2. LITERATURE REVIEW

Time-use data provides valuable insights into household economics. A common and not surprising finding from studies of various countries is that women spend more time doing unpaid work and men spend more time doing paid work [Sayer (Reference Sayer2005); Budlender (Reference Budlender2000); Stratton (Reference Stratton2015); Ruppanner and Maume (Reference Ruppanner and Maume2016)].Footnote ² This pattern helps explain the well-documented gender wage gap. However, little work has been done investigating time-use allocations within the households of same-sex couples. As such, this paper makes important contributions to the literature on the economics of the household. Further, investigating the experiences of members of same-sex households builds on the quickly growing literature on the economic outcomes of gay men and lesbians.

The literature on the economic outcomes of gay men and lesbians has largely been in response to the perplexing findings of Badgett (Reference Badgett1995). Badgett (Reference Badgett1995) found a significant wage penalty for gay men and an insignificant wage differential for lesbians. The asymmetry in the earnings effect of sexual orientation by sex is difficult to reconcile with most, but not all, economic theories of discrimination. While inconsistent with most theories of discrimination based on a prejudice toward sexual orientation, the asymmetric earnings effects are consistent with varying levels of labor market engagement for gay men and lesbians. Lesbians are more likely than heterosexual women to have a partner who will share non-market work thus giving lesbians more time to spend on market work, perceptions of which could motivate employers to favor lesbians [Peplau and Fingerhut (Reference Peplau and Fingerhut2004)]. The same logic would predict higher levels of non-market work among cohabiting gay men.

The need to investigate patterns of household labor by sexual orientation is highlighted by the growing literature that has repeatedly documented a significant penalty for gay men and a significant premium for lesbians in the US [Berg and Lien (Reference Berg and Lien2002); Baumle and Poston (Reference Baumle and Poston2011); Klawitter and Flatt (Reference Klawitter and Flatt1998); Cushing-Daniels and Yeung (Reference Cushing-Daniels and Yeung2009)], Sweden [Ahmed and Hammarstedt (Reference Ahmed and Hammarstedt2010); Ahmed et al. (Reference Ahmed, Andersson and Hammarstedt2011, Reference Ahmed, Andersson and Hammarstedt2013)], the UK [Arabsheibani et al. (Reference Arabsheibani, Marin and Wadsworth2004)], and the Netherlands [Plug and Berkhout (Reference Plug and Berkhout2008)].Footnote ³ Even though recent work suggests that these differentials are getting smaller for gay men [Cushing-Daniels and Yeung (Reference Cushing-Daniels and Yeung2009); Clarke and Sevak (Reference Clarke and Sevak2013); Carpenter (Reference Carpenter2007)], this may be the result of the expansion of protection from discrimination for gay men and lesbians [Klawitter (Reference Klawitter2011); Martell (Reference Martell2013b, Reference Martell2014)]. Explanations of varying patterns in household labor explaining the earnings effects of sexual orientation have risen in prominence [see, for example Black et al. (Reference Black, Sanders and Taylor2007) and Daneshvary et al. (Reference Daneshvary, Waddoups and Wimmer2009)], because evidence in favor of alternatives explanations has been inconclusive.Footnote ⁴

There are several potential sources for differences in labor market attachment by sexual orientation. Varying labor market patterns may be the result of different amounts of time spent in the household due to (1) different budget constraints or labor/leisure trade-offs (2) different patterns in household specialization, or (3) different levels of aggregate household production [Black et al. (Reference Black, Sanders and Taylor2007); Jepsen (Reference Jepsen2007); Clain and Leppel (Reference Clain and Leppel2001); Berg and Lien (Reference Berg and Lien2002)]. Our paper investigates the broad hypothesis that differences in time spent in the household can explain differing labor market outcomes for gays and lesbians compared to their heterosexual counterparts. While differences in household time use between same-sex and different-sex households are generally explained as a product of household specialization [see, for example, Jepsen (Reference Jepsen2007) and Black et al. (Reference Black, Sanders and Taylor2007)], it is important to note that such differences can occur without household specialization in same-sex households. Even if members of same-sex households do not specialize, time-use patterns will vary by sexual orientation due to the well-documented specialization by gender that occurs in different-sex households.

Existing empirical investigations into the household behavior of gay and lesbian couples have largely relied on indicators from the paid labor market to draw conclusions regarding the behavior of gays and lesbians within the household. To draw conclusions regarding specialization in gay and lesbian households, these investigations have tested if there are inequalities in labor market outcomes within gay and lesbian households. For example, the labor supply of members of same-sex households are responsive to proxies typically associated with bargaining power such as differences in age and education (Orrefice, Reference Orrefice2011). Cohabiting lesbians and gay men also exhibit intra-household inequalities in earnings, hours worked in paid labor, and the likelihood of working full-time suggesting that there is a primary worker and secondary worker even in same-sex households [Giddings et al. (Reference Giddings, Nunley, Schneebaum and Zietz2014); Jepsen and Jepsen (Reference Jepsen and Jepsen2015); Antecol et al. (Reference Antecol, Jong and Steinberger2008)]. That these differences are less pronounced in same-sex households than in different sex households suggest that specialization is less pronounced in same-sex households, which authors argue could explain observed wage differentials [Jepsen and Jepsen (Reference Jepsen and Jepsen2015)]. Inequalities in labor market outcomes among household members may in part be explained by specialization if specialization allows some members of a household to be more attached to the labor market than others.Footnote ⁵

Studies investigating labor market outcomes as proxies for household behavior are informative but incomplete in that they assume non-labor market behavior mirrors labor market behavior. Due to the possibility of differences in budget constraints, preferences and levels of household production, labor market patterns consistent with the household specialization hypothesis do not ensure that non-market time-use patterns will likewise be consistent. While an additional hour of paid labor can be offset by a decrease in an hour of unpaid labor indicating specialization, this is only one possible explanation. For example, lesbian women may spend more time on average working in the market than heterosexual women but may forgo leisure in order to spend an equal amount of time doing unpaid household and care labor due to preferences reflecting an internalization of gender norms. Furthermore, we may observe gay men on average spending less time on market work than heterosexual men, but this does not mean that gay men are necessarily spending more time on unpaid household work. There could be lower levels of household production in gay households. By looking directly at unpaid work and including an analysis of leisure, our approach offers a more precise investigation into the non-market behavior of gay men and lesbians and its relevance for market behavior and outcomes. We use time-use data to directly investigate the non-market behavior of cohabiting gay men and lesbians.

To our knowledge, Schneebaum (Reference Schneebaum2013) is the only research that utilizes time-use data in the US to study the economics of gay and lesbian households.Footnote ⁶ Schneebaum utilizes ATUS data to examine the responsiveness of time use to household characteristics of different-sex and same-sex couples. However, she does not test if patterns of non-market time use vary by sexual orientation after conditioning on observable characteristics.Footnote ⁷ Schneebaum’s descriptive statistics suggest that men in same-sex couples spend more time doing care work than men in different sex-couples and that women in same-sex couples also spend more time doing care work than women in different sex-couples (Schneebaum, Reference Schneebaum2013).Footnote ⁸ It is important to note that gay men and lesbians must exhibit different patterns in time use, after controlling for relevant characteristics, for the specialization hypothesis to explain labor market outcomes. We build on Schneebaum by asking: are patterns of household labor, care labor, and leisure systematically different for gay men and lesbians than heterosexual men and women? Our results give limited support to the specialization hypothesis as an explanation of labor market outcomes for gay men. Evidence in support of the specialization hypothesis for gay men is driven by households with children. For lesbians evidence regarding care labor that supports the specialization hypothesis does not hold when partitioning the sample by parenthood or when narrowing the definition of care labor, because children are key determinants to the division of labor in all household types (Giddings et al., Reference Giddings, Nunley, Schneebaum and Zietz2014). Furthermore, evidence contradicting the specialization hypothesis exist when considering women’s time spent on household labor in all specifications.

By comparing time use of gays and lesbians to heterosexual men and women, we are able to build on the literature by testing the specialization hypothesis. However, differences in time use between individuals in same-sex and heterosexual households, do not necessarily reflect differences in patterns of household specialization within these households. Below, we discuss possible intra-household behavior which could explain the time-use patterns we find, but leave a full investigation into the processes that occur within same-sex households to future work.

3. CONCEPTUAL FRAMEWORK

Economic models of the household such as Becker’s unitary household model, Chiaporri’s collective model, as well as the group of cooperative, non-cooperative, and cooperative conflict bargaining models are most widely used to understand dynamics and outcomes of heterosexual households [Manser and Brown (Reference Manser and Brown1980); Becker (Reference Becker1981); McElroy and Horney (Reference McElroy and Horney1981); Chiappori (Reference Chiappori1992); Sen (Reference Sen and Tinker1990); Lundberg and Pollak (Reference Lundberg and Pollak1994); Doss (Reference Doss1996); Lundberg and Pollak (Reference Lundberg and Pollak1996); Agarwal (Reference Agarwal1997); Carter and Katz (Reference Carter and Katz1997)].Footnote ⁹ One stylized fact that these models attempt to explain is heterosexual women’s greater contribution to unpaid household and care labor relative to their heterosexual male partners. Becker’s conclusions are based on the assumption of women’s comparative advantage in care and household labor and men’s comparative advantage in paid market labor. Chiapporri’s collective model considers asymmetries in earnings potential and preferences as determinants of labor supply outcomes. Bargaining models focus on differences in fall back positions and threat points for men and women that arise due to labor market inequalities and institutions. While typically these models are used to understand dynamics within households, they can also be used to predict systematic differences between households, namely those with same-sex partners versus different-sex partners.

Given the current gender differences in labor market outcomes in the US, it seems reasonable to expect the earnings potential of two same-sex partners, controlling for other demographic characteristics, to be more similar than those of different-sex partners. While the channels through which earnings potential affects time use varies by household model, all else held constant, more equal earnings potential translates into either more equal time spent in household labor or a more equal likelihood of being the partner who specializes in unpaid household labor.Footnote ¹⁰ Therefore, holding constant preferences, levels of household production, and budget constraints, existing economic models of the household predict that gay men will spend more time in the household and lesbians less time in the household relative to their heterosexual counterparts. The unequal division of labor stemming from the gender wage gap is particularly pronounced in different-sex households with children as the motherhood earnings penalty decreases earnings potential for women more than men. As such, the impact of parenthood, through its impact on earnings, should cause a more unequal distribution of household and care labor in different-sex households than same-sex households. It is important to note that the theoretical explanation for expected differences for individuals in same-sex versus heterosexual households is driven by inequality and social norms according to gender rather than sexual orientation. In the absence of such gender inequality in the labor market (and its effect on bargaining power), we would see equality in time use in different-sex and same-sex households.Footnote ¹¹ Systematic differences in time use by sexual orientation, then, result from widespread socio-economic gender inequalities.

Allowing for systematically different preferences in same-sex and heterosexual households can still yield a prediction consistent with the specialization hypothesis. Considering variation in preferences, it is possible that same-sex households (or individuals within same-sex households, depending on how utility maximization is modeledFootnote ¹² ) have stronger preferences for equality than different-sex households.Footnote ¹³ If this is the case, holding the level of production constant, we would still expect on average lesbians to spend less time and gay men to spend more time on unpaid work than their heterosexual counterparts.

The predictions of existing household models become less clear when we allow for systematically different levels of household production in same-sex and different-sex households. Household activities and market activities are jointly determined, because income from market activities is a necessary input for household production. However, income from market activities can also be used to purchase market substitutes for household goods. Members of lesbian households, despite an observed earnings advantage relative to heterosexual women, still experience the large gender wage gap. This gender gap in earnings causes lesbian households to face stricter budget constraints than heterosexual households. With a stricter budget constraint, the possibility of purchasing market substitutes for household and care labor is limited. Therefore, it is possible that the impact of a stricter budget constraint on household time use could dominate the effect of a more equitable distribution of household duties causing lesbians to spend more time in household production. Even in the absence of different budget constraints, lesbian households may choose a different level of household production from heterosexual households if social institutions, such as intolerance and gender norms, make the purchasing of markets substitutes less enjoyable or less feasible. On the other hand, gay men may be more likely to purchase market substitutes for household labor or may choose to produce fewer household goods. A full understanding of differences in non-market time use must consider the possibility of such systematic differences in household production.

Finally, the differential impact of norms and institutions may lead to outcomes inconsistent with the specialization hypothesis. The role of institutions and specific gender norms have been theorized to impact the division of labor between two partners in a heterosexual household.Footnote ¹⁴ Social norms, specifically those related to gender and family structure, can also explain systematic differences in time use and levels of household production in same-sex households compared to heterosexual households. Given the lack of sex-asymmetry within the household and unique challenges gay and lesbians face when constructing gender within a heteronormative society, the performance of gender is more difficult to predict in same-sex households. For example, both partners in a lesbian household may spend significant amounts of time on care labor attempting to meet societal expectations of motherhood. As such, the manifestation of gender in same-sex households may cause time use in same-sex households to be inconsistent with the specialization hypothesis. Another possibility is that pressures to emulate heteronormative conventions of the household may trump pressures to emulate traditional gender norms. For example, a gay man and a heterosexual man may have both been socialized as men, but pressures to create a heteronormative household may counteract pressure to conform to gender norms leading one gay partner to specialize in household and care labor. If this is the case, we would expect to find outcomes consistent with the specialization hypothesis for gay men.

4. DATA

We use data from the ATUS from 2003 to 2013. The ATUS is the only nation-wide time-use data source available for the US. The ATUS selects one member of a household from the Current Population Survey for inclusion in the survey. Survey respondents report the allocation of their time in minutes for one 24-hour period.Footnote ¹⁵ Only selecting one member from each household limits the ability to investigate different patterns in time use within households. However, selecting one member of each household does allow us to investigate if time-use patterns vary by sexual orientation. Further, we leverage the range of economic and demographic characteristics to investigate patterns of time use within households by investigating the impact of characteristics such as earnings that theoretically determine the intra-household distribution of time use. As we are interested in investigating patterns of time use as they relate to household specialization and labor market attachment, we are primarily interested in studying individuals who are of prime working age. Therefore, like Blau and Kahn (Reference Blau and Kahn2005) and Martell (Reference Martell2014), we limit our sample to respondents who are between the age of 25 years and 54 years to focus on workers who are likely to exhibit a higher degree of labor market attachment and a higher incidence of family obligations such as childcare. This restriction also allows us to focus on workers who are less likely to be choosing full-time education or retirement.Footnote ¹⁶

Gay men and lesbians may be identified in the survey if they cohabit with a member of the same sex.Footnote ¹⁷ As such, our analysis excludes single individuals. Our sample includes 102 cohabiting gay men, 16,539 married heterosexual men, 1,008 cohabiting unmarried heterosexual men, 136 cohabiting lesbians, 18,844 married heterosexual women, and 1,105 cohabiting unmarried heterosexual women. While small, the sample size of gay men and lesbians is larger than or comparable to the sample sizes used in many analyses of gay and lesbian labor market outcomes utilizing GSS data [see Badgett (Reference Badgett1995); Martell (Reference Martell2013a); Berg and Lien (Reference Berg and Lien2002)].

Descriptive statistics are shown in Tables 1 and 2. For the most part, the characteristics of cohabiting gay men and lesbians are consistent with existing demographic research [see Black et al. (Reference Black, Sanders and Taylor2007)]. Cohabiting gay men are more likely to live in an urban area, more likely to be white, are on average older,Footnote ¹⁸ more educated and have less children than married heterosexual men and cohabiting heterosexual men. Because gay men have higher levels of observable human capital such as education, descriptive statistics show a small but insignificant difference in earnings between cohabiting gay and married heterosexual men. The small difference here is consistent with research showing that the gay earnings penalty has been shrinking [Cushing-Daniels and Yeung (Reference Cushing-Daniels and Yeung2009)] as well as research showing that the observed gay earnings penalty is mitigated by higher levels of human capital [Martell (Reference Martell2013a)]. The earnings of the partners of gay men are higher than earnings of the partners of heterosexual men, which is not surprising given the gender wage gap. Total household earnings are similar for each type of household. In our sample, gay men are on average older, earn more and are more likely to own a home than cohabiting heterosexual men.

Table 1. Descriptive statistics for demographic variables

Note: Authors’ calculations from American Time Use Survey data, 2003–2013. * Difference from gay/lesbian average significant at 10% ** Difference significant at 5% *** Difference significant at 1%.

Table 2. Descriptive statistics for daily time use in minutes

Note: Authors’ calculations from American Time Use Survey data, 2003–2013. * Difference from gay/lesbian average significant at 10% ** Difference significant at 5% *** Difference significant at 1%.

The lesbians in our sample are more likely to live in urban areas, are on average younger, more educated, have higher earnings, and have fewer children than married heterosexual women. The earnings of partners of lesbians are found to be lower than those of the partners of married heterosexual women and household earnings are more equitable. While lesbians are less likely to own a home than married heterosexual women, lesbians are more likely to own their home than cohabiting heterosexual women. Consistent with existing analyses of the geography of same-sex households [Gates and Cooke (Reference Gates and Cooke2011)], lesbians are more likely than gay men as well as cohabiting and married heterosexual women to live in a state in which same sex marriage is legal. This residential pattern highlights the importance of controlling for not only regional variations in the residences of same-sex households but also variations in the legal institutions in which same-sex and different-sex households reside as we discuss below.Footnote ¹⁹ We do not find a similar significant difference when comparing cohabiting gay men to heterosexual men.

With the ATUS data, we investigate daily minutes spent in household labor, leisure, and care labor on the day of data collection. These time categories are as defined in the ATUS classifications.Footnote ²⁰ Consistent with standard practices in time-use research,Footnote ²¹ we define total care labor as including all care labor reported as a primary activity and childcare reported as a secondary activity. Secondary childcare is defined as time when at least one child under the age of 13 years is in the respondent’s care and at least one child is awake. It is particularly important to include secondary childcare as a component of care labor given our interest in time use and its relationship to labor market attachment. Even though it is possible to complete household labor or leisure activity while a child is present, it is much more difficult to do paid market work while a child is in one’s care. This difficulty is most pronounced when paid market work is outside the household as is generally the case in the US.Footnote ²²

Table 2 highlights the importance of considering the varying incidence of parenthood across household types in understanding time use in same-sex households. Gay men do considerably more household labor but less care labor than both married and cohabiting heterosexual men. While the difference in household labor is as expected given the traditional gender division of labor for heterosexual men, the difference in care labor is not. For married and cohabiting heterosexual men, about 80% of total care labor is secondary childcare and for gay men 63% of care labor is secondary childcare. When we partition the sample into men with children and men without children, the difference in time spent on care labor disappears. The difference in the amount of time gay and heterosexual men spend on household labor persists in the subsample of men with children. For the subsample without children, we see no statistically significant difference in average time spent on household and care labor. The differences in time spent in leisure are not statistically significant in the full sample or either subsample.

Without any controls, it appears that cohabiting lesbians exhibit different patterns of time spent in household labor and care labor than their heterosexual counterparts, as shown in Table 2. Cohabiting lesbians spend approximately 40 minutes less time in household labor and 90 fewer minutes in care labor than married heterosexual women. Relative to cohabiting heterosexual women, lesbians spend nearly 60 minutes less time in care labor but time spent in household labor and leisure are similar. Once again the differences in time spent in household labor are as expected for lesbians. The difference in care labor is as well. Similar to men, a large portion of total care labor is secondary childcare. Further, a higher share of lesbian women have small children than gay men, but they are still less likely than heterosexual women to have children. When the sample is partitioned into women with and without children, the differences in average daily time use between lesbians and heterosexual women become statistically insignificant. The one exception is that heterosexual married women without children spend a statistically significant greater amount of time on household labor than lesbians without children. The varying patterns of parenthood highlight the importance of controlling for children when investigating time-use data. We explore the impact of parenthood below and are among the first to incorporate the presence of children in comparing the time use of cohabiting gay men and lesbians to their heterosexual counterparts.Footnote ²³

5. EMPIRICAL SPECIFICATION

To estimate daily time use in household labor, leisure, and care labor, we follow the approaches of Kimmel and Connelly (Reference Kimmel and Connelly2007), while incorporating controls also included in Giddings et al. (Reference Giddings, Nunley, Schneebaum and Zietz2014) and Jepsen and Jepsen (Reference Jepsen and Jepsen2015). We are interested in estimating average time-use patterns across household types. Since reports of no time spent in any of these categories represent classical measurement error and not censoring, we model time use via seemingly unrelated OLS regressions (Stewart, Reference Stewart2013).Footnote ²⁴ Because the use of survey weights have the potential to lead to imprecise estimates in the context of predicting time use for same-sex households, we do not use the ATUS survey weights (Solon et al., Reference Solon, Haider and Wooldridge2015).Footnote ²⁵ Specifically, we estimate

(1) $$\begin{equation} \hat{y_i}=\alpha + \beta _1 H_i + \beta _2 T_i + \bm{\beta }X_i + \epsilon _i, \end{equation}$$

where y_i =household labor, care labor, and leisure for individual i. H_i is an indicator for cohabiting with a partner of the same-sex and is the key explanatory variable, T_i is a vector of year indicators, and X_i includes relevant demographic characteristics: age and its square, weekly earnings, spouse’s weekly earnings, the percent of household income that is comprised of the respondents earnings, years of education, and indicators for: being white, an urban residence, home ownership, the presence of a child under 6 years, the presence of children between 6 years and 18 years, the presence of non-own child, the presence of third adult, a summer time diary, a weekend time diary, a holiday time diary. We also include an indicator for respondents living in a state where same-sex marriage is legal to control for varying levels of tolerance of homosexuality as well as geographic variation in time use. The same-sex marriage indicator is used in place of regional controls as same-sex marriage legalization is regionally clustered.Footnote ²⁶ In some specifications discussed below, we interact the same-sex indicator with the presence of children to investigate any differences in the impact of parenthood on time use by sexual orientation.

We control for the weekly earnings of the respondent as well as those of the respondent’s partner as earnings potential affects time use in and outside of the household. However, as labor market outcomes such as earnings are in part the product of the intra-household distribution of labor, there are theoretical reasons for suspecting that weekly earnings may be endogenous. This potential endogeneity poses an empirical difficulty for researchers interested in estimating the impact of wages on household time use. To correct for this potential endogeneity, researchers can use predicted wages or instrumental variables to predict the exogenous effect of earnings on time use. Unfortunately, our data lack valid instruments for earnings as well as the ideal covariates necessary to make accurate earnings predictions. As such, we follow Bredtmann (Reference Bredtman2014) and use actual earnings in estimation. A priori, the extent to which the household process that determines weekly earnings varies by sexual orientation, and would bias our estimates, is unclear. Therefore, we do not believe its endogeneity will bias the estimate of our variable of interest: the member of a same-sex household indicator. Even though Hersch and Stratton (Reference Hersch and Stratton1997) find the results of instrumental variables estimates to be nearly identical to the standard OLS estimates of wages on time work suggesting that earnings endogeneity does not bias time-use estimates, we still caution readers in interpreting the coefficient estimate of earnings in isolation.

In addition to earnings, we also control for the percent of household income that is comprised of the individual’s earnings. The share of household income comprised of the respondents earnings or some measure of relative wages is a commonly used proxy for household bargaining power [see, for examples, Kimmel and Connoly (Reference Kimmel and Connelly2007), Masterson (Reference Masterson2012), and Hener (Reference Hener2015)]. Even though it is an imperfect proxy for bargaining power, controlling for the respondent’s earnings share allows us to test for differences in time use correcting for the distribution of relative earnings potential (the opportunity costs of time) of household members as well as the ability of household members to exert influence over the intra-household distribution of household time use. Both of these factors may vary by sexual orientation.

In testing, if time use varies by sexual orientation, it is unclear whether married or unmarried cohabiting heterosexuals is the most appropriate reference group for cohabiting gay men and lesbians a priori. Unfortunately, marital status of same-sex couples is not observable throughout the entire span of our data. At the beginning of our data, no states allowed gay men and lesbians to legally marry [Movement Advancement Project (2014)]. Throughout the time span of the data, in many states, gay men and lesbians were in marriage-like relationships but not allowed to marry. On the other hand, some of the cohabiting gay men and lesbians may be more akin to cohabiting heterosexuals. After conducting log likelihood ratio test comparing specifications that combine cohabiting and married respondents to specifications that partition the sample, we conclude that it is most appropriate to estimate time use separately allowing coefficient estimates to vary reflecting different time use and bargaining processes across household types.Footnote ²⁷

Investigating the impact of parenthood on time use among same-sex households is difficult, and children are an important determinant of time use and household specialization. Our descriptive statistics show that same-sex household are less likely to have children. If selection into parenthood varies by sexual orientation, which we discuss in detail below, conditioning on the presence of children may bias our estimates. To mediate this potential bias and examine the impact of parenthood on time use, we investigate differences in time use separately by the presence of children instead of simply conditioning on children as primarily done in the existing literature.