Money, Work, and Marital Stability

American Sociological Review

2016, Vol. 81(4) 696 –719

Association 2016

DOI: 10.1177/0003122416655340

http://asr.sagepub.com

Profound social and economic changes have

occurred in the U.S. family over the past half

century. For couples marrying for the first time

in the 1960s, approximately 30 percent divorced

within 15 years, and divorce risks are even

higher for couples in more recent marriage

cohorts (Stevenson and Wolfers 2007). A core

unresolved question is how trends in marital

stability relate to changing family and economic

circumstances. Have wives’ greater earnings

power and work experience increased divorce

by reducing the costs of exiting bad marriages?

Are strained household finances associated with

heightened risk of divorce? Or do spouses’ work

and earnings patterns alter marital stability by

conveying signals about whether each partner is

fulfilling the implicit, symbolic, gendered terms

of the marital contract?

Each of these questions expresses a differ-

ent perspective on the link between work,

money, and the risk of divorce. The economic

independence perspective predicts that

marriages will be more likely to dissolve

655340ASRXXX10.1177/0003122416655340American Sociological ReviewKillewald

2016

Harvard University

Corresponding Author:

Alexandra Killewald, 436 William James Hall,

Harvard University, 33 Kirkland St., Cambridge,

MA 02138

E-mail: [email protected]

Money, Work, and Marital

Stability: Assessing Change

in the Gendered Determinants

of Divorce

Alexandra Killewald

Abstract

Despite a large literature investigating how spouses’ earnings and division of labor relate to their

risk of divorce, findings remain mixed and conclusions elusive. Core unresolved questions

are (1) whether marital stability is primarily associated with the economic gains to marriage or

with the gendered lens through which spouses’ earnings and employment are interpreted and

(2) whether the determinants of marital stability have changed over time. Using data from the

1968 to 2013 waves of the Panel Study of Income Dynamics, I consider how spouses’ division

of labor, their overall financial resources, and a wife’s ability to support herself in the event

of divorce are associated with the risk of divorce, and how these associations have changed

between couples married before and after 1975. Financial considerations—wives’ economic

independence and total household income—are not predictive of divorce in either cohort.

Time use, however, is associated with divorce risk in both cohorts. For marriages formed

after 1975, husbands’ lack of full-time employment is associated with higher risk of divorce,

but neither wives’ full-time employment nor wives’ share of household labor is associated

with divorce risk. Expectations of wives’ homemaking may have eroded, but the husband

breadwinner norm persists.

Keywords

divorce, family, gender, earnings, working life

Killewald 697

when the costs of exiting the marriage are low.

The financial strain perspective argues that

limited financial resources stress marriages

and increase the risk of divorce. The gendered

institution perspective suggests that marriages

will be more stable when spouses conform to

the gendered expectations of husbands and

wives. The financial strain and economic

independence perspectives suggest that mate-

rial circumstances, within married couples

and in the event of divorce, respectively, shape

marital stability. The gendered institution per-

spective, by contrast, suggests that money and

work have implications for marital stability

primarily because of their symbolic content.

Given dramatic changes over the second half

of the twentieth century in women’s employ-

ment, education, household labor time, mar-

riage timing, divorce rates, and gender role

attitudes (Bianchi et al. 2012; Fitch and Ruggles

2000; Goldin 2006; Stevenson and Wolfers

2007; Thornton and Young-DeMarco 2001), the

circumstances that hold marriages together or

pull them apart may have changed substantially

during this period. I hypothesize that the effects

of financial characteristics—economic inde-

pendence and financial strain—on divorce risk

are likely to remain stable across time. How-

ever, I hypothesize that the gendered expecta-

tions of spouses have changed across marital

cohorts, with the result that behaviors perceived

as deviant in earlier cohorts may be normalized

in later cohorts. In particular, I predict that the

wife homemaker norm has become less impor-

tant for marital stability, while the husband

breadwinner norm has remained strong. Evalu-

ating change across marriage cohorts in the

determinants of marital stability recognizes that

marriage and its associated expectations for

spouses are embedded in broader, evolving gen-

der structures (Risman 2011).

Despite considerable research, empirical

support for the various perspectives linking

money and work to divorce has been character-

ized as “inconclusive” (Sayer and Bianchi

2000:910), “contradictory” (Dechter 1992:1),

“mixed” (Brines and Joyner 1999:338; Oppen-

heimer 1997:442; South 2001:226), and

“inconsistent” (Ono 1998:675; Sayer et al.

2011:1990), and conclusions regarding these

associations as “elusive” (Rogers 2004:59).

The difficulty adjudicating among competing

perspectives is in part due to the challenge of

separately measuring each. Wives’ economic

independence is often measured with either

wives’ earnings or their employment, which are

also measures of households’ financial strain

and conformity to a gender-traditional division

of labor. I instead construct a measure of wives’

economic independence based on the economic

well-being of divorced peers, allowing me to

distinguish among the competing theories of

marital stability.

THEORETICAL FRAMEWORK

Conceptually, divorce occurs when at least

one partner believes she will be better off

divorced than remaining married; the risk of

divorce depends on the gains from marriage

(Becker, Landes, and Michael 1977). The

economic independence perspective hypoth-

esizes that divorce rates increase when part-

ners depend less on marriage financially,

allowing spouses to exit unhappy marriages

(Ruggles 1997; Sayer et al. 2011; Schoen

et al. 2002). Wives are likely to be more eco-

nomically dependent on their husbands when

they have accumulated less work experience,

but economic independence also depends on

other aspects of women’s earnings potential,

such as education and occupation, as well as

factors such as child support policies or gov-

ernment support to low-income families. Evi-

dence for the economic independence

perspective for couples in the United States is

mixed, with some scholars finding support

(Dechter 1992; Heckert, Nowak, and Snyder

1998; Ruggles 1997; Sayer et al. 2011;

Schoen et al. 2002; South 2001; Teachman

2010) and others not (Rogers and DeBoer

2001; Sayer and Bianchi 2000). Although it

has received less attention, the economic

independence perspective suggests that men’s

divorce decisions are likewise affected by

their expected economic well-being in the

event of divorce (noted in Sayer et al.

2011:1987).

A second perspective suggests that the cur-

rent income of both spouses stabilizes

698 American Sociological Review 81(4)

marriages by reducing financial strain (Brines

and Joyner 1999; Dechter 1992), for example,

by allowing couples to outsource household

labor, reduce conflict, and increase leisure

time. Through this pathway, wives’ income

may reduce divorce risk, especially when

husbands’ earnings are low (Ono 1998). There

is some evidence that the risk of divorce is

higher for low-income couples (Brines and

Joyner 1999; Dechter 1992), but others find

no such association (Heckert et al. 1998; Sch-

oen et al. 2002; Tzeng and Mare 1995).

The economic independence perspective is

gendered in its implications, because wives

depend more on marriage for their financial

well-being, on average. Yet neither it nor the

financial strain perspective suggests that a

gendered interpretation of spouses’ earnings,

housework, or employment shapes marital

stability. In other words, given the economic

independence of two spouses and their house-

hold income, the preceding perspectives do

not suggest that the risk of divorce depends

on knowing which value belongs to which

spouse.

The gendered institution perspective, by

contrast, predicts that divorce is more likely

when spouses’ employment and earnings vio-

late gendered norms of behavior (Sayer et al.

2011). This is a cultural rather than purely

economic perspective; it draws on the logic of

“doing gender,” in which time in paid and

unpaid labor does not simply produce a pay-

check or clean home, but is a way that indi-

viduals produce and enact gender (Berk 1985;

West and Zimmerman 1987). The gendered

institution perspective is fundamentally dis-

tinct from the economic independence and

financial strain perspectives, because it theo-

rizes divorce as determined by work and

money not because of their financial implica-

tions, but because of the symbolic lens

through which they are interpreted and, spe-

cifically, whether they conform to the gen-

dered expectations of what it means to be a

good wife or a good husband.

Of course, what work behaviors or earnings

outcomes are considered gender-conforming

or gender-deviant is likely to change over

time (Risman 2011)—a point I discuss in

more detail in the next section. Thus, there is

no single test of the gendered institution per-

spective, which may take multiple forms. One

possibility is that wives’ employment is non-

normative. Some scholars have found a posi-

tive association between wives’ employment

and the risk of divorce in the United States

(Brines and Joyner 1999; South 2001; Spitze

and South 1985; Tzeng 1992), but others not

(Schoen, Rogers, and Amato 2006), or only

for couples in unhappy marriages (Sayer et al.

2011; Schoen et al. 2002).

Prior research on the determinants of

divorce disproportionately focuses on the

implications of wives’ employment and earn-

ings. But gendered expectations of men’s

labor affect marriages, too. Wage-earning

remains highly normative for married men

(Nock 1998), and men’s employment, earn-

ings, and economic potential are positively

associated with marriage formation (Oppen-

heimer, Kalmijn, and Lim 1997; Sweeney

2002; Xie et al. 2003) and stability (Ono

1998; Ruggles 1997; Sayer et al. 2011;

Schoen et al. 2002; Tzeng 1992). Beyond any

effects via financial strain, the gendered insti-

tution perspective hypothesizes that hus-

bands’ unemployment strains marriages by

violating the implicit terms of the marital

contract (Cherlin 1979).

In the domain of unpaid labor, the doing

gender perspective suggests that the perfor-

mance of housework is part of the production

of gender for women, and its avoidance is

associated with the production of masculinity.

Wives continue to perform the majority of

housework and childcare (Bianchi et al. 2012)

and to perceive that ultimate responsibility

for its completion falls to them (Hochschild

1989; Stone 2007). This suggests that mar-

riages are more stable when wives spend

more time in household labor and husbands

less. By contrast, Cooke (2006) finds that, in

the United States, a larger share of housework

done by the husband is associated with

increases in marital stability for all but the

most nontraditional divisions of unpaid labor.

In summary, I consider three possible man-

ifestations of the gendered institution perspec-

tive, with each behavior potentially decreasing

Killewald 699

the risk of divorce: wife’s nonemployment or

less than full-time employment, husband’s

full-time employment, and wife’s share of

household labor.

Other scholars sometimes use wives’ rela-

tive earnings, or whether the wife earns more

than her husband, as measures of the gendered

institution perspective, with some finding that

marriages in which wives earn more than half

the income are less stable (Bertrand, Kamen-

ica, and Pan 2015; Heckert et al. 1998; Teach-

man 2010), but others not (Schoen et al. 2002).

One challenge of this approach is that female-

breadwinner couples include both dual-earner

couples, in which she out-earns him, and

couples in which the wife is employed but not

the husband. These two types of couples may

be very different in financial circumstances

and the stability of the wife-breadwinner sta-

tus. Furthermore, the marital stability of the

latter group is affected both by wife-breadwinner

status and husband nonemployment. For sim-

plicity, I focus on employment status in the

main analyses. Supplementary analyses, dis-

cussed following the main results, show no

evidence that spouses’ relative earnings are

associated with the risk of divorce, net of

spouses’ employment statuses.

Change in Marriage Foundations

over Time

Marriage is a social institution, and thus both

rates of marital stability and its determinants

may vary across time and place. In the United

States, over the second half of the twentieth

century, women’s college completion rates

caught and passed those of men, their labor

force participation increased dramatically, and

the gender earnings gap among full-time work-

ers narrowed (Goldin 2006). At home, wives’

average time in unpaid labor declined substan-

tially (Bianchi et al. 2012). Couples married

later and were more likely to divorce (Fitch

and Ruggles 2000; Stevenson and Wolfers

2007), women’s earnings became positively

associated with marriage formation (Sweeney

2002), and gender role attitudes became more

egalitarian ( Thornton and Young-DeMarco

2001).

What are the likely implications of these

changes for the associations between the risk

of divorce and spouses’ labor, income, and

economic independence? I do not hypothe-

size any change in the importance of wives’

economic independence or financial strain.

Wives’ greater labor supply and education

among later marriage cohorts may increase

their economic independence, but there is no

reason to think that the consequences of these

economic factors have changed.

However, I hypothesize that changes in the

social and economic context in which mar-

riages take place have changed the terms of

the marital contract, changing the gendered

lens through which spouses’ labor is inter-

preted. Risman (2011:19–20) argues that each

society’s gender structure “shape[s] the social

roles women and men are expected to follow,

what ‘doing’ gender means in any given inter-

actional encounter, and how marriage is

understood and defined.” What it means to

“do gender” is therefore context-specific

(Cooke 2006). The last half-century was a

time of changing gender structures, suggest-

ing that the ways paid and unpaid labor contri-

butions to marriage by husbands and wives

are interpreted may have changed. For con-

temporary U.S. couples, the husband-bread-

winner/wife-homemaker household may be

neither economically strategic nor preferred

(Oppenheimer 1997; Sayer and Bianchi 2000;

Sayer et al. 2011). As women’s paid work

lives come to resemble men’s, wives’ partici-

pation in full-time paid labor may be one way

contemporary couples “undo” gender ( Risman

2011). For contemporary couples, wives’ full-

time employment is not rare, is not expected

to be inconsistent with gendered norms of

marriage, and is not expected to increase the

risk of divorce. For earlier marriage cohorts,

when wives’ employment was less common, it

is expected to increase the risk of divorce.

Couples’ division of unpaid labor has also

changed. In 2009 to 2010, wives’ average

time in housework was 1.7 times that of hus-

bands, compared to 3.9 times as much in

1975 (Bianchi et al. 2012). I expect that hus-

bands’ housework participation, like wives’

employment, has become more normative in

700 American Sociological Review 81(4)

recent marriage cohorts. For marriages

formed in recent decades, I expect that wives’

higher share of housework is not associated

with greater marital stability. In earlier

cohorts, however, I hypothesize that mar-

riages are more stable when the wife per-

forms a greater share of household labor.

Even in recent marriage cohorts, marriage

remains a gendered institution. As Sayer and

colleagues (2011) note, the male-breadwinner

norm has been more durable than the female-

homemaker norm. Focusing on the effects of

women’s behavior thus risks overlooking the

ongoing symbolic value of men’s employ-

ment for marital stability. Men’s economic

circumstances remain crucial for marriage

formation (Oppenheimer et al. 1997; Sweeney

2002; Xie et al. 2003), and I hypothesize that

men’s employment similarly remains impor-

tant for marital stability.

In summary, I expect that the male-bread-

winner norm has remained, and the female-

homemaker norm has eroded. For recent

marriage cohorts, I expect that husbands’

full-time employment remains positively

associated with marital stability, but wives’

full-time employment and responsibility for

household labor are not determinants of

divorce. For earlier marriage cohorts, I expect

that a gender-traditional division of labor,

with husbands spending time in paid labor but

not unpaid labor, and the reverse for wives, is

associated with greater marital stability.

Little research tests how the economic

determinants of divorce have changed across

marriage cohorts. In a sample of Dutch cou-

ples, Poortman and Kalmijn (2002) compare

couples married 1943 to 1970 versus 1971 to

1997 and find that wives’ employment is desta-

bilizing only in the earlier cohort, and a more

equal division of childrearing has become

more important for marital stability in recent

cohorts. By contrast, the positive association

between husbands’ employment and marital

stability did not decline across cohorts. On the

other hand, South (2001), using data on U.S.

couples from the Panel Study of Income

Dynamics (PSID), finds that wives’ employ-

ment is increasingly positively associated with

marital instability between 1969 and 1992.

Schwartz and Gonalons-Pons (forthcoming)

find that wives’ earnings relative to their hus-

bands’, and especially wives out-earning their

husbands, are positively associated with

divorce for couples marrying in the late 1960s

and 1970s, but not for couples marrying in the

1990s. I revisit the question of change across

marriage cohorts in the economic and time use

determinants of divorce after making several

improvements to the analytic approach.

An Empirical Challenge: Measuring

Economic Independence

Evaluating the economic independence,

financial strain, and gendered institution per-

spectives depends on developing measures

and tests that can distinguish among them. A

key obstacle is that economic independence

cannot be observed directly. Individuals are

more economically independent when their

financial well-being depends less on mar-

riage, which means that the gap between their

financial well-being in marriage versus in the

event of divorce is smaller. However, post-

divorce financial well-being is observed only

for couples who divorce. Therefore, every

attempt to evaluate the economic indepen-

dence perspective requires some decision

about how to approximate the missing value

for couples not observed to divorce.

The most common approximation of a wife’s

economic independence is her current earnings

(absolute or relative to her husband’s) (Heckert

et al. 1998; Rogers 2004; Sayer and Bianchi

2000; Teachman 2010) or employment (Sayer

et al. 2011; Schoen et al. 2002; South 2001).

These proxies, however, measure within-mar-

riage circumstances rather than the expected

costs of divorce—a particularly important limi-

tation given average increases in women’s earn-

ings and employment following divorce

( Tamborini, Iams, and Reznik 2012).

Furthermore, wives’ current employment and

earnings also measure economic resources and

the division of labor within the married house-

hold, requiring additional assumptions to distin-

guish the effects of economic independence

Killewald 701

from those of the financial strain or gendered

institution perspectives. Some scholars attempt

to distinguish between perspectives based on the

sign and shape of the association between wives’

earnings and their risk of divorce (e.g., Rogers

2004), but there is no consensus about which

functional form supports each theoretical per-

spective. Particular points of debate are whether

wives’ independence should be measured with

relative or absolute earnings and whether equal

earnings between spouses indicates mutual

dependence or maximum independence (see the

discussion in Oppenheimer 1997).

An alternative argument is that, if marital

quality moderates the association between

wives’ earnings or employment and divorce,

this supports the economic independence per-

spective, because under this perspective only

bad marriages are at risk of dissolution due to

wives’ economic resources (Sayer et al. 2011;

Schoen et al. 2002). But any determinants of

divorce should matter more for low-quality

marriages, which are closer to the threshold

for divorce. Several studies exploit informa-

tion on which spouse initiated the divorce,

arguing that wives’ economic independence

should particularly encourage wife-initiated

divorces (Rogers 2004; Sayer et al. 2011).

However, wives’ employment might dispro-

portionately increase wife-initiated divorces

not because it increases wives’ economic

independence, but for other reasons, such as

because it increases conflict over household

labor that disproportionately affects wives. I

directly model wives’ predicted post-divorce

economic well-being, making it possible to

distinguish economic independence from

other perspectives. Dechter (1992) uses a

similar strategy, but evaluates a smaller range

of alternative mechanisms.

The first column of Table 1 lists the vari-

ous theoretical perspectives, and the second

column summarizes the claims associated

with each. The key measures associated with

each theory, shown in the third column, are

discussed in detail in the Data section. The

fourth column lists my prediction of whether

a given association has changed across mar-

riage cohorts.

DATA

I model couples’ log odds of divorce using

discrete-time hazard models and data from

couples in the 1968 to 2013 waves of the Panel

Study of Income Dynamics (PSID) (Panel

Study of Income Dynamics 2015), which is

frequently used to study divorce (Brines and

Joyner 1999; Cooke 2006; Dechter 1992;

Heckert et al. 1998; Ono 1998; South 2001). In

1968, the PSID sampled approximately 5,000

U.S. households; it has subsequently collected

information on members of these households

and their descendants annually or biennially. I

restrict my sample to different-sex, married,

head-wife pairs in the PSID, both spouses age

18 to 55, with neither spouse previously mar-

ried. Details of the identification of remarried

couples, marriage start and end dates, and

cohabiting couples are provided in the online

supplement (http://asr.sagepub.com/supple-

mental). I exclude cohabiting and same-sex

couples because of insufficient sample size. I

exclude remarriages because their stability may

follow a different social process. The results of

models including remarried couples are shown

in the online supplement. Couples missing

information on the start or end date of their

marriage are dropped from the sample (227

couples, or 2 percent of the sample). The ana-

lytic sample includes 6,309 couples, of whom

1,684 are observed to divorce or permanently

separate. I apply sample weights, normalized to

one in each survey year, to all analyses.

The choice of a dividing line separating

“early” and “late” marriage cohorts is of

course subject to debate. Goldin (2006) marks

the late 1970s as the beginning of “the quiet

revolution,” when young women began to

accurately predict high labor force participa-

tion and plan for it, including investments in

college education and a career identity. Thus,

particularly for evaluating changing associa-

tions between wives’ employment and the risk

of divorce, 1975 is a reasonable dividing line.

Separating marriage cohorts in this fashion

reflects that change over time in the U.S. fam-

ily has not happened in a smooth, linear way,

but in successive stages.

702 American Sociological Review 81(4)

I allow marriage cohort to moderate all inde-

pendent variables in the divorce risk model,

dividing marriages by whether they occurred

prior to 1975. For ease of interpretation, I pre-

sent separate results for each cohort and indicate

whether the difference in associations between

the cohorts is statistically significant. Results

from models with alternative cutpoints at 1980

and 1985 are shown in the online supplement.

Divorced Women’s Economic

Well-Being: IPUMS Sample

I estimate ordinary least squares models of

economic well-being for divorced and sepa-

rated women age 18 to 55, using data from the

Integrated Public Use Microdata Series

(IPUMS), specifically, samples from the 1970

to 2000 decennial censuses and the 2005 to

2009, 2006 to 2010, and 2008 to 2012 five-year

files of the American Community Survey

(Ruggles et al. 2015). Each sample generates a

unique economic well-being equation for

divorced and separated women. I then apply the

estimated equations to married women in my

PSID sample to estimate their expected eco-

nomic well-being in the next year, were they to

divorce, based on their current characteristics

(incrementing age of the woman and her chil-

dren by one). Predicted post-divorce outcomes

for PSID wives are linear interpolations of the

predicted values based on the equations of the

two adjacent IPUMS years.

Because divorce timing is unknown in most

of the IPUMS samples, I include all currently

divorced and separated women in these mod-

els, although my goal is to estimate likely out-

comes immediately following divorce. This

may lead to biased estimates of women’s likely

outcomes immediately following divorce, if the

well-being of divorced women evolves in the

years following divorce in ways not captured

by the covariates, or if women who remain

divorced rather than remarrying have different

economic outcomes, net of covariates. I return

to this possibility following the main results.

Post-divorce economic well-being.

The outcome in the IPUMS models is the

divorced or separated woman’s economic

well-being, defined as her annual household

income from all sources, divided by the

square root of the number of family members

in the household. Economic well-being is top-

and bottom-coded at the 99th and 1st percen-

tiles of the year-specific distribution.

Human capital. Education and age are

strongly associated with earnings. I allow both

to predict divorced women’s economic well-

being, using interaction terms to allow different

age profiles by education (Heckman, Lochner,

Table 1. Theoretical Perspectives, Measures, and Hypotheses

Theory

Claim: Marriages

Are More Stable

When…

Key Independent

Variables

Hypothesized

Change across

Cohorts

Economic

independence

…spouses’ predicted

financial losses from

divorce are larger

Difference between couple’s

current economic well-being

and wife’s predicted post-

divorce economic well-being

No change

Financial strain … couples have high

economic well-being

Couple’s current economic

well-being

No change

Gendered institution /

wife employment

…wives are not employed

full-time

Dummy variable for full-time

wife employment

Weaken

Gendered institution /

husband employment

… husbands are em-

ployed full-time

Dummy variable for full-time

husband employment

No change

Gendered institution /

housework

…wives do a greater share

of household labor

Proportion of couple’s total

housework performed by the

wife

Weaken

Killewald 703

and Todd 2006). Education is specified in five

categories: (1) less than a high school degree

(less than 12 years of education); (2) a high

school degree but no college (exactly 12 years

of education); (3) some college but no bache-

lor’s degree (13 to 15 years of education); (4) a

bachelor’s degree (16 years of education); and

(5) some graduate education (more than 16

years of education). Age is included as a quad-

ratic. A dummy variable is also included for

whether the woman is currently a student,

based on her labor force status.

I also include three-digit 1990 occupation

and industry codes for the woman’s current

main job.

These variables are assumed to

affect the risk of divorce only through their

association with economic independence. If

the model of divorced women’s economic

well-being included only variables also

included in the model of the risk of divorce,

the economic independence measure would be

a linear combination of other covariates, mak-

ing it perfectly collinear with the other predic-

tors of divorce. As a result, I could not estimate

its unique association with the risk of divorce.

Family. To adjust for the negative associ-

ation between children and women’s wages

(Budig and England 2001), I include the

woman’s number of biological, step, or

adopted children in the household, separated

by whether they are under age 5, both top-

coded at four.

Demographic traits. I include the wom-

an’s race (Hispanic, non-Hispanic African

American, non-Hispanic white, and other

non-Hispanic), whether she is foreign-born,

region of residence, and whether she has a

health condition that limits the type or amount

of work she can do.

All variables except

occupation and industry are interacted with

race; even in the large IPUMS samples, some

occupation and industry categories are too

small for race-specific models.

Risk of Divorce: PSID Sample

Divorce. The outcome is whether a cou-

ple’s marriage terminates with divorce or

separation in the following year. In the biennial

period, divorce is allowed in either of the next

two years, and an offset in the hazard model

adjusts for the difference in exposure time.

Current economic well-being. Finan-

cial well-being is defined as total family

income in the prior calendar year, divided by

the square root of the number of family mem-

bers in the household.

Predicted financial consequences of

divorce. To measure economic dependence,

I take the difference between the married cou-

ple’s current economic well-being and the

wife’s predicted economic well-being in the

next year, were she to divorce, based on the

coefficients from the IPUMS models. If not

currently employed, I use the wife’s most

recent occupation and industry to predict her

post-divorce economic well-being.

Gendered institution / husband and

wife employment. For each spouse, full-

time employment (at least 1,500 hours) in the

prior calendar year is measured with an indi-

cator variable. For example, I use employment

from 1985, reported in survey year 1986, to

predict marital dissolution in 1987. The lag in

employment measures (which also applies to

the current economic well-being measure) is

important given prior evidence that wives’

employment and income are endogenous to

unhappiness in marriage (Rogers 1999; Rog-

ers and DeBoer 2001; Schoen et al. 2006).

Gendered institution / housework. I

constructed the share of the couple’s house-

work performed by the wife based on reports of

typical weekly hours of housework for each

spouse. Because housework hours were not

asked in 1968 to 1975 or in 1982, I rely entirely

on multiply imputed values for these years. The

results of a model estimated only on complete

cases are shown in the online supplement.

To reduce the possibility of omitted varia-

ble bias, I include a relevant set of control

variables. I control for a quadratic in the dura-

tion of the marriage in years, accounting for

higher marital stability later in marriage

704 American Sociological Review 81(4)

(Brines and Joyner 1999; Heckert et al. 1998;

Rogers 2004; Sayer and Bianchi 2000; Sayer

et al. 2011; Schoen et al. 2002; Schoen et al.

2006; Smock, Manning, and Gupta 1999;

South 2001). Given prior evidence that Afri-

can American couples experience higher rates

of marital disruption (Brines and Joyner

1999; Heckert et al. 1998; Ono 1998; Schoen

et al. 2006; Smock et al. 1999; Tzeng and

Mare 1995; Tzeng 1992), I control for the

wife’s race using the same categories as in the

IPUMS models.

I also include a binary vari-

able for whether the wife is foreign-born,

which in the PSID is approximated by where

she reports having grown up.

To reflect the stabilizing effect of spouses’

education (Cooke 2006; Rogers 2004; Schoen

et al. 2002; Schoen et al. 2006; Smock et al.

1999; South 2001; Teachman 2002, 2010;

Tzeng and Mare 1995), I include controls for

each spouse’s educational attainment, using

the same categories as in the IPUMS models.

I also include a binary variable for whether

the wife is currently a student, measured by

her labor force status.

Premarital fertility (Teachman 2002;

Tzeng and Mare 1995; Tzeng 1992), premari-

tal cohabitation (Schoen et al. 2002; Teach-

man 2002, 2010), and younger age at marriage

(Brines and Joyner 1999; Heckert et al. 1998;

Ono 1998; Sayer and Bianchi 2000; Sayer

et al. 2011; Schoen et al. 2002; Schoen et al.

2006; Smock et al. 1999; South 2001; Tzeng

and Mare 1995; Tzeng 1992) are associated

with increased risk of divorce. I include four

indicator variables set to one: (1) if either

spouse became a parent prior to marriage; (2)

if either spouse was under age 21 at marriage;

(3) if either spouse was under age 25 at mar-

riage; and (4) if the couple cohabited prior to

marriage. Cohabitation is not retrospectively

reported, so spouses are considered to have

cohabited prior to marriage if they were

observed as a cohabiting pair in a survey

wave before the year of their marriage. Thus,

the measure disproportionately excludes

short-term cohabitations and excludes all

cohabiting experiences of couples married

prior to 1968, when data collection began.

Shared assets, including homeownership

(Cooke 2006; Ono 1998; South 2001; Spitze

and South 1985) and children (Ono 1998;

Schoen et al. 2002; Schoen et al. 2006;

Teachman 2010; Tzeng and Mare 1995;

Tzeng 1992), are associated with heightened

marital stability. I measure homeownership

with a dummy variable. I include two varia-

bles capturing the number of children related

to either spouse in the household, separated

by whether they are under age 5, each top-

coded at 4.

I account for variation in divorce rates by

religion (Teachman 2002, 2010) using three

categories: (1) both spouses are Catholic; (2)

at least one spouse reports no religion; and (3)

all other. I also control for region (Northeast,

North Central, South, and West). Observa-

tions from couples currently living outside

the United States are dropped. For each

spouse, I include an indicator variable set to

one if in the present year the person has a

health limitation that affects the amount or

kind of work one can do. I measure the year

of marriage with a series of five-year inter-

vals, to allow a flexible, nonlinear association

between marriage cohort and the risk of

divorce.

As described previously, I exclude the

wife’s occupation and industry from the model

of the risk of divorce, so that the economic

independence measure is not perfectly col-

linear with other variables in the divorce

model. However, this is inappropriate if these

variables affect divorce risk through processes

other than economic independence. To address

one possible pathway, I include the percent of

the wife’s occupation and industry that are

female, as a proxy for the availability of alter-

native partners in the workplace (McKinnish

2007). I calculate the percent female in each

occupation and industry using the IPUMS

data, projecting forward one year as in the

measures of economic independence and lin-

early interpolating between IPUMS years.

All variables in the model are time-varying,

except race, foreign-born status, cohabitation

and birth prior to marriage, and year and age

at marriage. Race and foreign-born status are

Killewald 705

reported in multiple years, so I use the most

recent available valid report and attribute this

value to all observations from the individual.

To avoid understating the uncertainty in

the estimated association between the pre-

dicted financial consequences of divorce and

the risk of divorce, I calculate standard errors

by bootstrapping the IPUMS datasets with

500 replications, generating 500 distinct

models of divorced women’s economic well-

being, each of which is used to generate a

unique model of the hazard of divorce in a

panel-bootstrapped PSID sample.

Standard

errors are clustered at the couple level.

All continuous variables are top- and bottom-

coded at the 99th and 1st percentiles of the

weighted year-specific distributions. Missing

data rates are low for most variables, and

missing data are imputed. Details of the impu-

tation strategy, which includes multiple impu-

tation and the use of reports from adjacent

years, are provided in the online supplement,

along with information on the number of

observations and couples removed from the

sample for various reasons (e.g., age restric-

tions, no marriage history data, or remarried).

RESULTS

Descriptive statistics for the IPUMS samples

and results of the regression models predicting

divorced and separated women’s economic

well-being are shown in the online supple-

ment. Each model explains between 32 and 39

percent of the variation in divorced women’s

economic well-being. Occupation and indus-

try measures—the variables excluded from

the divorce models and used to facilitate sta-

tistical identification of the economic inde-

pendence measure as a distinct predictor—are

important contributors to these models: mod-

els that include all other predictors, but

exclude these, explain only between 22 and 29

percent of the variation in divorced and sepa-

rated women’s economic well-being.

How well do the predicted values match

what PSID women who divorce actually

experience? For the 1,159 women in the PSID

observed through divorce, the correlation

between predicted post-divorce economic

well-being and observed economic well-

being in the first full year (or, when not avail-

able, second year) following divorce is .48.

For comparison, wives’ own pre-divorce

earnings—a common proxy for economic

independence—are slightly less strongly cor-

related with post-divorce outcomes (.46).

Thus, predicted post-divorce economic well-

being is strongly associated with real varia-

tion in women’s post-divorce outcomes,

although it certainly does not explain all

individual-level variation.

The fact that the models do not explain all

the individual-level variation in post-divorce

economic well-being does not automatically

imply that coefficients will be biased toward

zero. However, the disruptive effect of eco-

nomic independence will be underestimated

if economic independence is systematically

underestimated for women who divorce or

systematically overestimated for women who

remain married. The median observed and

predicted outcomes of women who divorce

match fairly closely ($24,282 observed ver-

sus $23,433 predicted). Of course, I do not

know whether the post-divorce outcomes of

women who remain married would have been

less than predicted, but previous research

finds no evidence for selection bias of this

kind (Dechter 1992; Smock et al. 1999).

Table 2 shows descriptive traits of the PSID

sample of married couples by marriage cohort.

The economic well-being of married couples is

similar between cohorts (about $53,000), as is

the expected post-divorce well-being of wives

(around $30,000). As a result, there is little dif-

ference between cohorts in the expected

decline in wives’ financial well-being follow-

ing divorce (around $24,000). Likewise, there

has been little change in husbands’ employ-

ment rates: in both cohorts about 90 percent of

husbands are employed full-time.

Wives’ employment has changed more dra-

matically. In the earlier cohort, only 34 per-

cent of wives were employed full-time,

compared to 48 percent in the later cohort.

The average proportion of a couple’s house-

work performed by the wife declined across

706 American Sociological Review 81(4)

Table 2. PSID Sample Means and Proportions (Standard Deviations)

Married

before 1975

Married in

1975 to 2011

Financial Strain

Economic well-being (family income/sqrt(n)),

2013 dollars

$52,443.39 $54,090.21

($33,719.00) ($41,417.64)

Economic Independence

Wife’s predicted post-divorce well-being, 2013 dollars $29,195.56 $32,260.01

($12,710.79) ($15,545.25)

N = 24,816 N = 30,967

Wife’s economic dependence (current well-being –

predicted well-being)

$25,030.43 $23,060.83

($28,146.26) ($34,423.21)

N = 24,816 N = 30,967

Gendered Institution

Wife employed full-time .34 .48

Husband employed full-time .91 .90

Wife’s proportion of housework .81 .72

(.17) (.19)

N = 20,817 N = 32,448

Controls

Proportion female in wife’s

Most recent occupation .73 .68

(.24) (.23)

N = 25,389 N = 31,339

Most recent industry .62 .61

(.20) (.19)

N = 24,865 N = 31,088

Marital duration 19.13 9.91

(8.75) (7.45)

Religion

Other .67 .62

Both spouses Catholic .23 .22

At least one is not religious .10 .16

Race

Non-Hispanic white .88 .82

Non-Hispanic black .06 .07

Hispanic .05 .09

Other .01 .02

Foreign-born .03 .06

Region

Northeast .24 .23

North Central .31 .28

South .30 .29

West .15 .21

Owns home .83 .71

Children under age 5

Prop. with children under 5 .18 .37

Number among those with children under 5 1.19 1.22

(.42) (.44)

Children age 5 to 17

Prop. with children age 5 to 17 .62 .50

Number among those with children age 5 to 17 1.94 1.76

(.95) (.78)

(continued)

Killewald 707

Married

before 1975

Married in

1975 to 2011

Cohabited before marriage .003 .10

Premarital birth .02 .11

Married under age 21 .60 .27

Married under age 25 .93 .69

Wife’s health limits work .10 .08

Husband’s health limits work .10 .07

Wife’s age 38.88 32.72

(8.93) (8.09)

Husband’s age 41.09 34.41

(9.00) (8.09)

Wife currently enrolled in school .005 .02

Wife’s education

Less than high school .18 .10

High school .49 .32

Some college .17 .25

College .12 .23

Graduate school .05 .10

N = 32,850 N = 33,361

Husband’s education

Less than high school .20 .11

High school .34 .32

Some college .19 .23

College .17 .23

Graduate school .10 .11

N = 33,348

Couples 1,908 4,401

Divorces 381 1,303

Observations 32,853 33,470

Note: Samples include all couples with both spouses in a ﬁrst marriage, with individual variable

summaries restricted to the subsample not multiply imputed on that variable. Sample size matches

the number of observations given at the bottom of the table unless otherwise noted. All estimates are

weighted using the PSID family weight, normalized to average one in each year.

Table 2. (continued)

cohorts, although relatively modestly, from 81

to 72 percent.

Members of the earlier cohort have, on

average, higher marital durations, older

spouses, and older children. Some of this dis-

parity is likely due to the fact that couples

married prior to 1968 are included in the ear-

lier cohort, although they are not observed in

the first year of their marriage. I retain these

couples in the main analysis to increase statis-

tical power. Because I know the marriage

dates of these couples and can control for

marital duration, their inclusion will introduce

bias only if the proportional odds assumption

is violated—if marital duration moderates the

association between the other independent

variables and the risk of divorce. I return to

consideration of this possibility following the

main results.

Table 3 shows results from models predict-

ing couples’ log odds of divorce. Results for

the pre-1975 marriage cohorts are in the first

column, results for the later cohorts are in the

second column; statistically significant differ-

ences between cohorts in the associations are

marked with asterisks next to the variable

labels. For comparison, results from a pooled

model including all marriage cohorts are

708 American Sociological Review 81(4)

shown in the third column. There is no evi-

dence for either the financial strain or eco-

nomic independence perspective in either

cohort or in the pooled model: coefficients are

not statistically significant, and, for financial

strain, in the opposite direction from the theo-

retical prediction.

Thus, I do not find support

for the gender-symmetric, money-based

understandings of marital stability. There is

also no evidence for change in the importance

of these perspectives over time.

Turning to the gendered institution per-

spective, I first evaluate the association

between wives’ full-time employment and the

risk of divorce. In the pooled model, wives’

full-time employment is associated with an

18 percent increase in the odds of divorce

.166

= 1.18), and the association is statisti-

cally significant. In the cohort-specific mod-

els, estimates are less precise and wives’

full-time employment is not statistically sig-

nificantly associated with the risk of divorce

for either cohort. In terms of the point esti-

mates, wives’ full-time employment is associ-

ated with a 31 percent increase in the odds of

divorce in the early cohort, compared to 7

percent in the later cohort. The point esti-

mates are consistent with a decline in the

association between wives’ employment and

divorce risk in more recent cohorts, but the

cross-cohort change is not statistically

significant.

Next, I evaluate whether husbands’ lack of

full-time employment is associated with

heightened risk of divorce. In the pooled

model, couples in which the husband is

employed full-time have 21 percent lower

odds of divorce than couples in which he is

not. In the early cohort, husbands’ full-time

employment is associated with a 9 percent

reduction in the odds of divorce, compared to

a reduction of 25 percent in the later cohort.

The association is statistically significant in

the later but not the earlier cohort, and the

change across cohorts is not statistically sig-

nificant.

Consistent with predictions, there is

no evidence that the husband employment

aspect of the gendered institution perspective

has become less important for more recent

marriage cohorts. For both husbands and

wives, the observed associations between

employment and the risk of divorce are net of

effects on household income. Thus, the effect

of employment can be separated from its

financial consequences.

In the domain of unpaid labor, for the ear-

lier cohort, the share of housework done by

the wife is negatively and statistically signifi-

cantly associated with the risk of divorce.

This is consistent with the housework compo-

nent of the gendered institution perspective: a

more traditional division of household labor

is associated with greater marital stability in

the earlier cohort. In the more recent cohort,

however, the association is slightly positive

and not statistically significant. The estimate

in the pooled model is negative but not statis-

tically significant. As predicted, the stabiliz-

ing effect of the wife’s responsibility for

unpaid labor has eroded in the recent cohort,

and the change across cohorts is statistically

significant.

Supplementary analyses show that the

wife’s share of housework is less stabilizing

when wives are employed full-time, although

the variation in the association by employ-

ment status is only statistically significant in

the later cohort. This suggests that, at least

for the later cohort, the effect of wives’

housework burden on marital stability may

depend on whether it is part of a traditional

division of labor between spouses or a “sec-

ond shift” (Hochschild 1989) by full-time

employed wives. More flexible specifica-

tions of the division of housework suggest

that, even in the later cohort, wives’ greater

responsibility for household labor is associ-

ated with greater marital stability at least up

until the point of shared housework responsi-

bility. However, there is suggestive evidence

in the later cohort that husbands’ contribu-

tions to housework at low levels may be sta-

bilizing compared to all housework

responsibility falling to wives.

In summary, the pattern of results suggests

it is the division of labor, either paid or

unpaid, that is associated with the risk of

divorce for couples in both cohorts, not finan-

cial considerations. In the pooled model,

wives’ full-time employment is associated

Killewald 709

Table 3. Logistic Regression of Marital Dissolution, by Marriage Cohort (PSID)

1. Married in

1974 or Earlier

2. Married in

1975 or Later 3. Pooled

Financial Strain

Economic well-being (thousands of $) .014 .002 .003

(.009) (.005) (.004)

OR = 1.014 OR = 1.002 OR = 1.003

Economic Independence

Wife’s economic dependence (thousands of $) −.013 −.001 −.003

(.010) (.005) (.004)

OR = .987 OR = .999 OR = .997

Gendered Institution

Full-time employment

Wife .267 .069 .166

(.153) (.088) (.080)

OR = 1.306 OR = 1.071 OR = 1.181

Husband −.097 −.293

−.238

(.208) (.106) (.094)

OR = .908 OR = .746 OR = .788

Wife’s proportion of housework

−1.117

.050 −.244

(.463) (.233) (.224)

OR = .327 OR = 1.051 OR = .783

Controls

Marital duration

***

.139

***

−.044

−.008

(.041) (.022) (.018)

OR = 1.149 OR = .957 OR = .992

Marital duration-squared

−.004

***

.000 −.001

(.001) (.001) (.001)

OR = .996 OR = 1.000 OR = .999

Religion (other omitted)

Both spouses Catholic −.157 −.036 −.085

(.215) (.124) (.108)

OR = .854 OR = .964 OR = .918

At least one spouse is not religious

.812

***

.257

.395

***

(.176) (.101) (.091)

OR = 2.252 OR = 1.293 OR = 1.485

Wife’s race (non-Hispanic white omitted)

Non-Hispanic black

−.497 .152 .001

(.291) (.123) (.109)

OR = .609 OR = 1.164 OR = 1.001

Hispanic −.666 −.177 −.243

(.422) (.179) (.162)

OR = .514 OR = .838 OR = .785

Other −.581 .143 −.024

(.725) (.305) (.266)

OR = .559 OR = 1.154 OR = .976

Wife is foreign-born −.275 −.703

−.477

(.543) (.249) (.219)

OR = .759 OR = .495 OR = .620

(continued)

710 American Sociological Review 81(4)

1. Married in

1974 or Earlier

2. Married in

1975 or Later 3. Pooled

Home ownership −.061 −.398

***

−.328

***

(.172) (.082) (.071)

OR = .941 OR = .672 OR = .720

Number of children

Under age 5 .200 −.001 .004

(.123) (.062) (.054)

OR = 1.222 OR = .999 OR = 1.004

Age 5 and older .091 .152

.108

(.080) (.053) (.043)

OR = 1.095 OR = 1.164 OR = 1.114

Premarital cohabitation

1.308

.251

.331

(.413) (.120) (.115)

OR = 3.700 OR = 1.285 OR = 1.392

Premarital birth .490 .218 .417

***

(.424) (.117) (.114)

OR = 1.633 OR = 1.244 OR = 1.518

At marriage, at least one spouse

Under age 21 .628

***

.297

.434

***

(.181) (.096) (.085)

OR = 1.874 OR = 1.345 OR = 1.543

Under age 25 .666 .246

.337

***

(.462) (.105) (.100)

OR = 1.947 OR = 1.279 OR = 1.400

Wife is a student −.434 .373 .358

(.975) (.226) (.214)

OR = .648 OR = 1.453 OR = 1.431

Wife’s education (high school omitted)

Less than high school .191 .153 .113

(.212) (.117) (.104)

OR = 1.211 OR = 1.165 OR = 1.119

Some college .174 −.041 .041

(.200) (.097) (.090)

OR = 1.190 OR = .960 OR = 1.042

College .091 −.345

−.216

(.280) (.158) (.143)

OR = 1.096 OR = .708 OR = .806

Graduate school −.242 −.514

−.434

(.432) (.216) (.190)

OR = .785 OR = .598 OR = .648

Husband’s education (high school omitted)

Less than high school −.285 .032 −.093

(.215) (.116) (.102)

OR = .752 OR = 1.032 OR = .911

Some college −.000 −.168 −.100

(.178) (.108) (.094)

OR = 1.000 OR = .845 OR = .905

Table 3. (continued)

(continued)

Killewald 711

1. Married in

1974 or Earlier

2. Married in

1975 or Later 3. Pooled

College −.170 −.559

***

−.374

(.218) (.153) (.121)

OR = .843 OR = .572 OR = .688

Graduate school −.100 −.536

−.252

(.294) (.185) (.154)

OR = .905 OR = .585 OR = .777

Health limits work

Wife .104 .109 .091

(.242) (.132) (.115)

OR = 1.109 OR = 1.115 OR = 1.096

Husband .177 .216 .169

(.223) (.127) (.114)

OR = 1.193 OR = 1.241 OR = 1.184

Proportion female in wife’s

Occupation

.606 −.300 −.006

(.328) (.189) (.170)

OR = 1.832 OR = .740 OR = .994

Industry −.477 .054 −.145

(.333) (.210) (.173)

OR = .621 OR = 1.055 OR = .865

Region (Northeast omitted)

North Central .183 .037 .067

(.211) (.126) (.109)

OR = 1.200 OR = 1.038 OR = 1.069

South .531

.279

.324

(.228) (.124) (.111)

OR = 1.701 OR = 1.321 OR = 1.382

West .376 .181 .251

(.242) (.133) (.116)

OR = 1.456 OR = 1.198 OR = 1.286

Year began first marriage (before 1955 is

reference in Models 1 and 3)

1955 to 1959 .971

1.057

***

(.320) (.313)

OR = 2.641 OR = 2.879

1960 to 1965 1.208

***

1.268

***

(.296) (.292)

OR = 3.346 OR = 3.554

1965 to 1969 1.211

***

1.280

***

(.298) (.291)

OR = 3.358 OR = 3.597

1970 to 1974 .834

.910

(.296) (.289)

OR = 2.302 OR = 2.484

1975 to 1979 (reference category in Model 2) 1.359

***

(.289)

OR = 3.894

(continued)

Table 3. (continued)

712 American Sociological Review 81(4)

with heightened risk of divorce, and hus-

bands’ full-time employment is associated

with greater marital stability. Analyses of

change across cohorts are somewhat limited

by low statistical power. However, the stabi-

lizing association between wives’ share of

household labor and the risk of divorce has

declined significantly across cohorts. By con-

trast, there is no evidence that the stabilizing

role of husbands’ full-time employment has

diminished. Conclusions about changes in the

disruptive effect of wives’ full-time employ-

ment are more speculative; point estimates

are consistent with a decline in the associa-

tion between wives’ full-time work and the

risk of divorce, but the cross-cohort change is

not statistically significant.

To provide context for the magnitude of

these associations, Figure 1 shows the pre-

dicted probabilities of divorce, separately for

each cohort, by the employment status of

each spouse and whether the wife performs

50 or 75 percent of the couple’s housework

(the distribution of household labor is so

skewed that an equal division of housework is

the 10th percentile of the wife’s proportion of

housework). The bars show the predicted

probabilities holding all other covariates at

their cohort-specific sample means. The dif-

ferences between cohorts capture both pure

cohort effects and the changing composition

of married couples across cohorts, including

the fact that the earlier cohort disproportion-

ately includes couples of higher marital

1. Married in

1974 or Earlier

2. Married in

1975 or Later 3. Pooled

1980 to 1984 .244

1.610

***

(.120) (.286)

OR = 1.277 OR = 5.003

1985 to 1989 .602

***

1.965

***

(.132) (.293)

OR = 1.825 OR = 7.136

1990 to 1994 .291

1.637

***

(.142) (.308)

OR = 1.338 OR = 5.138

1995 to 1999 .559

***

1.918

***

(.147) (.301)

OR = 1.750 OR = 6.804

2000 to 2004 .609

***

1.976

***

(.159) (.316)

OR = 1.839 OR = 7.213

2005 to 2011 .425

1.831

***

(.178) (.321)

OR = 1.530 OR = 6.241

Constant

***

−7.476

***

−3.525

***

−5.393

***

(.884) (.393) (.435)

OR = .001 OR = .029 OR = .005

Observations 32,853 33,470 66,323

Note: Models are estimated with the PSID family weight, normalized to average one in each year.

Standard errors are bootstrapped to account for uncertainty in predicted well-being from the IPUMS

models. Signiﬁcance of change across cohorts is marked next to the variable names.

p < .05;

p < .01;

***

p < .001 (two-tailed tests).

Table 3. (continued)

Killewald 713

durations. The probability of divorce in a

given year is systematically higher for the

later cohort, regardless of employment status

or housework time.

In the early cohort, for an otherwise typi-

cal couple with a wife employed full-time, the

predicted probability of divorce in the next

year is 1.3 percent, compared to 1.0 percent if

she is not employed full-time. When wives do

only 50 percent of the housework in the early

cohort, the predicted probability of divorce in

the next year is 1.5 percent, compared to 1.1

percent if she does 75 percent. By contrast,

the predicted probability of divorce changes

little with the husband’s employment status

and is 1.0 percent if he is employed full-time

and 1.1 percent if he is not.

In the later cohort, an otherwise typical

couple with a husband not employed full-time

has a 3.3 percent predicted probability of

divorce the following year, compared to 2.5

percent if he is employed full-time. Across

wife’s employment and housework catego-

ries, the predicted probability of divorce var-

ies little for this cohort, ranging from 2.5 to

2.6 percent.

The results for control variables are largely

as expected. For both cohorts, at least one

spouse not identifying with a religion, early

marriage, premarital cohabitation, and living

in the South are all associated with higher

odds of divorce. However, for both non-

religious identification and premarital cohabi-

tation, the association has declined statisti-

cally significantly across cohorts. For the

most recent marriage cohorts, the heightened

divorce risk associated with cohabitation may

have disappeared entirely (Manning and

Cohen 2012), but the cohort range defined

here—marriages after 1974—is likely too

broad to detect this change.

LIMITATIONS AND

ROBUSTNESS OF RESULTS

How Robust Is the Lack of Support

for Economic Independence and

Financial Strain?

I considered the possibility that the lack of

support for the economic independence and

financial strain perspectives was due to

1.1%

1.0%

1.3%

1.5%

1.1%

3.3%

2.5%

2.6%

2.5%

2.6%

Husband not

employed full-

time

Husband

employed full-

time

Wife not

employed full-

time

Wife employed

full-time

Wife does 50%

of housework

Wife does 75%

of housework

Predicted Probability of Divorce

Marriage before 1975 Marriage in 1975 or later

Figure 1. Predicted Probabilities of Divorce in the Next Year

Note: Probabilities are predicted based on the logistic regression coefﬁcients in Table 3 for a respondent

with all other covariates set to the cohort-speciﬁc means. Error bars are 95% conﬁdence intervals,

estimated by calculating the bounds of the conﬁdence interval for the predicted log odds (non-

bootstrapped) and then transforming those bounds to predicted probabilities.

714 American Sociological Review 81(4)

specification error or low statistical power. I

experimented with more flexible specifica-

tions of couples’ economic well-being and the

wife’s economic independence and alternative

measures of these concepts. I allowed finan-

cial strain to interact with economic indepen-

dence, considering that wives’ economic

independence may have different effects

depending on household resources (Ono 1998;

Sayer and Bianchi 2000). To reduce collinear-

ity, I also re-estimated the models excluding

all of the time use and financial variables

except the economic independence measure,

and then excluding all except the financial

strain measure. The models provide no sup-

port for the economic independence or finan-

cial strain perspective in either cohort, with

one exception: in the early cohort, wives’

predicted post-divorce economic well-being

(rather than the difference between current

and predicted post-divorce well-being) was

positively and statistically significantly asso-

ciated with the risk of divorce. (Results for all

alternative specifications described in this sec-

tion are in the online supplement.) I also find

that the wife’s economic dependence is nega-

tively and statistically significantly associated

with the risk of divorce in an early cohort

sample restricted to couples married between

1968 and 1974. Because of the large number

of models estimated, it is expected that some

associations will be statistically significant

just by chance. However, it is possible that, in

earlier marriage cohorts, wives’ ability to sup-

port themselves post-divorce facilitated exit

from unhappy marriages.

I also considered the possibility that long-

term economic prospects, including for remar-

riage, rather than likely economic outcomes

immediately following divorce, may influence

divorce decisions. The Census did not collect

information on individuals’ number of mar-

riages between 1980 and 2008. However, I

performed a supplementary analysis using the

1970, 1980, and 2008 to 2012 IPUMS samples

to estimate wives’ long-term post-divorce eco-

nomic prospects, including the possibility of

remarriage. As in the main models, the eco-

nomic independence measure is not statistically

significantly associated with risk of divorce,

and the coefficient is close to zero.

In addition to the economic independence

and financial strain hypotheses, I considered

the possibility that spouses’ earnings, rather

than their employment, might shape the risk of

divorce for symbolic reasons, as an additional

manifestation of the gendered institution per-

spective. I found no evidence that husbands’

earnings stabilize marriages, net of employ-

ment, and no evidence that spouses’ relative

earnings are associated with the risk of divorce.

It is possible that unmeasured characteris-

tics associated with both the risk of divorce

and women’s post-divorce economic out-

comes have suppressed the association

between economic independence and the risk

of divorce, but prior research finds little evi-

dence for bias of this kind (Dechter 1992;

Smock et al. 1999). An additional limitation,

as previously discussed, is that the IPUMS

sample includes women who have been

divorced for more than one year, whose finan-

cial circumstances may not reflect immediate

post-divorce outcomes.

In summary, the additional analyses show

little support for the economic independence

or financial strain perspectives, although it is

of course possible that omitted variable bias

or measurement error may have artificially

suppressed the associations.

How Robust Are the Time Use

Findings?

As previously noted, spouses may adjust work

behaviors in anticipation of divorce, raising

concerns of reverse causality. Anticipatory

effects of this kind are less likely to explain

the association between husbands’ lack of full-

time employment and the risk of divorce:

there is no reason to expect that husbands

anticipating divorce preemptively quit their

jobs. However, wives anticipating divorce

may increase their paid work hours or decrease

household labor. In both cohorts, average paid

labor hours for wives not employed full-time

are no more than 10 hours per week, leaving

substantial room for anticipatory increases.

Killewald 715

To reduce the possibility for spurious asso-

ciations due to anticipation of divorce, I

repeated the analysis lagging measures of the

key independent variables by an additional

two years. For example, in the main models,

earnings and employment in 1985 are used to

predict divorce in 1987. In the supplemental

models, earnings and employment in 1985 are

used to predict divorce in 1989. The associa-

tions are weaker and not statistically signifi-

cant in these models. One interpretation is

that, in the early cohort, wives’ employment

and low responsibility for household labor

were responses to anticipation of divorce. An

alternative explanation is that it does not take

three to four years for gender-deviant behav-

iors to lead to divorce. Using the same exam-

ple, if a wife became employed full-time in

1985 and the couple divorced in 1987, this

couple would contribute to the association

between wives’ employment and the risk of

divorce in the main models, but not in the

lagged models: the couple would already

have divorced and be censored by the time the

outcome is measured in 1989.

I considered that the results for the early

cohorts may have been biased by including

couples married prior to 1968. As described

previously, because marriage dates are

known, the inclusion of these couples does

not bias results if the proportional odds

assumption holds, but it may bias results if

the association between the key independent

variables and the risk of divorce varies by

marital duration, because the early cohort is

disproportionately composed of couples of

longer marital duration. I performed a sup-

plemental analysis restricting the early cohort

sample to couples whose marriages began in

1968 to 1974. As expected, standard errors

increase, due to the smaller sample size. The

association between wives’ full-time employ-

ment and the risk of divorce is larger in mag-

nitude than in the model that includes couples

married before 1968, but, as in the main

sample, it is not statistically significant. For

the wife’s proportion of housework, the stand-

ard error is larger and the coefficient goes in

the opposite direction. The instability of the

housework coefficient is likely related to the

fact that, as mentioned previously, housework

was not reported in the early years of the

PSID.

I also tested the robustness of results by

experimenting with dividing cohorts at 1980

or 1985 instead of 1975. As expected, the

higher risk of divorce for couples with full-

time employed wives is muted when the

dividing line is set later. The patterns for

housework and husband’s employment are

similar to the main specification. Thus, the

late 1970s may have been a particularly

important time for changes in perceptions of

women’s employment, consistent with Gol-

din’s (2006) argument. However, given the

sample size, especially of the earlier cohort, it

is not possible to determine the exact shape of

the trends across cohorts in the determinants

of divorce, and the PSID is not well-suited to

evaluating changes across earlier marriage

cohorts, such as those of the 1960s versus

early 1970s. In general, future research is

needed to continue to evaluate the possibility

of change across cohorts in the determinants

of divorce.

Of course, omitted variable bias may be

responsible for the observed associations

between spouses’ paid and unpaid labor and

the risk of divorce, or for changes in the asso-

ciations across cohorts. For example, a posi-

tive association between wives’ employment

and the risk of divorce may be due to a joint

association with liberal gender ideology

(Sayer and Bianchi 2000), or other unob-

served characteristics associated with their

risk of divorce.

In summary, it is challenging to assess

whether wives’ employment caused higher risk

of divorce in even the earlier marriage cohorts,

because the association may be due to antici-

patory effects and the estimates are imprecise.

By contrast, the findings in the later cohort,

that wives’ employment and housework are

not associated with the risk of divorce, but

divorce is more likely when husbands are not

employed full-time, are not threatened by these

concerns: the later cohort is not affected by

left-censoring, the results are robust across

716 American Sociological Review 81(4)

different choices of cohort cutpoints, and

anticipatory effects should, if anything, over-

state the destabilizing effect of wives’ full-time

employment and stabilizing effect of wives’

household labor responsibility.

CONCLUSIONS

Previous research on the associations between

money, work, and marital stability has

attempted to determine whether marital sta-

bility is affected by wives’ ability to support

themselves in the event of divorce, couples’

financial resources, the gendered interpreta-

tion of spouses’ work and earnings, or some

combination of all three. By constructing a

measure of economic independence distinct

from wives’ current earnings or employment,

my analytic approach allows a more rigorous

test of the pathways by which spouses’ earn-

ings and employment are associated with

their risks of divorce, distinguishing among

the economic independence, financial strain,

and gendered institution perspectives. At the

same time, no single analysis can provide a

definitive assessment of the association

between spouses’ work, economic circum-

stances, and the risk of divorce. One of the

goals of this article is to articulate the consid-

erable analytic challenges to distinguishing

among perspectives and ascertaining causal

order. By making clear the assumptions

required to distinguish among perspectives

and to evaluate change across cohorts, I hope

to motivate future research that will evaluate

these same associations with alternative sam-

ples or assumptions.

Using a sample of married couples in the

United States observed between 1968 and

2013, I do not find support for the economic

independence or financial strain perspectives.

These findings suggest that material circum-

stances, within marriage or outside of it, are

not key determinants of marital stability. As a

consequence, the results cast doubt on the

claim that increases in divorce rates in the

mid-twentieth century were due to women’s

rising economic independence.

However, I find support for the gendered

institution perspective. Although prior research

on the economic determinants of divorce dis-

proportionately focuses on women’s charac-

teristics, the strongest evidence for the

gendered institution perspective is that, for

marriages begun in 1975 or later, divorce is

more likely when husbands are not employed

full-time. Consistent with my hypotheses,

there is no evidence that this association is

weaker for later than earlier marriage cohorts.

Just as male breadwinning has remained

important for marriage formation (Sweeney

2002), the results here demonstrate its endur-

ing importance for marital stability. The

results are consistent with claims that bread-

winning remains a central component of the

marital contract for husbands (Nock 1998).

It is possible that husbands’ less than full-

time employment is associated with marital

disruption more strongly than wives’, not

because of gendered interpretations of lack

of full-time employment, but because hus-

bands’ part-time employment or nonemploy-

ment is more likely to be involuntary.

Involuntary nonemployment may negatively

affect marriages more strongly than volun-

tary nonemployment, by affecting outcomes

like partners’ mental health. It is not possible

to evaluate this perspective with the current

data, because voluntary specialization by

men in unpaid labor is rare: in 2012, only

about one-fifth of stay-at-home fathers were

home primarily to care for the family

(Livingston 2014). Future research is needed

to explore the experiences of deliberately

nontraditional households, although their rar-

ity illustrates the consistency of the male-

breadwinner norm.

Assessing the association between wives’

employment and the risk of divorce and how

this association may have changed across

marriage cohorts is challenging. When all

marriage cohorts are pooled, wives’ full-time

employment is positively and statistically sig-

nificantly associated with the risk of divorce.

The magnitude of the association is smaller in

more recent cohorts, but the change across

cohorts is not statistically significant, nor is

the association in either of the cohort-specific

models. Furthermore, I cannot rule out the

possibility that the positive association

Killewald 717

between wives’ employment and the risk of

divorce is driven by anticipatory effects. The

results are consistent with change across

cohorts in the gender structure shaping expec-

tations for wives’ employment (Risman

2011), but conclusions about cross-cohort

change must remain speculative.

For unpaid labor, however, I find that the

association between wives’ share of house-

work and the risk of divorce has changed

across cohorts; for marriages founded since

1975, wives’ household labor responsibility is

not linearly associated with greater marital

stability. Supplemental models revealed that,

at least in the more recent cohort, the associa-

tion between wives’ household labor and the

risk of divorce is nonlinear and depends on a

wife’s employment status. This suggests that,

for more recent marriage cohorts, at least

some egalitarianism in the division of house-

work may increase marital stability. More

research is needed to investigate the precise

shape of the relationship between housework

contributions and marital stability for differ-

ent marriage cohorts.

The determinants of marital stability for

modern marriages are thus neither post-gender

nor entirely parallel to those of earlier mar-

riages. In both cohorts, marriage remains a

gendered institution, embedded in the larger

gender structure (Risman 2011), with the

division of labor, not financial resources, the

primary lens through which this gendered

nature is reflected. The differing results by

marital cohort suggest that the determinants

of marital stability are likely to continue to

change in tandem with changes in the broader

gender structure. The fact that, in recent

cohorts, wives’ employment is not associated

with the risk of divorce, while husbands’ lack

of full-time employment remains associated

with marital instability, suggests that changes

in the gender structure may not have pro-

ceeded evenly for men and women; fulfill-

ment of the male-breadwinner role appears to

be equally or more strongly associated with

marital stability in more recent marriage

cohorts. Thus, the results highlight the impor-

tance of incorporating men fully into study of

the effects of gender norms on family life.

Acknowledgments

I am grateful to Ian Lundberg and Cassandra Robertson

for research assistance and to the colleagues, seminar

audiences, and anonymous reviewers who provided valu-

able suggestions on earlier versions of this manuscript.

Funding

This research was supported in part by a grant from the

William F. Milton Fund, Harvard University. The collec-

tion of data used in this study (the Panel Study of Income

Dynamics) was partly supported by the National Insti-

tutes of Health under grant number R01 HD069609 and

the National Science Foundation under award number

1157698.

Notes

1. Throughout, I focus on evidence of the determinants

of divorce for U.S. couples, when possible, because

these determinants may vary across countries, espe-

cially when it comes to the gendered interpretation

of spouses’ paid and unpaid labor time and earnings

(Cooke 2006).

2. IPUMS harmonizes occupation and industry codes

to the classification system used in the 1990 Cen-

sus. The PSID classifies jobs according to the 1970

Census classification system in 1968 to 2001 and

according to the 2000 Census classification sys-

tem in 2003 to 2011. To harmonize the occupation

codes, I use a crosswalk provided by IPUMS to con-

vert all codes to 1990 codes (https://usa.ipums.org/

usa/volii/occ_ind.shtml). No such crosswalk exists

for industry codes, so I downloaded IPUMS data

from the 1970 and 2000 censuses, both of which

also give the 1990 industry codes. I then identified

the 1970 and 2000 codes that corresponded to each

1990 category and created a crosswalk to replicate

the mapping used in the IPUMS datasets.

3. Health limitation is not measured in the ACS sam-

ples and is therefore excluded from the models spe-

cific to those years.

4. The PSID has collected wealth information in too

few years to evaluate whether net worth is associ-

ated with the risk of divorce.

5. The determinants of marital stability may vary by

race (Teachman 2010) and parental status (Dechter

1992). I found no statistically significant variation

in the association between the key independent

variables and the risk of divorce between childless

couples and parents or between African Americans

and whites, for either cohort.

6. Because bootstrapped and non-bootstrapped stan-

dard errors were similar, robustness checks and

other supplemental models do not use bootstrapped

standard errors.

7. A model that examined husbands’ economic inde-

pendence instead of or in addition to wives’ also

found no statistically significant associations.

718 American Sociological Review 81(4)

8. I tested whether husbands’ lack of full-time employ-

ment was particularly disruptive if the wife was

not employed full-time when the husband was last

employed full-time. I also tested whether wives’

full-time employment was less disruptive when the

husband was not employed full-time. Results were

consistent with these perspectives, but not statisti-

cally significant for either cohort.

References

Becker, Gary S., Elisabeth M. Landes, and Robert T.

Michael. 1977. “An Economic Analysis of Marital

Instability.” Journal of Political Economy 85(6):

1141–87.

Berk, Sarah Fenstermaker. 1985. The Gender Factory:

The Apportionment of Work in American Households.

New York: Plenum.

Bertrand, Marianne, Emir Kamenica, and Jessica Pan.

2015. “Gender Identity and Relative Income within

Households.” Quarterly Journal of Economics

130(2):571–614.

Bianchi, Suzanne M., Liana C. Sayer, Melissa A. Milkie,

and John P. Robinson. 2012. “Housework: Who Did,

Does or Will Do It, and How Much Does It Matter?”

Social Forces 91(1):55–63.

Brines, Julie, and Kara Joyner. 1999. “The Ties That

Bind: Principles of Cohesion in Cohabitation

and Marriage.” American Sociological Review

64(3):333–55.

Budig, Michelle J., and Paula England. 2001. “The Wage

Penalty for Motherhood.” American Sociological

Review 66(2):204–225.

Cherlin, Andrew. 1979. “Work Life and Marital Dis-

solution.” Pp. 151–66 in Divorce and Separation:

Context, Causes, and Consequences, edited by G.

Levinger and O. C. Moles. New York: Basic Books.

Cooke, Lynn Prince. 2006. “‘Doing’ Gender in Context:

Household Bargaining and Risk of Divorce in Ger-

many and the United States.” American Journal of

Sociology 112(2):442–72.

Dechter, Aimée R. 1992. “The Effect of Women’s Eco-

nomic Independence on Union Dissolution.” CDE

Working Paper No. 92-28. Center for Demography

and Ecology, University of Wisconsin-Madison,

Madison, WI.

Fitch, Catherine A., and Steven Ruggles. 2000. “His-

torical Trends in Marriage Formation: The United

States 1850–1990.” Pp. 59–88 in The Ties That

Bind: Perspectives on Marriage and Cohabitation,

edited by L. J. Waite, C. Bachrach, M. Hindin, E.

Thomson, and A. Thornton. New York: Aldine de

Gruyter.

Goldin, Claudia. 2006. “The Quiet Revolution That

Transformed Women’s Employment, Education, and

Family.” American Economic Review 96(2):1–21.

Heckert, D. Alex, Thomas C. Nowak, and Kay A. Snyder.

1998. “The Impact of Husbands’ and Wives’ Relative

Earnings on Marital Disruption.” Journal of Mar-

riage and the Family 60(3):690–703.

Heckman, James J., Lance J. Lochner, and Petra E.

Todd. 2006. “Earnings Functions, Rates of Return

and Treatment Effects: The Mincer Equation and

Beyond.” Pp. 307–458 in Handbook of the Econom-

ics of Education, Vol. 1, edited by E. A. Hanushek and

F. Welch. Amsterdam: North-Holland.

Hochschild, Arlie. 1989. The Second Shift: Working Parents

and the Revolution at Home. New York: Viking Penguin.

Livingston, Gretchen. 2014. “Growing Number of Dads

Home with the Kids: Biggest Increase among Those

Caring for Family.” Washington, DC: Pew Research

Center’s Social and Demographic Trends project.

Manning, Wendy D., and Jessica A. Cohen. 2012. “Pre-

marital Cohabitation and Marital Dissolution: An

Examination of Recent Marriages.” Journal of Mar-

riage and Family 74(2):377–87.

McKinnish, Terra G. 2007. “Sexually Integrated Work-

places and Divorce: Another Form of On-the-Job

Search.” Journal of Human Resources 42(2):331–52.

Nock, Steven L. 1998. Marriage in Men’s Lives. New

York: Oxford University Press.

Ono, Hiromi. 1998. “Husbands’ and Wives’ Resources

and Marital Dissolution.” Journal of Marriage and

the Family 60(3):674–89.

Oppenheimer, Valerie Kincade. 1997. “Women’s

Employment and the Gain to Marriage: The Special-

ization and Trading Model.” Annual Review of Soci-

ology 23:431–53.

Oppenheimer, Valerie Kincade, Matthijs Kalmijn, and

Nelson Lim. 1997. “Men’s Career Development and

Marriage Timing During a Period of Rising Inequal-

ity.” Demography 34(3):311–30.

Panel Study of Income Dynamics. 2015. Public Use

Dataset. Ann Arbor: Survey Research Center, Insti-

tute for Social Research, University of Michigan.

Poortman, Anne-Rigt, and Matthijs Kalmijn. 2002.

“Women’s Labour Market Position and Divorce in

the Netherlands: Evaluating Economic Interpreta-

tions of the Work Effect.” European Journal of Popu-

lation 18(2):175–202.

Risman, Barbara J. 2011. “Gender as Structure or Trump

Card?” Journal of Family Theory & Review 3(1):18–22.

Rogers, Stacy J. 1999. “Wives’ Income and Marital Qual-

ity: Are There Reciprocal Effects?” Journal of Mar-

riage and the Family 61(1):123–32.

Rogers, Stacy J. 2004. “Dollars, Dependency, and Divorce:

Four Perspectives on the Role of Wives’ Income.”

Journal of Marriage and Family 66(1):59–74.

Rogers, Stacy J., and Danelle D. DeBoer. 2001. “Changes

in Wives’ Income: Effects on Marital Happiness, Psy-

chological Well-Being, and the Risk of Divorce.”

Journal of Marriage and Family 63(2):458–72.

Ruggles, Steven. 1997. “The Rise of Divorce and Separa-

tion in the United States, 1880–1990.” Demography

34(4):455–66.

Ruggles, Steven, Katie Genadek, Ronald Goeken, Josiah

Grover, and Matthew Sobek. 2015. “Integrated Pub-

lic Use Microdata Series: Version 6.0 [Machine-

Readable Database].” Minneapolis: University of

Minnesota.

Killewald 719

Sayer, Liana C., and Suzanne M. Bianchi. 2000. “Wom-

en’s Economic Independence and the Probability of

Divorce: A Review and Reexamination.” Journal of

Family Issues 21(7):906–943.

Sayer, Liana C., Paula England, Paul D. Allison, and

Nicole Kangas. 2011. “She Left, He Left: How

Employment and Satisfaction Affect Women’s and

Men’s Decisions to Leave Marriages.” American

Journal of Sociology 116(6):1982–2018.

Schoen, Robert, Nan Marie Astone, Kendra Rothert,

Nicola J. Standish, and Young J. Kim. 2002. “Wom-

en’s Employment, Marital Happiness, and Divorce.”

Social Forces 81(2):643–62.

Schoen, Robert, Stacy J. Rogers, and Paul R. Amato.

2006. “Wives’ Employment and Spouses’ Marital

Happiness: Assessing the Direction of Influence

Using Longitudinal Couple Data.” Journal of Family

Issues 27(4):506–528.

Schwartz, Christine R., and Pilar Gonalons-Pons. Forth-

coming. “Trends in Relative Earnings and Marital

Dissolution: Are Wives Who Outearn Their Husbands

Still More Likely to Divorce?” RSF: The Russell Sage

Foundation Journal of the Social Sciences.

Smock, Pamela J., Wendy D. Manning, and Sanjiv Gupta.

1999. “The Effect of Marriage and Divorce on Wom-

en’s Economic Well-Being.” American Sociological

Review 64(6):794–812.

South, Scott J. 2001. “Time-Dependent Effects of Wives’

Employment on Marital Dissolution.” American

Sociological Review 66(2):226–45.

Spitze, Glenna, and Scott J. South. 1985. “Women’s

Employment, Time Expenditure, and Divorce.” Jour-

nal of Family Issues 6(3):307–329.

Stevenson, Betsey, and Justin Wolfers. 2007. “Marriage

and Divorce: Changes and Their Driving Forces.”

Journal of Economic Perspectives 21(2):27–52.

Stone, Pamela. 2007. Opting Out? Why Women Really

Quit Careers and Head Home. Berkeley: University

of California Press.

Sweeney, Megan M. 2002. “Two Decades of Fam-

ily Change: The Shifting Economic Foundations

of Marriage.” American Sociological Review

67(1):132–47.

Tamborini, Christopher R., Howard M. Iams, and Gayle

L. Reznik. 2012. “Women’s Earnings before and after

Marital Dissolution: Evidence from Longitudinal

Earnings Records Matched to Survey Data.” Journal

of Family and Economic Issues 33(1):69–82.

Teachman, Jay D. 2002. “Stability across Cohorts in

Divorce Risk Factors.” Demography 39(2):331–51.

Teachman, Jay. 2010. “Wives’ Economic Resources

and Risk of Divorce.” Journal of Family Issues

31(10):1305–23.

Thornton, Arland, and Linda Young-DeMarco. 2001. “Four

Decades of Trends in Attitudes toward Family Issues in

the United States: The 1960s through the 1990s.” Jour-

nal of Marriage and Family 63(4):1009–37.

Tzeng, Jessie M., and Robert D. Mare. 1995. “Labor

Market and Socioeconomic Effects on Marital Stabil-

ity.” Social Science Research 24(4):329–51.

Tzeng, Meei-Shenn. 1992. “The Effects of Socioeco-

nomic Heterogamy and Changes on Marital Dissolu-

tion for First Marriages.” Journal of Marriage and

the Family 54(3):609–619.

West, Candace, and Don H. Zimmerman. 1987. “Doing

Gender.” Gender & Society 1(2):125–51.

Xie, Yu, James M. Raymo, Kimberly Goyette, and

Arland Thornton. 2003. “Economic Potential and

Entry into Marriage and Cohabitation.” Demography

40(2):351–67.

Alexandra Killewald is Professor of Sociology at Har-

vard University. Much of her research examines the

mutual relationships between family circumstances and

work outcomes in the United States, emphasizing how

gender shapes these associations and the role of the fam-

ily in social stratification.