ALEXANDRA KILLEWALD University of Michigan

Opting Out and Buying Out: Wives’ Earnings

and Housework Time

It has been proposed that the negativeassociation between wives’ earnings and theirtime in housework is due to greater outsourcingof household labor by households with high-earning wives, but this hypothesis has notbeen tested directly. In a sample of dual-earner married couples in the Consumptionand Activities Mail Survey of the Healthand Retirement Study (N = 796), use ofmarket substitutes for women’s houseworkwas found to be only weakly associated withwives’ time cooking and cleaning. Furthermore,expenditures on market substitutes explain lessthan 15% of the earnings – housework timerelationship. This suggests that use of marketsubstitutes plays a smaller role in explainingvariation in wives’ time in household labor thanhas previously been hypothesized.

Wives continue to spend more time than theirhusbands doing housework, even when bothspouses work full time (Kamo, 1988; Killewald& Gough, 2010). For those couples, domesticlabor is a source of gender stratification, asit contributes to unequal leisure time betweenspouses. Furthermore, wives’ time in houseworkis negatively associated with wages, thus making

Population Studies Center, Institute for Social Research,University of Michigan, 426 Thompson St., No. 2050, AnnArbor, MI 48106 (akillewa@umich.edu).

This article was edited by Suzanne Bianchi.

Key Words: families in middle and later life, Health andRetirement Study (HRS), housework/division of labor, timeuse, wives’ employment.

women’s greater domestic burden a contributorto the gender gap in wages (Hersch & Stratton,1997; Noonan, 2001).

It is therefore natural to ask what resourceswives may use to reduce their time in householdlabor. Existing studies indicate that wives’earnings are negatively associated with theirtime in housework, even after controlling fortime spent in market work (Gupta, 2006,2007; Killewald & Gough, 2010). Given thatwives’ earnings are positively associated withhousehold expenditures on market substitutes fortheir household labor and negatively associatedwith their time in housework, it has beenhypothesized that wives’ earnings allow them tooutsource household production (Cohen, 1998;de Ruijter, Treas, & Cohen, 2005; Gupta, 2006,2007; Gupta & Ash, 2008): Wives use theirearnings to buy out of time in housework.Nevertheless, the explanatory power of thebuying-out hypothesis has, to my knowledge,never been directly tested. An alternativecause of the negative earnings – houseworkrelationship is that higher earnings lead wivesto reduce their household labor hours, withoutpurchasing a market substitute for their owntime. In other words, high earners opt out bydoing less housework.

Testing the buying-out hypothesis has beendifficult because most data sets do not includeinformation on both housework time andhousehold expenditures. Brines (1994), usingdata from the Panel Study of Income Dynamics(PSID), found that expenditure on diningout relative to food consumed at homewas negatively associated with wives’ time

Journal of Marriage and Family 73 (April 2011): 459 – 471 459DOI:10.1111/j.1741-3737.2010.00818.x

460 Journal of Marriage and Family

in housework, but the PSID lacks data onexpenditures on other types of market substitutesfor housework. By linking time use andexpenditure measures from the Consumptionand Activities Mail Survey (CAMS) to earningsmeasures from its parent study, the Health andRetirement Study (HRS), I directly measuredthe extent to which use of market substitutesis associated with wives’ time in householdlabor. Furthermore, by comparing the results ofmodels of wives’ housework time before andafter the inclusion of a measure of use of marketsubstitutes, I measured the extent to which theexpenditures explain the negative relationshipbetween wives’ earnings and their houseworktime.

Understanding whether wives’ earningslessen the time they spend in household laborprimarily by increasing spending on marketsubstitutes has implications for understandingthe intersection between household responsi-bilities and market work. Acknowledging thatwives may not fully compensate for theirreduced housework time by increased use ofmarket substitutes introduces a new parameterinto the household decision-making model: thelevel of household production. Households makechoices about, for example, the degree of clean-liness of the home and the quality of the food thathousehold members consume. Household laborneed not be viewed as purely a task of allocationbetween spouses or between private productionand purchased commodities. Rather, householdsare also making trade-offs between the amountof domestic production and the amount of otheritems the household values, including leisuretime and consumption goods.

Buying Out and Opting Out

The relationship between wives’ earnings andtheir time in housework is both statisticallysignificant and practically large. Using datafrom the National Survey of Families andHouseholds (NSFH), Gupta (2006) found that,compared with wives in the lowest quartileof the earnings distribution, those in the topquartile spent 13 fewer hours per week inhousework, which implies a 40% reduction.Among couples in which both spouses workedfull time, wives in the second quartile of theearnings distribution spent 26 hours per week inhousework, compared with 18 hours for womenin the highest quartile, a 30% reduction. Thus,

the relationship is not entirely due to differencesin labor market work by high- and low-earningwives.

Households face decisions about both theuse of their members’ time and the ways tospend available financial resources. Domesticproduction, such as meals and a clean homeand clothes, can be created either directly,through the time inputs of household members,or indirectly, using the financial rewards frommarket work to purchase the services of others.If wives’ earnings are negatively related totheir housework hours, this may be becauseearnings are negatively correlated with the totalamount of household production or becausethey are positively correlated with the fractionof domestic production that is outsourced orallocated to other household members, or both.

The buying-out explanation for the negativerelationship between wives’ earnings and theirhousework hours suggests that wives’ earningsgive them the purchasing power to buy marketsubstitutes for their own household labor. Thisexplanation focuses on the positive correlationbetween wives’ earnings and the fraction ofdomestic production that is outsourced. Theassumption that higher income households usetheir financial resources to purchase marketsubstitutes that allow household members toreduce their own time in housework is frequentlygiven as a reason to control for household incomein models of individuals’ time in housework(e.g., Bittman, England, Sayer, Folbre, &Matheson, 2003; Brines, 1994; Evertsson &Nermo, 2004). The buying-out hypothesis isalso invoked in recent studies that recognize thedistinct effects of husbands’ and wives’ earningson wives’ housework time (Gupta, 2006, 2007;Gupta & Ash, 2008).

The buying-out hypothesis is appealingin part because it draws on a standardeconomic model of consumption: Conditionalon the amount of time spent in the laborforce, wives with higher earnings have greaterfinancial resources to outsource domestic labor,thereby ‘‘purchasing’’ additional leisure time forthemselves. Assuming that wives enjoy leisuremore than housework and again conditioning ontime in paid work, wives with greater earningsshould devote less time to housework andmore to leisure. If wives use their earningsto buy out of time in household labor, wewould expect that household expenditures onmarket substitutes for household labor rise with

Opting Out and Buying Out 461

household income, as high-income householdshave greater financial resources to purchasethe wife’s leisure time. This does not implythat there are no constraints on households thatprevent them from outsourcing household labor,merely that, all else equal, households with morefinancial resources are better able to overcomethose constraints and reduce the wife’s time inhousehold labor than are households with fewerfinancial resources.

Furthermore, the wife is likely to valueher own leisure time more highly thanher husband does. Even if spouses pooltheir incomes, existing evidence indicates thatspending on goods that wives value or arewithin wives’ sphere of traditional responsibilityrises more quickly with wives’ earnings thanwith husbands’ (Phipps & Burton, 1998). Asa result, the buying-out hypothesis predictsthat expenditures on market substitutes forwives’ household-labor time rise more quicklywith wives’ earnings than with husbands’.As expected, couples’ spending on marketsubstitutes for housework women typicallyperform, including child-care services andcleaning services, have been found to havea stronger positive association with wives’earnings than with husbands’ (Cohen, 1998;Oropesa, 1993; Phipps & Burton; Soberon-Ferrer & Dardis, 1991), although some studieshave found that dining out responds equallystrongly to husbands’ and wives’ earnings(Cohen; Oropesa). This is also consistent withpast evidence that wives’ time in houseworkfalls more rapidly with their own earnings thanwith the earnings of their husbands (Gupta,2006, 2007; Gupta & Ash, 2008; Killewald &Gough, 2010).

If use of market substitutes is an importantmechanism by which wives reduce their timein household labor as their earnings rise,use of market substitutes should be stronglynegatively related to wives’ time in housework.Furthermore, if the use of market substitutesfully explains the negative relationship betweenwives’ earnings and their housework hours, thenmodels of wives’ housework time that include ameasure of households’ use of market substitutesshould show no remaining association betweenwives’ earnings and their time in housework. Inother words, the link between wives’ earningsand their time in housework is entirely indirect,as it operates through increased expenditures onmarket substitutes.

Buying out, however, is not the only possiblesource of the negative relationship betweenwives’ earnings and their housework hours.It is possible that housework hours are lowerfor wives with higher earnings because theaverage level of domestic production in theirhouseholds is lower. This may occur for tworeasons: opting out and selection. Wives mayrespond to earnings increases by opting out ofhousework, forgoing time in household laborwithout purchasing a market substitute. Wiveswith higher earnings may have a preference forlower levels of domestic production because therewards for high levels of domestic productionare not as great for them. High-earning wivesmay face less social pressure to performthe traditionally female tasks of householdproduction, as has sometimes been suggested(Gupta, 2006, 2007; Gupta & Ash, 2008). Allwomen may feel pressure to perform houseworkas a way to ‘‘do’’ gender and to express affectionfor household members (Berk, 1985; DeVault,1991; Hochschild, 1989; West & Zimmerman,1987), but that pressure may not operate withequal force on all women. High-earning womenmay be particularly likely to derive morepersonal satisfaction and social status fromtheir labor-market roles, which makes the statusascribed for performing traditional householdproduction less important.

Furthermore, the effort required to procureadequate, trustworthy substitutes for wives’time in household labor, as well as reluctanceto trust service providers (especially whentheir employers do not easily observe theirbehavior) may reduce households’ desire tooutsource many aspects of market production(de Ruijter, van der Lippe, & Raub, 2003).Therefore, households that choose to reducewives’ time in housework still may not bewilling to bear the financial and nonfinancialcosts of purchasing market substitutes and maychoose instead to reduce household production.If wives are both opting out and buying out inresponse to earnings increases, expenditures onmarket substitutes will only partially mediate theearnings – housework relationship, and modelsthat include a control for expenditures on marketsubstitutes will continue to show a residualassociation between wives’ earnings and theirhousework time.

It is also possible that the observed negativerelationship between wives’ earnings and theirtime in housework is entirely spurious, as a

462 Journal of Marriage and Family

result of differences among women in tastefor housework or domestic production that arecorrelated with, but not caused by, differencesin earnings. In this case, too, we would expectto see lower levels of domestic production andless time spent in housework for wives withhigher earnings, and controlling for expenditureson market substitutes would not eliminate theearnings – housework relationship.

There is some evidence that use of marketsubstitutes may reduce women’s time inhousework. Van der Lippe, Tijdens, and deRuijter (2004), using data from the DutchNational Time Budget survey, found thatfrequency of takeout meals is associated withless time cooking for both men and women andthat use of cleaning services reduces women’stime cleaning. Similarly, Bittman, Rice, andWajcman (2004), using the Australian 1997Time Use Survey, found that having hiredsomeone to clean the home in the previous2 weeks was associated with significantly lesstotal time in housework for women, althoughthe number of times restaurant or takeout mealswere purchased over the same period was not.Nonetheless, neither study tested whether theuse of these market substitutes explained thenegative association between women’s earningsand their housework time.

METHOD

The sample was drawn from the Healthand Retirement Study (HRS) (Health andRetirement Study; RAND HRS Data, VersionJ). The Consumption and Activities MailSurvey (CAMS), which in each wave includeda subsample of households from the HRSsample, asked individuals about their own timein housework and household expenditures onvarious items. The 2003, 2005, 2007, and2009 waves of CAMS were merged with theHRS Core surveys of 2002, 2004, 2006, and2008. In the 2005 – 2009 waves of CAMS,the spouses of primary respondents were askedto complete a shorter interview that includedmeasures of the respondent’s time use but didnot repeat the household expenditure measures.In those cases, I imputed to both spousesthe level of household expenditure that theprimary household respondent reported. Forthe periods considered here, the HRS is arepresentative sample of the U.S. populationolder than age 50, although the sample

included younger spouses as well. This articlehighlights the experiences of a relativelyunderstudied group: mature couples, typicallyliving without young children in the household,but before retirement. As the population ages,understanding the experiences of this populationbecomes increasingly valuable.

In the analyses that follow, the dependentvariables are wives’ self-reported time in mealpreparation and cleanup (cooking) and cleaninghouse, washing, ironing, and mending (cleaning)in the week before the survey. Together, thesetasks—cooking, doing dishes, ironing, washing,and cleaning house—account for more than70% of women’s time in noncare householdactivities, including for women aged 45 – 64(Krantz-Kent, 2009).

Measuring Substitution

Before I discuss the measure of marketsubstitutes employed here, it is useful to thinkabout what such a measure would ideallycapture. By ‘‘market substitutes,’’ I mean anygood or service that an individual purchasesto increase household production, withoutperforming household labor herself. Purchasingtakeout food, hiring someone to clean the homeor mow the lawn, or sending shirts to a laundryare all examples of using market substitutes,sometimes also referred to as outsourcing.To test the extent to which wives substitutepurchased services for their own time inhousework, it would be ideal to determine theamount of wives’ own time that the purchasedgoods replace. For example, if a wife hires adomestic cleaner, how many hours would thewife have spent to accomplish the work thecleaner did?

There is not, of course, such an ideal measure.Instead, households’ expenditures on dining out(cooking) and housekeeping or laundry services(cleaning) are here considered expenditures onmarket substitutes for women’s housework time.This is consistent with existing research (Cohen,1998; de Ruijter et al., 2005; Oropesa, 1993;Treas & de Ruijter, 2008), although spending onlaundry services is sometimes excluded (Cohen;Oropesa).

In CAMS, individuals were asked to reporttheir expenditures on ‘‘Housekeeping, drycleaning and laundry services: hiring costsfor housekeeping or home cleaning, andamount spent at dry cleaners or laundries’’

Opting Out and Buying Out 463

and ‘‘Dining and/or drinking out: items inrestaurants, cafes, and diners, including take-outfood.’’ This measure is not without limitations.Expenditures in these areas may increase withoutreducing wives’ housework time. For example,consuming alcoholic beverages in restaurantsrather than at home would register as increasedspending on food away from home but wouldbe a poor measure of money spent reducingwives’ time cooking. Likewise, a householdmay spend more money on laundry servicessimply because it has purchased more clothesthat require dry cleaning.

In addition, particularly for food, variationin spending on market substitutes may reflectvariation in the quality of services purchasedrather than the amount of the wife’s timethat is replaced. Alternatively, wives’ greaterearnings may lead to greater tolerance of foodthat is purchased but ultimately goes uneatenby household members. These effects introducemeasurement error into the substitution measure,which will lead to a downward bias in theestimated relationship between the use of marketsubstitutes and wives’ housework time.

To the extent that increasing expenditureson market substitutes reflect quality increasesand tolerance for wastage, a similar effectshould also lead to increased spending ongroceries and, to a lesser extent, cleaningsupplies, which are complements to rather thansubstitutes for wives’ housework time. CAMSalso asked individuals about expenditures on‘‘Housekeeping supplies: cleaning and laundryproducts’’ and ‘‘Food and beverages: food anddrinks, including alcoholic, that you buy ingrocery or other stores.’’ To capture moreaccurately the extent of substitution for thewife’s housework, a measure of the share ofthe household’s expenditures in the domain ofcooking or cleaning that is spent on marketsubstitutes rather than on complements wasconstructed. This is given as follows: share =100 × (spending on substitutes)/(spending onsubstitutes + spending on complements).

Brines’s (1994) measure of spending onrestaurant meals relative to spending on foodprepared at home is similar in spirit. It isexpected that this measure is more stronglyrelated to wives’ housework time than is theabsolute level of expenditures on substitutes, asit more strongly indicates the extent to whichmarket substitutes, rather than complements,are used. Nonetheless, the measure is still not

perfect. For example, if wives choose to purchaseprepared foods that are more expensive than rawingredients but require relatively less time tocook, this inflates expenditures on groceries butin fact indicates greater use of market substitutes.

Model Specification

The analytic technique was ordinary leastsquares (OLS). Wives’ hours spent cookingand cleaning in the previous week were thedependent variables. Wives’ annual earnings inthe calendar year before the HRS survey wasthe primary independent variable. Husbands’earnings (also in the calendar year before theHRS survey), the usual weekly labor markethours of each spouse in main and secondaryjobs at the time of the HRS survey, anddummy variables for whether the wife isAfrican American and whether each spousehas a bachelor’s degree were included ascovariates. Because race, education, and labor-force participation are correlated with householdexpenditures on market substitutes and withhousework time (Baxter, Hewitt, & Haynes,2008; Bellante & Foster, 1984; Cohen, 1998; deRuijter et al., 2005; Pittman & Blanchard, 1996;Sanchez & Thomson, 1997; South & Spitze,1994) but are also associated with earnings,failure to control for those variables would riskconfounding their effects on expenditures andhousework time with those of earnings.

Two measures were used to adjust fordifferences across households in the demandsfor domestic production from people and space:the number of rooms in the family’s home anda dummy variable set to 1 if members of thehousehold include anyone other than the couple.

A measure of the wife’s employment historyand the number of children ever born to the wifewere included as indicators of her relative tastefor home production as opposed to market work.Furthermore, a wife’s employment and fertilityhistory may affect the household’s currentdivision of labor, net of current employmenthours, if household roles negotiated earlier inthe marriage shape her own and her husband’sexpectations of behavior later in life. Theemployment history measure was constructedby dividing the wife’s years of employmentto date by the difference between her presentage and 14. This approximates the share of thewife’s adult life that she has spent employed,although it is a coarse measure and subject

464 Journal of Marriage and Family

to measurement error in respondents’ reportedwork history.

Of the 12,052 observations of women fromCAMS, 237 (2.0%) were excluded becausetheir information could not be matched toHRS reports from the previous calendar year.To focus the analysis on married, working-age couples, women whose marital status wassomething other than married were excluded(5,513 observations, 45.7%), as were couplesin which either spouse was older than age 65(4,084 observations, 33.9%). Couples in whicheither spouse was not in the labor force at thetime of the HRS survey or reported no earningsin the previous calendar year were excludedfrom the analysis (1,374 observations, 11.4%).This restricts the focus of the analysis to dual-earner couples. Retired couples were excludedbecause their earnings are a poor measure of theirfinancial resources. Household sample weightsof 0 led to the exclusion of 11 observations(0.09%). An additional 11 observations (0.09%)were excluded because they were missing dataon one of the variables used to define the sample:marital status or age, earnings, or retirementstatus of either spouse.

After excluding couples who did not meetthe age or employment criteria, 822 couple-yearobservations remained in the sample. Of those,12 observations (1.5% of the remaining sample),were removed because of reporting no expendi-ture at all in the domains of either cooking orcleaning, suggesting that the couples have someother means of household production that directspending on household goods or services doesnot capture. Missing values on the dependentvariable—the wife’s time in housework—led toa loss of 14 observations, or 1.7% of the remain-ing sample. The final analytic sample included796 observations from 449 wives.

In the analytic sample, 2.1% of the obser-vations were missing data on the usual hoursspent in paid work by the husband, and 2.3%were missing this information for the wife. Thenumber of rooms in the house was missing for7.2% of the sample and the number of childrenever born was missing for 0.1% of the sample.Last, 5.6% of the sample was missing data onone of the expenditure variables related to clean-ing, and 5.1% was missing data on one of thecooking expenditure variables. For each of thesevariables, an indicator variable was created thatwas set to 1 if the observation was missing dataon this covariate. The indicator for missing data

was included in any model that includes theassociated covariate.

The HRS household-level weights, normal-ized to average one in each year in the full sampleof HRS households in each wave, were used toweight the sample in all analyses. Because wivesmay be represented in multiple waves, all analy-ses clustered the standard errors at the individuallevel. The top 5% of both time-use and financialvariables were recoded to the 95th percentile,as were the variables for the number of roomsin the home and the number of children everborn. To adjust for inflation during the period,financial variables were scaled to 2009 dollars.

RESULTS

Table 1 presents descriptive statistics for thesample. The mean annual earnings were $43,111for wives and $68,352 for husbands. Husbandsalso spent somewhat more time in the labor mar-ket—an average of 44.9 hours per week, com-pared with 38.0 hours for wives. Thus, in thissample of dual-earner couples, husbands bothsubstantially outearned their wives and weremore engaged in the labor market. A bachelor’sdegree was held by 36% of wives and 40%of husbands. African Americans comprised 4%of the sample. Couples had an average of 7.2rooms in their homes, and 44% of couples livedwith at least one other resident. On average,wives had spent 73% of their years since age 14in the labor market and had given birth to 2.2children. Wives were 54.1 years old on average,compared with 56.8 years old for husbands.

Wives spent an average of 6.6 hours per weekcooking and 8.0 hours per week cleaning anddoing laundry. Almost all households reportedsome spending on groceries, cleaning supplies,and dining out: More than 97% of the samplehouseholds reported some expenditure on eachof those items. By contrast, only 56% of house-holds reported any spending on house-cleaningor laundry services in the previous year. Amongthose who buy such goods or services, meanmonthly expenditures were $66 for laundry andhousecleaning services, $35 for cleaning andlaundry supplies, $217 for dining out, and $485for groceries.

Multivariate Results

Table 2 presents two models of wives’ time inhousehold labor. Model 2 is identical to Model 1,

Opting Out and Buying Out 465

Table 1. Spouses’ Characteristics: Descriptive Statistics(N = 796)

Variables M SD

Wife’s annual earnings $43,111.17 $31,828.80Husband’s annual earnings $68,351.66 $43,050.04Wife has bachelor’s degreea 0.36 0.48Husband has bachelor’s

degreeb0.40 0.49

Wife is African Americanc 0.04 0.20Wife’s weekly labor-market

hours38.03 11.67

Husband’s weeklylabor-market hours

44.93 11.63

Residents other than coupled 0.44 0.50Rooms in the home 7.23 1.86Share of wife’s adult years

employed0.73 0.20

Number of children ever bornto wife

2.16 1.31

Wife’s age 54.05 4.91Husband’s age 56.81 4.51Wife’s weekly hours cooking 6.62 4.29Wife’s weekly hours cleaning 7.97 5.44Monthly spending on

cleaning/laundry services,if positive

$66.32 $77.85

Monthly spending oncleaning/laundry supplies,if positive

$35.33 $27.41

Monthly spending on diningout, if positive

$217.32 $177.50

Monthly spending ongroceries, if positive

$485.34 $255.98

aWife has bachelor’s degree: 0 = no; 1 = yes. bHusbandhas bachelor’s degree: 0 = no; 1 = yes. cWife is AfricanAmerican: 0 = no; 1 = yes. dResidents other than couple:0 = no; 1 = yes.

except that it also includes the measure of thehousehold’s level of use of market substitutes.The reduction in the size of the coefficient onwives’ earnings between Model 1 and Model2 indicates the extent to which greater use ofmarket substitutes in households in which wiveshave higher earnings explains the relationshipbetween wives’ earnings and their time inhousework.

In Model 1, wives’ earnings were significantlynegatively associated with their time in bothcooking and cleaning, consistent with existingevidence. For each $10,000 increase in a wife’sannual earnings, her weekly time in cleaning was

predicted to be 0.21 hours (13 minutes) lower,and her weekly time cooking was predicted to be0.19 hours (11 minutes) lower. Consistent withpast research, in models of wives’ time in bothcleaning and cooking, it was possible to reject thenull hypothesis that the coefficients on husbands’and wives’ earnings are equal, F(1,448) = 7.49,p < 0.01, and F(1,448) = 8.65, p < 0.01, andin fact husbands’ earnings were positively andnot significantly associated with wives’ timeboth cooking and cleaning. This suggests thatincreased husbands’ earnings do not translateinto less housework for wives.

Model 2, of Table 2, presents the results fromthe models that include the measure of thehousehold’s use of market substitutes. For bothcooking and cleaning, wives’ housework timewas predicted to be lower when their householdsmade greater use of market substitutes, butthe relationship was weak and only marginallysignificant in the model of time cleaning. Anincrease of 1 percentage point in a household’sreliance on market substitutes was associatedwith a predicted decline of 0.01 hours (1minute) in a wife’s weekly time cleaning andof 0.02 hours (1 minute) in her weekly timecooking. Even considerable changes in the useof market substitutes were therefore associatedwith small changes in wives’ housework time.

Controlling for use of market substitutesreduces the negative association between wives’earnings and their time spent in housework, butthe reduction was moderate in both models: 12%in the model of time spent cleaning and 2% inthe model of time spent cooking. The greaterrole of market substitutes in mediating thecleaning – earnings relationship as compared tothe cooking – earnings relationship is consistentwith the results of models of households’expenditures on market substitutes, whichshowed that wives’ earnings were positively andsignificantly associated with their reliance onsubstitutes for wives’ time cleaning but not withreliance on substitutes for wives’ time cooking(see Appendix Table 1 in the online materialassociated with this article).

After controlling for the use of marketsubstitutes, wives’ earnings and the level of useof substitutes were the only significant predictorsin the model of wives’ time cooking, and themodel explained only 7% of the variation inwives’ time cooking. A larger number of controlvariables were significant in the model of wives’time cleaning, which explained 16% of the

466 Journal of Marriage and Family

Tab

le2.

Sum

mar

yof

OLS

Reg

ress

ion

Ana

lyse

sfo

rW

ives

’W

eekl

yH

ouse

wor

kH

ours

(N=

796)

Cle

anin

gC

ooki

ng

Mod

el1

Mod

el2

Mod

el1

Mod

el2

BSE

BB

SEB

BSE

BB

SEB

Wif

e’s

annu

alea

rnin

gs($

10K

s)−0

.21∗

∗0.

08−0

.19∗

0.08

−0.1

9∗∗

0.06

−0.1

8∗∗

0.07

Hus

band

’san

nual

earn

ings

($10

Ks)

0.07

0.05

0.09

†0.

050.

060.

050.

060.

05W

ife

has

bach

elor

’sde

gree

a−2

.21∗∗

∗0.

48−2

.12∗∗

∗0.

48−0

.35

0.44

−0.2

60.

43H

usba

ndha

sba

chel

or’s

degr

eeb

0.09

0.54

0.17

0.54

−0.1

80.

41−0

.20

0.41

Wif

eis

Afr

ican

Am

eric

anc

1.09

1.60

1.26

1.62

0.20

1.18

0.16

1.17

Wif

e’s

wee

kly

labo

r-m

arke

thou

rs−0

.03

0.02

−0.0

30.

02−0

.02

0.02

−0.0

20.

02H

usba

nd’s

wee

kly

labo

r-m

arke

thou

rs0.

010.

020.

010.

02−0

.01

0.02

−0.0

00.

02R

esid

ents

othe

rth

anco

uple

d0.

680.

470.

600.

460.

72†

0.38

0.59

0.37

Roo

ms

inth

eho

me

−0.3

3∗∗0.

12−0

.34∗∗

0.12

−0.1

70.

11−0

.16

0.10

Shar

eof

wif

e’s

adul

tyea

rsem

ploy

ed1.

641.

181.

751.

180.

131.

040.

271.

04N

umbe

rof

child

ren

ever

born

tow

ife

0.44

∗0.

200.

44∗

0.20

0.17

0.15

0.19

0.15

Shar

eof

expe

nditu

res

onsu

bstit

utes

——

−0.0

1†0.

01—

—−0

.02∗

0.01

Con

stan

t9.

83∗∗

∗1.

369.

64∗∗

∗1.

368.

63∗∗

∗1.

388.

94∗∗

∗1.

41R

20.

150.

160.

060.

07

Not

e:A

llm

odel

sal

soin

clud

efo

urin

dica

tor

vari

able

sfo

rw

heth

erth

ere

spon

dent

ism

issi

ngva

lidda

taon

the

labo

r-m

arke

tho

urs

ofth

ehu

sban

d,th

ela

bor-

mar

ket

hour

sof

the

wif

e,th

enu

mbe

rof

room

sin

the

hom

e,an

dth

enu

mbe

rof

child

ren

ever

born

toth

ew

ife.

Mod

el2

incl

udes

anin

dica

tor

vari

able

for

whe

ther

the

resp

onde

ntis

mis

sing

valid

data

onth

esh

are

ofex

pend

iture

son

subs

titut

es.A

llst

anda

rder

rors

are

clus

tere

dat

the

indi

vidu

alle

vel.

a Wif

eha

sba

chel

or’s

degr

ee:0

=no

;1=

yes.

bH

usba

ndha

sba

chel

or’s

degr

ee:0

=no

;1=

yes.

c Wif

eis

Afr

ican

Am

eric

an:0

=no

;1=

yes.

dR

esid

ents

othe

rth

anco

uple

:0=

no;

1=

yes.

†p<

.10.

∗ p<

.05.

∗∗p<

.01.

∗∗∗ p

<0.

001.

Opting Out and Buying Out 467

variation in wives’ time cleaning, after includingthe measure of use of market substitutes. Wiveswith bachelor’s degrees spent, on average, 2.1fewer hours per week cleaning than other wives,even after including controls for earnings, labor-force hours, and use of market substitutes. Thenumber of children ever born was positivelyand significantly associated with wives’ timecleaning, as expected. Each additional child bornto the wife was associated with a predictedincrease of 0.4 hours per week in her timecleaning, net of other controls. Each additionalroom in the house was associated with apredicted decline of 0.3 hours per week in thewife’s time cleaning, contrary to expectations.A possibility for this counterintuitive finding isthat larger homes are less cluttered and thereforeeasier to clean. Neither the share of the wife’sadult life spent working nor the presence ofresidents other than the couple in the householdwas significantly associated with the wife’shousework time.

Husbands’ Housework

The main results may be limited by their neglectof husbands’ time in housework. First, it ispossible that wives’ earnings enable them tonegotiate greater time in housework by theirhusbands. If this is the case, including husbands’housework hours in the model of wives’ timeuse should reduce the negative relationshipbetween wives’ earnings and their time spenton housework. In addition, increased spendingon substitutes may be used to reduce husbands’housework time, rather than wives’, thus leadingto a weak association between households’ use ofmarket substitutes and wives’ housework time.

To test these hypotheses, I used the subsetof 362 wives (n = 609) from the 2005 – 2009CAMS for whom husbands’ housework timewas also available. The husbands in thissubsample reported spending an average of3.1 hours per week cleaning and 3.0 hours perweek cooking. Including husbands’ houseworkhours as a covariate in the models of wives’housework time revealed that husbands’ timecleaning was positively and not significantlyassociated with wives’ time cleaning, whereashusbands’ time cooking was negatively andmarginally significantly associated with wives’time cooking (full results available in AppendixTable 2 in the online material associatedwith this article). Nonetheless, controlling

for husbands’ time in housework slightlyincreased the negative association betweenwives’ earnings and their time both cooking andcleaning. Therefore, although wives cooked lesswhen their husbands cooked more, the negativeassociation between wives’ earnings and theirtime in housework does not appear to be dueto high-earning wives reallocating housework totheir husbands.

Second, I repeated the analysis of Table 2using husbands’ time in housework as theoutcome (full results available in AppendixTable 3 of the online material associatedwith this article). If high-earning wives werereallocating housework to husbands, we wouldexpect to see a positive association betweenwives’ earnings and their husbands’ houseworktime. If high-earning wives use their earningsto purchase substitutes for their husbands’housework time, we would expect to see anegative relationship between wives’ earningsand their husbands’ housework time, and wewould expect that the association would besubstantially reduced after controlling for thehousehold’s use of market substitutes. Wives’earnings were not significantly associated withhusbands’ time cooking but were marginallysignificantly and negatively associated withhusbands’ time cleaning once the measureof use of market substitutes was included.Despite this, the household’s use of marketsubstitutes was not significantly associated withthe husband’s time cooking or cleaning, and thenegative relationship between wives’ earningsand their husbands’ cleaning time was slightlygreater after the measure of use of marketsubstitutes is added to the model. Thus, althoughhusbands married to high-earning wives spendless time cleaning, this does not appear to bebecause wives’ earnings are used to purchasesubstitutes for their husbands’ household labortime. Instead, it is possible that high-earningwives not only opt out of housework themselvesbut also allow their husbands to do so.

Alternative Specifications

This section considers various possible sourcesof heterogeneity that may be associated withboth wives’ earnings and their time in house-hold labor, as well as alternative specificationsof several independent variables. Under eachalternative specification, the models presentedin Table 2 were repeated to test whether the

468 Journal of Marriage and Family

inclusion of the additional variables alters themain conclusions. Spouses’ ages, whether eitherspouse is younger than age 50, whether eitherspouse is in poor health, household income otherthan the earnings of spouses, and year of thereport were all tested as possible omitted vari-ables. The fraction of the earnings – cleaningrelationship explained by the inclusion ofthe market substitute measure ranged from11% to 16%, and from 1% and 3% for theearnings – cooking relationship. The results aretherefore quite similar to those in the mainmodels.

Alternative specifications of the finan-cial variables were also considered. First,I used absolute rather than relative expen-ditures on market substitutes in the time-use models but found that the inclusion ofthose variables reduced the magnitude of theearnings – cleaning relationship by only 4% andthe earnings – cooking relationship by less than1%, and the level of expenditure was not asignificant predictor of time in either cleaningor cooking. Second, because spouses’ wagesrather than earnings determine the opportunitycost of each hour of housework, I repeated themodels replacing spouses’ earnings with theirhourly wage. Wives’ wages were negativelyand marginally significantly related to theirtime cleaning, and the use of market substitutesexplained 8% of the relationship. Wives’ wageswere positively and not significantly related totheir time cooking. Third, I included a linearmeasure of the share of the couple’s earningsthat the wife earns, because spouses’ bargain-ing positions may affect both their division ofhousehold labor and their expenditure decisions.The relative earnings measure was not signif-icant in any of the models. It is also possiblethat spouses’ relative earnings may be nonlin-early related to their time in housework (Bittmanet al. 2003; Brines 1994), but when I added aquadratic term for spouses’ relative earnings tothe linear term, neither term was significant, norwere they jointly significant.

DISCUSSION

The results presented here go beyond existingwork by explicitly considering the relationshipsamong wives’ earnings, household expenditureson market substitutes for wives’ time inhousehold labor, and wives’ housework time.Household expenditures on domestic substitutes

were negatively related to wives’ time inhousehold labor, although the association is farweaker than might be supposed. The inclusionof measures of the use of market substitutesin models of wives’ time in household laborexplained 12% of the negative associationbetween wives’ time in cleaning and theirearnings and 2% of the relationship betweenwives’ time in cooking and their earnings.

Throughout, I found stronger associations inthe domain of cleaning than cooking. The modelexplained a greater share of the variation inwives’ time cleaning than in their time cooking,and the inclusion of the measure of use of mar-ket substitutes reduced the cleaning – earningsassociation more than the cooking – earningsassociation. A reason for this may be that clean-ing is a more uniformly undesirable activity,whereas some forms of cooking are enjoyablefor at least some wives, so wives are more likelyto use their earnings to outsource cleaning. Fur-thermore, dining out is a recreational activityfor spouses, as well as a market substitute forwives’ time cooking (Cohen, 1998; Oropesa,1993). At a minimum, the results suggest that itmay be inappropriate to treat women’s time inhousehold labor as homogenous.

This study has several limitations. First,the sample size is small, which limits thepower of the statistical tests. Second, althoughthe analyses presented suggest that buyingout is not the only explanation for thenegative relationship between wives’ earningsand their time in housework, they do notdistinguish between opting out and unobservedheterogeneity among wives as the source of theresidual earnings – housework association, net ofuse of market substitutes. As the primary focus ofthis analysis was to call into question the powerof the buying-out hypothesis, this ambiguitydoes not threaten the main conclusion. Moreextensive panel data on earnings, houseworktime and use of market substitutes wouldprovide an even richer test of the source ofthe earnings – housework relationship.

The measures available also limit analyses.First, housework hours and household expen-ditures were measured in the calendar yearfollowing the collection of information onspouses’ earnings, labor-force hours, and house-hold composition, so the measures are not con-temporaneous. Second, the available measureof substitution is limited. Although cooking,cleaning, and laundry constitute a substantial

Opting Out and Buying Out 469

share of women’s noncare housework time, theanalyses did not include all forms of householdlabor. In addition, not all household expendi-tures intended to reduce wives’ housework timeare included in the measure of expenditures onmarket substitutes. For example, purchases oflabor-saving devices have the potential to reducewives’ household labor time, although the evi-dence of their effectiveness in this regard ismixed (Bittman et al., 2004; van der Lippe et al.,2004). Furthermore, among the types of expen-ditures considered, households may report theirexpenditures with error and, as discussed earlier,individuals report the dollar amount spent ongoods rather than the goods’ capacity to reducehousehold labor time. Such measurement errormay partially explain the low predictive powerof use of market substitutes in models of wives’housework time. If this is true, it minimallysuggests that the documented positive relation-ship between wives’ earnings and householdexpenditures on domestic services and restau-rant meals cannot be interpreted as evidence thatwives are using those goods to substitute fortheir own time in housework.

The extent to which the findings presentedhere generalize to younger samples of womenis unknown. Younger women, particularly thosewith children in the household, may experiencestronger time pressure that motivates them to useincreased earnings to buy some relief in terms ofhousehold work. In addition to age differences,cohort differences may exist. Women born inlater birth cohorts may be more willing tooutsource household labor when they possessthe financial resources to do so.

Despite the limitations, the results of thisstudy shed light on the degree to which use ofmarket substitutes explains variation in wives’time in housework and, in particular, to whatextent increased use of market substitutes byhigh-earning women explains the negative rela-tionship between wives’ earnings and theirtime in housework. For those hypothesizinga role of market substitutes in mediating theearnings – housework relationship, this providesthe first direct evidence in favor of the buying-out story. At the same time, the results indicatethat the use of market substitutes is insufficient toexplain all, or even the majority, of the differencebetween high-earning and low-earning wives’time in household labor. Although it is not pos-sible in this study to determine the reason for theresidual negative relationship between wives’

earnings and their housework time, the resultssuggest that future research should recognize thepotential for wives to opt out of housework.

Furthermore, although wives’ time in house-work falls as their earnings rise, the rate isquite slow. A $10,000 increase in wives’ annualearnings is associated with a predicted declinein her combined cooking and cleaning time ofonly 0.4 hours. The fact that wives do not makeextensive use of market substitutes to replacetheir own time in household labor suggests thatnonfinancial concerns may motivate even high-earning women to invest time in domestic pro-duction, even though this reduces the amount oftime available for either market work or leisure.Even high-earning wives may feel compelledto perform at least some household labor them-selves, whether because they or their families donot perceive goods purchased in the market asadequate substitutes for the wife’s own time orbecause norms of doing gender suggest that itis appropriate for women to spend some time inhousework, regardless of the economic logic ofthis behavior. The idea that gendered norms ofbehavior operate to give greater responsibilityfor household labor to women than to men isnot new (Berk, 1985; Hochschild, 1989; West& Zimmerman, 1987), but the results presentedhere suggest that these norms may also operateto keep domestic production in the householdrather than outsourcing it to market goods andservices.

This analysis suggests a need for moreresearch understanding how couples make deci-sions about the level of household productionin their home. Just as the allocation of house-work between spouses provides insight aboutprocesses of household decision making andgender inequality in the household, explorationof the circumstances in which households reducedomestic production provides insight about themore general question of work – family trade-offs that households make. Existing research hastoo often treated the amount of housework tobe done as exogenous. As a result, we knowlittle about the social processes by which indi-viduals and households construct definitions ofappropriate levels of domestic production andmake decisions about time in housework asopposed to other activities or expenditures ondomestic production as opposed to other goods.Households are not bound to produce a fixedlevel of domestic production, with the onlydecisions being how much to outsource and how

470 Journal of Marriage and Family

to allocate the remaining labor among house-hold members. Instead, households also makedecisions about how much they value domes-tic production as opposed to other goods andactivities as well as which forms of householdproduction to retain and which to let go. Cer-tainly, individuals do not have perfect freedom tochoose the level and kind of domestic productionthey prefer, but it is equally erroneous to assumethat households have no control over such deci-sions. An understanding of domestic productiondecisions as households experience them musttherefore take seriously the full complement ofdomestic production options that families andindividuals face, including the decision to optout of certain types of domestic production.

NOTE

I acknowledge support from a National Institute on Agingtraining grant to the Population Studies Center at theUniversity of Michigan (T32 AG000221) and from theGerald R. Ford School of Public Policy and the QuantitativeMethodology Program in the Survey Research Center, both atthe University of Michigan. The HRS (Health and RetirementStudy) is sponsored by the National Institute on Aging(grant number NIA U01AG009740) and is conducted by theUniversity of Michigan. This analysis uses Early Releasedata from the Health and Retirement Study, RAND HRSData File (v.J), sponsored by the National Institute on Aging(grant number NIA U01AG009740) and conducted by theUniversity of Michigan. These data have not been cleanedand may contain errors that will be corrected in the FinalPublic Release version of the dataset. Amy Cooter, MaryCorcoran, Sheldon Danziger, Margaret Gough, Anju Paul,Jane Rochmes, Jeffrey Smith, Pamela Smock, Jessi Streib,Jessica Wiederspan, and Yu Xie provided helpful commentson earlier versions of this article.

REFERENCES

Baxter, J., Hewitt, B., & Haynes, M. (2008). Lifecourse transitions and housework: Marriage,parenthood, and time on housework. Jour-nal of Marriage and Family, 70, 259 – 272.doi:10.1111/j.1741-3737.2008.00479.x

Bellante, D., & Foster, A. C. (1984). Working wivesand expenditure on services. Journal of ConsumerResearch, 11, 700 – 707.

Berk, S. F. (1985). The gender factory: The appor-tionment of work in American households. NewYork: Plenum Press.

Bittman, M., England, P., Sayer, L., Folbre, N., &Matheson, G. (2003). When does gender trumpmoney? Bargaining and time in household work.American Journal of Sociology, 109, 186 – 214.doi:10.1086/378341

Bittman, M., Rice, J. M., & Wajcman, J. (2004).Appliances and their impact: The ownership of

domestic technology and time spent on householdwork. British Journal of Sociology, 55, 401 – 423.doi:10.1111/j.1468-4446.2004.00026.x

Brines, J. (1994). Economic dependency, gender, andthe division of labor at home. American Journal ofSociology, 100, 652 – 688.

Cohen, P. N. (1998). Replacing housework in theservice economy: Gender, class, and race-ethnicityin service spending. Gender and Society, 12,219 – 231. doi:10.1177/089124398012002006

De Ruijter, E., Treas, J. K., & Cohen, P. N. (2005).Outsourcing the gender factory: Living arrange-ments and service expenditures on femaleand male tasks. Social Forces, 84, 305 – 322.doi:10.1353/sof.2005.0124

De Ruijter, E., van der Lippe, T., & Raub, W.(2003). Trust problems in household outsourc-ing. Rationality and Society, 15, 473 – 507.doi:10.1177/1043463103154003

DeVault, M. L. (1991). Feeding the family: The socialorganization of caring as gendered work. Chicago:University of Chicago Press.

Evertsson, M., & Nermo, M. (2004). Dependencewithin families and the division of labor:Comparing Sweden and the United States.Journal of Marriage and Family, 66, 1272 – 1286.doi:10.1111/j.0022-2445.2004.00092.x

Gupta, S. (2006). Her money, her time: Women’searnings and their housework hours. Social ScienceResearch, 35, 975 – 999. doi:10.1016/j.ssresearch.2005.07.003

Gupta, S. (2007). Autonomy, dependence, or dis-play? The relationship between married women’searnings and housework. Journal of Marriageand Family, 69, 399 – 417. doi:10.1111/j.1741-3737.2007.00373.x

Gupta, S., & Ash, M. (2008). Whose money,whose time? A nonparametric approach tomodeling time spent on housework in theUnited States. Feminist Economics, 14, 93 – 120.doi:10.1080/13545700701716664

Health and Retirement Study, RAND HRS Data File(v.J), 1992 HRS, 1994 HRS, 1996 HRS Core, 1998HRS Core, 2000 HRS Core, 2002 HRS Core, 2004HRS Core, 2006 HRS Core, 2008 HRS Core,2003 HRS Mailout CAMS, 2005 HRS MailoutCAMS, 2007 HRS Mailout CAMS, and 2009 HRSMailout CAMS, public use data sets. Producedand distributed by the University of Michiganwith funding from the National Institute on Aging(grant number NIA U01AG009740). Ann Arbor,MI, 2009.

Hersch, J., & Stratton, L. S. (1997). Housework, fixedeffects, and wages of married workers. Journal ofHuman Resources, 32, 285 – 307.

Hochschild, A. (1989). The second shift: Workingparents and the revolution at home. New York:Viking.

Opting Out and Buying Out 471

Kamo, Y. (1988). Determinants of household divi-sion of labor: Resources, power, and ideol-ogy. Journal of Family Issues, 9, 177 – 200.doi:10.1177/019251388009002002

Killewald, A., & Gough, M. (2010). Money isn’teverything: Wives’ earnings and houseworktime. Social Science Research, 39, 987 – 1003.doi:10.1016/j.ssresearch.2010.08.005

Krantz-Kent, R. (2009). Measuring time spent inunpaid household work: Results from the Amer-ican Time Use Survey. Monthly Labor Review,132(7), 46 – 59.

Noonan, M. C. (2001). The impact of domesticwork on men’s and women’s wages. Jour-nal of Marriage and Family, 63, 1134 – 1145.doi:10.1111/j.1741-3737.2001.01134.x

Oropesa, R. S. (1993). Using the service econ-omy to relieve the double burden: Femalelabor force participation and service pur-chases. Journal of Family Issues, 14, 438 – 473.doi:10.1177/019251393014003006

Phipps, S. A., & Burton, P. S. (1998). What’s mine isyours? The influence of male and female incomeson patterns of household expenditures. Economica,65, 599 – 613. doi:10.1111/1468-0335.00148

Pittman, J. F., & Blanchard, D. (1996). The effectsof work history and timing of marriage on the divi-sion of household labor: A life-course perspective.

Journal of Marriage and the Family, 58,78 – 90.

RAND HRS Data, Version J. Produced by the RANDCenter for the Study of Aging, with fundingfrom the National Institute on Aging and theSocial Security Administration. Santa Monica, CA(March 2010).

Sanchez, L., & Thomson, E. (1997). Becomingmothers and fathers: Parenthood, gender, andthe division of labor. Gender and Society, 11,747 – 772. doi:10.1177/089124397011006003

Soberon-Ferrer, H., & Dardis, R. (1991). Determi-nants of household expenditures for services.Journal of Consumer Research, 17, 385 – 397.

South, S. J., & Spitze, G. (1994). Housework inmarital and nonmarital households. AmericanSociological Review, 59, 327 – 347.

Treas, J., & de Ruijter, E. (2008). Earnings andexpenditures on household services in marriedand cohabiting unions. Journal of Marriageand Family, 70, 796 – 805. doi:10.1111/j.1741-3737.2008.00522.x

Van der Lippe, T., Tijdens, K., & de Ruijter, E.(2004). Outsourcing of domestic tasks and time-saving effects. Journal of Family Issues, 25,216 – 240. doi:10.1177/0192513X03257425

West, C., & Zimmerman, D. H. (1987). Doing gender.Gender and Society, 1, 125 – 151.