gender identity and relative income within households - faculty

GENDER IDENTITY AND RELATIVE INCOME WITHINHOUSEHOLDS*

Marianne Bertrand

Emir Kamenica

Jessica Pan

We examine causes and consequences of relative income within households.We show that the distribution of the share of income earned by the wifeexhibits a sharp drop to the right of 1

2, where the wife’s income exceeds the hus-band’s income. We argue that this pattern is best explained by gender identitynorms, which induce an aversion to a situation where the wife earns more thanher husband. We present evidence that this aversion also impacts marriageformation, the wife’s labor force participation, the wife’s income conditionalon working, marriage satisfaction, likelihood of divorce, and the division ofhome production. Within marriage markets, when a randomly chosen womanbecomes more likely to earn more than a randomly chosen man, marriage ratesdecline. In couples where the wife’s potential income is likely to exceed thehusband’s, the wife is less likely to be in the labor force and earns less thanher potential if she does work. In couples where the wife earns more thanthe husband, the wife spends more time on household chores; moreover,those couples are less satisfied with their marriage and are more likely to di-vorce. These patterns hold both cross-sectionally and within couples over time.JEL Codes: D10, J12, J16.

I. Introduction

We begin by establishing the following fact: among marriedcouples in the United States, the distribution of the share ofhousehold income earned by the wife drops sharply at 1

2 —wherethe wife starts to earn more than the husband. Standard eco-nomic models of the marriage market cannot account for thispattern. Instead, we argue that gender identity norms play animportant role in marriage.

Akerlof and Kranton (2000, 2010) import ideas about identityfrom sociology and social psychology into economics. They defineidentity as a sense of belonging to a social category, coupled with

*We thank John Abowd for pointing us to the SIPP/SSA/IRS Public Use FileProject. Amanda Chuan and David Toniatti provided excellent research assis-tance. We thank the George J. Stigler Center for the Study of the Economy andthe State and the Initiative on Global Markets, both at the University of ChicagoBooth School of Business, and the Sloan Research Fellowship, for financial sup-port. We thank the editors, anonymous referees, and numerous seminar partici-pants for helpful comments and suggestions.

! The Author(s) 2015. Published by Oxford University Press, on behalf of Presidentand Fellows of Harvard College. All rights reserved. For Permissions, please email:[email protected] Quarterly Journal of Economics (2015), 571–614. doi:10.1093/qje/qjv001.Advance Access publication on January 29, 2015.

571

a view on how people in that category should behave. Theypropose that identity influences economic outcomes because de-viating from the prescribed behavior is inherently costly. In oneapplication of this model, the social categories are man andwoman, and these categories are associated with specific behav-ioral prescriptions, such as ‘‘a man should earn more than hiswife.’’1 Survey responses indeed suggest the prevalence of suchprescriptions. For example, 38 percent of the U.S. respondents tothe 1995 World Value Survey agree with the claim ‘‘If a womanearns more money than her husband, it’s almost certain to causeproblems.’’ Moreover, this attitude persists in the more recentcohorts.2

In this article, we analyze how the behavioral prescriptionthat ‘‘a man should earn more than his wife’’ affects social andeconomic outcomes. First we demonstrate that considerations ofrelative income affect whether people get married. Using aBartik-style instrument, we show that when a randomly chosenwoman within a marriage market becomes more likely to outearna randomly chosen man, the marriage rate declines.3 This resultsuggests a potential link between two important social develop-ments over the past several decades: the increase in women’sincome relative to that of men and the decline in the prevalenceof marriage. Our estimates imply that aversion to having the wifeearn more than the husband explains 29 percent of the decline inmarriage rates over the last thirty years.

We then assess the impact of gender identity on women’s laborsupply. For each married woman in our sample, we estimate the

1. Baker and Jacobsen (2007) attempt to endogenize such social norms. Theypoint out that a rule that mandates the customary division of labor within thehousehold can be Pareto improving if it eliminates wasteful acquisition of humancapital aimed at improving one’s bargaining position.

2. The overall sample includes U.S. respondents, both male and female, to the1995 World Values Survey Question: ‘‘If a woman earns more money than herhusband, it’s almost certain to cause problems. Do you agree strongly, agree, dis-agree, or disagree strongly?’’ We exclude those with missing education (since welater examine how attitudes vary by education) and those who responded ‘‘Don’tknow’’ to the question. Out of the remaining 1,483 respondents, 38 percent ‘‘agree’’or ‘‘agree strongly.’’ Among those born since 1965, 37 percent ‘‘agree’’ or ‘‘agreestrongly.’’

3. We flexibly control for the distribution of men’s income and the distributionof women’s income. Hence, we are not simply picking up the fact that women withhigher incomes are less likely to get married.

QUARTERLY JOURNAL OF ECONOMICS572

distribution of potential earnings based on her demographics. Weshow that if the probability that her income would exceed the hus-band’s income is high, she is less likely to participate in the laborforce. Moreover, if she does work, the gap between her realized andpotential income is greater (in part due to fewer hours of work).These patterns suggest that women reduce their labor supply so asto avoid gender role reversals in earnings. Of course, an importantconcern is that high-skill women who marry low-income men mayhave unobservable characteristics that keep them out of the laborforce. We consider several approaches to deal with this issue. First,we show that the key coefficient is stable as we include variouscontrols for characteristics of the couple. Second, controlling for (aproxy for) relative income at marriage does not affect our esti-mates. Most important, we document the same patterns withincouples over time.

Even though it seems that couples try to avoid the situationwhere the wife earns more than the husband, this situation hasnonetheless become more common. For example, in the AmericanCommunity Survey 2008–2011, the wife earns more than the hus-band in 27 percent of the couples.4 Among these couples, the vio-lation of the gender identity norm seems to influence the quality ofmarriage. Using data from the National Survey of Families andHouseholds, we find that the couples where the wife earns morethan the husband are less happy, report greater strife in theirmarriage, and are ultimately more likely to get a divorce.

We also examine how the violation of the gender identitynorm affects the division of home production. We find that thegender gap in home production—how much more time the wifespends on non-market work than the husband—is larger in cou-ples where she earns more than he does. This suggests that a‘‘threatening’’ wife takes on a greater share of housework so asto assuage the husband’s unease with the situation. The wife, ofcourse, may ultimately get tired of working this ‘‘second shift’’(Hochschild and Machung 1989), which might be one of the mech-anisms behind our results on divorce.

Most of these results rely on cross-sectional variation acrosscouples. To further alleviate concerns about omitted variablebias, we draw on data from the Panel Study of IncomeDynamics (PSID) to examine how changes in relative income

4. Throughout the article, we focus exclusively on couples where both thehusband and the wife are between 18 and 65 years of age.

GENDER IDENTITY AND RELATIVE INCOME 573

within a couple affect the aforementioned outcomes. We replicateall of our results: when the wife’s income exceeds the husband’s,the wife becomes more likely to exit the labor force and she takeson more chores; moreover, there is suggestive evidence that di-vorce becomes more likely.

The fact that traditional gender role attitudes still influencebehavior has important consequences. Over the last half a cen-tury, women have experienced substantial labor market gains;the gender gap in labor force participation and the gender gapin earnings have both declined.5 Despite these gains, substantialgender gaps remain, both in labor force participation and in earn-ings. Female labor force participation appears to have plateauedsince the early to mid-1990s (Blau and Kahn 2006). Among full-time, full-year workers, the gender gap in earnings remains at 25percent. This halted progress has led researchers to consider lesstraditional (within economics, at least) factors that might influ-ence the gender gap in labor market outcomes (Bertrand 2010).The results we present in this article support the view that slow-moving identity norms are an important factor that limits furtherconvergence in labor market outcomes. Women are bringingpersonal glass ceilings from home to the workplace.

Since the initial work by Becker (1973, 1974), the economicanalysis of marriage markets has made great strides by develop-ing tractable models that abstract from issues like tradition andidentity. Consequently, while the economics literature on mar-riage markets is vast, little of it examines the role of gender iden-tity. A few papers (e.g., Fernandez, Fogli, and Olivetti 2004;Fortin 2005, 2009;) have examined how the variation in genderattitudes (across countries, across time, or across couples) corre-lates with women’s labor force participation. In contrast, thisarticle examines the extent to which the overall prevalence oftraditional attitudes affects a wide range of outcomes. In additionto women’s labor force participation and the gender gap in

5. Several factors have been identified as contributing to these gains. First andforemost has been the reversal of the gender gap in education (Blau and Kahn 2006;Goldin, Katz, and Kuziemko 2006). Various technological innovations, such as thecontraceptive pill, have favored women (Goldin and Katz 2002; Greenwood,Seshadri, and Yorukoglu 2005). Labor demand has shifted toward industrieswhere female skills are overrepresented (Black and Juhn 2000; Weinberg 2000).Finally, better regulatory controls and greater competitiveness have reduced labormarket discrimination against women (Black and Strahan 2001; Black andBrainerd 2004).


income, we study the distribution of relative income withinhouseholds, marriage rates, division of home production, mar-riage satisfaction, and divorce.6 Some of these latter issues—especially the impact of relative income on household productionand divorce—have been explored in the sociology literature (Westand Zimmerman 1987). We discuss the relationship of our resultsto the sociology literature in the relevant parts of the article.Fisman et al. (2006) document a pattern related to our resultsin a dating context: men value women’s intelligence or ambition ifand only if it does not exceed their own.

II. Relative Income within Households

II.A. Distribution of Relative Income

Figure I depicts the distribution, across married couples inthe United States, of the share of the household income earned bythe wife. Specifically, we use the Survey of Income and ProgramParticipation (SIPP), which is linked to administrative data onincome from the Social Security Administration and the InternalRevenue Service. The SIPP consists of a series of national panels,each representative of the U.S. civilian, noninstitutionalized pop-ulation. For each married couple, we use the observation from thefirst year that the couple is in the panel, which ranges from 1990to 2004. This leaves us with 73,654 couple-level observations.

We define relativeIncomei as wifeIncomei

wifeIncomei þ husbIncomeiwhere i in-

dexes the couple, and wifeIncomei and husbIncomei are the totallabor and self-employment income of the wife and the husband,respectively. Figure I depicts the frequency distribution ofrelativeIncomei grouped in 20 bins, along with a lowess (locallyweighted scatterplot smoothing) estimate of the distribution oneach side of relativeIncomei ¼

12.

The distribution exhibits a sharp drop at the point where thewife starts to earn more than the husband. The McCrary (2008)

6. Using administrative data from Denmark, Pierce, Dahl, and Nielsen (2012)employ a regression discontinuity design to argue that a husband is more likely touse erectile dysfunction medication if he earns less than his wife. Stuart, Moon, andCasciaro (2011) find that winning an Academy Award is associated with a greaterrisk of divorce for Best Actresses but not for Best Actors.


test for the discontinuity of the distribution function estimatesthat the distribution drops by 12.3 percent ( p< .01).7

We next turn to examining how the distribution of relativeincome varies by decade, by whether the couple has any children,and by how long the couple has been married. We cannot studythese questions using the administrative data from the SIPPbecause there are too few observations. Instead we turn todata from the U.S. Census Bureau. Specifically, we use the1970–2000 U.S. Censuses and the American Community

.02

.04

.06

.08

Frac

tion

of c

oupl

es

0 .2 .4 .6 .8 1Share earned by the wife

FIGURE I

Distribution of Relative Income (SIPP Administrative Data)

The data are from the 1990 to 2004 SIPP/SSA/IRS gold standard files. Thesample includes married couples where both the husband and wife earn posi-tive income and are between 18 and 65 years of age. For each couple, we usethe observation from the first year that the couple is in the panel. Each dotis the fraction of couples in a 0.05 relative income bin. The vertical line indi-cates the relative income share = 0.5. The dashed line is the lowess smootherapplied to the distribution allowing for a break at 0.5.

7. Since there is a positive—albeit small, about 0.0026—fraction of coupleswith relative income exactly at 1

2, we estimate the discontinuity to the right of 12.

We also conduct a placebo exercise, testing for the presence of a discontinuity to theright of each point in the set 0:100; 0:105; :::; 0:445f g [ 0:555; 0:560; :::; 0:900f g. Wefind that there is a significant discontinuity ( p< .05) at roughly 5%, namely,8

140 ¼ 0:057, of these points.


Survey (ACS) 2008 to 2011. Our sample includes household headsand their spouses.8

One important issue with these data is that there is a largenumber of couples—a few percent—with relative income exactlyat 1

2. We know from the administrative data that the actual dis-tribution of relative income does not exhibit such a spike. Thereare at least four reasons the spike might be present in the CensusBureau data. First, incomes are top-coded, so any couple whereboth individuals earn more than the upper bound will have arelative income of 1

2. Second, a nontrivial fraction of incomes inthe census data are imputed. Third, income for both spouses canbe reported by a single respondent. Fourth, except for the 1990census, individual incomes are rounded for confidentiality rea-sons, which increases the chance that two incomes exactly coin-cide. To deal with these issues, we apply the following steps.First, we drop couples where either individual’s income is im-puted. Second, we drop couples where both spouses have top-coded incomes; for couples where one spouse has a top-codedincome, we multiply his or her income by 1.5 and leave theother individual’s income as is. Even after following these steps,a substantial mass at 1

2 remains (around 3 percent of couples),which could be a reflection of either rounding or misreporting.Hence, in a final step, we ‘‘de-round’’ the data, save for 0.26 per-cent of couples (the fraction of couples with relative income of 1

2 inthe administrative data). To de-round the data, we replace eachindividual’s income with a uniform draw from the range of in-comes that would have been rounded to that income.9 All of thefigures that draw on the census and ACS data have been‘‘cleaned’’ in the sense that the spike at 1

2 has been dealt withthrough the procedure described above.

Figure II shows the cleaned distribution of relative income inthe census, restricting the data to the 1990 and 2000 censusesand the 2008–2011 ACS to roughly match the years representedin the SIPP. The distribution is quite similar to the one based onthe administrative data. Most important, the extent of the dis-continuity at 1

2 is also similar. The drop in the distribution in the

8. We do not include other household members who are married because thespouse links for these are based on IPUMS imputed values.

9. Our results are similar if we instead assign incomes based on demographicsor based on a triangular kernel.


Census Bureau data is 13.6% ( p< .01), compared to the 12.3%drop in the SIPP.10

Using the Census Bureau data, we can examine how the dis-tribution of relative income has evolved over time. As Figure IIIshows, the drop in the distribution at equal incomes is present ineach decade. That said, the size of the discontinuity seems to havegotten smaller in the past 20 years. In 1970, the drop is 20.6%; in1980, 26.2%; in 1990, 24.4%; in 2000, 10.8%; finally in the ACS2008–2011 aggregate, the drop in the distribution is 10.0%. Allestimates are significant ( p< .01).

We next turn to the presence of children. There are a numberof reasons the presence of children in the household might affect

FIGURE II

Distribution of Relative Income (Census Bureau Data)

The data are from the 1990 and 2000 censuses and 2008 to 2011 ACSsingle-year files. The sample includes married couples where both the husbandand the wife earn positive income and are between 18 and 65 years of age. Eachdot is the fraction of couples in a 0.05 relative income bin. The vertical lineindicates the relative income share = 0.5. The dashed line is the lowesssmoother applied to the distribution allowing for a break at 0.5.

10. We investigated the potential influence of the tax code on the distribution ofrelative income. We focused on the 1990 census (where income data are notrounded) and couples without children (whose tax liability is easier to impute).We computed the fraction of couples within $200 of a tax bracket cutoff acrosslevels of relative income. There is no evidence that bunching at tax bracket cutoffsis more prevalent when relative income is 0.5.


FIGURE III

Distribution of Relative Income over Time (Census Bureau Data)

The data are from the 1970 to 2000 censuses and 2008 to 2011 ACS single-year files. The sample includes married couples where both the husband andthe wife earn positive income and are between 18 and 65 years of age. Each dotis the fraction of couples in a 0.05 relative income bin. The vertical line indi-cates the relative income share = 0.5. The dashed line is the lowess smootherapplied to the distribution allowing for a break at 0.5.


the overall distribution of relative income, but the discontinuityat equal incomes turns out to be present both for couples withand without a child: among those with a child, the drop is 15.2%( p< .01); among those without, the drop is 11.6% ( p< .01).11

These distributions are depicted in Figure A.1 in the OnlineAppendix.12

Overall the observed discontinuity could be due to the factthat couples avoid getting married if she earns more than he does,due to evolution of relative income during marriage, or due to theimpact of relative income on divorce. Examining how the size ofthe discontinuity varies with marriage tenure sheds some lighton the relative importance of these channels. For this exercise, werestrict our attention to the 2008–2011 ACS since earlier censusdata do not include information on how long the couple has beenmarried.

We find that the discontinuity is present, albeit with a smal-ler drop of 8.4%, among couples who have been married for oneyear or less. Among couples married 2–5 years, the drop is 10.1%;among those married 6–10 years, 12.9%; and finally, among cou-ples who have been married for more than 10 years, the drop inthe distribution at equal incomes is 10.5%. All estimates arehighly statistically significant ( p< .01).13 In the next subsection,we argue that the discontinuity at equal incomes is driven bygender identity norms. Under this interpretation, the discontinu-ity among the newlyweds implies that gender identity affects whomarries whom, while the fact that the discontinuity grows withmarriage tenure (over the first decade of marriage) suggests thatidentity considerations also influence the evolution of relativeincome within a couple and/or the likelihood of divorce. InSections IV and V we explore these two channels in greater detail.

II.B. Relative Income in Standard Models of the MarriageMarket

In standard models of the marriage market, marriages arepartnerships formed for the purpose of joint production and joint

11. For these calculations, we restrict attention to the 1990 and 2000 censusesand the 2008–2011 ACS for ease of comparison with the discontinuity in the overalldistribution.

12. The Online Appendix contains all supplemental figures and tables and thedetailed description of the data.

13. Figure A.2 in the Online Appendix depicts the distributions of relativeincome by marriage tenure.


http://qje.oxfordjournals.org/lookup/suppl/doi:10.1093/qje/qjv001/-/DC1






consumption (see Weiss 1997 for a survey). A widely studied classof models is one where men and women are endowed with asingle-dimensional attribute that positively affects the familyoutput. If utility is nontransferable, equilibrium induces positiveassortative matching. If utility is transferable, equilibrium isconsistent with either positive or negative assortative matchingdepending on whether the individuals’ attributes are comple-ments or substitutes. Whether positive or negative, assortativematching relates the ranks of individuals in their gender-specificdistribution of attributes. For example, if matching is positiveand income is the attribute of interest, a woman in the 30th per-centile of women’s income distribution will match with the manwho is in the 30th percentile of men’s income distribution.Whether the woman earns more or less than the man in absoluteterms has no significance in these models.14

A second important class of models is one where marriagesenable division of labor to exploit comparative advantage or in-creasing returns. In case of increasing returns, it is optimal foronly one spouse to work, so for purposes of studying relativeincome we focus on comparative advantage. With decreasing re-turns and comparative advantage, both spouses may participatein the labor force; their contributions to household income and tohousehold production are determined by their relative productiv-ity in those two activities. In these models, women tend to domore chores and men tend to earn higher incomes becausewomen are relatively more productive at home while men arerelatively more productive in the labor force. That said, thepoint where the wife and the husband earn the same incomeagain plays no special role—the gains from marrying a man arecontinuous in his labor market productivity with no discrete jumpat the point where endogenous incomes would be equal.Moreover, in Section VI we show that the amount of time a wifespends on household production increases when she earns morethan the husband, which stands in sharp contrast with the basicprediction of this class of models.

Existing literature also distinguishes modes of decisionmaking within households. A key distinction is between modelswith a common objective where all decisions are taken to

14. As we discuss in Section V, difference in income percentiles between thespouses does not predict divorce, whereas wife earning more than the husband inabsolute terms does.


maximize a shared utility function, and models where the hus-band and the wife have distinct preferences and bargain theirway to a Pareto efficient outcome. Relative income could clearlyinfluence the bargaining positions within the couple, but onceagain equilibrium outcomes are continuous in the outsideoption with no special significance of the point where the twoincomes are equal.15

Standard models of the marriage market thus cannot ac-count for the observed distribution of relative income with itsdiscontinuous drop at the point where the wife earns more thanthe husband.16 One natural interpretation of this discontinuity issimply that some couples try to avoid the circumstance where thewife earns more than the husband. Recall that a substantial frac-tion of U.S. respondents to the World Values Survey agrees withthe claim that ‘‘If a woman earns more money than her husband,it’s almost certain to cause problems.’’ The tendency to agree withthis claim turns out to be much stronger among less educatedrespondents: only 28% of the couples where both the husbandand the wife have at least some college education agree withthe claim, compared with 45% of the couples where neitherspouse went beyond high school. Hence, if gender role attitudesare indeed the source of the cliff in the distribution of relativeincome, we should expect the discontinuity to be greater amongless-educated couples. This is indeed the case. Among less-edu-cated couples, the distribution drops by 20.1% ( p< .01). Amongmore educated couples, there is only a (statistically insignificant)drop of 5.53% (see Figure A.3 in the Online Appendix). This dif-ference in the discontinuity across these education groups is sig-nificant ( p< .05).

III. Marriage Rates and Relative Income

For the past 40 years, marriage rates in the United Stateshave been steadily declining. Between 1970 and 2008, the frac-tion of young adults who are currently married decreased by 30 to

15. See Lundberg and Pollak (1996) for a survey of bargaining models of thehousehold.

16. We discuss here only a fraction of the vast literature on marriage markets,but similar arguments can be used to show that other motives for marriageformation—credit constraints, public consumption, risk pooling, and so on—alsocannot account for the observed distribution of relative income.




50 percentage points among all race, gender, and educationgroups (Autor 2010).17 Over the same period, women’s incomehas greatly increased relative to that of men. Results from theprevious section suggest a potential link between these twotrends: if couples dislike unions where the husband earns lessthan the wife, then, as women start to command a greatershare of labor income, marriages may become less appealing.

In this section, we analyze how the share of individuals whoare currently married varies with the distributions of men’sand women’s (potential) income. Throughout the section, we use1980 to 2000 data from the U.S. Census and the 2008–2011American Community Survey.18 We assign individuals to mar-riage markets based on the pattern of homophily: most marriagesoccur between men and women who are of the same race and areof similar age and education.19 Moreover, marriages tend to formbetween individuals who live close to each other. Accordingly, wedefine marriage markets based on race, age group, educationgroup, and the state of residence. The three race groups we con-sider are (non-Hispanic) whites, (non-Hispanic) blacks, andHispanics.20 The three age groups are (i) 22–31 for women and24–33 for men; (ii) 32–41 for women and 34–43 for men; and (iii)42–51 for women and 44–53 for men. The two education groupsare (i) high school degree or less, and (ii) some college or more.Online Appendix Table A.1 documents sorting along these dimen-sions. For example, 98% of wives who are white are married to ahusband who is white,21 72% of wives with a high school degree orless are married to a husband with similar educational qualifica-tions, and 76% of wives aged 22–31 are married to a husband aged24–33. Overall, 59% of all marriages are between a man andwoman from the same marriage market.22

Given a particular marriage market, we wish to know howthe changes in women’s income relative to that of men affect

17. Part of this decrease is due to delay in marriage, but the fraction of olderadults who are married has also been declining.

18. For convenience, we refer to the 2008–2011 ACS as 2010 data.19. Our approach to assigning individuals to marriage markets is similar to

those used in Charles and Luoh (2010) and Loughran (2002).20. We drop individuals of other races.21. For a broader discussion of same-race marriages in the United States, see

Fisman et al. (2008).22. This share has not changed much over the time period we consider (it ranges

from a minimum of 57% in 2000 to a maximum of 61% in 1980).



marriage rates. For each marriage market m and year t2 {1980,1990, 2000, 2010} we compute how likely it is, when a womanencounters a man, that her income exceeds his. Specifically,given woman i and man j, consider a binary variable that takesvalue 1 if i’s income exceeds j’s. We define PrWomanEarnsMoremt

as the mean of this variable taken across all possible couples.Operationally, we construct this variable by randomly drawing50,000 women and men with replacement and computing theshare of couples where the woman earns more than the man.

We consider several measures of income. First we use indi-viduals’ actual earnings, where we code an individual as havingzero income if he or she is not in the labor force. Second we con-struct a measure of predicted earnings based on demographiccharacteristics. In particular, we assign each woman and manin a census year to a demographic group defined based on race,age,23 education,24 and state of residence. We then assign poten-tial income to each individual by drawing from the earnings dis-tribution of those in the demographic group who have positiveincome. Finally, for our preferred specification, we construct dis-tributions of income based on a Bartik-style instrument that iso-lates the variation in relative income, which is plausiblyunrelated to the factors that directly affect the marriage market.

Across all census years and marriage markets, the likelihoodthat a randomly chosen woman earns more than a randomlychosen man is about 0.25. This likelihood has increased steadilyover time, going from 17–20% in 1980 to about 31–33% in 2010.25

More important for our purposes, these dynamics have variedacross marriage markets.26 Thus, there is ample variation inPrWomanEarnsMoremt even when we include marriage marketand year fixed effects. Note that this residual variation stemsfrom both compositional shifts within a marriage market overtime and from shocks that differentially affect men and womenwithin a marriage market. When we turn to our instrumentalvariables approach, we isolate the component of the latter varia-tion which stems from nationwide changes in labor demandacross industries.

23. We group age into three-year intervals.24. We group education into the following categories: less than high school, high

school, some college, college, more than college.25. See Online Appendix Table A.2 for summary statistics.26. See Online Appendix Table A.3.




Our baseline OLS specification regresses shareMarriedmt—the share of males who are currently married27—onPrWomanEarnsMoremt, controlling for the logs of the averagefemale and male income and the logs of female and maleincome at each income decile in the marriage market that year.All specifications include marriage market fixed effects as well asyear fixed effects interacted with race, age group, educationgroup, and the state of residence. We include these interactionswith year fixed effects because the relationship between demo-graphic variables and marriage rates may have changed overtime. The unit of observation is a marriage market in a censusyear. Standard errors are clustered by state, and each observa-tion is weighted by the number of women in the marriage market.

This baseline specification is in column (1) of Table I. Theestimated impact of PrWomanEarnsMoremt on shareMarriedmt

is –0.080, but is not statistically significant. Column (2) adds acontrol for average women’s income divided by the sum of averagemen’s and women’s income. With this specification, the estimatedeffect remains negative but is smaller and still not significant. Incolumn (3), we include several additional marriage market byyear controls: the sex ratio, male and female incarcerationrates, average years of schooling for men and women, and thenumber of men and women in the market. The estimated effectbecomes stronger and significant.

In columns (4)–(6) we consider the same three specifications,but we construct the variable PrWomanEarnsMoremt (as well asall other relevant income variables in the regression) using pre-dicted income. As we show later, relative income considerationshave a direct impact on the realized income of males and females,so predicted income is likely to be a cleaner measure of the rele-vant distributions. In these specifications, estimated impact ofPrWomanEarnsMoremt on shareMarriedmt is consistently nega-tive, stable (ranging from –0.236 to –0.266 across the threespecifications), and always highly significant ( p< .01).

The identifying assumption behind these specifications isthat PrWomanEarnsMoremt is uncorrelated with unobservedshocks that influence marriage rates. The fact that the coefficientof interest is stable across columns (4)–(6) somewhat amelioratesconcerns about omitted variables, but to provide further support

27. We get similar results if we use the share of women who are married.


TA

BL

EI

PO

TE

NT

IAL

RE

LA

TIV

EIN

CO

ME

AN

DM

AR

RIA

GE

RA

TE

S

Inco

me

mea

sure

:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Act

ual

Pre

dic

ted

Bart

ik

Dep

end

ent

vari

able

:sh

are

Ma

rrie

d

PrW

oma

nE

arn

sMor

e–0.0

80

–0.0

46

–0.2

09**

*–0.2

66**

*–0.2

52**

*–0.2

36**

*–0.5

15**

*–0.3

43*

–0.3

51*

[0.0

75]

[0.0

80]

[0.0

74]

[0.0

68]

[0.0

66]

[0.0

62]

[0.1

89]

[0.1

83]

[0.1

81]

lnA

ver

age

Wom

en’s

Inco

me

0.0

55*

0.1

71**

0.0

88

0.0

66*

0.2

66**

0.1

51

0.2

70

0.9

43**

*0.4

61

[0.0

30]

[0.0

71]

[0.0

74]

[0.0

36]

[0.1

08]

[0.1

08]

[0.1

77]

[0.3

33]

[0.3

71]

lnA

ver

age

Men

’sIn

com

e0.0

23

–0.0

92

0.0

05

–0.0

01

–0.2

01**

–0.0

63

0.1

14

–0.5

58*

–0.0

97

[0.0

32]

[0.0

70]

[0.0

73]

[0.0

53]

[0.0

84]

[0.0

93]

[0.1

40]

[0.2

92]

[0.3

48]

Sex

Rati

o–0.0

30**

*–0.0

27**

*–0.0

06

[0.0

07]

[0.0

07]

[0.0

07]

Fem

ale

Inca

rcer

ati

onR

ate

–0.3

69

–0.2

92

–0.0

48

[0.2

41]

[0.2

32]

[0.1

72]

Male

Inca

rcer

ati

onR

ate

0.4

33**

*0.2

10**

*0.0

56

[0.0

89]

[0.0

71]

[0.0

69]

Fem

ale

Aver

age

Yea

rsof

Ed

uca

tion

0.0

09

0.0

05

–0.0

02

[0.0

08]

[0.0

07]

[0.0

07]

Male

Aver

age

Yea

rsof

Ed

uca

tion

–0.0

31**

*–0.0

23**

–0.0

10

[0.0

10]

[0.0

08]

[0.0

07]

Nu

mber

ofF

emale

s(p

erm

illi

on)

0.0

01

0.0

03

–0.0

03

[0.0

05]

[0.0

06]

[0.0

08]

Nu

mber

ofM

ale

s(p

erm

illi

on)

0.0

04

0.0

02

0.0

05

[0.0

05]

[0.0

06]

[0.0

07]


TA

BL

EI

(CO

NT

INU

ED)

Inco

me

mea

sure

:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Act

ual

Pre

dic

ted

Bart

ik

Dep

end

ent

vari

able

:sh

are

Ma

rrie

d

Obse

rvati

ons

3,6

35

3,6

35

3,6

35

3,5

78

3,5

78

3,5

78

3,6

36

3,6

36

3,6

36

R-s

qu

are

d0.9

90

0.9

90

0.9

91

0.9

90

0.9

90

0.9

91

0.9

91

0.9

91

0.9

91

Con

trol

for

rela

tive

inco

me

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

Con

trol

for

dec

iles

ofm

en’s

an

dw

omen

’sin

com

eY

esY

esY

esY

esY

esY

esY

esY

esY

es

Not

es.

Data

are

from

the

1980

to2000

U.S

.C

ensu

ses

an

d2008

an

d2011

AC

Ssi

ngle

-yea

rfi

les.

Th

eu

nit

ofob

serv

ati

onis

am

arr

iage

mark

etby

dec

ad

e(1

980,

1990,

2000,

an

d2010,

usi

ng

2008–2011).

Am

arr

iage

mark

etis

defi

ned

by

the

inte

ract

ion

ofage

gro

up

,ed

uca

tion

gro

up

,ra

ce,

an

dst

ate

ofre

sid

ence

.S

eete

xt

for

furt

her

det

ail

son

the

con

stru

ctio

nof

the

marr

iage

mark

et.

sha

reM

arr

ied

isth

esh

are

ofm

ale

sw

ho

are

marr

ied

.P

rWom

an

Ea

rnsM

ore

isth

ep

robabil

ity

that

ara

nd

omly

chos

enw

oman

earn

sm

ore

than

ara

nd

omly

chos

enm

an

.S

exR

ati

ois

the

rati

oof

the

male

pop

ula

tion

toth

efe

male

pop

ula

tion

ina

marr

iage

mark

et.

All

spec

ifica

tion

sin

clu

de

marr

iage

mark

etfi

xed

effe

cts,

dec

ad

efi

xed

effe

cts,

an

dth

ed

ecad

ein

tera

cted

wit

hth

eage

gro

up

,th

eed

uca

tion

gro

up

,th

era

ce,

the

state

ofre

sid

ence

an

dd

ecil

esof

men

’san

dw

omen

’sin

com

e.‘‘C

ontr

olfo

rre

lati

ve

inco

me’

’is

defi

ned

as

the

rati

oof

aver

age

wom

en’s

inco

me

toth

esu

mof

aver

age

men

’san

dw

omen

’sin

com

e.C

olu

mn

s(1

)–(3

)u

seact

ual

earn

ings

toca

lcu

late

PrW

oma

nE

arn

sMor

e(a

nd

all

oth

erre

levan

tin

com

evari

able

sin

the

regre

ssio

n);

colu

mn

s(4

)–(6

)u

sep

red

icte

dea

rnin

gs

toca

lcu

late

PrW

oma

nE

arn

sMor

e(a

nd

all

oth

erre

levan

tin

com

evari

able

sin

the

regre

ssio

n)

an

dco

lum

ns

(7)–

(9)

use

the

Bart

iksp

ecifi

cati

onto

calc

ula

teP

rWom

an

Ea

rnsM

ore

(an

dall

oth

erre

levan

tin

com

evari

able

su

sed

inth

ere

gre

ssio

n).

Reg

ress

ion

sare

wei

gh

ted

by

the

nu

mber

ofw

omen

inth

em

arr

iage

mark

et.

Sta

nd

ard

erro

rscl

ust

ered

at

the

state

level

are

inbra

cket

s.**

*sig

nifi

can

tat

1%

,**

at

5%

,*a

t10%

.


for our causal interpretation, we now turn to an instrumentalvariables approach.

Historically men and women have tended to work in differ-ent industries (e.g., women are overrepresented in servicesand men in construction and manufacturing). Based on the in-dustry composition of the state and the industry-wide wagechanges at the national level, we can thus isolate gender-specificvariation in local wages that is driven solely by aggregate la-bor demand (which is presumably uncorrelated with the charac-teristics of workers in a given marriage market). This approachbuilds on previous work by Bartik (1991) and Aizer (2010). Incontrast to previous uses of the ‘‘Bartik instrument,’’ whichfocus on changes in average wages, we construct an instrumentfor the entire distribution of potential income in each marriagemarket.

We begin by instrumenting for average yearly wages bygender and marriage market as follows:

�wgmt �

X

j

�grejs;1980 �wg

reajt;�s

where g indexes gender, r race, e education group, a age group,j industry,28 t census year, and s state. Variable wg

reajt;�s is theaverage wage in year t in industry j for workers of a given gender,race, education, and age group in the nation, excluding state s.Variable �g

rejs;1980 is the fraction of individuals with gender g, racer, and education e in state s who are working in industry j, as ofthe base year 1980.29

Variable �wgmt is strongly correlated with the actual mean

income of gender g in marriage market m in year t: states thatinitially had relatively more women in industries that subse-quently experienced wage growth at the national level tend tohave more growth in women’s income relative to that of men.

28. We consider 12 industry groups: agriculture; mining; construction;manufacturing; transportation; wholesale trade; retail trade; finance, insurance,and real estate; business, personal, and repair services; entertainment and recre-ation services; professional services; and public administration.

29. We choose to use 1980 as the base year instead of an earlier decade (e.g.,1970) as the earlier census data sets are only 1% samples. Thus, compared to the1980 census data, there are many fewer observations, which would result in noisierestimates of �g

rejs.


But unlike the variation in actual income, variation in �wgmt over

time is driven by aggregate shocks and is thus plausibly orthog-onal to factors that might directly influence marriage rates inmarket m.

Similarly, we wish to construct a measure of the entire dis-tribution of income by gender which is driven solely by aggregateshocks. We modify the standard Bartik instrument to computepredicted yearly wages at the p = {5th, 10th, 15th, . . . 90th, 95th}percentile.

Specifically, let

�wg;pmt �

X

j

�grejs;1980 �wg;p

reajt;�s

where wg;preajt;�s is the pth percentile of the national income distri-

bution in year t in industry j for workers of a given gender, race,education, and age group, excluding state s. A priori it is not clearthat �wg;p

mt will be correlated with the pth percentile of gender g’sdistribution in market mt. For example, if half the women in ademographic group m work in some industry jhigh where the min-imum income is yhigh and half the women in m work in someindustry jlow where the maximum income is ylow< yhigh, increasein the 5th percentile of wages in industry jhigh will not raisethe 5th percentile of wages of women in market m. This example,however, has little empirical relevance—a posteriori, thedistributions defined by �wg;p

mt

� �p

indeed correlate with the actual

distributions of income. In other words, the Bartik instrumenthas a strong ‘‘first stage’’ when it is used to predict how the dis-tribution of income varies across markets (see Online AppendixTable A.4).

This modification to the standard Bartik approach allows usto construct a measure of PrWomanEarnsMoremt whose variationover time is orthogonal to local labor market conditions.Specifically, we draw from the distributions defined by �wg;p

mt

� �p,

and calculate the likelihood that a randomly chosen womenearns more than a randomly chosen man. Column (7) repeatsthe baseline specification from column (1), but using �wg

mt and�wg;p

mt to construct all measures of income. The estimated impactof PrWomanEarnsMoremt on the marriage rate is negative(–0.515) and significant ( p< .01). In column (8) we control forrelative income. The estimate declines to �0.343 and becomessignificant only at the 10% level. Finally, in column (9), we




include a set of marriage market by year controls. The estimate issimilar at -0.351 and remains significant at the 10% level.30

We also estimate the specification in column (9) separatelyfor the less educated (high school or less) and more educated(some college or more) marriage markets. The impact ofPrWomanEarnsMoremt on the marriage rate is –0.631 (std.err. = 0.174) in the less educated markets, compared to 0.141(std. err. = 0.177) in the more educated markets. This pattern fur-ther supports the view that the observed relationship is driven bytraditional attitudes toward gender roles.

Taken together, these results highlight the importance of therelative distribution of men and women’s income in marriagemarkets. The estimate from our preferred specification (column(9)) implies that the secular increase in the income of womenrelative to that of man explains 29 percent of the overall declinein marriage rates from 1980 to 2010.31

Note that the relative distribution of men and women’sincome might influence the formation of marriage even in theabsence of gender identity considerations. In Beckerian modelsof the marriage market, one of the key benefits of marriage isspecialization. Specialization, in turn, is more valuable if a manand a woman have different opportunities in the labor market. AsPrWomanEarnsMore increases, there are smaller ‘‘gains fromtrade’’ that can be achieved through marriage.32 This forcealone might account for our results. That said, the evidence wepresent in other sections of this article is in direct conflict withthe standard models of the marriage market and, as we alreadydiscussed, standard models cannot easily explain the observeddistribution of relative income. Thus, the view that coupleshave an aversion to the wife earning more than the husband

30. Variability in the income of potential spouses can also lead to a lower mar-riage rate. This fact, however, cannot account for our findings because we includecontrols for the deciles of the distributions of men and women’s income in the mar-riage market in all specifications. In other words, all specifications in Table I al-ready control for any linear effect of variability in income on marriage rates.

31. The coefficient –0.351 multiplied by the 14 percentage point increase inPrWomanEarnsMore is 29% of the 17 percentage point decrease in marriagerates (Online Appendix Table A.2).

32. Oppenheimer (1997) argues against the view that reduced gains from tradeaccount for the decrease in marriage rates, but she focuses on the impact of women’sincome per se rather than the relative income of men and women.



provides a more parsimonious explanation of the various patternswe present herein.

IV. Women’s Labor Supply and Relative Income

The previous sections establish that couples are less likely toform if the wife’s income would exceed the husband’s. When suchcouples do form, we might expect gender identity to distort labormarket outcomes. A wife whose income would exceed her hus-band’s may choose to stay at home so as to be less threatening.Or she may distort her labor supply in other ways—for example,work fewer hours or take a job that is less demanding and paysless. In this section we analyze such potential distortions in thewife’s labor force participation and labor market outcomes.

This analysis complements existing work on how the varia-tion in attitudes toward gender influences women’s labor supply.Fortin (2005) uses data from the World Values Surveys in 25Organisation for Economic Co-operation and Development(OECD) countries over a 10-year period and shows that thesocial representation of women as homemakers and men asbreadwinners is associated with a low labor force participationby women and a large gender gap in income. Fortin (2009) exam-ines a similar question in a single country (the United States)over a longer time period (1977–2006). She shows that the evolu-tion of gender role attitudes over time correlates with the evolu-tion of female labor force participation. In particular, whilewomen’s gender role attitudes steadily became less traditionaluntil the mid-1990s (e.g., more and more women came to disagreewith the notion that husbands should be the breadwinners andwives should be the homemakers), these trends reversed in themid-1990s, precisely at the time that coincides with the slowdownin the closing of the gender gap in labor force participation.

Throughout this section, we use data on married couplesfrom the 1970 to 2000 U.S. Censuses and the ACS single-yearfiles from 2008 to 2011 (which we refer to as the 2010 ACS forsimplicity). We restrict the sample to those couples where thehusband is working. For each couple i, we estimate the distribu-tion of the wife’s potential earnings as follows. For p2 {5, . . . , 95},we define wp

i as the pth percentile of earnings among workingwomen in the wife’s demographic group that year. We assignthe demographic group based on race, age group, education


level, and state of residence.33 We then define a variable PrWife

EarnsMorei ¼119

Pp 1 wp

i>husbIncomeif g where husbIncomei is the

husband’s income.34 Thus, whether the wife works or not,PrWifeEarnsMorei captures the likelihood that she would earnmore than her husband if her income were a random draw fromthe population of working women in her demographic group.35

Summary statistics for this sample are presented in OnlineAppendix Table A.5. Across all census years, the mean ofPrWifeEarnsMorei is 0.18. Not surprisingly, this probability hasincreased over time, from 0.09 in the 1970 census to 0.27 in the2010 ACS. Across all years, about 67% of wives participate in thelabor force. As we mentioned in the introduction, wives’ laborforce participation increased steeply between 1970 and 1990(going from 43% to 70%), but has essentially plateaued since1990, with 74% of wives in the labor force in the 2010 ACS.

IV.A. Labor Force Participation

We first examine wives’ labor force participation. One of thestrongest ways to conform to traditional gender roles is for thewife to stay at home while the husband plays the role of bread-winner. Might it be the case that some couples retreat to tradi-tional gender roles when gender identity is threatened by thepossibility that the wife would be the primary provider?

Given a couple i, let wifeLFPi be a binary variable equal to 1if the wife is in the labor force. In column (1) of Table II, we con-sider, as the baseline specification, a linear probability model:

wifeLFPi ¼ �0 þ �1 � PrWifeEarnsMorei

þwpi þ �2 � lnHusbIncomei þ �3 � Xi þ "i;

where lnHusbIncomei is the log of husband’s income, wpi are con-

trols for the wife’s potential income at each of the vigintiles, andXi represents nonincome controls: year fixed effects, state fixed

33. Throughout this section, by ‘‘age group’’ we mean five-year intervals and by‘‘education level’’ we mean the following categories: less than high school, highschool, some college, college, more than college.

34. Since we estimate the distribution based on every 5th percentile from the5th to the 95th, PrWifeEarnsMore is based on 19 percentiles.

35. As we discuss in the next subsection, the income of the women who do workmay also be distorted by gender identity considerations. By ‘‘potential income’’ wemean the (possibly distorted) income that the wife would likely earn were she to jointhe labor force.




TA

BL

EII

PO

TE

NT

IAL

RE

LA

TIV

EIN

CO

ME

AN

DW

IFE’S

LA

BO

RF

OR

CE

PA

RT

ICIP

AT

ION

(1)

(2)

(3)

(4)

(5)

(6)

Dep

end

ent

vari

able

:W

ife

inth

ela

bor

forc

e

PrW

ifeE

arn

sMor

e–0.1

78**

*–0.1

42**

*–0.1

39**

*–0.1

43**

*–0.1

48**

*–0.1

52**

*[0

.004]

[0.0

04]

[0.0

04]

[0.0

04]

[0.0

05]

[0.0

05]

Obse

rvati

ons

7,3

84,1

76

1,3

75,1

21

1,3

75,1

21

R-s

qu

are

d0.0

97

0.1

03

0.1

04

0.1

45

0.0

87

0.0

90

Ad

dit

ion

al

con

trol

s:C

ubic

inln

Hu

sbIn

com

en

oyes

yes

yes

yes

yes

lnM

edia

nW

ifeP

oten

tia

l�

lnH

usb

Inco

me

no

no

yes

yes

no

no

an

yCh

ild

ren

no

no

no

yes

no

no

Wif

e’s

dem

ogra

ph

icgro

up�

Hu

sban

d’s

dem

ogra

ph

icgro

up

no

no

no

yes

no

no

PrW

ifeE

arn

sMor

eA

tMa

rria

ge

no

no

no

no

no

yes

Vig

inti

les

ofth

ew

ife’

san

dth

eh

usb

an

d’s

pot

enti

al

inco

me

at

marr

iage

no

no

no

no

no

yes

Marr

iage

du

rati

onfi

xed

effe

cts

no

no

no

no

no

yes

Sam

ple

rest

rict

ion

non

en

one

non

en

one

2010su

b2010su

b

Not

es.

Data

are

from

the

1970

to2000

U.S

.C

ensu

san

d2008

to2011

AC

Ssi

ngle

-yea

rfi

les.

Th

esa

mp

leco

nsi

sts

ofco

up

les

wh

ere

bot

hth

ew

ife

an

dth

eh

usb

an

dare

bet

wee

n18

an

d65

yea

rsol

dan

dth

eh

usb

an

dis

wor

kin

g.

Th

ed

epen

den

tvari

able

Wif

ein

the

labor

forc

eis

abin

ary

vari

able

that

equ

als

1if

the

wif

eis

inth

ela

bor

forc

e,0

oth

erw

ise.

Th

ek

eyin

dep

end

ent

vari

able

PrW

ifeE

arn

sMor

eis

the

pro

babil

ity

that

the

wif

e’s

inco

me

wou

ldex

ceed

the

hu

sban

d’s

ifh

erin

com

ew

ere

dra

wn

from

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inh

erd

emog

rap

hic

gro

up

.ln

Hu

sbIn

com

eis

the

log

ofh

usb

an

d’s

inco

me.

lnM

edia

nW

ifeP

oten

tia

lis

the

log

ofth

em

edia

nof

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

ew

ife’

sd

emog

rap

hic

gro

up

.a

nyC

hil

dre

nis

abin

ary

vari

able

that

equ

als

1if

the

wif

ere

por

tsh

avin

gan

ych

ild

ren

,0

oth

erw

ise.

PrW

ifeE

arn

sMor

eAtM

arr

iage

isth

ep

robabil

ity

that

inco

me

dra

wn

from

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

ew

ife’

sd

emog

rap

hic

gro

up

exce

eds

inco

me

dra

wn

from

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

eh

usb

an

d’s

dem

ogra

ph

icgro

up

inth

ecl

oses

tce

nsu

syea

rto

the

yea

rof

thei

rm

arr

iage.

Marr

iage

du

rati

onfi

xed

effe

cts

are

du

mm

ies

for

the

nu

mber

ofyea

rsth

eco

up

leh

as

bee

nm

arr

ied

.A

llre

gre

ssio

ns

incl

ud

eco

ntr

ols

for

log

hu

sban

d’s

inco

me,

vig

inti

les

ofth

ew

ife’

sp

oten

tial

inco

me,

wif

e’s

an

dh

usb

an

d’s

edu

cati

on(fi

ve

cate

gor

ies)

,w

ife’

san

dh

usb

an

d’s

five-

yea

rage

gro

up

,w

ife’

san

dh

usb

an

d’s

race

(th

ree

cate

gor

ies)

,yea

r,an

dst

ate

fixed

effe

cts.

Sam

ple

rest

rict

ion

‘‘2010su

b’’

ind

icate

sa

subsa

mp

leof

AC

Sw

her

eth

ed

iffe

ren

cebet

wee

nyea

rof

marr

iage

an

dth

ecl

oses

tce

nsu

syea

ris

no

mor

eth

an

five

yea

rsan

dP

rWif

eEa

rnsM

oreA

tMa

rria

ge

cou

ldbe

com

pu

ted

.S

tan

dard

erro

rsare

rep

orte

din

bra

cket

s.**

*sig

nifi

can

tat

1%

level

,**

at

5%

,*a

t10%

.


effects, the wife’s and the husband’s race, the wife and the hus-band’s age group, and the wife’s and the husband’s educationlevel. Standard errors are clustered by the wife’s demographicgroup (which pins down the distribution of her potentialincome). The baseline estimate of �1 is –0.178 ( p< .01).

The husband’s income might affect the marginal utility ofhousehold income non-linearly, so in column (2) we include acubic polynomial in lnHusbIncomei. The estimate of �1 falls some-what to –0.142 but remains statistically significant ( p< .01).Another issue is that the impact of the wife’s potential incomeon her labor supply might interact with the husband’s income forreasons that are separate from the couple’s concern that she willearn more than he does. To deal with this, in column (3) we add acontrol for the median of the wife’s predicted income w50

i

� �inter-

acted with the income of the husband. The estimate of �1 is stableat –0.139.

This estimated effect is economically significant: a 10 per-centage point increase in the probability that a wife would earnmore than her husband reduces the likelihood that she partici-pates in the labor force by around 1.4 percentage points. Put dif-ferently, a 1 standard deviation increase (across all years) in theprobability that a wife would earn more than her husband redu-ces the likelihood that she participates in the labor force by about3.5 percentage points.

The main concern in this subsection is that a woman who iswilling to marry a man whose income is below her potentialincome might have unobservable characteristics that keep herout of the labor force. For example, highly educated women thatmarry men with lower education and low earnings might be sys-tematic underachievers or systematically lack the confidence toparticipate in the labor market; such women might be relativelymore drawn toward home production and child-rearing activities.We consider two approaches to deal with this concern.

First, we examine the sensitivity of �̂1 to the inclusion ofother controls. In column (4), we include as controls an indicatorvariable for whether the wife ever had a child and indicator var-iables for the full interaction of the wife’s demographic group andthe husband’s demographic group. The inclusion of these addi-tional controls barely affects the estimate of �1. The fact that ourestimate appears stable across specifications suggests that to theextent that the observable characteristics in our data are


representative of unobservables, the negative value of �̂1 is notdue to an omitted variable bias (Altonji, Elder, and Taber 2005).36

Second, in the 2010 ACS, we can partly isolate the variationin PrWifeEarnsMorei that is driven by changes in relative in-come that took place after the couple got married. Unlike theearlier census data, the ACS contains information on theyear the current marriage started. We can thus proxy for rela-tive income of spouses at the time of marriage as follows. Foreach couple i, let mi be the census year that is the closest to theyear of marriage (we drop couples for whom the difference be-tween the year of marriage and the closest available censusyear is more than five years). We then construct the distributionof potential income for both the husband hp

i

� �and the wife

wpi

� �in year mi, using the same procedure as before.

Based on these two distributions, we define a variable

PrWifeEarnsMoreAtMarriagei ¼1

361

Pp

Pq 1 wp

i>hq

if g. This vari-

able captures the probability that based on the couples’ demo-graphics, the wife’s potential income exceeded that of thehusband at the time of marriage.37

We first show that the focus on the new subsample does notmake a difference: column (5) replicates the specification from col-umn (2) but using only data on couples for whom we could constructrelative earnings at marriage. The effect of PrWifeEarnsMorei onlabor force participation is very similar as in the overall sample.38

Our main specification is in column (6), where we include as con-trols PrWifeEarnsMoreAtMarriagei and the vigintiles of the poten-tial earnings for both the husband and the wife at marriage.39 Inthis specification, we are relying only on the variation that stemsfrom changes in the wife’s and the husband’s relative income sincemarriage.40 Yet the estimate of �1 is unaffected.

36. The coefficient is also stable if we include the additional controls one by one.37. Since we estimate the distribution based on every 5th percentile from the

5th to the 95th, there are 19 percentiles for each spouse and thus 19�19 = 361overall comparisons in the expression

Pp

Pq 1 wp

i>hq

if g.38. In Online Appendix Table A.6, we consider the specification separately

decade by decade. In each period we find a negative and statistically significanteffect.

39. We also control nonparametrically for marriage duration in years.40. Two concerns remain. First, our proxy for relative income at marriage is

imperfect as we do not know the couples’ actual income at the time. Second, to theextent that couples can predict how their relative income will evolve after marriage,some concerns about selection are still present.



Overall, although we do not have an exogenous source ofvariation in PrWifeEarnsMorei, the data suggest that marriedwomen sometimes stay out of the labor force so as to avoid a sit-uation where they would become the primary breadwinner.Moreover, we find that the effect is stronger among less educatedcouples, which is consistent with the view that it reflects genderidentity norms.41 Finally, in Section VII, we document a similareffect using panel data: within couples, when a wife outearns herhusband, she becomes less likely to be in the labor force the fol-lowing year.

IV.B. Gap between Potential and Realized Income

Having the wife leave the labor force is a financially costlyway to restore traditional gender roles. It would be less costly forthe wife to simply reduce her earnings to a level that does notthreaten the husband’s status as the primary breadwinner. Inthis subsection, we present evidence for such behavior.

Given a couple i, let incomeGapi ¼wifeIncomei � wifePotentiali

wifePotentiali

where wifePotentiali is simply the mean of the distribution ofpotential earnings for the wife, as defined in the previous subsec-tion. To emphasize the distortions in income for women who donot leave the workforce, we focus on the sample of couples wherethe woman is working. These results are reported in Table III,which follows exactly the same structure as Table II, but withincomeGapi rather than wifeLFPi as the dependent variable.

Column (1) of Table III considers the baseline specificationand yields an estimate of �1 equal to�0.031 ( p< .01). Including acubic polynomial of lnHusbIncomei in column (2) strengthens theeffect: �̂1 ¼ �0:095 p < :01ð Þ. A 10 percentage point increase inthe probability that a wife would earn more than her husbandincreases the gap between her actual earnings and her potentialearnings by about 1 percentage point. Columns (3)–(6) considerthe same robustness checks as in the previous subsection. Theeffect persists and the estimate is reasonably stable across speci-fications, though somewhat larger in the recent years.42

41. We consider the specification in column (2) separately for less educatedcouples (both the wife and the husband have only a high school education or less)and more educated couples (both the wife and the husband have at least somecollege education). Among the former, the estimate of �1 is –0.18 (std.err. = 0.007); among the latter, it is –0.129 (std. err. = 0.004).

42. Online Appendix Table A.7 considers the specification separately decade bydecade. In each period we find a negative and statistically significant effect.



TA

BL

EII

I

PO

TE

NT

IAL

RE

LA

TIV

EIN

CO

ME

AN

DW

IFE’S

RE

AL

IZE

DE

AR

NIN

GS

(1)

(2)

(3)

(4)

(5)

(6)

Dep

end

ent

vari

able

:in

com

eGa

p

PrW

ifeE

arn

sMor

e–0.0

31**

*–0.0

95**

*–0.0

95**

*–0.1

09**

*–0.1

68**

*–0.1

76**

*[0

.007]

[0.0

06]

[0.0

06]

[0.0

07]

[0.0

09]

[0.0

09]

Obse

rvati

ons

5,3

06,6

82

5,3

06,6

82

5,3

06,6

64

5,3

06,6

64

1,0

49,7

93

1,0

49,7

93

R-s

qu

are

d0.0

04

0.0

06

0.0

06

0.0

50

0.0

07

0.0

13

Ad

dit

ion

al

con

trol

s:C

ubic

inln

Hu

sbIn

com

en

oyes

yes

yes

yes

yes

lnM

edia

nW

ifeP

oten

tia

l�

lnH

usb

Inco

me

no

no

yes

yes

no

no

an

yCh

ild

ren

no

no

no

yes

no

no

Wif

e’s

dem

ogra

ph

icgro

up�

Hu

sban

d’s

dem

ogra

ph

icgro

up

no

no

no

yes

no

no

PrW

ifeE

arn

sMor

eAtM

arr

iage

no

no

no

no

no

yes

Vig

inti

les

ofth

ew

ife’

san

dh

usb

an

d’s

pot

enti

al

inco

me

at

marr

iage

no

no

no

no

no

yes

Marr

iage

du

rati

onfi

xed

effe

cts

no

no

no

no

no

yes

Sam

ple

rest

rict

ion

non

en

one

non

en

one

2010su

b2010su

b

Not

es.

Data

are

from

the

1970

to2000

U.S

.C

ensu

san

d2008

to2011

AC

Ssi

ngle

-yea

rfi

les.

Th

esa

mp

leco

nsi

sts

ofco

up

les

wh

ere

bot

hth

ew

ife

an

dth

eh

usb

an

dare

bet

wee

n18

an

d65

yea

rsol

dan

dth

eh

usb

an

dis

wor

kin

g.

Th

ed

epen

den

tvari

able

inco

meG

ap

mea

sure

sth

ed

iffe

ren

cebet

wee

nth

ew

ife’

sre

ali

zed

an

dp

oten

tial

earn

ings.

Th

ek

eyin

dep

end

ent

vari

able

PrW

ifeE

arn

sMor

eis

the

pro

babil

ity

that

wif

e’s

inco

me

wou

ldex

ceed

the

hu

sban

d’s

ifh

erin

com

ew

ere

dra

wn

from

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

ew

ife’

sd

emog

rap

hic

gro

up

.ln

Hu

sbIn

com

eis

the

log

ofh

usb

an

d’s

inco

me.

lnM

edia

nW

ifeP

oten

tia

lis

the

log

ofth

em

edia

nof

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

ew

ife’

sd

emog

rap

hic

gro

up

.a

nyC

hil

dre

nis

abin

ary

vari

able

that

equ

als

1if

the

wif

ere

por

tsh

avin

gan

ych

ild

ren

,0

oth

erw

ise.

PrW

ifeE

arn

sMor

eAtM

arr

iage

isth

ep

robabil

ity

that

inco

me

dra

wn

from

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

ew

ife’

sd

emog

rap

hic

gro

up

exce

eds

inco

me

dra

wn

from

the

dis

trib

uti

onof

pos

itiv

eea

rnin

gs

inth

eh

usb

an

d’s

dem

ogra

ph

icgro

up

inth

ecl

oses

tce

nsu

syea

rto

the

yea

rof

thei

rm

arr

iage.

Marr

iage

du

rati

onfi

xed

effe

cts

are

du

mm

ies

for

nu

mber

ofyea

rsth

eco

up

leh

as

bee

nm

arr

ied

.A

llre

gre

ssio

ns

incl

ud

eco

ntr

ols

for

the

log

ofh

usb

an

d’s

inco

me,

vig

inti

les

ofth

ew

ife’

sp

oten

tial

inco

me,

wif

e’s

an

dh

usb

an

d’s

edu

cati

on(fi

ve

cate

gor

ies)

,w

ife’

san

dh

usb

an

d’s

five-

yea

rage

gro

up

,w

ife’

san

dh

usb

an

d’s

race

,yea

r,an

dst

ate

fixed

effe

cts.

Sam

ple

rest

rict

ion

‘‘2010su

b’’

ind

icate

sa

subsa

mp

leof

AC

Sw

her

eth

ed

iffe

ren

cebet

wee

nyea

rof

marr

iage

an

dth

ecl

oses

tce

nsu

syea

ris

no

mor

eth

an

five

yea

rsan

dP

rWif

eEa

rnsM

oreA

tMa

rria

ge

cou

ldbe

com

pu

ted

.S

tan

dard

erro

rsare

rep

orte

din

bra

cket

s.**

*sig

nifi

can

tat

1%

level

,**

at

5%

,*a

t10%

.


In Online Appendix Table A.8, we show that part of the effect ismoderated through the number of hours the wife works. Thischannel accounts for about a third of the negative relationshipbetween incomeGapi and PrWifeEarnsMorei.

In summary, women’s labor supply decisions seem to be dis-torted in situations where there is a threat that they mightbecome the primary breadwinner. In the next section we docu-ment some of the costs that arise when the woman does earn morethan the husband. The presence of those costs provides a poten-tial rationalization for the labor market distortions that we doc-ument here.

V. Marital Stability and Relative Income

Does relative income affect marital stability? To address thisquestion, we exploit the rich information on marital satisfactionand marital outcomes from the National Survey of Families andHouseholds (NSFH). The NSFH is a nationally representativesurvey of U.S. households and includes approximately 9,500households that were followed over three waves from 1988 to2002. We use data from the first two waves (1987–88 and 1992–94).43 We restrict our analysis to couples where at least oneperson in the household has positive labor income. Our sampleconsists of approximately 4,000 married couples.

The NSFH has three questions on marital stability. The firstquestion asks: ‘‘Taking things all together, how would you de-scribe your marriage?’’ Respondents can choose answers on ascale of 1 (very unhappy) to 7 (very happy). Close to 50 percentreport being ‘‘very happy,’’ however, so we define a binary vari-able happyMarriagei that simply indicates whether the answer is‘‘very happy.’’ The second question asks: ‘‘During the past year,have you ever thought that your marriage might be in trouble?’’We define a binary variable marriageTroublei that indicates anaffirmative response. The third question asks: ‘‘During the pastyear, have you and your husband/wife discussed the idea of sep-arating?’’ We define a binary variable discussSeparationi thatindicates the answer is affirmative.

43. We do not use Wave 3 since, due to budgetary constraints, it is based on adifferent sampling rule that substantially alters the profile of the respondents.



The NSFH also provides information on the wife’s andthe husband’s annual labor income, on the basis of which wedefine self-explanatory variables lnWifeIncomei, lnHusbIncomei,and lnTotIncomei.

44 For each couple we also computerelativeIncomei, the share of the household income earned bythe wife. In Wave 1, the mean of relativeIncomei is 0.27 and itexceeds 1

2 in 15 percent of households. We define wifeEarnsMorei

as a binary variable equal to 1 if relativeIncomei >12. Summary

statistics for the main variables used in the analysis are in OnlineAppendix Table A.9.

In Table IV, we examine how the relative income within thehousehold affects answers to survey questions about marriagequality. Our baseline specification is a linear probability model

Yi ¼ �0 þ �1 �wifeEarnsMorei

þ �2 � lnWifeIncomei þ �3 � lnHusbIncomei

þ �4 � lnTotIncomei þ �5 � Xi þ "i;

where Yi is the answer to the survey question and Xi representsnonincome controls: region fixed effects,45 indicator variables forwhether the wife is working, whether the husband is working, thewife and the husband’s race and education groups, and a qua-dratic in the wife’s and the husband’s age.46

We pool the wives’ and the husbands’ responses (and includeas a control an indicator variable for whether the respondent isthe wife or the husband). We cluster standard errors at the levelof the couple. In Online Appendix Table A.10, we examine thewives’ and the husbands’ responses separately. In short, we findthe same patterns as here for each spouse (with somewhat lessconsistent precision). One might be tempted to use the differencebetween the coefficients on the wives’ and the husbands’ re-sponses to determine whether it is the wife or the husband whodislikes the reversal of traditional gender roles. We suspect, how-ever, that such a comparison is not particularly useful. If, say, the

44. In this and the next section, we set lnWifeIncome = 0 if the wife’s income isequal to zero, and in all regressions we include an indicator variable for whether thewife’s income was zero. We apply the same procedure for the husband’s income.

45. State identifiers are not available in the public-use version of the NSFH.46. We weight the observations using the couple-level weights. The NSFH pro-

vides two sets of weights: a person-level weight and a couple-level weight. Resultsare similar whether we use no weights, person-level weights, or couple-levelweights.





husband is initially the one who is unhappy, he may start tobehave in ways that make the wife unhappy, perhaps evenmore so.47 Such a possibility echoes Al Roth’s Iron Law ofMarriage: you cannot be happier than your spouse (Roth 2008).

As column (1) of Table IV shows, spouses tend to be lesshappy with the marriage, are more likely to report that the

TABLE IV

RELATIVE INCOME AND MARITAL SATISFACTION

(1) (2) (3) (4)

Panel A: dependent variable: happyMarriagewifeEarnsMore –0.068** –0.060* –0.070* –0.065*

[0.031] [0.032] [0.036] [0.037]Observations 7,659 7,659 7,659 7,659R-squared 0.025 0.026 0.025 0.025

Panel B: dependent variable: marriageTroublewifeEarnsMore 0.082*** 0.078*** 0.079** 0.086**


Panel C: dependent variable: discussSeparationwifeEarnsMore 0.068*** 0.064*** 0.060** 0.065**


Additional controls:Cubic in lnWifeIncome and lnHusbIncome no yes no norelativeIncome no no yes yesjWife-Husb Income Rankj no no no yes

Notes. The data are from Wave 1 of the National Survey of Family and Households (NSFH). Thesample is restricted to couples where both the wife and the husband are between 18 and 65 years old andat least one person in the household has positive income. The sample includes observations from bothhusbands and wives. The dependent variables, happyMarriage, marriageTrouble, and discussSeparationare binary variables based on respondents’ answers about their marriage (details in the text).relativeIncome is the share of the household income earned by the wife. wifeEarnsMore is an indicatorvariable for whether relativeIncome> 0.5. lnWifeIncome and lnHusbIncome are the logs of the wife’s andhusband’s income, respectively. jWife-Husb Income Rankj is the absolute difference in the income ranks ofhusbands and wives in their respective income distributions. All regressions include the log of the wife’sincome, log of the husband’s income, log of the total household income, a quadratic in wife and husband’sage, indicator variables for wife and husband’s race and education (five categories), region fixed effects,and an indicator variable for whether only the wife is working or only the husband is working. Eachregression includes an indicator variable for whether the wife or the husband is the respondent and havestandard errors clustered at the level of the couple. All regressions are weighted using the Wave 1 coupleweights from NSFH. Robust standard errors are reported in brackets. ***significant at 1%, **at 5%, *at10%.

47. Sayer et al. (2011) analyze a postdivorce survey that ascertains whichspouse wanted the divorce more, but they do not examine the role of relative income.


marriage is in trouble, and are more likely to have discussed sep-aration if the wife earns more than the husband. In column (2),we add more flexible income controls, namely, cubic polynomialsof lnWifeIncomei and lnHusbIncomei. The estimates of �1 aremostly unaffected. Since gender identity is more plausibly asso-ciated with a prescription that ‘‘the husband should earn morethan the wife’’ than with a prescription that ‘‘it is better for thewife to earn 20% rather than 30% of the household income,’’ incolumn (3) we include relativeIncomei as a control. This variabledoes not predict survey responses and the estimates of �1 areagain unaffected. Hence, relative income within a household mat-ters if and only if it makes the wife the primary breadwinner.

Finally, in column (4) we include a control for the absolutedifference in income rank of the wife and the husband (definedwithin gender-specific distributions of income). Recall that stan-dard models with assortative matching predict this is the variablethat should matter. We again find no impact on �̂1: differences inincome ranks have no predictive power; what matters is whetherthe wife earns more than the husband in absolute terms.

The effects are economically significant. In our preferred spe-cification (column (3)), we find that if the wife earns more than thehusband, spouses are 7 percentage points (15%) less likely toreport that their marriage is very happy, 8 percentage points(32%) more likely to report marital troubles in the past year,and 6 percentage points (46%) more likely to have discussed sep-arating in the past year.

Next we turn away from survey data to the revealed stabilityof marriage. For each couple in Wave 1 (1987–88), we constructa binary variable divorcedi which is equal to 1 if the couple isseparated or divorced when they are reinterviewed in Wave 2(1992–94).48 In Table V, we consider the same specifications asin Table IV but with divorcedi rather than the survey responsesas the dependent variable.49 Column (1) considers the baselinespecification. Column (2) controls more flexibly for the wife’s andthe husband’s earnings. In both specifications we find that when

48. One concern is that there may be selective attrition by divorce status. Ifdivorced couples are less likely to remain in the panel, we would underestimatethe overall tendency to divorce, but the estimate of our key coefficient would beunaffected. Moreover, we find that there is no relationship between attrition andmeasures of marital stability. The overall attrition rate is about 10 percent.

49. All of the independent variables are measured in Wave 1. We use Wave 2only to identify whether the couple is still together.


the wife earns more than the husband, the likelihood of divorceincreases by about 6 percentage points ( p< .05). Since 12 percentof couples in the sample get divorced, this estimate implies thathaving the wife earn more than the husband increases the like-lihood of divorce by 50 percent. In column (3), we include a controlfor relative income. The estimate decreases slightly to about 5percentage points and becomes statistically insignificant( p = .11). Including a control for the absolute difference in wifeand husband’s income rank does not change the estimate appre-ciably (column (4)).50 Hence, as before, what seems to matter isnot relative income per se or the difference in ranks but whetherthe norm that the husband should earn more than the wife hasbeen violated.

TABLE V

RELATIVE INCOME AND DIVORCE

(1) (2) (3) (4)Dependent variable: divorced

wifeEarnsMore 0.062** 0.060** 0.048 0.051*[0.025] [0.026] [0.030] [0.030]

Observations 3,439 3,439 3,439 3,439R-squared 0.080 0.086 0.080 0.080

Additional controls:Cubic in lnWifeIncome and

lnHusbIncomeno yes no no

relativeIncome no no yes yesjWife-Husb Income Rankj no no no yes

Notes. The data are from Waves 1 and 2 of the National Survey of Family and Households (NSFH).The sample is restricted to couples where both the wife and the husband are between 18 and 65 years old(in Wave 1) and at least one person in the household has positive income. The dependent variable divorcedis an indicator for whether the couple is divorced or separated as of Wave 2. relativeIncome is the share ofthe household income earned by the wife. wifeEarnsMore is an indicator variable for whetherrelativeIncome> 0.5. lnWifeIncome and lnHusbIncome are the logs of the wife’s and husband’s income,respectively. jWife-Husb Income Rankj is the absolute difference in the income ranks of husbands andwives in their respective income distributions. All regressions include the log of the wife’s income, log ofthe husband’s income, log of the total household income, a quadratic in wife and husband’s age, indicatorvariables for wife and husband’s race and education (five categories), region fixed effects, and an indicatorvariable for whether only the wife is working or only the husband is working. All regressions are weightedusing the Wave 1 couple weights from NSFH. Robust standard errors are reported in brackets. ***signif-icant at 1%, **at 5%, *at 10%.

50. Separation or divorce occurs when the marriage fully breaks down and canbe regarded as the end-point of marital instability. Among couples who remainmarried in both survey waves, we can also examine whether wifeEarnsMore inWave 1 is associated with a deterioration in reported marital stability. We findsome suggestive evidence in this direction but most of the point estimates are notstatistically significant (see Online Appendix Table A.11).



The main concern with the interpretation of these results isthe heterogeneity of the effect by education. When we considerthe specification in column (3) separately by couples’ education,we find that wifeEarnsMore increases divorce only among themore educated couples (where both the husband and the wifehave at least some college education).

Our results in this section contribute to an existing literature insociology that examines the relationship between relative incomeand divorce. The results in this literature are somewhat mixed. Forexample, Heckert, Nowak, and Snyder (1998), Jalovaara (2003),and Liu and Vikat (2004) report that couples are more likely todivorce if the wife earns more than the husband, while Rogers(2004) argues that divorce is most likely when the couples earnaround the same amount of money.51 In contrast to this literature,we examine marital satisfaction in addition to divorce and, moreimportant, we analyze the impact of relative income conditional oneach individual’s income (and the total household income.) None ofthe aforementioned papers considers such specifications.52

VI. Home Production and Relative Income

Traditional gender roles also contain prescriptions about thedivision of chores within the household. In this section, we ex-plore whether, when the wife earns more than the husband, thecouple changes behavior at home so as to alleviate the sense ofgender role reversal.

We use data from the American Time Use Suvey (ATUS) andthe Current Population Survey (CPS), covering the years 2003 to2011. As in the previous section, we restrict our analysis to coupleswhere at least one person has positive labor income. FollowingAguiar and Hurst (2007), we define chores as ‘‘core’’ nonmarketwork (meal preparation and cleanup, doing laundry, vacuuming,etc.), time spent ‘‘obtaining goods and services’’ (e.g., grocery shop-ping), and time spent in ‘‘other’’ home production (home mainte-nance, vehicle repair, gardening, etc.). Our measure of childcareincludes primary child care (e.g., changing diapers and feeding thechild), educational child care (such as helping a child with

51. Kalmijn, Loeve, and Manting (2007) argue that the impact of relativeincome on dissolution differs across married and cohabiting couples.

52. Jalovaara (2003) reports divorce rates across all possible combinations ofthe wife’s and the husband’s income, but she employs a very rough measure thatdistinguishes between only five levels of income.


homework), and recreational child care (playing games with thechildren, taking them to the zoo, etc.). For each individual in thesample, we define totalNonMarketWorki as the weekly number ofhours spent on chores and childcare.

Variables lnWifeIncomei, lnHusbIncomei, and lnTotIncomei

are based on the weekly earnings at the main job reported in theCPS or ATUS interviews (see Online Appendix for details).53

Based on these earnings we define relativeIncomei andwifeEarnsMorei as before. Summary statistics are presented inOnline Appendix Table A.12. Wives spend an average of 36 hoursa week on chores and childcare, compared with 22.5 hours forhusbands. Mean relative income is 0.34 and the wife earnsmore than the husband in 25 percent of the couples.

Ideally, we would like to compare the wife–husband gap intime spent on home production across couples where the wifeearns more and those where she does not. Unfortunately, theATUS/CPS only includes one respondent per household. Thus,to analyze how relative income impacts the division of home pro-duction, we focus on the interaction between the impact of genderand the impact of relative income on time use. Specifically, incolumn (1) of Table VI, we consider the baseline OLS model:

totalNonMarketWorki

¼ �0 þ �1 � femalei �wifeEarnsMorei

þ �2 � femalei þ �3 �wifeEarnsMorei

þ �4 � lnWifeIncomei þ �5 � femalei � lnWifeIncomei

þ �6 � lnHusbIncomei þ �7 � femalei � lnHusbIncomei

þ �8 � lnTotIncomei þ �9 � femalei � lnTotIncomei

þ �10 � Xi þ �11 � femalei � Xi þ "i;

where Xi includes year, state, and day of the week fixed effects;indicator variables for whether only the wife is working; whetheronly the husband is working; the wife and the husband’s race andeducation groups; and a quadratic in the wife’s and the husband’sage. Our coefficient of interest is �1. A positive estimate of �1

53. The reliance on a weekly earnings measure here, which deviates from theannual labor market earnings measures we use in other sections, is due to the lackof annual earnings for the substantive subsample of ATUS respondents who did notcomplete the March CPS.




TA

BL

EV

I

RE

LA

TIV

EIN

CO

ME

AN

DT

HE

GE

ND

ER

GA

PIN

NO

NM

AR

KE

TW

OR

K

(1)

(2)

(3)

(4)

(5)

(6)

Dep

end

ent

vari

able

:T

ota

ln

onm

ark

etw

ork

(hou

rsp

erw

eek)

fem

ale�

wif

eEa

rnsM

ore

1.0

87

1.2

63*

2.1

83**

*2.2

97**

*2.9

61**

*2.7

57**

*[0

.740]

[0.7

62]

[0.7

82]

[0.7

56]

[0.8

44]

[0.9

98]

wif

eEa

rnsM

ore

0.4

60

0.1

32

–0.0

31

–0.1

47

–0.5

46

–0.4

39

[0.5

23]

[0.5

44]

[0.5

57]

[0.5

38]

[0.6

00]

[0.7

07]

Obse

rvati

ons

37,6

65

37,6

65

37,6

65

37,6

65

22,3

90

18,2

27

R-s

qu

are

d0.2

33

0.2

33

0.2

34

0.2

85

0.2

24

0.3

75

Ad

dit

ion

al

con

trol

s:C

ubic

inln

Wif

eIn

com

ean

dln

Hu

sbIn

com

en

oyes

yes

yes

yes

yes

rela

tive

Inco

me

no

no

yes

yes

yes

yes

Ch

ild

ren

con

trol

sn

on

on

oyes

yes

yes

Sam

ple

rest

rict

ion

non

en

one

non

en

one

bot

hsp

ouse

sh

ave

pos

itiv

ein

com

ew

eek

-day

only

Not

es.

Data

sou

rce

are

AT

US

/CP

Sfr

om2003

to2011.

Th

esa

mp

leis

rest

rict

edto

marr

ied

ind

ivid

uals

wh

oare

bet

wee

n18

an

d65

yea

rsol

dan

dw

hos

esp

ouse

isals

obet

wee

n18

an

d65

yea

rsol

d.

Th

esa

mp

leis

furt

her

rest

rict

edto

cou

ple

sw

her

ebot

hsp

ouse

sare

pre

sen

tat

the

tim

eof

the

AT

US

inte

rvie

w,

an

dw

her

eat

least

one

ofth

esp

ouse

sis

emp

loyed

at

the

tim

eof

the

AT

US

inte

rvie

w.

wif

eIn

com

e(h

usb

an

dIn

com

e)is

the

wif

e’s

(hu

sban

d’s

)w

eek

lyea

rnin

gs

at

main

job.

Earn

ings

are

dra

wn

from

the

AT

US

inte

rvie

wfo

rth

esp

ouse

that

com

ple

tes

that

inte

rvie

wan

dfr

omth

eC

PS

inte

rvie

wfo

rth

eot

her

spou

se;

earn

ings

from

the

CP

Sin

terv

iew

are

chan

ged

to0

ifth

at

spou

seis

not

wor

kin

gat

the

tim

eof

the

AT

US

inte

rvie

w.

rela

tive

Inco

me

isth

esh

are

ofth

eh

ouse

hol

din

com

eea

rned

by

the

wif

e.T

he

key

ind

epen

den

tvari

able

wif

eEa

rnsM

ore

isan

ind

icato

rvari

able

for

wh

eth

erre

lati

veIn

com

e>

0.5

.‘‘C

hil

dre

nco

ntr

ols’

’in

clu

de

ind

icato

rvari

able

sfo

rw

het

her

the

resp

ond

ent

has

no

chil

dre

n,

wh

eth

erth

eyou

nges

tch

ild

is3

oryou

nger

,bet

wee

n4

an

d6,

oris

old

erth

an

6.

All

regre

ssio

ns

incl

ud

eth

elo

gof

the

wif

e’s

inco

me,

log

ofth

eh

usb

an

d’s

inco

me,

log

ofth

eto

tal

hou

seh

old

inco

me,

yea

r,st

ate

,an

dd

ay

ofth

ew

eek

fixed

effe

cts,

the

wif

ean

dth

eh

usb

an

d’s

race

,a

qu

ad

rati

cin

wif

ean

dh

usb

an

d’s

age,

ind

icato

rvari

able

sfo

rth

ew

ife’

san

dth

eh

usb

an

d’s

edu

cati

ongro

up

s(fi

ve

cate

gor

ies)

,an

din

dic

ato

rvari

able

sfo

rw

het

her

only

the

hu

sban

dis

wor

kin

g,

an

dw

het

her

only

the

wif

eis

wor

kin

g.

Th

esp

ecifi

cati

ons

als

oin

clu

de

the

inte

ract

ion

ofall

thes

eco

ntr

ols

wit

han

ind

icato

rvari

able

for

wh

eth

erth

eA

TU

Sre

spon

den

tis

fem

ale

.T

he

esti

mate

dco

effi

cien

tson

fem

ale

AT

US

resp

ond

ent

ina

regre

ssio

nth

at

incl

ud

esth

esa

me

con

trol

sas

inco

lum

n(2

)bu

ton

lyin

tera

cts

wif

eEa

rnsM

ore

wit

hfe

male

AT

US

resp

ond

ents

are

,re

spec

tivel

y:

16.7

0[0

.26]

(non

mark

etw

ork

+ch

ild

care

);11.0

0[0

.22]

(non

mark

etw

ork

).E

ach

obse

rvati

onis

wei

gh

ted

usi

ng

the

AT

US

/CP

Sw

eigh

t.S

tan

dard

erro

rsare

rep

orte

din

bra

cket

s.**

*sig

nifi

can

tat

1%

level

,**

at

5%

,*a

t10%

.


would indicate that, ceteris paribus, in couples where the wifeearns more than the husband, she also spends more time onchores and childcare. In the baseline specification, the estimateof �1 is 1.09 (statistically insignificant). In column (2) we includemore flexible cubic polynomial controls for the wife’s and the hus-band’s income. The estimate of �1 is similar at 1.26.

In column (3) we include a control for relativeIncomei. Thatis particularly important in this context because standardBeckerian forces will lead the wife to do less housework whenher contribution to family income increases. With this control,the estimate of �1 increases to 2.18 and becomes highly significant( p< .01): Beckerian forces are present, but once a wife earnsmore than her husband she starts to compensate for it by spend-ing more time on chores and childcare.54

In column (4), we further control for the presence of childrenof various ages.55 The estimate remains largely unchanged.Finally, our results are robust to restricting the sample to coupleswhere both the wife and the husband have strictly positive earn-ings (column (5)) as well as time-use during weekdays only (col-umn (6)). In Online Appendix Table A.13, we separately analyzethe impact on various components of total nonmarket work. Ingeneral we find positive coefficients, but only the effects on choresare statistically significant. When we examine the effect by edu-cation group, we find a larger estimate among the couples thatare more educated but standard errors are much too large to drawany strong inference from this comparison.

In summary, it seems that women who are overly successfulin the labor force pay for this success at home to abate the rever-sal of the traditional gender roles. Akerlof and Kranton (2000)report that women do not undertake less than half of the house-work even if they work or earn more than the husband.

54. One potential source of an omitted variable bias would be that in couplesthat are ‘‘more traditional,’’ women are less likely to earn more than their husbandsand are more likely to take on a larger share of housework. This force would bias theestimate of �1 downward, in the opposite direction of our finding. Similarly, onemight be concerned that there are unobservable factors that lower the husband’sincome (below his potential and below the wife’s income) and simultaneously lowerhis ability to do household chores. This is unlikely to be important, however: whenthe husband’s realized income is further below his potential income, he spends moretime on chores.

55. Specifically, we add indicator variables for whether there is no child, theyoungest child is younger than 3, the youngest child is between 4 and 6, or theyoungest child is older than 6.



Our finding is even more striking: the gender gap in nonmarketwork is greater when the wife earns more than the husband.

As we mentioned earlier, the idea that individuals modifytheir behavior so as to fulfill gender roles has been present inthe sociology literature at least since West and Zimmerman(1987). This idea was first applied to the division of householdchores by Fenstermaker Berk (1985). More recently, Bittmanet al. (2003) report that the extent of the wife’s houseworkdecreases in relative income when she makes less than the hus-band, but this relationship reverses when relative income exceedsone half. In contrast, Gupta (2007) and Gupta and Ash (2008)argue that the number of hours the wife spends on chores issolely determined by her level of income, without any regard forher relative income in the household.56 Our findings are more inline with Bittman et al. (2003), though our approach varies fromtheirs on two dimensions: we use a different data set57 and, moreimportant, our specification includes controls for each individ-ual’s income.58 Finally, in the next section we build on the exist-ing evidence by documenting how relative income impactshousehold production within couples over time.

VII. Panel Data Evidence

In this final section, we analyze labor force participation,marital stability, and home production using within-couple vari-ation in relative income. We combine data from the family files(1968–2009) and the marital history file (2009) of the PSID (seeOnline Appendix for details of sample construction). We focus ouranalysis on couple-year observations where at least one of thespouses has positive earnings in the year(s) where relative earn-ings measures are used as independent variables. Summary sta-tistics are in Online Appendix Table A.14.

56. Cooke (2006) finds that among U.S. couples where the wife earns more thanthe husband, the likelihood of divorce is lower if the wife engages in ‘‘compensatorybehavior’’—that is, if she does a greater share of the housework. Greenstein (2000)argues that both men and women adjust their household production to neutralizedeviance from traditional gender roles.

57. Bittman et al. (2003) draw on an Australian time-use survey from 1992.58. Including these controls is particularly important given the results in

Gupta (2007) and Gupta and Ash (2008).




VII.A. Wife’s Labor Force Participation

Our analysis of wives’ labor force participation in Section IVwas based on an imputed likelihood that the wife would earnmore than her husband. The panel setting offers an alternativeto this imputation exercise. Because we observe couples overtime, we can ask whether an actual realization of the wife earningmore than her husband is predictive of the wife subsequentlyleaving the workforce. Specifically, we estimate a linear probabil-ity model with couple fixed effects regressing the wife’s laborforce participation in year t on whether the wife earned morethan the husband in year t – 1 (controlling for lnWifeIncome,lnHusbandIncome, and lnTotalIncome in t – 1 as well as for anumber of other covariates).59

As shown in Table VII, Panel A, column (1), if a wife earnsmore than her husband, she is 1.9 percentage points less likely bein the labor force the following year ( p< .01). Moreover, in resultsnot reported here, we find that if a wife earns more than herhusband, she starts to work fewer hours the following year.These results echo Winslow-Bowe’s (2006) finding that whenthe wife earns more than the husband, that situation typicallydoes not persist for more than a year or two.

In column (2), we include cubics of each spouse’s income ascontrols; the estimate is unaffected. When we add relativeIncomein t – 1 as a control in column (3), the effect falls down to 1.3 per-centage points but remains highly significant ( p< .01). In column(4) we add a vector of controls for the presence of children; theestimate is barely affected. Finally, as column (5) shows, the spe-cifications with couple fixed effects yields the same estimate asthe cross-sectional relationship within the PSID. This furthersupports the view that selection is not driving the results inSection IV.

VII.B. Divorce

We study the impact of relative income on divorce using sim-ilar specifications as in the previous subsection, but we allow for alonger lag structure since divorce may take time to be completed.

59. Namely, year and state fixed effects, an indicator variable for only the wifeworking in t – 1, an indicator variable for only the husband working in t – 1, andquadratic functions of the wife’s and the husband’s age.


Hence, we regress divorce in year t on wifeEarnsMore in both t – 1and t – 2.60 When we construct the dependent variable, we codethe couple as having divorced if they are no longer listed as mar-ried and there is no explicit mention of widowhood.61

Panel B of Table VII shows the results. In all of thespecifications, the estimated effect of wifeEarnsMore on divorceis positive but typically imprecise. The lack of statistical power isalso evident in the fact that the relationship is not significanteven in the cross section. With this important caveat in mind,the point estimates are substantial: the likelihood of divorce in-creases by 25 percent if the wife earned more than her husbandtwo years ago.

VII.C. Home Production

The PSID asks the following question of both husbands andwives: ‘‘About how much time do you spend on housework in anaverage week—I mean time spent cooking, cleaning, and otherwork around the house?’’ Relative to ATUS, this question is morevague about the exact content of the home production activities,but the PSID has two important advantages: (i) both spouses areasked to answer the question, which allows us to directly measurethe gender gap in home production within a household, and (ii)the panel nature of the data allows us to examine how changes inrelative income affect this gap.

We estimate the exact same specifications as in subsectionVII.A but with the gap in housework as the dependent variable.As Panel C of Table VII shows, within a given couple, whenthe wife starts to earn more than the husband she takes on rel-atively more housework. In Online Appendix Table A.15, weshow that the effect is driven by the wife doing more and thehusband doing less housework, though the former channelplays a larger role.

60. Accordingly, we control for each spouse’s income and total income in t – 2 aswell as t – 1, and for year and state fixed effects, indicator variables for only the wifeworking in t – 1 and t – 2, indicator variables for only the husband working in t – 1and t – 2, and quadratic functions of the wife’s and the husband’s age.

61. See the Online Appendix for a detailed discussion of how we construct thedivorce variable. Our results are similar if we alternatively code the couple ashaving divorced only if divorce or separation was explicitly reported. We preferour measure since it yields an average divorce rate of 2% per year, roughly similarto that observed in other data sets. (Under the alternative measure, average divorcerate is only 0.8% per year.)




TA

BL

EV

II

RE

LA

TIV

EIN

CO

ME

AN

DL

AB

OR

FO

RC

EP

AR

TIC

IPA

TIO

N,

DIV

OR

CE,

AN

DG

EN

DE

RG

AP

INH

OU

SE

WO

RK,

PA

NE

LD

AT

AE

VID

EN

CE

(1)

(2)

(3)

(4)

(5)

Pan

elA

:re

lati

ve

inco

me

an

dla

bor

forc

ep

art

icip

ati

onD

epen

den

tvari

able

:W

ife

inth

ela

bor

forc

ein

tw

ifeE

arn

sMor

ein

t–1

–0.0

19**

*–0.0

18**

*–0.0

13**

*–0.0

12**

*–0.0

14**

*[0

.004]

[0.0

04]

[0.0

04]

[0.0

04]

[0.0

03]

Obse

rvati

ons

110,1

23

110,1

23

110,1

23

110,1

23

110,1

23

R-s

qu

are

d0.7

18

0.7

20

0.7

20

0.7

21

0.6

62

Pan

elB

:re

lati

ve

inco

me

an

dm

ari

tal

stabil

ity

Dep

end

ent

vari

able

:D

ivor

ced

int

wif

eEa

rnsM

ore

int–

10.0

03

0.0

03

0.0

04

0.0

04

0.0

00

[0.0

02]

[0.0

03]

[0.0

03]

[0.0

03]

[0.0

03]

wif

eEa

rnsM

ore

int–

20.0

05*

0.0

05*

0.0

05

0.0

05

0.0

02

[0.0

03]

[0.0

03]

[0.0

03]

[0.0

03]

[0.0

03]

Obse

rvati

ons

72,1

69

72,1

69

72,1

69

72,1

69

72,1

69

R-s

qu

are

d0.4

18

0.4

18

0.4

18

0.4

18

0.0

13


TA

BL

EV

II

(CO

NT

INU

ED)

(1)

(2)

(3)

(4)

(5)

Pan

elC

:re

lati

ve

inco

me

an

dgen

der

gap

inh

ouse

wor

kD

epen

den

tvari

able

:W

ife

hou

sew

ork

int

–H

usb

an

dh

ouse

wor

kin

tw

ifeE

arn

sMor

ein

t0.0

59

0.4

64*

1.0

48**

*0.9

04**

*1.1

41**

*[0

.250]

[0.2

61]

[0.2

75]

[0.2

73]

[0.3

10]

Obse

rvati

ons

67,6

03

67,6

03

67,6

03

67,6

03

67,6

03

R-s

qu

are

d0.5

86

0.5

87

0.5

87

0.5

95

0.3

76

Ad

dit

ion

al

con

trol

s:C

oup

lefi

xed

effe

cts

yes

yes

yes

yes

no

Cu

bic

inln

Wif

eIn

com

ean

dln

Hu

sba

nd

Inco

me

no

yes

yes

yes

yes

rela

tive

Inco

me

no

no

yes

yes

yes

Ch

ild

ren

con

trol

sn

on

on

oyes

yes

Not

es.

Th

ed

ata

are

from

the

PS

ID,

1968

to2009.

See

On

lin

eA

pp

end

ixfo

rsa

mp

leco

nst

ruct

ion

for

each

pan

el.

Th

esa

mp

leis

rest

rict

edto

cou

ple

sw

her

ebot

hth

ew

ife

an

dth

eh

usb

an

dare

bet

wee

n18

an

d65

yea

rsof

age

an

dw

her

eat

least

one

ofth

esp

ouse

sis

wor

kin

gin

the

yea

rin

com

ean

dre

lati

ve

inco

me

mea

sure

sare

use

din

the

regre

ssio

n.

Wif

ein

the

labor

forc

ein

tis

ad

um

my

vari

able

that

equ

als

1if

the

wif

eis

inth

ela

bor

forc

eat

tim

et,

0ot

her

wis

e.D

ivor

ced

int

isa

du

mm

yvari

able

that

equ

als

1if

the

cou

ple

isd

ivor

ced

orse

para

ted

at

tim

et,

0ot

her

wis

e.W

ife

hou

sew

ork

int

–H

usb

an

dh

ouse

wor

kin

tis

the

dif

fere

nce

bet

wee

nth

ew

ife’

san

dth

eh

usb

an

d’s

wee

kly

hou

rsof

hom

ep

rod

uct

ion

at

tim

et.

Wif

eh

ouse

wor

kin

tis

the

wif

e’s

wee

kly

hou

rsof

hom

ep

rod

uct

ion

at

tim

et.

Hu

sba

nd

hou

sew

ork

int

isth

eh

usb

an

d’s

wee

kly

hou

rsof

hom

ep

rod

uct

ion

at

tim

et.

wif

eEa

rnsM

ore

int

isan

ind

icato

rvari

able

that

equ

als

1if

rela

tive

Inco

me>

0.5

at

tim

et.

rela

tive

Inco

me

isth

esh

are

ofth

eh

ouse

hol

din

com

eea

rned

by

the

wif

e.ln

Wif

eIn

com

eis

the

log

ofth

ew

ife’

sla

bor

mark

etea

rnin

gs;

lnH

usb

an

dIn

com

eis

the

log

ofth

eh

usb

an

d’s

labor

mark

etea

rnin

gs.

‘‘Ch

ild

ren

con

trol

s’’

incl

ud

ein

dic

ato

rvari

able

sfo

rw

het

her

the

resp

ond

ent

has

no

chil

dre

n,

wh

eth

erth

eyou

nges

tch

ild

is3

oryou

nger

,bet

wee

n4

an

d6,

old

erth

an

6.

All

regre

ssio

ns

incl

ud

eth

elo

gof

the

wif

e’s

inco

me,

the

log

ofth

eh

usb

an

d’s

inco

me,

the

log

ofth

eco

up

le’s

inco

me,

an

ind

icato

rfo

rw

het

her

only

the

wif

eis

wor

kin

g,

an

ind

icato

rfo

rw

het

her

only

the

hu

sban

dis

wor

kin

g,

aqu

ad

rati

cin

wif

e’s

an

dh

usb

an

d’s

age,

yea

rfi

xed

effe

cts

an

dst

ate

fixed

effe

cts.

All

ind

epen

den

tvari

able

sre

late

dto

inco

me

ina

regre

ssio

n(i

ncl

ud

ing

wh

eth

eron

lyth

ew

ife

isw

ork

ing

orw

het

her

only

the

hu

sban

dis

wor

kin

g)

are

mea

sure

din

the

sam

eyea

ror

yea

rsas

Wif

eEa

rnsM

ore

inth

at

regre

ssio

n.

Sta

nd

ard

erro

rsare

clu

ster

edat

the

cou

ple

level

an

dare

rep

orte

din

bra

cket

s.**

*sig

nifi

can

tat

1%

level

,**

at

5%

,*a

t10%

.



VIII. Conclusion

We show that the social norm ‘‘a man should earn more thanhis wife’’ influences the distribution of relative income withinhouseholds, the patterns of marriage and divorce, women’slabor supply, and the division of home production activities be-tween husbands and wives.

By definition, the norm that we focus on here would be of nopractical relevance in a world where a woman could never earnmore than her husband. The relative gains in women’s labormarket opportunities over the past half century have thusmade this aspect of gender identity increasingly relevant. Wesuspect that these changes were particularly important becausethey happened quickly in comparison to the slow-moving socialnorms and concepts of gender.

While our empirical work focuses on the United States, rapidgains in women’s labor market opportunities are not unique tothis country. Even more rapid changes have taken place in devel-oped Asian countries, such as Korea and Japan. At the same time,these Asian countries have experienced large declines in mar-riage rates and fertility among educated women. As suggestedby Hwang (2012), the interaction of economic growth and inter-generational transmission of gender attitudes might play an im-portant part in these developments.

In future work, we would like to better understand thelong-run determinants of gender identity. While the evidencein this article suggests that the behavioral prescription that‘‘a man should earn more than his wife’’ helps explain economicand social outcomes even in the most recent decade, this doesnot imply that this prescription is as strong today as it wasin the past. How are gender identity norms evolving in theface of market forces that are making those norms increasinglycostly?

University of Chicago

University of Chicago

National University of Singapore

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qje.oxfordjournal.org).



References

Aguiar, Mark, and Erik Hurst, ‘‘Measuring Trends in Leisure: The Allocation of TimeOver Five Decades,’’ Quarterly Journal of Economics, 122 (2007), 969–1006.

Aizer, Anna, ‘‘The Gender Wage Gap and Domestic Violence,’’ American EconomicReview, 100 (2010), 1847–1859.

Akerlof, George A., and Rachel E. Kranton, ‘‘Economics and Identity,’’ QuarterlyJournal of Economics, 115 (2000), 715–753.

———, Identity Economics: How Our Identities Affect Our Work, Wages, and Well-Being (Princeton, NJ: Princeton University Press, 2010).

Altonji, Joseph G., Todd E. Elder, and Christopher R. Taber, ‘‘Selection onObserved and Unobserved Variables: Assessing the Effectiveness ofCatholic Schools,’’ Journal of Political Economy, 113 (2005), 151–184.

Autor, David, ‘‘U.S. Labor Market Challenges over the Longer Term,’’ Paper pre-pared for the Federal Reserve Board of Governors (2010).

Baker, Matthew J., and Joyce P. Jacobsen, ‘‘Marriage, Specialization, and theGender Division of Labor,’’ Journal of Labor Economics, 25 (2007), 763–793.

Bartik,Timothy, Who BenefitsFrom State and Local EconomicDevelopment Policies?(Kalamazoo, MI: W. E. Upjohn Institute for Employment Research, 1991).

Becker, Gary S., ‘‘A Theory of Marriage: Part I,’’ Journal of Political Economy, 81(1973), 813–846.

———, ‘‘A Theory of Marriage: Part II,’’ Journal of Political Economy, 82 (1974),S11–S26.

Bertrand, Marianne, ‘‘New Perspectives on Gender,’’ in Handbook of LaborEconomics, O. Ashenfelter, and D. Card, eds. Amsterdam: North-Holland,1545–1592.

Bittman, Michael, Paula England, Liana Sayer, Nancy Folbre, andGeorge Matheson, ‘‘When Does Gender Trump Money? Bargaining and Timein Household Work,’’ American Journal of Sociology, 109 (2003), 186–214.

Black, Sandra E., and Elizabeth Brainerd, ‘‘The Impact of Globalization onGender Discrimination,’’ Industrial and Labor Relations Review, 57 (2004),540–559.

Black, Sandra E., and Chinhui Juhn, ‘‘The Rise of Female Professionals: Women’sResponse to Rising Skill Demand,’’ American Economic Review, 90 (2000),450–455.

Black, Sandra E., and Philip E. Strahan, ‘‘The Division of Spoils: Rent-Sharingand Discrimination in a Regulated Industry,’’ American Economic Review, 91(2001), 814–831.

Blau, Francine D., and Lawrence M. Kahn, ‘‘The U.S. Gender Pay Gap in the1990s: Slowing Convergence,’’ Industrial and Labor Relations Review, 60(2006), 45–66.

Charles, Kerwin K., and Ming Ching Luoh, ‘‘Male Incarceration, the MarriageMarket, and Female Outcomes,’’ Review of Economics and Statistics, 92(2010), 614–627.

Cooke, Lynn Prince, ‘‘ ‘Doing’ Gender in Context: Household Bargaining and Riskof Divorce in Germany and the United States,’’ American Journal ofSociology, 112 (2006), 442–472.

Fenstermaker Berk, Sarah, The Gender Factory: The Apportionment of Work inAmerican Households (New York: Plenum Press, 1985).

Fernandez, Raquel, Alessandra Fogli, and Claudia Olivetti, ‘‘Mothers and Sons:Preference Development and Female Labor Force Dynamics,’’ QuarterlyJournal of Economics, 119 (2004), 1249–1299.

Fisman, Raymond, Sheena S. Iyengar, Emir Kamenica, and Itamar Simonson,‘‘Gender Differences in Mate Selection: Evidence from a Speed DatingExperiment,’’ Quarterly Journal of Economics, 121 (2006), 673–697.

———, ‘‘Racial Preferences in Dating,’’ Review of Economic Studies, 75 (2008),117–132.

Fortin, Nicole, ‘‘Gender Role Attitudes and Women’s Labour Market Outcomes acrossOECD Countries,’’ Oxford Review of Economic Policy, 21 (2005), 416–438.

Fortin, Nicole, ‘‘Gender Role Attitudes and Women’s Labor Market Participation:Opting Out, AIDS, and the Persistent Appeal of Housewifery,’’ University ofBritish Columbia Working Paper, 2009.


Goldin, Claudia, and Lawrence F. Katz, ‘‘The Power of the Pill: OralContraceptives and Women’s Career and Marriage Decisions,’’ Journal ofPolitical Economy, 110 (2002), 730–770.

Goldin, Claudia, Lawrence F. Katz, and Ilyana Kuziemko, ‘‘The Homecoming ofAmerican College Women: The Reversal of the Gender Gap in College,’’Journal of Economic Perspectives, 20 (2006), 133–156.

Greenstein, Theodore, ‘‘Economic Dependence, Gender, and the Division of Laborin the Home: A Replication and Extension,’’ Journal of Marriage and Family,62 (2000), 322–335.

Greenwood, Jeremy, Ananth Seshadri, and Mehmet Yorukoglu, ‘‘Engines ofLiberalization,’’ Review of Economic Studies, 72 (2005), 109–133.

Gupta, Sanjiv, ‘‘Autonomy, Dependence, or Display,’’ Journal of Marriage andFamily, 69 (2007), 399–417.

Gupta, Sanjiv, and Michael Ash, ‘‘Whose Money, Whose Time? A NonparametricApproach to Modeling Time Spent on Housework in the United States,’’Feminist Economics, 14 (2008), 93–120.

Heckert, D. Alex, Thomas C. Nowak, and Kay A. Snyder, ‘‘The Impact ofHusbands’ and Wives’ Relative Earnings on Marital Disruption,’’ Journalof Marriage and the Family, 60 (1998), 690–703.

Hochschild, Arlie, and Anne Machung, The Second Shift (New York: Viking, 1989).Hwang, Jisoo ‘‘Housewife, ‘Gold Miss,’ and Equal: The Evolution of Educated

Women’s Role in Asia and the U.S.,’’ Working Paper, Harvard University, 2012.Jalovaara, Marika, ‘‘The Joint Effects of Marriage Partners’ Socioeconomic

Positions on the Risk of Divorce,’’ Demography, 40 (2003), 67–81.Kalmijn, Matthijs, Anneke Loeve, and Dorien Manting, ‘‘Income Dynamics in

Couples and the Dissolution of Marriage and Cohabitation,’’ Demography,44 (2007), 159–179.

Liu, Guiping, and Andres Vikat, ‘‘Does Divorce Risk Depend on Spouses’ RelativeIncome? A Register-Based Study of First Marriages in Sweden in 1981–1998,’’ MPIDR Working Paper WP 2004-010, 2004.

Loughran, David S., ‘‘The Effect of Male Wage Inequality on Female Age at FirstMarriage,’’ Review of Economics and Statistics, 84 (2002), 237–251.

Lundberg, Shelly, and Robert A. Pollak, ‘‘Bargaining and Distribution inMarriage,’’ Journal of Economic Perspectives, 10 (1996), 139–158.

McCrary, Justin, ‘‘Testing for Manipulation of the Running Variable in theRegression Discontinuity Design: A Density Test,’’ Journal ofEconometrics, 142 (2008), 698–714.

Oppenheimer, Valerie K., ‘‘Women’s Employment and the Gain to Marriage: TheSpecialization and Trading Model,’’ Annual Review of Sociology, 23 (1997),431–453.

Pierce, Lamar, Michael S. Dahl, and Jimmi Nielsen, ‘‘In Sickness and in Wealth:Psychological and Sexual Costs of Income Comparison in Marriage,’’Personality and Social Psychology Bulletin, 39 (2013), 359–374.

Rogers, Stacy J., ‘‘Dollars, Dependency, and Divorce: Four Perspectives on theRole of Wives’ Income,’’ Journal of Marriage and Family, 66 (2004), 59–74.

Roth, Alvin E., ‘‘Deferred Acceptance Algorithms: History, Theory, Practice, andOpen Questions,’’ International Journal of Game Theory, 36 (2008), 537–569.

Sayer, Liana C., Paula England, Paul Allison, and Nicole Kangas, ‘‘She Left, HeLeft: How Employment and Satisfaction Affect Men’s and Women’s Decisionsto Leave Marriage,’’ American Journal of Sociology, 116 (2011), 1982–2018.

Stuart, Colleen H., Sue Moon, and Tiziana Casciaro, ‘‘The Oscar Curse: StatusDynamics and Gender Differences in Marital Survival,’’ Working Paper,Carnegie Mellon University, 2011.

Weinberg, Bruce A., ‘‘Computer Use and the Demand for Female Workers,’’Industrial and Labor Relations Review, 53 (2000), 290–308.

Weiss, Yoram, ‘‘The Formation and Dissolution of Families: Why Marry? WhoMarries Whom?,’’ in Handbook of Population and Family Economics, M.R. Rosenzweig, and O. Stark, eds. Amsterdam: North-Holland, 81–123.

West, Candace, and Don H. Zimmerman, ‘‘Doing Gender,’’ Gender & Society, 1(1987), 125–151.

Winslow-Bowe, Sarah, ‘‘The Persistence of Wives’ Income Advantage,’’ Journal ofMarriage and Family, 68 (2006), 824–842.


gender identity and relative income within households - faculty

Documents