. 771–99.

Auster, Richard; Irving Leveson and DeborahSarachek. 1969. “The Production of Health: An Ex-ploratory Study,” J. Human Res. 4, pp. 411–36.


Article 3 2/21/03 4:20 PM Page 152

Backlund, E.; Paul D. Sorlie and Norman J. Johnson.1999. “A Comparison of the Relationship of Educa-tion and Income with Mortality: the National Longi-tudinal Mortality Study,” Soc. Sci. Med.. 49, pp.1373–84.

Bardhan, Pranab; Samuel Bowles and Herbert Gintis.2000. “Wealth Inequality, Wealth Constraints, andEconomic Performance,” in Handbook of IncomeDistribution. Anthony B. Atkinson and FrançoisBourguignon, eds. Vol. 1, Amsterdam: North-Hol-land, pp. 541–603.

Barker, David J. P. 1994. Mothers, Babies, and Diseasein Later Life. London: BMJ Publishing.

———. 1995. “Fetal Origins of Coronary Heart Dis-ease,” Brit. Med. J. 311, pp. 117–4.

Baumbusch, Andrew P. 1995. “Income Inequality andInfant Mortality,” Princeton U., junior independentwork.

Benabou, Roland. 1996. “Inequality and Growth,” inNBER Macroeconomics Annual. Ben Bernanke andJulio Rotemberg, eds. Cambridge: MIT Press.

Ben-Shlomo, Yoav; Ian R. White and Michael Marmot.1996. “Does the Variation in Socio-Economic Char-acteristics of an Area Affect Mortality?” Brit. Med. J.312, pp. 1013–14.

Benzeval, Michaela and Ken Judge. 2001. “Income andHealth: The Time Dimension,” Soc. Sci. Med. 52,pp. 1371–90.

Benzeval, Michaela; Ken Judge and Sue Shouls. 2001.“Understanding the Relationship Between Incomeand Health: How Much Can Be Gleaned fromCross-Sectional Data?” Soc. Policy Admin. 35, pp.376–96.

Binswanger, Hans and Mark Rosenzweig. 1984. Con-tractual Arrangements, Employment, and Wages inRural South Asia. New Haven: Yale U. Press.

Bliss, Christopher J. and Nicholas H. Stern. 1978.“Productivity, Wages, and Nutrition,” J. Devel. Econ.5, pp. 331–98.

Blakely, Tony A.; Kimberley Lochner and IchiroKawachi. 2002. “Metropolitan Area Income In-equality and Self-Rated Health—A MultilevelStudy,” Soc. Sci. Med. 54, pp. 65–77.

Bowles, Samuel and Herbert Gintis. 1998. “Is EqualityPassé?” Boston Rev. 23, pp. 4–35.

Brodish, Paul; Mark Massing and Herman Tyroler.2000. “Income Inequality and All-Cause Mortality inthe 100 Counties of North Carolina,” South. Med. J.93, pp. 386–91.

Brunner, Eric and Michael Marmot. 1999. “Social Or-ganization, Stress, and Health,” in Social Determi-nants of Health. Michael Marmot and Richard G.Wilkinson, eds. Oxford: Oxford U. Press.

Card, David and Alan B. Krueger. 1993. “Trends inRelative Black-White Earnings Revisited,” Amer.Econ. Rev., Papers and Proceedings 83, pp.85–91.

Case, Anne. 2001. “Does Money Protect Health Sta-tus? Evidence from South African Pensions,” NBERwork. paper 8496, Cambridge, MA.

———. 2002. “Health, Income, and Economic Devel-opment,” Proceedings of the World Bank Conferenceon Development Economics, 2001/2002, WorldBank, pp. 221–41.

Center for Disease Control. 2002. National Vital Sta-tistics Report, 50:6, March.

Chandra, Amitabh and Jonathan Skinner. 2002. “Geog-raphy and Racial Health Disparities,” econ. dept.,Dartmouth College, processed.

Chattopadhyay, Raghabendra and Esther Duflo. 2001.“Women’s Leadership and Policy Decisions: Evi-dence from a Nationwide Randomized Experimentin India,” I.I.M. Calcutta and econ. dept. MIT.

Chiang, Tung-liang. 1999. “Economic Transition andChanging Relation between Income Inequality andMortality in Taiwan: Regression Analysis,” Brit.Med. J. 319, pp. 1162–65.

Cohen, Sheldon; S. Line, S. B. Manuck, B. S. Rabin,E. R. Heise and J. R. Kaplan. “Chronic Social Stress,Social Status, and Susceptibility to Upper Respira-tory Infections in Nonhuman Primates,” Psychosom.Med. 59:3, pp. 213–21.

Cohen, Sheldon. 1999. “Social Status and Upper Res-piratory Infection,” in Socioeconomic Status andHealth in Industrialized Nations: Social, Psychologi-cal, and Biological Pathways. Nancy Adler et al.,eds. Annals NY Academy Science 896, pp. 246–53.

Daly, Mary C.; Greg J. Duncan, George A. Kaplan andJohn W. Lynch. 1998. “Macro to Micro Links in theRelation Between Income Inequality and Mortality,”Milbank Quart. 76, pp. 315–39.

Dasgupta, Partha. 1993. An Inquiry into Well-beingand Destitution. Oxford: Clarendon Press.

Dasgupta, Partha and Debraj Ray. 1986. “Inequality asa Determinant of Malnutrition and Unemployment:theory,” Econ. J. 96, pp. 1011–34.

———. 1987. “Inequality as a Determinant of Malnutritionand Unemployment: Policy,” Econ. J. 97, pp. 177–88.

Deaton, Angus. 2001a. “Inequalities in Income and In-equalities in Health,” in The Causes and Conse-quences of Increasing Inequality. Finis Welch, ed.Chicago: U. Chicago Press, pp. 285–313.

———. 2001b. “Relative Deprivation, Inequality, andMortality,” Princeton U. Research Program Devel-opment Studies and Center for Health and Wellbe-ing (NBER work. paper 8099).

———. 2002. “Policy Implications of the Gradient ofHealth and Wealth,” Health Affairs 21, pp. 13–30.

Deaton, Angus and Christina Paxson. 2001a. “Mortal-ity, Education, Income, and Inequality amongAmerican Cohorts,” in Themes in the Economics ofAging. David Wise, ed. Chicago U. Press for NBER(NBER work. paper 7140)

———. 2001b. “Mortality, Income, and Income In-equality among British and American Cohorts,”Center for Health and Wellbeing, Princeton U.,processed.

Deaton, Angus and Darren Lubotsky. 2003. “Mortality,Inequality and Race in American Cities and States,”Soc. Sci. Med., forthcoming.

Deininger, Klaus and Lyn Squire. 1996. “A New DataSet Measuring Income Inequality,” World BankEcon. Rev. 10, pp. 565–91.

Doblhammer, Gabriele. 2002. “Differences in Life-span by Month of Birth for the United States: TheImpact of Early Life Events and Conditions on LateLife Mortality,” Max Planck Institute for Demo-graphic Research, work. paper 2002–19.


Article 3 2/21/03 4:20 PM Page 153

Doblhammer, Gabriele and James W. Vaupel. 2001.“Lifespan Depends on Month of Birth,” ProceedingsNat. Academy Sci. 98, pp. 2934–9.

Donohue, John J. and James J. Heckman. 1991. “Con-tinuous Versus Episodic Change: The Impact of Af-firmative Action and Civil Rights Policy on the Eco-nomic Status of Blacks,” J. Econ. Lit. 29, pp.1603–44.

Drèze, Jean and Reetika Khera. 2000. “Crime, Gender,and Society in India: Insights from Homicide Data,”Population Devel. Rev. 26, pp. 335–52.

Drèze, Jean and Mamta Murthi. 2001. “Fertility, Edu-cation, and Development: Evidence from India,”Population Devel. Rev. 27, pp. 33–63.

Easterlin, Richard A. 1974. “Does Economic GrowthImprove the Human Lot? Some Empirical Evi-dence,” in Nations and Households in EconomicGrowth. P. A. David and W. R. Levin, eds. Stanford:Stanford U. Press, pp. 98–125.

———. 1996. Growth Triumphant: the Twenty-firstCentury in Historical Perspective. Ann Arbor: U.Michigan Press.

Ehrlich, Isaac. 1973. “Participation in Illegitimate Ac-tivities: A Theoretical and Empirical Investigation,”J. Polit. Econ. 81, pp. 521–65.

Eibner, Christine. 2001. “Relative Deprivation, PoorHealth Habits, and Mortality,” Princeton, Center forHealth and Wellbeing work. Paper, May (http://www.wws.princeton.edu/~chw/papers.html)

Ellen, Ingrid Gould. 2000. “Is Segregation Bad forYour Health? The Case of Low Birth Weight,”Brookings-Wharton Pap. Urban Affairs pp. 203–38.

Elo, Irma and Samuel H. Preston. 1996. “EducationalDifferentials in Mortality: United States, 1979–85,”Soc. Sci. Med.42, pp. 47–57.

Erdal, David, and Andrew Whiten. 1996. “Egalitarian-ism and Machiavellian Intelligence in Human Evo-lution,” in Modelling the Early Human Mind. PaulMellars and Kathleen Gibson, eds. McDonald Insti-tute Monographs, pp. 139–50.

Fajnzylber, Pablo; Daniel Lederman and NormanLoayza. 2000. “Crime and Victimization: An Eco-nomic Perspective,” Economía 1, pp. 219–78.

Farmer, Paul. 1999. Infections and Inequalities: themodern plagues. Berkeley: U. California Press.

Fehr, Ernst, and Klaus M. Schmidt. 1999. “A Theory ofFairness, Competition, and Cooperation,” Quart. J.Econ. 114, pp. 817–68.

Feinstein, Jonathan S. 1993. “The Relationship Be-tween Socioeconomic Status and Health: a review ofthe literature,” Milbank Quart.71, pp. 279–322

Filmer, Deon, and Lant Pritchett. 1999a. “The Impactof Public Spending on health: does money matter?”Soc. Sci. Med. 49, pp. 1309–23.

———. 1999b. “The Effect of Household Wealth onEducational Attainment: evidence from 35 coun-tries,” Population Devel. Rev. 25, pp. 85–120.

Fiscella, Kevin, and Peter Franks. 1997. “Poverty orIncome Inequality as a Predictor of Mortality: lon-gitudinal cohort study,” Brit. Med. J. 314, pp.1724–28.

———. 2000. “Individual Income, Income Inequality,Health and Mortality: what are the relationships?”Health Services Res. 35, pp. 307–18.

Flegg, A.T. 1982. “Inequality of Income, Illiteracy andMedical Care as Determinants of Infant Mortality inDeveloping Countries,” Population Stud. 36, pp.441–58.

Fogel, Robert W. 1994 “Economic Growth, PopulationTheory and Physiology: the bearing of long-termprocesses on the making of economic policy,” Amer.Econ. Rev. 84, pp. 369–95.

———. 1997. “New Findings on Secular Trends inNutrition and Mortality: some implications for pop-ulation theory,” in Handbook of Population andFamily Economics. Mark Rosenzweig and OdedStark, eds. Vol 1A, Amsterdam: Elsevier.

Foster, James and Amartya K. Sen. 1997. “On Eco-nomic Inequality After a Quarter of a Century,” an-nex to Sen, Amartya K., On Economic Inequality.Oxford: Clarendon Press.

Fuchs, Victor. 1993. “Poverty and Health: asking theRight Questions,” in Medical Care and the Health ofthe Poor. David E. Rogers and Eli Ginzberg, eds.Boulder: Westview Press.

Fuchs, Victor; Mark McClellan and Jonathan Skinner.2001. “Area Differences in Utilization of MedicalCare and Mortality among U.S. Elderly,” NBERwork. paper 8628, Cambridge, MA.

Franzini, Luiza; John Ribble and William Spears. 2001.“The Effects of Income Inequality and IncomeLevel on Mortality Vary by Population Size in TexasCounties,” J. Health Soc. Behav. 42, pp. 373–87.

Gaarder, Marie M. 2001. “Can Population Characteris-tics Account for the Variation in Health Impacts ofAir Pollution? A Meta Analysis of PM10–MortalityStudies,” University College, London, processed.

Garber, Alan M. 1989. “Pursuing the Links betweenSocioeconomic Status and Health: critique, policyimplications, and directions for future research,” inPathways to Health: The Role of Social Factors. JohnP. Bunker, Deanna S. Gombey and Barbara Kehrer,eds. Menlo Park, CA. Kaiser Family Foundation, pp.271–315.

Galor, Oded and Omer Moav. 2000. “From Physical toHuman Capital Accumulation: inequality in theprocess of development,” econ. dept. Brown U. andMIT, processed.

Gerdtham, Ulf-G. and Magnus Johannesson. 2001.“Absolute Income, Relative Income, Income In-equality and Mortality,” Stockholm School Econ.,processed.

Gertler, Paul and Jonathan Gruber. 2002. “InsuringConsumption against Illness,” Amer. Econ. Rev. 92,pp. 51–70.

Gersovitz, Mark. 1983. “Savings and Nutrition at LowIncomes,” J. Polit. Econ. 91, pp. 841–55.

Goldin, Claudia and Lawrence F. Katz. 2001. “De-creasing (and Then Increasing) Inequality in Amer-ica: A Tale of Two Half-Centuries,” in The Causesand Consequences of Increasing Inequality FinisWelch, ed. Chicago: U. Chicago Press, pp. 37–82.

Gottshcalk, Peter and Timothy Smeeding. 2000. “Em-pirical Evidence on Income Inequality in IndustrialCountries,” in Handbook of Income Distribution.Anthony B. Atkinson and François Bourguignon,eds. Vol. 1, Amsterdam: North-Holland, pp.261–307.


Article 3 2/21/03 4:20 PM Page 154

Gravelle, Hugh. 1998. “How Much of the Relation be-tween Population Mortality and Unequal Distribu-tion of Income Is a Statistical Artefact?” Brit. Med. J.316, pp. 382–85.

Gravelle, Hugh; John Wildman and Matthew Sutton.2002. “Income, Income Inequality and Health:What Can We Learn from the Aggregate Data?”Soc. Sci. Med. 54, pp. 577–89.

Grossman, Michael. 1972. “On the Concept of HealthCapital and the Demand for Health,” J. Polit. Econ.80, pp. 223–55.

———. 1975. “The Correlation between Health andSchooling,” in Household Production and Consump-tion. Nestor Terleckyj, ed. NY: Columbia U. Pressfor NBER, pp. 147–211.

———. 2000. “The Human Capital Model,” in Hand-book of Health Economics. Anthony J. Culyer andJoseph P. Newhouse, eds. Vol 1A, Amsterdam: Else-vier, pp. 347–408.

Groves, Robert M. and Mick P. Couper. 1998. Nonre-sponse in Household Interview Surveys. NY: Wiley.

Gwatkin, Davidson R. 2000. “Health Inequalities andthe Health of the Poor: What Do We Know? WhatCan We Do?” Bull. World Health Org. 78, pp. 3–18.

Haines, Michael R. and Hallie J. Kintner. 2000. “TheMortality Transition in Germany, 1860–1935,” Hist.Methods 33, pp. 83–104.

Hales, Simon; Philippa Howden-Chapman, ClareSalmond, Alistair Woodward and Johan Macken-bach. 1999. “National Infant Mortality Rates in Re-lation to Gross National Product and Distribution ofIncome,” Lancet 354, pp. 2047.

Hammond, Richard J. 1951. Food. History of the Sec-ond World War, United Kingdom Civil Series, Lon-don, H. M. Stationery Office.

Hsieh, C. C. and M. D. Pugh. 1993. “Poverty, Inequal-ity, and Violent Crime: A Meta-Analysis of RecentAggregate Data Studies,” Criminal Justice Rev. 18,pp. 182–202.

Idler, Ellen L. and Yael Benyamini. 1997. “Self-ratedHealth and Mortality: A Review of Twenty-SevenCommunity Studies,” J. Health Soc. Behav. 38, Pp.21–37.

Jain, Shail B. 1975. Size Distribution of Income: ACompilation of Data. Washington, DC: World Bank.

Jones, Arthur F. and Daniel H. Weinberg. 2000. “TheChanging Shape of the Nation’s Income Distribu-tion,” P60–204, Washington, DC, Bureau of Census.http://www.census.gov/ftp/pub/hhes/www/p60204.html

Judge, Ken. 1995. “Income Distribution and Life Ex-pectancy: A Critical Appraisal,” Brit. Med. J. 311, pp.1282–85.

Judge, Ken and Iain Patterson. 2001. “Poverty, IncomeInequality, and Health,” work. paper No. 01/29,New Zealand Treasury.

Judge, Ken; Jo-Ann Mulligan and Michaela Benzeval.1997. “Income Inequality and Population Health,”Soc. Sci. Med. 46, pp. 567–79.

———. 1998. “Reply to Richard Wilkinson,” Soc. Sci.Med. 47, pp. 983–85.

Kahn, Robert S.; Paul H. Wise, Bruce P. Kennedy andIchiro Kawachi. 2000. “State Income Inequality,Household Income, and Maternal Mental and Phys-

ical Health: Cross Sectional National Survey,” Brit.Med. J. 321, pp. 1311–15.

Kaplan, George A.; Elsie R. Pamuk, John W. Lynch,Richard. D. Cohen and Jennifer L. Balfour. 1996.“Inequality in Income and Mortality in the UnitedStates: Analysis of Mortality and Potential Path-ways,” Brit. Med. J. 312, pp. 999–1003.

Kaplan, George A. and John W. Lynch. 2001. “Editor-ial: Is Economic Policy Health Policy?” Amer. J.Public Health 91, pp. 351.

Kawachi, Ichiro, and Tony A Blakely. 2001. “WhenEconomists and Epidemiologists Disagree,” J.Health. Politics, Policy, Law 26, pp. 533–41.

Kawachi, Ichiro and Bruce P. Kennedy. 1997. “The Re-lationship of Income Inequality to Mortality: Doesthe Choice of Indicator Matter?” Soc. Sci. Med. 45,pp. 1121–27.

Kawachi, Ichiro; Bruce P. Kennedy and Deborah Pro-throw-Stith. 1997. “Social Capital, Income Inequal-ity, and Mortality,” Amer. J. Public Health 87, pp.1491–98.

Kawachi, Ichiro; Bruce P. Kennedy and Richard G.Wilkinson. 1999. The Society and Population HealthReader: Volume 1: Income Inequality and Health.NY: New Press.

Kennedy, Bruce P.; Ichiro Kawachi and Deborah Pro-throw-Stith. 1996a. “Income Distribution and Mor-tality: Cross-Sectional Ecological Study of the RobinHood Index in the United States,” Brit. Med. J. 312,pp. 1004–07.

Kennedy, Bruce P.; Ichiro Kawachi, Roberta Glass andDeborah Prothrow-Stith. 1998. “Income Distribu-tion, Socioeconomic Status and Self-Rated Health inthe United States,” Brit. Med. J. 317, pp. 917–21.

Kennedy, Bruce P.; Ichiro Kawachi and Deborah Pro-throw-Smith. 1996b. “Important Correction,” Brit.Med. J. 312, pp. 1194.

Kitigawa, Evelyn and Philip Hauser. 1973. Differen-tial Mortality in the United States: A Study in So-cioeconomic Epidemiology. Cambridge: HarvardU. Press.

Krieger, Nancy. 2001. “Theories for Social Epidemiol-ogy in the 21st Century: An Ecosocial Perspective,”Int. J. Epidemiology 30, pp. 668–77.

Kunst, Anton; Feikje Groenhof, Johan P Mackenbachand EU Working Group on Socioeconomic Inequal-ities in Health. 1998. “Occupational Class and CauseSpecific Mortality in Middle Aged Men in 11 Euro-pean Countries: Comparison of Population BasedStudies,” Brit. Med. J. 316, pp. 1636–42.

Lantz, Paula; James House, J. Lepowski, DavidWilliams, R. Mero and J. Chen. 1998. “Socioeco-nomic Factors, Health Behaviors, and Mortality,” J.Amer. Med. Assoc. 279, pp. 1703–708.

Laveist, Thomas A. 1993. “Segregation, Poverty, andEmpowerment: Health Consequences for AfricanAmericans,” Milbank Quart. 71, pp. 41–64.

Lynch, John; George Davey-Smith, Marianne Hille-meier, Mary Shaw, Trivellore Raghunathan andGeorge Kaplan. 2001. “Income Inequality, Psy-chosocial Environment and Health: Comparisonsacross Wealthy Nations,” Lancet 358, pp. 194–200.

LeClere, Felicia B. and Mah-Jabeen Soobader. 2000.“The Effect of Income Inequality on the Health of


Article 3 2/21/03 4:20 PM Page 155

Selected US Demographic Groups,” Amer. J. PublicHealth 90, pp. 1892–97.

Leibenstein, Harvey. 1957. Economic Backwardnessand Economic Growth. NY: Wiley.

Lindert, Peter H. 2000. “Three Centuries of Inequalityin Britain and America,” in Handbook of IncomeDistribution. Anthony B. Atkinson and FrançoisBourguignon, eds. Vol. 1. Amsterdam: North-Holland,pp. 167–216.

Lynch, John W.; George A. Kaplan, Elsie R. Pamuk,Richard D. Cohen, Katherine E. Heck, Jennifer L.Balfour and Irene H. Yen. 1998. “Income Inequal-ity and Mortality in Metropolitan Areas of theUnited States,” Amer. J. Public Health 88, pp.1074–80.

Lochner, Kim; Elsie R. Pamuk, Diane Makuc, Bruce P.Kennedy and Ichiro Kawachi. 2001. “State-level In-come Inequality and Individual Mortality Risk: AProspective Multilevel Study,” Amer. J. PublicHealth, 91:3, pp. 385–91.

Macintyre, Sally. 1997. “The Black Report and Beyond:what are the issues?” Soc. Sci. Med. 44, pp. 723–45.

Marmot, Michael G. 1994. “Social Differences inHealth Within and Between Populations,” Daedalus123, pp. 197–216.

———. 2001. “Editorial: inequalities in health,” NewEngland J. Med. 345, pp. 134–6.

———. 2002. “The Influence of Income on Health:Views of an Epidemiologist,” Health Affairs 21, pp.31–46.

Marmot, Michael and Martin Bobak. 2000. “Interna-tional Comparators and Poverty and Health in Eu-rope,” Brit. Med. J. 321, pp. 1124–28.

Marmot, Michael G.; H. Bosma, H. Hemingway, EricBrunner and S. Stansfeld. 1997. “Contribution ofJob Control and Other Risk Factors to Social Varia-tions in Coronory Heart Disease Incidence,” Lancet350:9073, pp. 235–9.

Marmot, Michael G.; Martin Shipley and GeoffreyRose. 1984. “Inequalities in Death—Specific Expla-nations of a General Pattern,” Lancet 1:8384, pp.1003–1006.

Marmot, Michael G.; George Davey-Smith, S. Stans-feld, C. Patel, F. North, J. Head, I White, Eric Brun-ner and A. Feeney. 1991. “Health Inequalitiesamong British Civil Servants: The Whitehall IIstudy,” Lancet 337:8754, pp. 1387–93.

Massey, Douglas M. 1990. “American Apartheid: Seg-regation and the Making of the Underclass,” Amer. J.Sociology 96, pp. 329–57.

McEwen, Bruce S. 1998. “Protective and DamagingEffects of Stress Mediators: Allostatis and Allostaticload,” New England J. Med. 338, pp. 171–79.

McIsaac, Sandra J. and Richard G. Wilkinson. 1997.“Income Distribution and Cause-specific Mortality,”Eur. J. Public Health 7, pp. 45–53.

McKeown, Thomas. 1976. The Modern Rise of Popula-tion. NY: Academic Press.

———. 1979. The Role of Medicine: Dream, Mirage,or Nemesis. Princeton: Princeton U. Press.

McLaughlin, Diane K. and C. Shannon Stokes. 2002.“Income Inequality and Mortality in US Counties:Does Minority Racial Concentration Matter?” Amer.J. Public Health 92, pp. 99–104.

Mealli, Fabrizia and Donald Rubin. 2003. “Commen-tary: Assumptions Allowing the Estimation of DirectCausal Effects,” J. Econometrics, 112, pp. 79–87.

Mellor, Jennifer M. and Jeffrey Milyo. 2001. “Re-examining the Evidence of an Ecological Associationbetween Income Inequality and Health,” J. Health,Politics, Policy, Law 26, pp. 487–522.

———. 2002. “Income Inequality and IndividualHealth: Evidence from the Current Population Sur-vey,” J. Human Res., 37, pp. 510–39.

Miller, Douglas. 2000. “Income Inequality and Mortal-ity in the US: Aggregate Data and Micro Relation-ships,” Princeton U. econ. dept., Ph.D. thesis, ch. 1.

Mirrlees, James. 1975. “A Pure Theory of Underdevel-oped Economies,” in Agriculture in DevelopmentTheory. Lloyd Reynolds, ed. New Haven, CT: YaleU. Press.

Mujica, Oscar J.; Carlos Castillo-Salgado, Jorge Ba-callao, Enrique Loyola-Elizondo, Maria ChristinaSchneider, Manuel Vidaurre and George A. O. Al-leyne. 2000. “Income Inequality and Suicide Risk inthe Americas,” Pan-American Health Org.

Murray, Christopher J.L. 1996. “Epidemiologic andMorbidity Transitions in India,” in Health, Poverty,and Development in India. Monica das Gupta, Lin-coln C. Chen and T. N. Krishnan, eds. Delhi: OxfordU. Press.

Murray, Christopher J.L.; M.C. Michaud, M.McKenna and J. Marks. 1998. “US Patterns of Mor-tality by County and Race, 1965–94,” Cambridge,MA. Harvard Center Population DevelopmentStudies.

Murthi, Mamta; Anne-Catherine Guio and JeanDrèze. 1995. “Mortality, Fertility, and Gender Biasin India: A District Level Analysis,” Population De-vel. Rev. 21, pp. 745–82.

Newhouse, Joseph P. and Lindy Friedlander. 1980.“The Relationship between Medical Resources andMeasures of Health: Some Additional Evidence,” J.Human Res. 15, pp. 200–18.

Nielsen, Jacob Arendt. 2002. “The Relationship Be-tween Health, Education, and Income: EmpiricalEvidence,” U. Copenhagen dept. econ.

Osler, Merete; Eva Prescott, Morten Grønbœk, UllaChristensen, Pernille Due and Gerda Engholm.2002. “Income Inequality, Individual Income, andMortality in Danish Adults: Analysis of Pooled Datafrom Two Cohort Studies,” Brit. Med. J. 324, pp.1–4.

Over, Mead. 1998. “The Effects of Societal Variableson Urban Rates of HIV Infection in DevelopingCountries: An Exploratory Analysis,” in ConfrontingAIDS: Evidence from the Developing World. Euro-pean Commission, ch. 2. http://europa.eu.int/comm/development/aids/limelette/index.htm

Pampel, Fred C. and Vijayan K. Pillai. 1986. “Pat-terns and Determinants of Infant Mortality in De-veloped Nations 1950–75,” Demography 23, pp.525–41.

Paukert, Felix. 1973. “Income Distribution at Differ-ent Levels of Development: A Survey of Evidence,”Int. Lab. Rev. 108, pp. 97–125.

Peña, Rodolfo; Stig Wall and Lars- ke Persson. 2000.“The Effect of Poverty, Social Inequity, and Maternal


Article 3 2/21/03 4:20 PM Page 156

Education on Infant Mortality in Nicaragua,1988–1993,” Amer. J. Public Health 90, pp. 64–69.

Preston, Samuel H. 1975. “The Changing Relation be-tween Mortality and Level of Economic Develop-ment,” Population Stud. 29, pp. 231–48.

———. 1996 “American Longevity: Past, Present, andFuture,” Syracuse U., Maxwell School, Center Pol-icy Research, policy brief 7/1996.

Preston, Samuel H. and Michael R. Haines. 1991. Fa-tal Years: Child Mortality in Late Nineteenth-Century America. Princeton: Princeton U. Press.

Preston, Samuel H. and Paul Taubman. 1994. “Socio-economic Differences in Adult Mortality and HealthStatus,” in Demography of Aging. Linda G. Martinand Samuel H. Preston, eds. Washington, DC: Na-tional Academies Press, pp. 279–318.

Pritchett, Lant and Lawrence H. Summers. 1996.“Wealthier Is Healthier,” J. Human Res. 31, pp.841–68.

Putnam, Robert D., with Robert Leonardi and Raf-faella Nanetti. 1993. Making Democracy Work:Civic Traditions in Modern Italy. Princeton: Prince-ton U. Press.

Ray, Debraj. 1998. Development Economics. Prince-ton: Princeton U. Press.

Regidor, Enrique; Pedro Navarro, Vicente Dominguezand Carmen Rodriguez. 1997. “Inequalities in In-come and Long-term Disability in Spain: Analysis ofRecent Hypotheses Using Cross-Sectional StudyBased on Individual Data,” Brit. Med. J. 319, pp.898–902.

Roberts, Kevin W. S. 1977. “Voting Over Income TaxSchemes,” J. Public Econ. 8, pp. 329–40.

Robson, Arthur. 2002. “Evolution and Human Na-ture,” J. Econ. Perspect. 16, pp. 89–106.

Rodgers, G. B. 1979 “Income and Inequality as Deter-minants of Mortality: An International Cross-sectionAnalysis,” Population Stud. 33, pp. 343–51.

Rogers, Richard G.; Robert A. Hummer and CharlesB. Nam. 2000. Living and Dying in the USA: Behav-ioral, Health, and Social Differentials in Adult Mor-tality. NY: Academic Press.

Rogot, Eugebe; Paul D. Sorlie, Norman J. Johnson andCatherine Schmitt, eds. 1992. A Mortality Study of1.3 Million Persons by Demographic, Social, andEconomic Factors: 1979–1985 Follow-up. Bethesda:NIH.

Ross, Nancy A.; Michael C. Wolfson, James R. Dunn,Jean-Marie Berthelot, George A. Kaplan and JohnW. Lynch. 2000. “Relationship between Income In-equality and Mortality in Canada and in the UnitedStates: Cross Sectional Assessment Using CensusData and Vital Statistics,” Brit. Med. J. 319, pp. 989.

Rosenzweig, Mark. 1988. “Labor Markets in Low In-come Countries,” in Handbook of Development Eco-nomics. Hollis Chenery and T. N. Srinivasan, eds.Vol 1. Amsterdam: North-Holland, pp. 713–62.

Ruhm, Christopher. 2000. “Are Recessions Good forYour Health?” Quart. J. Econ. 115, pp. 617–50.

Sachs, Jeffrey D. 2001. “Tropical Underdevelopment,”Cambridge, MA. NBER work. paper 8119.

Sapolsky, Robert M. 1993. “Endocrinology Alfresco:Psychoendocrine Studies of Wild Baboons,” RecentProgress in Hormone Research 48, pp. 437–68.

———. 1998. Why Don’t Zebras Get Ulcers? An Up-dated Guide to Stress, Stress-Related Diseases, andCoping. NY: Freeman.

Sawyer, Malcolm. 1976. “Income Distribution inOECD Countries,” OECD Economic Outlook, Oc-casional Studies. Paris, OECD.

Schofield, Roger and David Reher. 1991. “The Declineof Mortality in Europe,” in The decline of mortalityin Europe. Roger Schofield, David Reher and AlainBideau, eds. Oxford: Clarendon Press, pp. 1–17.

Sen, Amartya K. 1973. On Economic Inequality. Ox-ford: Clarendon Press.

———. 1999. Development as Freedom. NY: Knopf.Shibuya, Kenji; Hideki Hashimoto and Eiji Yano. 2002.

“Individual Income, Income Distribution, and Self-rated Health in Japan: Cross-Sectional Analysis ofNationally Representative Sample,” Brit. Med. J.324, pp. 1–5.

Smedley, Brian D.; Adrienne Y. Stith and Alan R. Nel-son, eds. 2002. Unequal Treatment: ConfrontingRacial and Ethnic Disparities in Health Care. Insti-tute of Medicine, Washington, DC: National Acad-emy Press.

Smith, James P. 1999. “Healthy Bodies and Thick Wal-lets: The Dual Relation between Health and Eco-nomic Status,” J. Econ. Perspect. 13, pp. 145–66.

Soobader, Mah-Jabeen and Felicia LeClere. 1999.“Aggregation and Measurement of Income Inequal-ity: Effects on Morbidity,” Soc. Sci. Med. 48, pp.733–44.

Stanistreet, Debbie; Alex Scott-Samuel and Mark A.Bellis. 1999. “Income Inequality and Mortality inEngland,” J. Public Health Med. 21, pp. 205–207.

Steckel, Richard. 1995. “Stature and Standard of Liv-ing,” J. Econ. Lit. 33, pp. 1903–40.

Stiglitz, Joseph E. 1976. “The Efficiency Wage Hy-pothesis, Surplus Labor, and the Distribution of In-come in L.D.C.s,” Oxford Econ. Papers 28, pp.185–207.

Strauss, John and Duncan Thomas. 1998. “Health, Nu-trition, and Economic Development,” J. Econ. Lit.36:2, pp. 766–817.

Sturm, Roland and Carole Roan Gresenz. 2002. “Rela-tions of Income Inequality and Family Income toChronic Medical Conditions and Mental HealthDisorders: National Survey,” Brit. Med. J. 324, pp.1–5.

Subramanian, Shankar and Angus Deaton. 1996. “TheDemand for Food and Calories,” J. Polit. Econ. 104,pp. 133–62.

Swamy, Anand V. 1997. “A Simple Test of the Nutri-tion-based Efficiency Wage Model,” J. Devel. Econ.53, pp. 85–98.

Sydenstricker, Edgar and Willford I. King. 1921. “TheMeasurement of the Relative Economic Status ofFamilies,” Quart. Pub. Amer. Statist. Assoc. 17, pp.842–57.

Szreter, Simon. 1988. “The Importance of Social Inter-vention in Britain’s Mortality Decline c. 1850–1914:A Reinterpretation of the Role of Public Health,”Soc. Hist. Med., 1, pp. 1–37.

Victora, Cesar G.; J. Patrick Vaughan, Fernando C.Barros, Anamaria C. Silva and Elaine Tomasi. 2000.“Explaining Trends in Inequities, Evidence from


Article 3 2/21/03 4:20 PM Page 157

Brazilian Child Health Studies,” Lancet 356, pp.1093–98.

Wagstaff, Adam. 2000. “Socioeconomic Inequalities inChild Mortality: Comparisons across Nine Develop-ing Countries,” Bull. World Health Org. 78, pp.19–29.

———. 2002. “Poverty and Health,” Bull. WorldHealth Org., 80, pp. 97–105.

Wagstaff, Adam and Eddy van Doorslaer. 2000. “In-come Inequality and Health: What Does the Litera-ture Tell Us?” Annual Rev. Public Health 21, pp.543–67.

Walberg, Peter; Martin McKee, Vladimir Shkolnikov,Laurent Chenet and David A. Leon. 1998. “Eco-nomic Change, Crime, and Mortality Crisis in Rus-sia: Regional Analysis,” Brit. Med. J. 317, pp.312–8.

Waldmann, Robert J. 1992. “Income Distribution andInfant Mortality,” Quart. J. Econ. 107, pp.1283–302.

Wennemo, Irene. 1993. “Infant Mortality, Public Pol-icy and Inequality—a comparison of 18 industrial-ized countries, 1950–1985,” Sociology Health Illness15, pp. 429–46.

Wilkinson, Richard G. 1989. “Class Mortality Differen-tials, Income Distribution and Trends in Poverty1921–1981,” J. Soc. Policy 18, pp. 307–35.

———. 1992. “Income Distribution and Life-Expectancy,” Brit. Med. J. 304, pp. 165–68.

———. 1994. “The Epidemiological Transition: FromMaterial Scarcity to Social Disadvantage?” Daedalus123, pp. 61–77.

———. 1996. Unhealthy Societies: The Affliction of In-equality. London: Routledge.

———. 1997. “Commentary: Income Inequality Sum-marizes the Health Burden of Individual RelativeDeprivation,” Brit. Med. J. 314, pp. 1727–28.

———. 1998. “Letter to the Editor,” Soc. Sci. Med. 47,pp. 411–12.

———. 2000. Mind the Gap: Hierarchies, Health, andHuman Evolution. London: Weidenfeld and Nicol-son.

Wolfson, Michael C.; George A. Kaplan, John W. Lynch,Nancy A. Ross and Eric Backlund. 1999. “Relation-ship between Income Inequality and Mortality: Em-pirical Demonstration,” Brit. Med. J. 319, pp. 953–55.

Yitzhaki, Shlomo. 1979. “Relative Deprivation and theGini Coefficient,” Quart. J. Econ. 93, pp. 321–4.


Article 3 2/21/03 4:20 PM Page 158

health, inequality, and economic...

Documents