Market Earnings and Household Work: New Tests of Gender Performance Theory
This article was edited by Suzanne Bianchi.
Abstract
I examine the contested finding that men and women engage in gender performance through housework. Prior scholarship has found a curvilinear association between earnings share and housework that has been interpreted as evidence of gender performance. I reexamine these findings by conducting the first such analysis to use high-quality time diary data for a U.S. sample in the contemporary period. Drawing on data on 11,868 married women and 10,770 married men in the American Time Use Survey (2003–2007), I find no evidence that married men “do gender” through housework. I do, however, find strong evidence of gender performance among women as evidenced by a curvilinear association between earnings share and women's housework time.
Prior research has led to near unanimity among scholars that what married men and women earn in the market affects the amount of housework they do at home. Nevertheless, there is substantial ambiguity and debate about just how earnings affect housework time (Bittman, England, Folbre, Sayer, & Matheson, 2003; Evertsson & Nermo, 2004; Gupta, 2007). Household bargaining theory posits that earnings share should be negatively related to housework time, as the higher earning spouse can be expected to use his or her position of superior earnings to negotiate a smaller housework burden (Lundberg & Pollak, 1996). Though some research bears out this prediction in couples in which the husband earns the majority of couple earnings (Brines, 1994; Greenstein, 2000), scholars have detected a surprising relationship between earnings share and housework in couples in which the wife earns more than half of couple earnings. Women in these couples actually appear to do more housework than otherwise similar women who have earnings that are roughly equal to their husbands', and men in these couples appear to do less housework than otherwise similar men in couples with approximately equal earnings (Brines; Bittman et al.; Greenstein). The literature then documents an unexpected curvilinear relationship between earnings share and housework time.
Scholars have interpreted this relationship through the prism of gender performance theory, arguing that housework can serve as a way in which men and women enact gender and create social meaning (Coltrane, 2000; Shelton & John, 1996; West & Zimmerman, 1987). This type of gender performance may be particularly important in the context of gender deviance, such as when couples do not adhere to the male-breadwinner norm, with housework being used to neutralize deviance and reconstruct gender (Bittman et al., 2003; Greenstein, 2000).
Though a number of findings in the housework literature seem to provide support for gender performance theory, the broader empirical record is substantially more ambiguous. One set of findings suggests that men may in fact “do gender” by reducing housework time when they earn less than their wives (Brines, 1994; Greenstein, 2000). But this evidence in support of gender performance theory has been challenged by scholars who found that these results are driven primarily by outliers—the lowest earning men in married couples (Bittman et al., 2003; Gupta, 1999b). Other research provides some evidence that, although men may not be doing gender, women may be engaging in gender performance through housework (Bittman et al.; Evertsson & Nermo, 2004). This finding has also been challenged, in this case by recent work that draws on autonomy theory to suggest that women's absolute levels of earnings, not their share of earnings, may predict the amount of time they spend on housework (Achen & Gough, 2009; Gupta, 2006, 2007). This uncertainty in the literature has created a substantial gap in our understanding of housework.
In this paper I provide new evidence to help resolve the empirical debate. Drawing on time diary data collected between 2003 and 2007 by the American Time Use Survey (ATUS), I assessed the relationship between men's and women's housework time and their earnings. In contrast to the predictions of gender performance theory and bargaining theory, I found no evidence of an association between men's relative earnings share and the amount of time men spend on housework. But, in confirmation of findings by Bittman and colleagues (2003), I did find evidence that women's housework time exhibits patterns of association with relative earnings share that indicate both gender performance and bargaining. Moreover, in contrast to work by Gupta (2007) supporting autonomy theory, I found that adjusting separately for men's and women's absolute earnings generally does not attenuate the curvilinear association between earnings share and housework time that has been interpreted as evidence of gender performance and bargaining.
Theoretical Background
Bargaining theorists draw a connection between earnings and housework by arguing that income earned in the market serves as a bargaining resource in negotiations over housework (Lundberg & Pollak, 1996). Spouses who contribute larger shares of couple earnings have more power in the couple and so are better able to bargain to avoid the unpleasant work of household chores. Bargaining theory then predicts a strong negative relationship between each spouse's relative earnings and the amount of time each spends on housework. A key feature of the bargaining model is that this negative relationship should be identical for men and for women. Men who are the primary breadwinners in a couple can be expected to do less housework than otherwise similar men who earn less than their wives, and the same type of relationship can be expected to hold for women. The utility of earnings as a bargaining resource does not vary depending on the gender of the earner.
Yet, bargaining theory cannot adequately explain the puzzling empirical finding of a quadratic relationship between earnings share and housework time. Gender performance theory, explicitly focused on the idea that both earnings and housework have important cultural meanings that vary by gender, appears to have greater explanatory power. Originating in the theoretical work of West and Zimmerman (1987), this perspective suggests that spending time on housework (or avoiding it) is a way in which women and men actually produce gender. Men and women must constantly perform gender through activities that conform to social expectations (West & Zimmerman, 2009), and when they do not, their actions simply do not make sense to others (England & Folbre, 2005). Housework, then, is not a gender-neutral activity that is bargained over, but rather is a set of actions imbued with cultural meaning, actions that serve as a resource for the construction of femininity for women and as a threat to masculinity for men.
Scholars have extended this theory to argue that when men and women are gender deviant in one domain, they may attempt to neutralize that deviance by making particularly pronounced efforts at gender performance in another (Bittman et al., 2003; Greenstein, 2000; Kroska, 2008). Research to date that links gender deviance and housework focuses on situations in which men earn less than their wives (e.g., Bittman et al.; Brines, 1994; but see Schneider, 2009) and so fail to meet the social expectation that they be the primary economic providers for their families. In these situations, the predictions of bargaining theory and gender performance theory diverge. Although bargaining theory would predict that relatively low-earning men would spend more time on housework and that their wives would spend less time, gender performance theory predicts the opposite. Men who are gender deviant in their relative contributions to couple earnings will seek to neutralize that deviance by doing less housework and their wives will actually do more housework in response to their deviant earnings.
Both bargaining theory and gender performance theory hypothesize that there is an important relationship between housework time and earnings share. These theories predict that the share of earnings contributed by men and women will affect housework time apart from the absolute amount of earnings, either because share is the relevant measure of bargaining power or the relevant cultural marker of gender deviance. Gupta (2006, 2007) argued that this focus on earnings share is misplaced. Instead, explicating what he termed “autonomy theory,”Gupta (2006, 2007) suggested that the absolute amount of earnings, particularly women's absolute earnings, actually matters most for housework time, particularly women's housework time.
The theoretical mechanism connecting the absolute amount of women's earnings with the amount of time women spend on housework is uncertain. Perhaps the clearest moderating process involves the purchase of housework services. Women who attain some level of earnings may be able to purchase housework services and so substitute hired labor for their own. This kind of substitution may only occur when women have some sufficient minimum level of earnings, because men's earnings may not be deployed to purchase housework services. Consequently, adjusting for total couple earnings may not capture the effect of women's own earnings (Gupta, 2007). Alternative explanations for the relationship between absolute earnings and women's housework time include the possibility that housework has a higher opportunity cost for women with higher absolute earnings or that women with higher earnings have different preferences for household cleanliness (Gupta, 2007). The association could also be driven by reverse causality, if women who did less housework tended to earn more (Gupta, 2007).
Empirical Evidence
Several studies have reported evidence of a relationship between earnings share and men's housework time that would seem to support the predictions of gender performance theory. Drawing on data from the 1985 wave of the Panel Survey of Income Dynamics, Brines (1994) found that wives' housework time exhibited a roughly linear negative relationship with earnings share and that husbands appeared to increase their housework time as couple earnings shares approached equality. Yet, in contrast to the predictions of bargaining theory, when husbands were the minority breadwinners, they actually appeared to spend less time on housework. Brines argued that this quadratic relationship between earnings share and men's housework time appeared to better fit the predictions of gender performance theory. Married men interviewed for the first wave of the National Survey of Families and Households (NSFH1) appeared to have similar patterns of housework. Men who earned less than their wives actually did less housework, not more, than men in couples where spouses had equal earnings (Greenstein, 2000).
Subsequent research suggested that both of these results, showing evidence of gender performance among men, were primarily driven by the lowest earning men in the samples. After excluding the 2%–3% of lowest earning men from the Panel Survey of Income Dynamics (PSID) and NSFH1 samples used by Brines (1994) and Greenstein (2000) respectively, Gupta (1999b) and Bittman and colleagues (2003) showed that there was no quadratic relationship between men's housework time and earnings share, and Evertsson and Nermo (2004) found evidence only of a simple linear negative relationship.
More recent research provides somewhat more persuasive evidence that although men may not “do gender” through housework to neutralize gender deviance in earnings, women may. Married Australian women appeared to reduce their housework time as their relative earnings shares approached equality, but, as predicted by gender performance theory, past the point of equality, these women actually increased the amount of time spent on housework (Bittman et al., 2003). Further, this finding was robust to the exclusion of outliers and to the use of alternative measures of earnings share. Additionally, analyzing data from the 1973, 1981, 1991, and 1999 waves of the PSID, Evertsson and Nermo (2004) found evidence that women did less housework as their earnings increased toward equality with their husbands' earnings, but that in the 1981, 1991, and 1999 waves, once women's earnings exceeded their husbands' earnings, women actually increased the amount of time they spent on housework (though this curvalinearity was only strongly evident in the 1991 wave). This curvilinear relationship suggests a pattern of gender performance by women to neutralize the deviance created by earning more than their husbands.
These results have been subject to challenge in the recent work of Gupta and his coauthors. Using data from the NSFH1 and from the second wave of the National Survey of Families and Households (NSFH2), these scholars reported evidence in support of the autonomy argument, finding that women's own earnings had a strong negative relationship with women's housework time (Gupta, 2006; Gupta, Sayer, & Cohen, 2009). Additionally, though Gupta (2007) found a quadratic relationship between earnings share and housework time among married women employed full time in the NSFH2, he showed that this relationship disappeared when men's and women's absolute levels of earnings were controlled for separately and showed that women's own earnings had a large negative relationship with housework. Though compelling, Gupta (2007) limited his sample to married women who were employed full time. That restriction may have allowed for cleaner comparisons between women, but it limited the external validity of his analysis much more than was the case in prior work. Additionally, Gupta (2007) used a definition of housework that was limited to cooking, cleaning, laundry, and doing the dishes. This fairly restrictive definition makes it more difficult to compare his results to other research that used broader measures of housework (e.g., Bittman et al., 2003; Greenstein, 2000; Gupta, 1999a).
Unanswered Questions
This study takes up three unanswered questions in this dynamic literature. First, though it appears that there is in fact little evidence in the NSFH1 and PSID that men do gender through housework in response to gender deviance in earnings, that conclusion rests on data gathered with survey-based interview methods rather than with the time diary method, which may be a more reliable data-collection strategy (Bianchi, Milkie, Sayer, & Robinson, 2000; Bonke, 2005; Kroska, 2008; Marini & Shelton, 1993). Additionally, with the exception of Evertsson and Nermo's (2004) research, all of this work is based on data collected in the 1980s or early 1990s. Before concluding that men really do not “do gender” through housework, it would be useful to revisit the issue, drawing on up-to-date data with analyses that assess the sensitivity of the results to outliers.
Second, there is a similar need to examine the relationship between earnings share and women's housework time using up-to-date time diary data for the United States. Though Bittman and colleagues (2003) use a recent source of time diary data, their analysis was confined to Australia, and so we lack a current accounting with high-quality data of gender display in the United States.
Third, though Bittman and colleagues (2003) found evidence of gender performance among married Australian women, their work predates Gupta's (2007) research on autonomy theory, and so their models do not adjust for men's and women's absolute earnings. Work is needed that uses time diary data to examine the relationship between absolute individual earnings and housework time in order to test if autonomy, more than bargaining or gender performance, is the link between market earnings and household work.
Method
Data
To attempt to answer these questions, I made use of a relatively new data source, the 2003–2007 waves of the American Time Use Survey (ATUS), a nationally representative repeated cross-sectional survey of the United States noninstitutionalized population aged 15 and older. The ATUS draws its sample from the outgoing rotation group of the Current Population Survey (CPS), reinterviewing selected CPS respondents 2 to 5 months after the completion of their final CPS interview. In all, the 2003–2007 ATUS samples include 72,922 respondents for whom there is time diary data, reflecting an average response rate of about 58% (Bureau of Labor Statistics, 2009). In the case of most surveys, it is difficult to discern if nonresponses result in samples that differ from the sampling frame. But, because the ATUS is embedded in the CPS, it is possible to use respondent characteristics collected through the CPS to model nonresponses to the ATUS and look for patterns of bias. Abraham, Maitland, and Bianchi (2006) took this approach and examined data from the 2004 wave of the ATUS to investigate whether nonrespondents were busier or less socially integrated than respondents. Evidence of the former would be of concern for the study of housework because busier respondents might have less available time and so could have distinctive patterns of housework. Though Abraham and colleagues found evidence of possible links between social integration and survey response, there was little evidence that busier respondents, as measured by work hours, children in the household, or spouse's work hours, were less likely to respond.
The time diary method involved asking each ATUS respondent to provide detailed information about all of the activities he or she engaged in throughout the course of a randomly chosen day (with weekend days overweighted), starting at 4:00 a.m. and continuing for the next 24 hours. Individuals are asked at what time they began each activity, how long it lasted, who they were with, and where they were. This iterative process creates a detailed record of how each respondent spent his or her day.
I pooled data from the 2003–2007 waves of the ATUS, drawing on the time-diary data and the accompanying survey-based data collected on demographic and economic characteristics. In addition, I joined the ATUS data with supplemental data collected by the CPS about married ATUS respondents' spouses. I accessed both the ATUS and the CPS data through the American Time Use Survey Extract System (Abraham, Flood, Sobek, & Thorn, 2008) maintained by the University of Maryland and the University of Minnesota. The data are weighted in all analyses to take account of sample design and nonresponses.
I imposed four restrictions on the sample, including only those respondents (a) in married heterosexual couples, (b) in which both spouses were between the ages of 18 and 65, (c) in which at least one spouse had nonzero earnings, and (d) for whom there is complete data on all covariates. Of the full ATUS sample of 41,335 female respondents and 31,587 male respondents, I first limited my sample to the 19,969 women and 17,820 men in married couples. I then further restricted the sample to the 16,914 women and 14,981 men who were between the ages of 18 and 65 and whose spouses were also in that age range. I next excluded the 833 female and the 716 male respondents in married couples in which neither spouse had labor market earnings because they were both unemployed or out of the labor force. Finally, I eliminated 4,213 female and 3,495 male respondents (26% and 25%, respectively, of the 16,081 women and 14,265 men in married couples in which both spouses were age 18–65 and at least one had earned income) who lack complete data on any of the model covariates. These procedures yielded an analysis sample of 11,868 women and 10,770 men.
I examined whether having missing data on earnings or work hours was associated with respondent reports of housework time. I found no relationship between housework time and missing data for women and only an association between housework time and missing earnings for men. Such men do approximately nine fewer minutes of housework on average than men not missing earnings data. I next examined whether the association between usual work hours and housework time varied depending on whether the respondent was missing data on earnings and whether the association between earnings share and housework time varied depending on whether the respondent was missing data on work hours. I found no evidence of such variation.
The data are somewhat unusual, however, for having clear patterns of missing data. The respondents who were excluded through listwise deletion were primarily missing data on their (3,204) or their partner's (3,388) earnings; nearly all (99%) of this missing data on respondents' earnings and spouses' earnings is the result of questionnaire design, as respondents and spouses who were self-employed were not asked to report their earnings. An additional 1,060 respondents had complete earnings data but were missing data on their or their spouses' usual hours of work per week. All of the missing data on usual work hours was the result of respondents and spouses reporting that their usual hours of work “vary,” rather than reporting a number of usual hours. Finally, a very small number of additional respondents, 56, were deleted because of missing data on homeownership. Consequently, though there is limited evidence that excluded respondents spent less time on housework and no evidence that the relationship between my key covariates varied depending on missingness, I should caution that given the reasons for missing data, my estimates are best interpreted as being generalizable only to respondents who are not self-employed and whose usual work hours do not vary.
Dependent Variable
I calculated total housework time as the sum of minutes per day spent on nine types of housework: (a) cleaning, laundry, sewing, (b) meal preparation and clean up, (c) shopping, (d) interior maintenance, (e) exterior maintenance, (f) lawn, garden, and yard care, (g) auto maintenance and repair, (h) household management, and (i) care of pets. I top-coded the total time spent on these housework tasks at the 99th percentile for the analysis sample. Although the definition of housework that scholars have used varies by study, the fairly inclusive measure used here is quite similar to that used by Bittman and colleagues (2003) and Greenstein (2000) in work on earnings share and housework and to the measure used by South and Spitze (1994), Gupta (1999a), and Bianchi and colleagues (2000) in work on other dimensions of housework. This broad definition should also overlap with the measure used by Brines (1994), Evertsson and Nermo (2004), and Achen and Gough (2009) that is based on a single catch-all question about housework in the PSID. I also tested the robustness of my results to two alternative definitions of housework, one that also included child care and one that was limited to only “female-typed” tasks (cleaning, laundry, sewing, meal preparation, and clean up). I found that my results were robust to the use of either of these alternative measures.
Independent Variables
Following Sorenson and McLanahan (1987) and much of the housework literature to date (e.g., Bittman et al., 2003; Evertsson & Nermo, 2004; Greenstein, 2000), I defined earnings share as husband's earnings less wife's earnings divided by total couple earnings. The resulting variable ranges from −1 (wife contributes all of couple earnings) to 1 (husband contributes all of couple earnings), and I rescaled this variable to range from 0 (wife contributes all of couple earnings) to 1 (husband contributes all of couple earnings). Additionally, I included controls for total couple earnings and, in a second set of models, separate controls for men's and women's individual absolute earnings.
These variables were designed to separate the relationship between earnings share and housework from the relationship between absolute earnings and housework. But all three of these theoretically important variables may also be associated with other characteristics, confounding the relationship between earnings and housework. The leading candidate for this kind of omitted variable bias is the amount of time that men and women spend at paid market work (Shelton & John, 1996). To avoid this threat to validity, I included measures of respondents' and spouses' usual hours of paid work.
Other demographic and economic variables may have readily apparent associations with housework time, but less intuitively clear associations with earnings share. Nevertheless, I included linear and squared terms for age, a dichotomous measure of race (White vs. non-White), a dichotomous indicator of homeownership, a measure of the number of children under age 18 in the household, indicators for completed education (high school, some college, college completion or more—relative to not graduating from high school), a dichotomous indicator of current enrollment in school, and dichotomous indicators of unemployment and labor force participation. In addition, I included dichotomous indicators for time diary interviews conducted on Saturday, Sunday, and holidays (relative to nonholiday weekdays).
Analyses
I used Ordinary Least Squares regression to estimate the relationship between earnings share and the amount of time men and women spend on housework. In my first set of analyses, I use the ATUS data to estimate models of the relationship between earnings share and men's housework time. These analyses revisit the findings (Brines, 1994; Greenstein, 2000) and challenges to those findings (Bittman et al., 2003; Gupta, 1999b) in the literature on whether there is a curvilinear association between earnings share and men's housework time. I assess the sensitivity of my results to the exclusion of men with zero earnings, men married to women with zero earnings, and nondual-earner couples. In the Appendix (published as supplemental online material), I report the results of additional sensitivity tests.
Second, shifting from a focus on men's housework time, I regress a linear and squared term for earnings share on women's housework time, controlling for total couple earnings, usual work hours, and economic and demographic characteristics. Bargaining theory would predict that women would do less housework as their earnings shares increase. Empirically, bargaining theory would lead us to expect a linear negative relationship between this measure and housework time for both men and women. But, gender performance theory would predict that women who earn more than their husbands would actually do more housework than women in couples with equal earnings. Empirically, gender performance theory would predict a quadratic relationship between earnings share and women's housework time, with a negative coefficient on the linear term and a positive coefficient on the squared term. In addition to examining the relationship between earnings share and women's housework time in the full analysis sample, I also assess the robustness of my results to estimation on samples limited to women with earnings, women married to men with earnings, and dual-earner couples. I report the results of additional sensitivity tests in the Appendix.
Finally, I re-estimated these models substituting separate measures of women's and men's own absolute earnings for total couple earnings. This modeling strategy allows me to investigate Gupta's (2007) recently articulated autonomy theory of housework.
Descriptive Statistics
Married women do far more housework each day than married men. On average, as shown in Table 1, married women (Column 5) in my sample spend about two and a half hours (154 minutes) on housework and married men (Column 1) spend a bit more than half that amount of time, 87 minutes, on similar tasks each day. Summed over the length of a week, married women report doing about 18 hours of housework as compared with 10 hours for married men. These averages, for the period 2003—2007, track the time trend in housework described by Bianchi and colleagues (2000) in their time series of time diary measurements of married men's and women's housework. The average hours per week spent by married men on housework rose from 4.7 in 1965 to 6.7 in 1975 before leveling off at 10.4 in 1985 and 1995. My results show a continuation at that level through the period 2003–2007. Similarly, Bianchi and colleagues describe a decline in married women's time spent on housework from 33.9 hours in 1965 to 26.1 hours in 1975 to 21.9 hours in 1985 to 19.4 hours in 1995. This more recent sample of the ATUS data show that trend continuing, with housework time dropping to an average of 18 hours per week for married women in the more recent period.
| Men by Husband's Share of Couple Earnings | Women by Husband's Share of Couple Earnings | |||||||
|---|---|---|---|---|---|---|---|---|
| All | > = .9 | .55 to .45 | < = .1 | All | > = .9 | .55 to .45 | < = .1 | |
| [1] | [2] | [3] | [4] | [5] | [6] | [7] | [8] | |
| Housework time (minutes per day) | 86.70 | 73.04 | 81.38 | 152.78 | 153.83 | 216.41 | 110.24 | 122.23 |
| Economic attributes | ||||||||
| Share of couple earningsa (%) | 0.67 | 1.00 | 0.50 | 0.00 | 0.34 | 0.01 | 0.50 | 1.00 |
| Couple weekly earnings ($ per week) | 1,413 | 1,120 | 1,685 | 665 | 1,388 | 1,083 | 1,652 | 723 |
| Own earnings ($ per week) | 948 | 1,115 | 847 | 0.48 | 482 | 84 | 821 | 722 |
| Own home (%) | 0.83 | 0.78 | 0.86 | 0.82 | 0.83 | 0.76 | 0.84 | 0.80 |
| Respondent's usual work hours | 41.89 | 46.10 | 44.11 | 1.18 | 26.72 | 1.40 | 41.39 | 38.38 |
| Spouse's usual work hours | 26.12 | 0.91 | 41.09 | 39.93 | 40.28 | 45.26 | 43.03 | 0.72 |
| Unemployed (%) | 0.02 | 0.00 | 0.00 | 0.25 | 0.03 | 0.10 | 0.00 | 0.00 |
| Not in the labor force (%) | 0.06 | 0.00 | 0.00 | 0.723 | 0.26 | 0.82 | 0.00 | 0.00 |
| Education (%) | ||||||||
| Less than high school | 0.10 | 0.16 | 0.07 | 0.13 | 0.09 | 0.16 | 0.06 | 0.07 |
| High school | 0.30 | 0.28 | 0.31 | 0.40 | 0.31 | 0.32 | 0.26 | 0.37 |
| Some college | 0.25 | 0.20 | 0.28 | 0.27 | 0.26 | 0.25 | 0.26 | 0.26 |
| College or more | 0.34 | 0.36 | 0.34 | 0.20 | 0.34 | 0.28 | 0.42 | 0.30 |
| Currently enrolled in school (%) | 0.04 | 0.03 | 0.04 | 0.05 | 0.06 | 0.06 | 0.07 | 0.04 |
| Demographic attributes | ||||||||
| Age (years) | 43.54 | 42.55 | 43.00 | 49.23 | 41.55 | 40.48 | 40.24 | 46.5 |
| White (%) | 0.72 | 0.67 | 0.71 | 0.72 | 0.73 | 0.67 | 0.75 | 0.73 |
| Number of children | 1.13 | 1.46 | 0.92 | 0.74 | 1.12 | 1.45 | 0.98 | 0.74 |
| n b | 10,770 | 3,713 | 1,418 | 854 | 11,868 | 3,820 | 1,637 | 1,026 |
- aHusband's (for men) and wife's (for women) earnings divided by couple's earnings. bThe ATUS data do not contain reports of housework for both members of a married couple. Consequently, there can be unequal numbers of men and women and couple-level measures, such as couple earnings or home ownership, may not match exactly for men and women.
Columns 1 and 5 of Table 1 also present basic descriptive statistics about the economic and demographic characteristics of the men and women in the analysis sample. Though men and women have approximately equal levels of educational attainment, men have higher rates of labor force participation and have higher earnings. These higher average absolute earnings translate into a male advantage in terms of relative contributions to couple earnings. On average, men contribute two thirds of couple earnings and women contribute just one third.
Table 1 also presents some descriptive evidence to support the hypothesis that women who earn more than their husbands might actually spend more time on housework than women who earn the same amount as their husbands. Women whose husbands contribute 10% or less of couple earnings (Column 8) perform 122 minutes of housework per day on average, about 12 minutes more than women whose husbands earn between 45% and 55% of couple earnings (Column 7), but still substantially less than women whose husbands earn at least 90% of couple earnings (Column 6). Men who earn between 45% and 55% of couple earnings spend a similar amount of time on housework (Column 3) as men who earn more than 90% of couple earnings (Column 2). But men who earn less than 10% of couple earnings do substantially more housework (Column 4) than these higher earning men, which would seem to accord with the predictions of bargaining theory.
But Table 1 also reveals that women and men in households where husbands earn 10% or less of couple earnings are also less affluent than their counterparts in couples in which husbands earn at least 90% of earnings or in couples in which spouses' earnings are approximately equal. Focusing on women's reports, these couples have weekly earnings of $723 on average, as compared to $1,388 for all married couples, $1,652 for couples in which women earn about the same amount as their husbands, and $1,083 for couples in which women's husbands earn at least 90% of couple earnings. Further, both women who earn large and small shares of couple earnings are much less likely to have completed college than women whose earnings are roughly equal to their husbands' earnings. Men who earn less than 10% of couple earnings are also much less likely to have completed college than men who earn larger shares of couple earnings.
These differences raise the question of whether any relationship between housework time and earnings share may in part be driven by class differences that shape both earnings share and men's and women's housework time. Perhaps less-affluent men and women simply complete more housework and are more likely to be in couples where the wife earns a large share of couple earnings. Additionally, the relationship between earnings share and housework time could be confounded by work hours or by the presence of dual- versus single-earner household structures. In the following regression analyses, I take care to control for these class and work attributes in an attempt to isolate the relationship between earnings share and housework time.
Results
Do Men Do Gender Through Housework?
Whether men do gender through housework in response to gender deviance in earnings persists as a key puzzle in the literature on earnings and housework. Table 2 presents the results of analyses that revisit this puzzle, using data from the ATUS. Employing this new source of time diary data, I found no evidence that men's housework time is responsive to their relative earnings shares. There is neither a linear relationship (Model 1) nor a quadratic relationship (Model 2) between men's relative earnings share and men's time spent on housework.
| Linear | Quadratic | ||||
|---|---|---|---|---|---|
| Full Sample | Full Sample | Men With Wives With Earningsa | Men With Earningsb | Dual-Earnerc | |
| [1] | [2] | [3] | [4] | [5] | |
| Earnings shared | 5.76 | 0.28 | −0.43 | −31.05 | −37.46 |
| Earnings share squared | — | 4.58 | 5.55 | 33.62 | 39.94 |
| Total couple earnings | 0.05 | 0.04 | −0.07 | −0.03 | −0.17 |
| Husband's usual work hours | −0.94*** | −0.94*** | −0.93*** | −0.93*** | −0.92*** |
| Wife's usual work hours | 0.26 | 0.28 | 0.31 | 0.43* | 0.48* |
| Own home | 18.16*** | 18.17*** | 13.12** | 19.08*** | 13.33** |
| Educatione | |||||
| High school | 0.06 | 0.08 | 6.56 | −2.64 | 2.18 |
| Some college | 6.37 | 6.39 | 10.73 | 5.37 | 8.45 |
| College or more | −0.21 | −0.21 | 5.45 | −3.66 | −0.34 |
| Respondent enrolled in school | −25.27*** | −25.29*** | −24.80*** | −20.02*** | −17.53** |
| Respondent agef | 0.55*** | 0.55*** | 0.67*** | 0.55*** | 0.67*** |
| Respondent age squared | −0.03* | −0.03* | −0.04* | −0.03** | −0.04** |
| White | 18.30*** | 18.31*** | 20.27*** | 17.54*** | 19.30*** |
| Number of children | 0.46 | 0.46 | 1.48 | 0.50 | 1.61 |
| Unemployed | 57.72*** | 56.20** | 55.21* | — | — |
| Not in labor force | 23.89* | 22.38 | 21.76 | — | — |
| Saturday | 57.54*** | 57.54*** | 57.41*** | 64.59*** | 68.08*** |
| Sunday | 58.29*** | 58.29*** | 60.26*** | 68.06*** | 75.47*** |
| Holiday | 33.32** | 33.31** | 43.74** | 34.34** | 47.17** |
| Intercept | 68.91*** | 69.88 | 67.00*** | 72.35*** | 72.07*** |
| N | 10,770 | 10,770 | 7,265 | 9,928 | 6,423 |
| R 2 | 0.098 | 0.098 | 0.097 | 0.093 | 0.098 |
- aModel 3 replicates Model 2 on a restricted subsample of the ATUS respondents. In addition to being limited to married men aged 18–65 with complete data on all covariates, the subsample also excludes men whose wives reported no earnings. bModel 4 replicates Model 2 on a restricted subsample of the ATUS respondents. In addition to being limited to married men aged 18–65 with complete data on all covariates, the subsample also excludes men who reported no earnings. This restriction results in unemployed men and men who are out of the labor force being dropped from the analysis sample. cModel 5 replicates Model 2 on a restricted subsample of the ATUS respondents. In addition to being limited to married men aged 18–65 with complete data on all covariates, the subsample also excludes nondual-earner couples. This restriction results in unemployed men and men who are out of the labor force being dropped from the analysis sample. dEarnings share is calculated as (husband earnings − wife earnings)/(husband earnings + wife earnings) and is then rescaled to range from 0 to 1. eRelative to respondents with less than a high school diploma. fAge is centered on its mean to reduce multivariate collinearity.
- * p < .05. **p < .01. ***p < .001.
In the existing literature, scholars have erased nonlinear findings by excluding outliers (Bittman et al., 2003; Gupta, 1999b). In this case, there are no such curvalinearities to question. Excluding men with zero earnings (Model 3), men whose wives' have zero earnings (Model 4), or restricting the sample to dual-earner couples (Model 5) does not change that null finding. I describe additional robustness tests and model diagnostics in the Appendix.
Earnings Share and Women's Housework Time
Although I found no evidence of a relationship between relative earnings share and men's housework time, I found strong evidence of a relationship between earnings share and housework time for women. As shown in Model 6 (Table 3), there is a statistically significant relationship between the linear measure of earnings share and women's housework time (β = 26.76, p < .01). As men's share of couple earnings increases, women appear to increase the amount of time they spend on housework. But Model 7 demonstrates that there is in fact a quadratic relationship between those two variables. The squared term for earnings share is large, positive, and statistically significant (β = 102.79, p < .001) and the linear term is negative and significant (β = −69.25, p < .05). Further, the 9.5 difference in the Bayesian Information Criteria (BIC) between the two models provides strong evidence for concluding that including the quadratic term for earnings share improves model fit (Raftery, 1995). Plotting the predicted values from Model 7 (holding all other covariates at the sample means) shows a pronounced curvilinear relationship (Figure 1). Even after controlling for a host of economic and demographic characteristics, it appears that married women decrease the amount of time that they spend on housework as their share of couple earnings rises towards equality. That relationship fits with the predictions of bargaining theory. In contrast, and in better accord with the predictions of gender performance theory, it appears that when women earn more than their husbands they actually increase the amount of time they spend on housework. Women who are the majority breadwinners do more, not less, housework than women who earn about the same amount of money as their husbands.
| Linear | Quadratic | ||||
|---|---|---|---|---|---|
| Full Sample | Full Sample | Women With Husbands With Earningsa | Women With Earningsb | Dual Earnerc | |
| [6] | [7] | [8] | [9] | [10] | |
| Earnings shared | 26.76** | −69.25* | −110.26* | −70.93* | −116.93* |
| Earnings share squared | — | 102.79*** | 138.41** | 107.76*** | 148.69** |
| Total couple earnings | −0.51* | −0.51* | 0.50* | −0.70** | −0.71** |
| Wife's usual work hours | −0.96*** | −0.73*** | −0.61*** | −0.73*** | −0.59** |
| Husband's usual work hours | 0.12 | 0.36* | 0.34 | 0.33 | 0.28 |
| Own home | 2.88 | 2.72 | 2.15 | 6.14 | 6.45 |
| Educatione | |||||
| High school | −19.50** | −19.99** | −20.50** | −15.98* | −18.36* |
| Some college | −26.19*** | −26.56*** | −27.35*** | −19.68** | −21.89** |
| College or more | −26.61*** | −27.17*** | −28.46*** | −16.75* | −19.33* |
| Respondent enrolled in school | −26.49*** | −26.92*** | −27.72*** | −16.38** | −16.20* |
| Respondent agef | 1.58*** | 1.52*** | 1.62*** | 1.33*** | 1.42*** |
| Respondent age squared | −0.04* | −0.04* | −0.03 | −0.05** | −0.04* |
| White | 1.22 | 0.63 | −2.10 | 5.65 | 2.26 |
| Number of children | 11.26*** | 11.05*** | 11.19*** | 9.18*** | 8.94*** |
| Unemployed | 43.61*** | 28.65* | 26.76 | — | — |
| Not in labor force | 35.64*** | 20.38* | 18.54* | — | — |
| Saturday | 35.39*** | 35.6*** | 34.45*** | 62.44*** | 64.57*** |
| Sunday | 27.75*** | 27.71*** | 23.42*** | 64.22*** | 63.60*** |
| Holiday | 26.71* | 26.37 | 25.22 | 38.28* | 38.61* |
| Intercept | 159.17*** | 157.8*** | 168.59 | 140.79*** | 153.66*** |
| N | 11,868 | 11,868 | 10,857 | 8,326 | 7,315 |
| R 2 | 0.137 | 0.138 | 0.140 | 0.099 | 0.099 |
- aModel 8 replicates Model 7 on a restricted subsample of the ATUS respondents. In addition to being limited to married women aged 18–65 with complete data on all covariates, the subsample also excludes women whose husbands reported no earnings. bModel 9 replicates Model 7 on a restricted subsample of the ATUS respondents. In addition to being limited to married women aged 18–65 with complete data on all covariates, the subsample also excludes women who reported no earnings. This restriction results in unemployed women and women who are out of the labor force being dropped from the analysis sample. cModel 10 replicates Model 7 on a restricted subsample of the ATUS respondents. In addition to being limited to married women aged 18–65 with complete data on all covariates, the subsample also excludes nondual-earner couples. This restriction results in unemployed women and women who are out of the labor force being dropped from the analysis sample. dEarnings share is calculated as (husband earnings − wife earnings)/(husband earnings + wife earnings) and is then rescaled to range from 0 to 1. eRelative to respondents with less than a high school diploma. fAge is centered on its mean to reduce multivariate collinearity.
- * p < .05. **p < .01. ***p < .001.
Women's Minutes of Housework per Day, Predicted Values From OLS Regression (ATUS, 2003–2007).
Further, as reported in the results of Model 8 in Table 3, excluding the 8% of female respondents whose husbands reported no earnings produces substantively similar results to those in Model 7 and, if anything, the curvalinearity is more pronounced, a result that actually mirrors Evertsson and Nermo's (2004) results from similar analyses of the PSID. I also re-estimated the model, excluding women who had no earnings (Model 9) and excluding women who were not in dual-earner couples (Model 10). The results from these models are also quite similar, with the linear and quadratic terms for earnings share in each of the models retaining statistical significance. Comparing the BICs for these three models with alternative models that omit the quadratic term for earnings share, but are otherwise identical, indicates positive evidence that the quadratic term improves model fit for Models 8 and 10 and very strong evidence for Model 9. I describe additional robustness tests and model diagnostics in the Appendix.
Bargaining, Performance, or Autonomy?
Although the results presented in Table 3 and described above are robust to the inclusion of a large number of economic and demographic controls, they do not address the objections to bargaining and gender performance theory that have been raised by autonomy theory (Gupta, 2007). That theory suggests that women's absolute market earnings affect housework time, separately from a couple's total combined earnings and separately from earnings share. This relationship appears most likely to be driven by the ability of higher earning women to deploy their own earnings to purchase housework services to substitute for their own labor (Gupta, 2006).
Table 4 presents the results of several analyses designed to investigate this possible confounding relationship. First, in Model 11, I show the simple unadjusted relationship between each member of the couple's own earnings and women's housework time. In confirmation of prior work (Achen & Gough, 2009; Gupta, 2007) I found a negative relationship between women's earnings and women's housework time and found that husband's earnings actually have a positive relationship with women's housework time. Although the magnitude of the coefficient on women's own earnings is substantially reduced after the inclusion of additional controls (Model 12), there is still a negative and statistically significant relationship between the two variables (β = −1.08, p < .01).
| Model 11 | Model 12 | Model 13 | |
|---|---|---|---|
| Wife's weekly earnings ($ hundred) | −6.31*** | −1.08** | −0.23 |
| Husband's weekly earnings ($ hundred) | 1.34*** | −0.12 | −0.67 |
| Earnings sharea | — | — | −69.43* |
| Earnings share squared | — | — | 108.42*** |
| Controlsb | — | yes | yes |
| Intercept | 172.12*** | 171.00*** | 155.57 |
| N | 11,868 | 11,868 | 11,868 |
| R 2 | 0.057 | 0.136 | 0.138 |
- Note: The sample is restricted to married women aged 18–65 with complete data on all covariates.
- aEarnings share is calculated as (husband earnings − wife earnings) / (husband earnings + wife earnings) and is then rescaled to range from 0 to 1. bThe controls are (a) respondent's and respondent's spouse's usual work hours; (b) a dichotomous measure of homeownership; (c) dichotomous measures of high school completion, completion of some college, and completion of college or more; (d) a dichotomous measure of respondent's school enrollment; (e) respondent's age and age squared; (f) a dichotomous indicator of race (1 = White); (g) a measure of the number of children (< age 18) present in the household; (h) a dichotomous measure of unemployment; (i) a dichotomous indicator of labor force participation; and (j) dichotomous indicators whether the survey was administered on a Saturday, Sunday, or holiday. These are the same controls included in the models presented in Table 3.
- * p < .05. **p < .01. ***p < .001.
In Model 13, I include both measures of absolute individual earnings of wives and husbands and linear and squared terms for relative earnings (as well as the same set of economic and demographic controls as used in the models presented in Table 3, except excluding total couple earnings). In contrast to the predictions of autonomy theory, it is the coefficients on relative earnings share that are statistically significant, not the coefficients on individual absolute earnings. The coefficient on the linear term for earnings share is −69.43 (p < .05), and the coefficient on the squared term is 108.42 (p < .001), nearly the same values as for Model 7 (Table 3) (which did not include adjustments for individual absolute earnings). Plotting the predicted values for housework time against relative earnings share with the other covariates set at their sample means shows an identical U-shaped relationship between relative earnings share and women's housework time as was seen in the models that do not control separately for men's and women's own absolute earnings. Further, since Model 12 (Table 4) is nested in Model 13 (Table 4), I can perform a formal test of the significance of the relative earnings variables against Model 12 (which excludes them). The F statistic is 9.08 and is statistically significant (p < .001). These results provide no substantiation of autonomy theory and seem instead to show continued support for bargaining and gender performance theory. I describe additional sensitivity tests in the Appendix.
Discussion
In this paper, I report on new evidence concerning the relationship between market earnings and housework time. I drew on a recently available data set, the American Time Use Survey, to bring high-quality time diary data to bear on a longstanding puzzle: Do men and women do gender when they do housework?
I found that the answer to that puzzle is relatively straightforward, at least for men. There is no evidence of a relationship between earnings share and men's housework time in the ATUS data. This result stands in contrast to earlier literature that found a curvilinear association between earnings share and men's housework time (Brines, 1994; Greenstein, 2000) and even to work that found a simple linear negative relationship (Evertsson & Nermo, 2004). This null finding is, however, the same as that previously reported in the Australian context in research using time diary data (Bittman et al., 2003). But, though I found no relationship between men's earnings share and housework time, I did find evidence that men who are unemployed spend more time on housework per day than men who are currently working. This relationship is large and statistically significant. Similarly, there is a negative relationship between men's work hours and housework time. This evidence seems to provide support for the idea that, rather than avoiding housework to mitigate any gender deviance created through unemployment, these men are responding to reduced work time by increasing housework, behavior that appears more consonant with the predictions of time availability theory.
In contrast, my results provide strong evidence that the proportion of couple earnings that women provide affects the amount of housework that they do. Consistent with the predictions of bargaining theory, it does appear that as married women increase their earnings share toward equality with their husbands, they reduce the amount of time they spend each day on routine housework tasks. This negative relationship persists even after adjusting for a large number of possibly confounding variables including usual work hours, education, and demographic characteristics. The evidence strongly suggests that increased relative earnings, not just differences in time availability or household composition, equates with reduced housework for women.
Up to the point of equality of earnings with their husbands, women's earnings appear to function as a bargaining resource and housework simply as a set of tasks to be avoided if possible. But examining the relationship between housework and earnings share for women who out-earn their husbands suggests a more complex relationship. Women who earn more than their husbands, and thus deviate from the male-breadwinner norm, appear to increase the amount of time they spend on housework. This U-shaped association fits with what we would expect from a pattern of behavior rooted in gender deviance neutralization and is consonant with the predictions of gender performance theory. Further, this curvilinear relationship is robust to adjustment for women's absolute income under most model specifications. My results suggest that far from being just a simple power resource used in bargaining or an economic resource used to outsource housework, earnings and breadwinning are imbued with a particular set of gendered meanings. Rather than simply being a set of undesirable tasks to be avoided, housework is an important venue for the construction of gender.
These results help to resolve an empirical ambiguity in the literature. I found that women do gender through housework but that men do not. These results help to reconcile the strong evidence of gender performance among women that Bittman and colleagues (2003) found in the Australian context with the prior null and contested findings in the United States. Though Bittman and colleagues argued that these divergent results are indicative of a “real national difference [between Australia and the United States] that centers on the nonlinearity in the effect of earnings on women's housework” (p. 207), my results suggest that these apparent behavioral differences may simply be the artifact of different data-collection methodologies. In this paper, I used data collected with the same type of time diary methodology as used by Bittman and colleagues rather than the survey-collection methodology used in other prior work on this topic in the United States. With data-collection methodology in effect held constant, I found no evidence of a national difference between the United States and Australia in the relationship between women's earnings share and housework time.
Yet my results also present a new theoretical puzzle: Why is it that women do gender through housework but men do not? This result is all the more puzzling because, in some respects, it runs counter to what we might expect from theory. Brines (1994) suggested men might be more likely to “do gender,” because masculinity is culturally regarded as achieved and femininity is seen as a “natural condition” and so less subject to doubt (p. 683). Further, Thorne (1993) described how gender deviance among girls, seen as tomboys, is less censored than among boys, seen as sissies. This work would suggest that it should be men, not women, who do gender through housework.
An alternative perspective may help to explain why gender performance through housework may be evident for women. Blair-Loy (2001) argued that professionally successful women in particular are caught between two dominant gendered schemas, the “family devotion schema” and the “work devotion schema.” The first captures the continued expectation that women are the primary managers of home and family and the second an expectation of total commitment to work. Blair-Loy (2001) argued that though earlier cohorts of women may have simply opted out of the family devotion schema, forgoing children and marriage, later cohorts have sought to navigate the two schemas, pursuing demanding careers and fulfilling social expectations of home and family involvement. Blair-Loy's work, though not focused specifically on housework and limited empirically to a group of women in executive positions, does suggest an explanation for the gender performance puzzle. Later-cohort women find themselves at once pursuing demanding and financially rewarding careers while still trying to satisfy a cultural family schema that emphasizes the performance of tasks such as housework. These women encounter a tension that is not experienced by men.
This explanation does not provide a full accounting of why men do not do gender through housework. I propose one possible hypothesis for why this might be the case. We might expect that women who seek to neutralize gender deviance by increasing the amount of time they spend on housework would face relatively little spousal resistance. Such women are taking on a larger amount of housework, which would be unlikely to be a burden to their husbands. In contrast, men seeking to reduce their housework time as a gender deviance neutralization strategy might face significant opposition from their wives. These men are seeking to do less work around the home, very possibly creating a burden for their wives. This burden might be especially hard because, although men's housework hours did increase from an average of 4.7 in 1965 to 10.4 in 1985, it is notable that there have been no further shifts in the years since (Bianchi et al., 2000). For men to reduce their housework time from this already relatively low level might well lead to spousal conflict. In this sense, if men and women have a shared preference for relationship quality, it might operate to constrain men from doing gender through housework (because of opposition to reducing their housework time) but not constrain women (who would conceivably encounter little resistance to increasing their housework time). In short, because doing more housework may be less likely to lead to conflict than doing less housework, relationship harmony may trump gender deviance for men.
My results are subject to several limitations. First, unlike the data collected in the two waves of the NSFH, the ATUS only contains reports of housework time from one member of a married couple. It is not possible therefore to construct measures of the percentage of total housework completed by a couple that is completed by the husband or by the wife. Using the ATUS, I can only create measures of the total time spent on housework by either the husband (in the case of male respondents) or the wife (in the case of female respondents). Though much of the recent literature on housework and earnings has used individual reports of housework time (e.g., Bittman et al., 2003), other work has used relative measures (Greenstein, 2000). A key empirical benefit of the latter approach is that it in some ways accounts for between-couple heterogeneity in preferences for cleanliness and other correlates of total housework time. Second, my analyses exclude respondents who are self-employed and those who report that their usual work hours vary. Consequently, my results are only properly generalizable to married couples who are not self-employed and whose usual work hours do not vary. Finally, Gupta (2007) has suggested that the principal mechanism connecting women's absolute earnings and housework time is the purchase of housework services that substitute for women's time and labor. But it is not possible to test this hypothesis in the ATUS, as the data do not contain any measures of expenditures.
My findings help to clarify the role of gender performance in housework. Although recent work has raised serious questions about gender performance theory and provided evidence in favor of autonomy theory, my results, generated using a large sample of respondents interviewed using time diary methods, show that there is a strong curvilinear relationship between women's housework time and earnings share. These results suggest that, even as increasing numbers of women earn more than their husbands, that behavior seems to remain an act of gender deviance that some married women attempt to neutralize through the performance of housework.
Note
This paper has benefited from the helpful advice and comments of Devah Pager, Cristobal Young, Sara McLanahan, Audrey Beck, and participants in the Princeton University Women and Gender Graduate Student Colloquium. A previous version of this paper was presented at the 2010 Annual Meeting of the Population Association of America. The author is grateful for support from the National Science Foundation Graduate Research Fellowship (Grant DGE-0646086) and Princeton University.
