While the division of domestic labor among couples has evolved, it remains central to the (in)equality between women and men. The situation differs according to union status, whether cohabiting or formalized. Are these differences related to the socioeconomic characteristics of partners in cohabiting relationships, civil unions, and marriages? Further, does the very nature of these unions reflect the more or less egalitarian values of the people who enter into them? Using three waves of the INSEE time-use survey, the authors explore these questions from a long-term perspective, taking into account social, economic, and legal developments.
1Despite the increase in female labor market participation, women still perform a greater proportion of housework and childcare than men. Research on the domestic division of labor within couples suggests that cohabiting couples are less traditionally organized than married ones. Cohabiting women perform less domestic labor than married women, whereas cohabiting men report performing more domestic labor than married men (Blair and Lichter, 1991; Baxter, 2001; Davis et al., 2007; Domínguez-Folgueras, 2012; Bianchi et al., 2014). Three main factors can explain the differences in shared housework according to union type. First, the socioeconomic and demographic characteristics of couples might differ from one union type to another (hereafter called the socioeconomic characteristics effect), as cohabiting couples can have different characteristics from those of married couples (educational level, involvement in paid work, number of children, etc.), thus leading to more egalitarian organization (Domínguez-Folgueras, 2012). Constituting a second factor are the differences in legal regulation of unions (hereafter called the marital status effect). The degree of institutional support of marriage relative to cohabitation can create greater incentives for married partners (especially women) to invest in domestic activities. Conversely, when one partner works more than the other in the labor market, such specialized couples may opt for marriage because it offers a legal framework conducive to their organization (Barg and Beblo, 2012). The gender gap in paid and domestic work between partners is then expected to be larger in the most regulated unions (Bianchi et al., 2014). The third factor pertains to values, such as religion and preferences regarding parenthood or gender norms (hereafter called the gender values effect). Indeed, couples with egalitarian values share domestic work more equally (Greenstein, 2000; Nitsche and Grunow, 2016). Consequently, if a specific union type attracts individuals holding gender equality values, then the share of domestic work performed by the woman within these couples is expected to be lower than in other types of union. Cohabiting couples may be committed to more egalitarian values than married couples, thus reducing women’s share of domestic work.
2This paper contributes to the literature in three ways. First, it draws upon three waves of the French time-use surveys (1985–1986, 1998–1999, and 2009–2010) to provide insight into the gender gap in domestic labor within couples according to marital status, as well as into the ways this gender division has changed. We account for the self-selection of couples in different types of union based on their observable characteristics. Ignoring this effect may bias the analysis of the legal framework’s impact. Secondly, it contributes to understanding the gendered division of labor within couples by examining links between gender values and the legal regulation of unions, whereas most research focuses on one or the other. Thirdly, it reveals how a legal modification to the regulation of unions has impacted couples’ choices. The French case is particularly enlightening because the civil partnership contract (pacte civil de solidarité [PACS]), implemented in 1999, serves as an intermediary regulation between cohabitation and marriage (Appendix Table A.1). This form of civil union is gender-neutral and not restricted to same-sex couples. It can reveal how gender values contribute to the ways couples divide domestic labor according to union type. Similar legislation has been enacted in other countries, such as the Netherlands (1998), Portugal (1999), Belgium (2000), and Luxembourg (2004).
3Using the ordinary least squares (OLS) method, we analyze the differences in the division of domestic labor according to union type while controlling for the characteristics observed in time-use surveys. Yet, because the OLS estimation does not account for couples self-selecting into different union types based on their socioeconomic characteristics and gender values, this method cannot help in disentangling the three effects listed above. Therefore, we use the matching method to control for the socioeconomic characteristics effect. Then, the remaining gap in sharing domestic tasks based on marital status can be attributed to the two other effects. To disentangle the marital status effect and the gender values effect, we test the following two hypotheses using changes over time and the impact of the PACS civil partnership:
- Hypothesis 1: According to the marital status effect, the more regulated a union is, the stronger the division of labor is within the couple. Marriage is the most regulated union, followed by the PACS, then cohabitation, which confers few legal protections and obligations. We therefore expect to observe that the woman’s share of domestic work will be highest in marriages, followed by civil partnerships and cohabiting unions.
- Hypothesis 2: The gender values effect implies that couples with egalitarian values opt to share domestic work more equally. We expect that if a specific form of union attracts more egalitarian couples, then the proportion of work performed by the woman within these couples will be lower than in other types of union.
4Throughout the analysis, we consider the three effects as independent, although intertwined. For instance, couples may be dual earners because both partners hold gender equality values. The gender values effect is linked to the socioeconomic characteristics effect, as the most educated couples share more egalitarian norms (Domínguez-Folgueras, 2012).
I – Specialization within couples
1 – Why do couples specialize?
5Economic resources and gender norms are key factors that affect the division of housework among couples. Bargaining power impacts the degree to which one partner specializes in household labor, depending on each partner’s actual or potential contribution to the family income. Men with higher-earning partners do more housework than other men, although women still do more than their partners (Lyonette and Crompton, 2015). Women’s housework is negatively associated with their own earnings, and this effect is greater than that of their partner’s earnings (Gupta, 2006). Each partner’s contribution to household income determines their respective bargaining power and thus the sharing of domestic labor. The empirical literature on the time-availability factor has evaluated the effect of both partners’ paid work on the distribution of domestic work, and it has found that women working full-time contribute less to domestic work than inactive women or those working part-time, while dual-earner couples share domestic work more equally (Gershuny, 2000). Men who spend less time in paid work spend more time on domestic work. When both partners work full-time, the distribution of domestic and family work becomes less unequal, although women still perform a larger share than their partners do (Ponthieux and Schreiber, 2006). Similarly, the association between unemployment and time reallocated to housework proves to be more disadvantageous for unemployed wives than for unemployed husbands (Gough and Killewald, 2011).
6A large body of literature has analyzed how individuals “do gender” in the domestic sphere. Some of this research focuses on how gender norms impact couples’ division of labor through the daily construction of identity (Brines, 1994; Greenstein, 2000; Evertsson and Nermo, 2004). According to Brines (1994), the link between economic dependence and performance of housework is consistent with the empirical findings of the literature on economic resources and time availability, but it is observed only for wives. Husbands dependent on their partners do the least housework, whereas husbands earning the same as their wives do the most. The concept of “gender display” explains that couples who do not comply with the male breadwinner model might opt for a more traditional division of housework. The theory of “gender deviance neutralization” (Greenstein, 2000) states that men who are financially dependent on their partners compensate for this deviance from gender norms by investing less in domestic chores and, for the same reasons, women working full-time tend to increase their contribution to domestic duties if their partner loses his job. Killewald and Gough (2010) showed that the relationship between wives’ earnings and their housework is not linear: high-income women do not decrease the time they spend on housework when their earnings increase, as they already outsource a large part of domestic work, but low-income women do (Killewald and Gough, 2010). Finally, the social incentive to perform gender has less impact on decisions to share housework than does each partner’s relative bargaining power via their wages (Bianchi et al., 2000; England, 2011; Sullivan, 2011; Ponthieux and Meurs, 2015). Beyond the economic and gender dimensions, the legal structures of different types of union might affect the division of housework.
2 – Marital status and division of labor within couples
7In general, cohabiting couples are more egalitarian than married couples in adopting patterns of sharing domestic and family work (Baxter, 2005; Davis et al., 2007; Domínguez-Folgueras, 2012; Bianchi et al., 2014). Socioeconomic and demographic factors impact both the division of labor within the couple and the choice of marital status (socioeconomic characteristics effect). Compared to cohabiting couples, the greater specialization of married couples observed in many countries is partly due to differences in couples’ socioeconomic characteristics (educational level and wages, among others). For example, relative to married women, cohabiting women in Italy are more educated, better integrated into the labor market, and they perform less housework (Kiernan, 2002; Bianchi et al., 2014). However, economic resources alone cannot explain the gender division of labor based on union type, since married women bear a greater burden of the domestic work compared to cohabiting women, regardless of their respective incomes. Having children is also an important factor for consideration, as married couples can specialize more than cohabiting couples due to their greater likelihood of having a child (Barg and Beblo, 2012).
8The legal regulation of each union type is another factor that explains the differences in sharing domestic labor according to union type. Marriage tends to confer more legal rights and obligations than cohabitation. In cases of separation, the risk of a decrease in standard of living for partners specializing in domestic work then becomes higher among cohabiting individuals relative to their married counterparts. In this context, a greater specialization is expected in the most regulated unions like marriage, while it should be lower in cohabiting relationships. Bianchi et al. (2014) tested this hypothesis using the differences in marital regimes in three countries (France, USA, and Italy). They expected that the differences in time spent on paid and domestic work between cohabiting and married couples will be largest in Italy, where cohabitation is not legally recognized; followed by the United States, where regulation varies by state; and smallest in France, where cohabitation is close to marriage, including PACS unions. They found that cohabiting women do less domestic work than married women, but after controlling for observable characteristics, the difference remains only in Italy. In Nordic countries, where the rights of cohabiting couples are almost similar to those granted to married couples (Sánchez Gassen and Perelli-Harris, 2015), the latter are less specialized than in other countries (Davis et al., 2007).
9The direction of the causal relationship between marital status and specialization (marital status effect) is undetermined a priori, as couples can specialize after marrying in response to the social protections and benefits associated with marriage. In this case, the regulatory framework for marriage creates incentives for couples to adopt gender specialization. Couples can also specialize and then marry to benefit from the protections adapted to their household organization. Marriage is thereby the response to, rather than the cause of, the couple’s specialization. The institution of marriage would then be viewed as an insurance that guarantees the working partner’s, i.e. the man’s, protection and commitment to the partner specializing in domestic work, i.e. the woman. This protection extends beyond the union itself by granting the right to financial compensation in case of divorce. These two causal relationships may coexist as couples begin to specialize before marriage, then marry as a result, and ultimately reinforce this specialization (Barg and Beblo, 2012).
10The gender values approach provides another perspective of the division of labor within couples. The gender values of couples can be measured on a scale ranging from egalitarian (favorable to sharing domestic and family responsibilities) to conservative (favorable to the male breadwinner model). Men with egalitarian values are more involved in housework than men with conservative values (Greenstein, 2000). From a life course perspective, both partners having egalitarian values leads to dynamics of more egalitarian division in Germany (Nitsche and Grunow, 2016). Furthermore, the social perceptions associated with each union type explain the link between gender values and union type (gender values effect). In cases where cohabitation is considered a simple prelude to marriage, it should then be associated with the same values as marriage. If it represents an alternative for rejecting the view of marriage as a patriarchal institution (Heuveline and Timberlake, 2004), it should then be associated with more egalitarian values. Married couples behave more in accordance with gender norms and specialize more than unmarried couples do (Shelton and John, 1993). In Italy, for example, cohabiting women not only adhere to more egalitarian values and perform less housework than married women, but they also have a higher labor force participation rate (Domínguez-Folgueras, 2012; Bianchi et al., 2014).
3 – France as a case study
11In France, women account for 71% of the time dedicated to housework (cleaning, cooking, and laundry, among other chores) and 65% of parental time (caring for children). While the time women spend on domestic work has decreased since the 1980s, the amount men spend has remained stable (Champagne et al., 2015). Women in couples perform more housework and family work than do other women (Roy, 2012), while the arrival of children reinforces the unequal sharing of tasks within the couple. Even though fathers are devoting more time to the children’s education than previously, the amount mothers spend has also increased since the 1980s (Régnier-Loilier and Hiron, 2010; Ricroch, 2012).
12Since the late 1990s, cohabitation has become a socially accepted mode of union, and the arrival of a child no longer triggers a marriage. As in the Nordic countries, the proportion of children born outside marriage in France is among the highest in Europe, at 61% in 2019 (Prioux, 2009; Breton et al., 2019). However, the two unions are subject to differences in social, tax, and legal regulations that are much more marked in France than in the Nordic countries (Sánchez Gassen and Perelli-Harris, 2015). For example, the French welfare state provides protections for married women and compensation for their investment in domestic activities (such as survivors’ benefits, rights and obligations between spouses, and financial compensation in case of divorce). In contrast, cohabitation is not subject to any compensation or obligation toward the partner specializing in domestic activities and family work, not even partially.  If the partnership ends, the risk associated with the gender division of labor for cohabiting couples is then borne entirely by the partner who was more involved in the family sphere, in most cases the woman. The 1999 enactment of PACS civil partnerships partially and gradually incorporated some protections initially reserved for marriage (Appendix Table A.1), and since 2005, PACS couples are subject to the same joint taxation as married couples, which has increased PACS rates (Leturcq, 2012). More heterosexual couples are progressively opting for PACS unions. In 2011, 75% of couples were married, 20% were cohabiting, and 5% were contracted under the PACS legislation (Buisson and Lapinte, 2013).
13The few studies that exist show that French women perform more domestic work than men, regardless of marriage or cohabitation. Domínguez-Folgueras (2012) found that unmarried couples distribute the domestic chores in a more egalitarian fashion than married couples do, but the data used (multinational time-use surveys) do not allow cohabiting and PACS couples to be distinguished. Bianchi et al. (2014) showed that the paid working hours of married and cohabiting women are roughly the same (with a difference of 4 minutes only). They focused on the role that institutional differences between marriage and cohabitation play in the gender gap between paid and unpaid work at the individual level. The results show that, relative to either France or the United States, Italy presents the greatest gender differences in time allocated to market and non-market work. We examine this matter in greater depth by looking at couples—and not individuals—to show the links between the gender division of labor and marital status in France. By taking into account that civil partnerships are a form of union distinct from marriage and cohabitation, we can clarify the existing literature’s study of couples’ behaviors regarding the division of household labor.
II – Data and methodology
1 – The French time-use surveys
14Time-use surveys for 1985–1986, 1998–1999, and 2009–2010 (hereafter named the 1985, 1998, and 2009 surveys) are used to explore changes in the distribution of domestic labor within couples according to their marital status. The French National Institute of Statistics and Economic Studies (INSEE) implemented these three time-use surveys within metropolitan France over 12-month periods overlapping 2 calendar years. They include a questionnaire describing the household composition, a second questionnaire providing information about the surveyed individual, and an individual diary (detailing activities at 5-minute intervals for the 1985 survey and at 10-minute intervals for the two others). In the 1985 and 1998 surveys, individuals filled in diary entries for a day of the week (of either the workweek or weekend), while each respondent in the 2009 survey reported on a weekday and in some cases an additional weekend day. Surveying an individual in a coresident partnership implies that, in most cases, the partner is also surveyed and that both fill in their diaries on the same day. 
15The literature identifies this methodology for collecting time-use data as the most reliable (Geist, 2010; Ponthieux, 2015), even though some gender biases can still exist. For the 1985 and 1998 surveys, two types of couples are distinguished: married and cohabiting. Since the PACS law was passed in 1999, the 2009 survey added PACS couples. Married couples are those who declare their status as married (or remarried); civil partners are those who declare they have entered a PACS; and cohabiting couples live together but have not declared being either married or in a civil partnership. The sample in this analysis includes households where the reference person lives with another household member as a couple, whatever the length of this union.  The sample retains only couples in which (a) both partners are between ages 25 and 55, (b) at least one member of the couple is active in the labor force, and (c) both partners have filled in the diaries. The analysis excludes couples in which one partner is a student, retired, or disabled, as well as complex households with other housemates (such as elder parents or relatives). Same-sex couples have also been excluded. In addition, we have removed from the sample the few couples for which we lack all the requested information for running the analysis.  The initial data contain 5,598 couples in the 1985 survey, 5,152 couples in the 1998 survey, and 7,252 couples in the 2009 survey. After basing the selection on the criteria described above, the final sample consists of 3,334 couples for the 1985 survey, 2,715 for 1998, and 2,873 for 2009.
16The datasets contain detailed information on the tasks performed by each partner during the day. The scope of domestic work used for the analysis includes the most burdensome everyday activities, such as cooking, dish-washing, doing the laundry, putting things away, cleaning, shopping, and stocking. These also cover household management, such as doing accounts, administrative correspondence, seeking administrative services, trips, caring for children, caring for other adults, and other miscellaneous activities.  This definition is used by Roy (2012), who excludes tasks considered intermediate or semi-leisure: sewing, repairing, gardening, fishing, and time spent on leisure with children or educating them, etc. The definition of both housework and childcare used for the analysis is restricted to the most routine tasks (Appendix Table A.2).
2 – Methodology: OLS regression and matching
17The first step in the econometric strategy consists of estimating a simple model in which the dependent variable is the share of domestic work performed by the woman in the couple. It is defined as the ratio between the time spent by the woman and by both members of the couple on domestic work. Using OLS regression, the woman’s share of domestic work is explained by the socioeconomic characteristics available in the data. The explanatory variables include each partner’s individual characteristics, such as educational level and employment status.
18Bargaining power constitutes an important explanatory variable for the woman’s share of domestic work. As this is not directly observed in the data, it must be approximated. Our estimations use the relative wages of both partners as a proxy,  defined as the hourly wage gap between the partners relative to the sum of their hourly wages.  Hourly wage is a better indicator than monthly or yearly wages that depend on the number of hours spent in the labor market, and thus they are endogenous with respect to the time devoted to domestic work. The hourly wage represents here an individual’s potential gains in the labor market, which gives him or her bargaining power despite not working. Therefore, a measure of bargaining power is also assigned to non- working individuals.
19Since the wages of non-working individuals are not observed, we estimated a wage equation to assign them a potential predicted wage (Appendix, ‘Estimating the wage equation’). Our wage equation estimation uses Heckman’s method to factor in the selection effect on the labor market, specifically by simultaneously estimating the equations for wage and for participation in the labor market (Heckman, 1979). To calculate bargaining power, we based the predicted wages on our estimated wage equations and assigned them to individuals with unobserved salaries (due to unemployment, inactivity, or uncompleted pay field in the survey), as well as to their partners. The observed wage was used for everyone else. The 1985 time-use survey lacks information on wage income, which explains why the empirical analysis of this survey does not use bargaining power as a control variable.
20We have added variables to control for the couple’s total income, the type of household appliances (such as a dishwasher, washing machine, or microwave), the use of paid or unpaid cleaning services, and the total amount of domestic work the couple performed. We also include the number of children and the presence of a child under 3 years old, as having children in the household increases the woman’s share of domestic work. This group of variables approximates task-sharing pressure on the couple. We have introduced the couple’s average age and added the age gap between partners because it might influence bargaining power. Finally, rural versus urban habitations are distinguished due to their influence on the possibility of outsourcing part of the domestic work.
21Additionally, as the use of time and its distribution between partners vary between weekends and weekdays, we use an indicator to control for the day on which they filled in their diaries. For the 2009 survey, we introduce a diary variable that distinguishes the following situations: (a) only two diaries per household, and both partners completed their diaries on a weekday; (b) only two diaries per household, and both partners completed their diaries on the weekend; (c) three diaries per household, with one partner filling in two diaries and the other only one; and (d) both partners filling in two diaries each, i.e. four diaries per household (reference situation). For the 1985 and 1998 surveys, we introduce a dichotomous variable set to 1 if the diaries were filled in on the weekend and 0 otherwise. Recall that we are interested in the domestic work time the couple spends on an average day. So, because the 2009 survey provided fairly accurate information on the time partners spent on weekdays and weekends, we could use all this information to measure the couple’s division of domestic work. We take the weighted average of weekends and weekdays based on the weights of the corresponding diaries. This allows us to consider both partners’ different distributions of time spent on domestic work throughout the week rather than treating the weekend and weekday diaries separately and clustering couples.
22To account for the self-selection of couples into different types of union, we then use a matching method. This method associates each married couple with one or more non-married couples with similar characteristics. By doing so, we can compare a married woman’s share of domestic work with that of her comparable unmarried counterpart. Unfortunately, we cannot control for gender values, as the time-use surveys do not include variables that correspond to these. Thus, the matching method can only control for the socioeconomic characteristics observed in the surveys. The matching method has an underlying identification assumption based on conditional independence (or unconfoundedness). Given an assumed vector of observable characteristics xi that captures the self-selection bias, then, conditional on xi, the “marriage treatment” is random. This implies that, again conditional on xi, the expected division of domestic work in unmarried couples is equal to that of married couples before their marriage.
23If it were possible to use the same characteristics for exact matching between married and unmarried couples (cohabiting or in civil partnerships for the 2009 survey), we could build a perfect counterfactual. However, this matching is not feasible. The matching problem is therefore reduced to a single dimension known as propensity score matching (Rosenbaum and Rubin, 1983), defined as the probability of marrying p(x). This propensity score verifies the theoretical balancing property: the conditional distribution of xi, given p(xi), is orthogonal to the choice of marital status. This property implies that the distribution of x within subgroups of couples having the same propensity scores p(x) should be identical between different types of couples. Once conditioned on the propensity score, the independence between division of labor and choice of marital status is also satisfied.
24Accordingly, we estimate the probability of marrying across all couples and compare the distributions of the propensity score for married and cohabiting couples (and civil partnerships in 2009). Our estimates retain only couples with common support of distributions.  The matching is then performed between the married and unmarried couples (cohabiting or civil partners) with identical propensity scores, and the balancing property assumption is verified. Once satisfying these two conditions, we dispose of subgroups of paired couples (married vs. non-married) whose mean propensity scores are not statistically different and for whom the set of variables conditioning the score has an identical distribution.  The results presented below are based on the Epanechnikov kernel matching method: every married couple is paired with all the unmarried couples, weighted by their propensity score distance.  Furthermore, the woman’s share of domestic work in each married couple is compared with that of the counterfactual.
25Using this procedure, the estimated task distribution differences between different union types cannot result from couples self-selecting based on their observable characteristics. The set of variables xi used to estimate the propensity score p(xi) is a subset of the OLS explanatory variables: household income, both partners’ labor market status, bargaining power, partners’ average age, their age difference, their educational level, the presence of a child, the presence of a child under age 3, and their area of residence. Then, the difference between each married woman’s share of domestic work and that of the counterfactual is estimated, controlling for all the same variables used in the OLS method. The final treatment effect is obtained by averaging over the estimated differences.
26Therefore, it is worth noting that OLS is distinct from the matching method in two important ways that impede directly comparing the magnitudes of estimates. First, matching provides local effects because it must verify the existence of a suitable counterfactual, while OLS gives a global measure over the whole sample. Second, matching produces effects that are the average difference between each married couple and their counterfactual. This means that married couples are given more weight because they have higher scores, while OLS puts more weight on control variables, which have identical values between the two types of marital status.
27As mentioned previously, once we control for the socioeconomic characteristics effect, the remaining gap between couples can be explained by the marital status effect (Hypothesis 1) and the gender values effect (Hypothesis 2).
III – Results and analysis
1 – Descriptive statistics
28Table 1 describes the sample according to the different characteristics. For the three periods under review, married couples had more children than cohabiting and PACS couples. Except for the 1985 survey, the age gap between partners was similar across all union types, but married couples were older than cohabiting and PACS couples. In 1985, married couples were less educated (both husbands and wives) than cohabiting couples, but the educational gap narrowed in 1998. In 2009, civil partners were more educated than married and cohabiting couples.
29Table 2 describes the weekly time spent on domestic and paid work by women and men living in unions, according to their marital status and the woman’s average share for each union category in the three surveys. A married woman’s average share of domestic work in 1985 was 80.9% versus 75.1% for a cohabiting woman. However, in 2009 a woman’s average share of domestic work was much the same whether married (73.5%) or cohabiting (72%). This convergence was the result of two trends: the share borne by married women declined significantly, part of a general downward trend in the time women spend on domestic work. Thus, the extent of the gender division of labor in married couples grew closer to the level observed in cohabiting couples. Simultaneously, the 1999 introduction of the PACS changed the legal framework: women in civil partnerships performed the lowest share of domestic work (65.1%). At the same time, men in PACS couples spent 2 hours 30 minutes more on domestic tasks than married men, compared to a gap of only 3 minutes between married and cohabiting men.
30Data on the time dedicated to paid work are incomplete (see note to Table 2), so the results must be interpreted with caution. Nevertheless, in 2009, we observe that women in PACS couples worked more than others: around 31 hours a week versus 28 hours for cohabiting women, and 27 hours 50 minutes for married women. Combined with the trend in domestic work, this explains why the woman’s average share of total work (paid and domestic work) is higher in cohabiting couples (52%) compared with married (50.4%) and PACS couples (50.7%). Married women spent 1 hour and 53 minutes more on domestic work compared to women in civil partnerships and 56 minutes more than cohabiting women. They performed 3 hours and 10 minutes less paid work than PACS women and 22 minutes less than cohabiting women. On average, the share of total work performed by the woman is comparable in married couples and PACS couples, but the repartition between paid and domestic work differs.
Table 1: Characteristics of individuals and households based on union type
Table 1: Characteristics of individuals and households based on union typeNote: Subsamples are in parentheses. Minimum and maximum observed values are in square brackets; n/a = not available.
Coverage: Couples in which both members filled in diaries and at least one is active.
Table 2: Division of labor within couples according to union type
Table 2: Division of labor within couples according to union typeInterpretation: The woman’s average share of “total work” and “domestic work” corresponds to her average share at the couple level. It differs slightly from the share of average total or domestic work performed by women in the sample.
Note: Hours of paid work were missing for some working persons in the samples, to whom we assigned the average paid work hours of all observed working persons, by sex and marital status. Paid work values were missing for the following: 1985, 212 out of 3,189 working men, 183 out of 2,137 working women; 1998, 458 out of 2,571 working men, 282 out of 1,963 working women; 2009, 685 out of 2,687 working men, 664 out of 2,370 working women.
Coverage: Couples in which both members filled in diaries, and at least one is active.
2 – Detailed results of the estimates
31Table 3 displays the results of the OLS regression model, upon which we base our interpretation. The reference couple is married, with no dependent children, residing in an urban area, both working full-time, and the educational level of both is less than a French high school diploma (baccalauréat).
32The effect of marital status variables changes according to the year of observation. In 1985 and 2009, a cohabiting woman’s share of domestic work did not significantly differ from that of a married woman, while in 1998 it was about 3.5 percentage points lower. The result for 2009 is consistent with the findings of Bianchi et al. (2014). However, all else being equal, the share of domestic work carried out by women in civil partnerships is about 4.5 percentage points lower than a married woman’s share.
33The estimates indicate that a woman’s share of domestic work declines as household income increases. Indeed, couples with higher incomes outsource more domestic tasks, especially those performed by women (cleaning and laundry in particular). The variables related to household equipment and appliances indicate that, in 1985, having a washing machine cut a woman’s share of domestic work by 4.3 percentage points. In 1998 and 2009, this variable lost its significance due to the increase in levels of household equipment; by the late 1990s, most households with this sample’s characteristics had a washing machine.
34The main results of the OLS estimation are consistent with the literature. The paid working time of both partners plays an important role in the division of the domestic work. When one partner is less involved in the labor market (inactive, unemployed, or part-time), then he or she tends to perform more of the work. In couples where the woman has no job, her share of domestic work rises by about 14 percentage points. Similarly, when the man is inactive, then the woman’s share of domestic work falls by 8.5 percentage points in 1985 and around 18 percentage points in 2009.  Over a 25-year period, gender identity becomes potentially less important, and men accept more involvement in domestic work. For women, part-time work or unemployment increases their amount of domestic work; conversely, the woman’s share of domestic work decreases if the man works part-time or is unemployed. The woman performs a smaller share of the domestic work when her bargaining power in the couple increases, but this effect is significant only for 2009.
35As expected, men’s participation in domestic work increases with their level of education. Conversely, the woman’s share decreases with her level of education. More educated women have greater bargaining power, and more educated men generally have more egalitarian values (Domínguez-Folgueras, 2012).  This is consistent with the results of other studies on this topic (Bianchi et al., 2000; Gershuny, 2000; Anxo et al., 2007). In 1985 and 1998, having children significantly increased a woman’s share of domestic work, with an impact of around 2 percentage points. For the year 2009, this effect was reversed, as the presence of a child reduced the domestic work performed by women (effect of 2 percentage points), a trend that reflects men’s greater investment in the family.  On the other hand, the coefficient of the variable for the presence of a child under age 3 is not significant for the three surveys. To explore this specific point, we have examined the correlation between this information and other model variables, finding that inactivity correlates highly with having a young child. Thus, the woman’s labor market status variable partially captures a young child’s effect on her share of domestic work.
Table 3: Estimates of the woman’s share of domestic work (OLS regression results)
Table 3: Estimates of the woman’s share of domestic work (OLS regression results)
Note: n/a = not available; n/o = no observation.
Coverage: Couples in which both members filled in diaries and at least one is active.
36The restricted definition of childcare in this analysis limits the risk of grouping tasks with different subjective values for men and women. Nevertheless, we test our results by withdrawing childcare from domestic activities. Indeed, Sullivan (2013) suggested that housework is not enjoyable for either partner, whereas childcare is a rewarding task valued by both men and women. Civil partnerships possibly share childcare more equally because PACS fathers choose to be more involved than married fathers. If so, it would not be gender norms driving more equal sharing but rather a different relationship to fatherhood, although childcare covers only routine tasks in our analysis. In 2009, the estimates for activity excluding childcare indicate that the share of domestic work of women in PACS unions is 5 percentage points lower than that of married women (significant at the 1% threshold).  Couples in civil partnerships therefore opt for a more egalitarian distribution of housework than married couples, independently of any activity related directly to their children. A cohabiting woman’s share of domestic work persists as not significantly different from that of married women.
37The marital trajectory of couples (number of unions preceding the one observed, past union type, duration of observed union, and other factors) influences their distribution of domestic work (Baxter, 2005; Nitsche and Grunow, 2016). Questions on a respondent’s marital past were included only in the 2009 survey’s Decision-making Within Couples supplementary module, which provides a subsample of 1,454 couples. Of the 976 couples in this subsample who had never been in a union before, alternative regressions on them show that the length of the relationship has a positive (although not significant) effect on the woman’s share of domestic work (Appendix Table A.4). The PACS effect remains negative and significant at the 10% threshold when considering the observed length of the relationship, as civil partnership couples are more egalitarian than married couples, while cohabiting couples are not. Similarly, no difference exists between cohabiting and married couples in the 1985 survey, although by 1998 cohabiting couples are more egalitarian than married couples (Table 3).
38Overall, the matching method’s results are consistent with those obtained using OLS (Table 4). In 1998, a woman’s share of domestic tasks would have been 5.9 points lower had she been cohabiting rather than married. In 2009, no statistically significant differences in the share of housework exist between the two types of union, whereas a married woman’s share would have been 8.6 points lower had she been living in a PACS union. Several reasons, not exclusive of one another, can explain the difference in results between both methods. It could be the way we measured the average effect, as mentioned previously. It could also be due to sample restriction in the matching. However, more interestingly, this could be interpreted as a non-biased measure of the marital status effect purged from self-selection on some observable characteristics. By using the matching method, the estimated differences in how different union types share tasks cannot be due to couples self-selecting based on their observable characteristics. These differences can result from either the marital status effect or unobserved characteristics, which include values and, more specifically, gender values.
3 – Interpreting the results: the role of gender values
39To disentangle these two effects, we contrast the results with the two hypotheses tested for each year. In 1985, a woman’s share of domestic work is unaffected by the fact of cohabiting. The observed gap between both types of couples is explained by differences in observable characteristics (for example, cohabiting partners are younger than married couples). This contradicts Hypothesis 1, which implies a lower share of domestic work performed by cohabiting versus married women, because in the early 1980s, cohabitation as an option for long-term union was still a marginal (but growing) practice and viewed as a “prelude to marriage” or a “pre-marriage test” (Villeneuve-Gokalp, 1990; Toulemon, 1996). The gender division of labor preceded formalization of the union because either the couples organized themselves in anticipation of marriage or they simply wound up marrying after having established specialization. Thus, no significant differences in the gender division of labor were observed between the two union types.
Table 4: Estimated differences in the woman’s share of domestic work based on marital status
Table 4: Estimated differences in the woman’s share of domestic work based on marital status*** p < 1%. ** p < 5%. * p < 10%.
Note: Values are expressed in percentage points. Standard errors (in parentheses) for the matching pairs model are obtained by bootstrapping.
Coverage: Couples in which both members filled in diaries and at least one is active. Differences between the two methods in numbers of couples are due to the matching method’s sample restriction.
40The 1998 results indicate that a married woman’s domestic work would have been 5.9 percentage points lower had she been cohabiting. This result is consistent with Hypothesis 1. In the late 1990s, cohabitation was spreading as a socially accepted alternative to marriage. Couples stabilized their relationships outside marriage, and the arrival of children did not necessarily lead to formalizing their unions. The gender division of labor in these couples is more egalitarian than in married couples, and not because of their socioeconomic profiles. In accordance with Hypothesis 1, the marital status effect can explain this result, although the causal direction cannot be established. Marriage reinforces the degree of the couples’ gender division of labor, and couples anticipating such specialization opt for marriage. Unobserved heterogeneity may also explain this gap. As Hypothesis 2 specifically states, the gender values effect is likely to influence one’s choice of union type. Cohabitation in the 1990s may have attracted people seeking a type of union distinct from the conservative norms of marriage and which reflected egalitarian values.
41In 2009, no significant difference is observed between married and cohabiting couples. Thus, unlike what was observed for 1998, cohabiting couples were not more egalitarian than married couples in the distribution of domestic work. In contrast, the woman’s share of domestic work in a PACS couple was significantly lower (about 8.6 percentage points) than the share observed in married couples. This result contradicts Hypothesis 1, whereby the marital status effect predicts that married couples are the least egalitarian, followed by civil partners, then cohabiting couples as the most egalitarian. The gender values effect offers another explanation in accordance with Hypothesis 2, namely that the values of cohabiting couples are less egalitarian than those of civil partners.
42This hypothesis is further reinforced by the International Social Survey Programme: Family and Changing Gender Roles, III (2002) and IV (2012), a survey that provides information on the evolution of individuals’ gender values according to their marital status.  We have selected a sample with the same characteristics as those used in the time-use survey analysis.  The 2002 survey contains no information on PACS status and therefore identifies those couples as unmarried; couples in civil partnerships are contained in the cohabiting category. The lack of information in this database is not an issue, as few people were opting for the PACS in 2002. 
43We use the respondents’ indicated level of agreement with the following statement as a proxy for their gender values: ‘‘A man’s job is to earn money; a woman’s job is to look after the home and family.” Between 2002 and 2012, the conservative values of married couples declined, as the proportion of the respondents who disagreed or strongly disagreed with this statement increased from 75% to 82% (due mainly to an increase in those who strongly disagreed), and the proportion who agreed or strongly agreed dropped from 13% to 7%. In contrast, these proportions remained stable at 87% for cohabiting respondents during that decade (Figure 1). The 2012 proportion of respondents living in a PACS union who rejected this statement is higher, especially those who strongly disagreed (82% compared to 67% of cohabiting persons and 62% of married respondents).
44In addition to these descriptive statistics, we have estimated a logistic model with the following dependent variable: the respondent strongly disagrees with the statement, relative to all other degrees of agreement (Appendix Table A.5). The results indicate that, all else being equal in 2012, PACS respondents more often strongly disagree with the conservative statement than married respondents, whereas no statistically significant difference occurs for cohabiting respondents.  This result is corroborated by the sociological literature on the PACS. The decision to opt for a civil union is associated with a value system based on an egalitarian view of male and female roles in society and in the family (Rault and Letrait, 2010). Meanwhile, all else being equal, couples with the most egalitarian values are attracted to civil partnerships, whereas they were more attracted to cohabitation before this institutionalized form of union was implemented.
Figure 1: Percentage in France who say they _____ with the statement: “A man’s job is to earn money; a women’s job is to look after the home and the family,” by marital status
Figure 1: Percentage in France who say they _____ with the statement: “A man’s job is to earn money; a women’s job is to look after the home and the family,” by marital statusNote: Sample sizes: in 2002, 482 married and 155 cohabiting couples (including PACS, despite its implementation in 1999); in 2012, 630 married, 202 cohabiting, and 88 PACS couples.
Coverage: Individuals aged between 25 and 55 years, living in a couple, with at least one active partner, and neither is retired, a student, in training, or disabled.
45This result implies that civil union attracts individuals holding the most egalitarian values, and thus the woman’s share of domestic labor is lower in PACS unions. This confirms Hypothesis 2: in 2009, after having controlled for the socioeconomic characteristics effect, the gender values effect seems to more than offset the marital status effect.
46In 1985 and in 1998, married women performed a greater proportion of domestic work than cohabiting women. While the 1985 gap is explained by differences in socioeconomic characteristics, the reason in the late 1990s is that cohabitation became less of a transition and turned into a widespread and socially accepted alternative to marriage (Toulemon, 1996; Prioux, 2009). In 2009, the average domestic work performed by women is about the same whether cohabiting or married (72% and 73.5%), but it is significantly lower for women in civil partnerships (65.1%), a gap not due to differences in socioeconomic characteristics. Two non-exclusive interpretations are possible. The marital status effect can explain the 1998 gap between cohabiting and married couples, but it does not explain the 2009 results showing more equal organization among PACS than cohabiting couples and comparable organization among cohabiting and married couples. A second interpretation is self-selection of couples based on their gender values, the gender values effect. Individuals who opt for a PACS adhere to more gender egalitarian values than those who cohabit or are married. Therefore, our findings can be interpreted as follows: civil partnerships in 2009 attracted the most egalitarian couples in terms of their gender values, while such couples opted for cohabitation before the introduction of the PACS.
47This paper enriches the literature on the sharing of domestic tasks based on marital status. The case of France shows convergence in the proportion of domestic tasks performed by married and cohabiting women. In 2009, after controlling for the differences in observable characteristics of these two types of couples, no significant difference remains. Couples in civil partnerships were more egalitarian than other couples in the way they organized domestic chores—a result that reflects two phenomena. The first is the trend toward women reducing time spent on domestic work, which is well documented in the literature. Thus, the extent of married couples’ gender division of labor is reduced, and it gradually converges to the level observed in cohabiting couples. The second trend is due to the late 1990s introduction of the PACS civil partnership, which attracted the most egalitarian couples. Indeed, estimates indicate that this difference is not the result of couples self-selecting according to their socioeconomic characteristics but is instead more likely to be due to a gender values effect: couples choosing the PACS hold more egalitarian values than those opting for the other two forms of union.
48We have focused on the woman’s share of domestic work in order to analyze the degree of specialization in couples. Regarding the definition of gender equality, the total amount of work performed by each partner may be another relevant indicator. Cohabiting women perform the largest share of total work, around 52% (paid and domestic work) compared to both married and PACS women, whose share of total work is around 50%. Further, both married and PACS women perform a larger share of domestic than paid work, but the imbalance is largest among married women. This result raises the issue of gender equality from a long-term perspective, as domestic work grants no direct social rights and cohabitation grants no derivative social rights (in terms of pensions) or rights to alimony (in case of separation), by which cohabiting women might be exposed to greater precariousness. The gap between gendered behaviors in the distribution of domestic work is especially problematic for cohabiting women, as suggested by Martin and Théry (2001), who found that cohabiting women performed the same share of domestic work as married women but do not benefit from specific protections or compensation for it. In light of changes in behavior and marital choices over recent decades, the French social welfare state has still not been overhauled and lies between two models. The first centers on marriage and the male breadwinner model, which is associated with compensating the cost of specialization through protections and transfers. This encourages at least in part a gendered division of roles and attracts couples with more conservative values, although it benefits the wife by including safeguards in case of separation. The second model centers on cohabitation, which does not address the observed persistence of gendered division of labor in the family. The PACS only partially responds in terms of regulation, and its effects on reducing gender inequality are limited because this form of union attracts couples who already hold egalitarian values.
49Taking a broader view beyond the specific case of France, the results show an interrelationship between two major effects that explain the differences in couples’ gender division of labor according to their union type: the marital status effect and the gender values effect. Due to changes in the legal framework introduced through France’s PACS union, which is less regulated than marriage but more so than cohabitation, a shift has occurred in the behavior of couples who hold the most egalitarian values. Similarly, policies that regulate unions in Europe have changed recently as the result of an increasing trend toward cohabitation, with some countries reinforcing the rights of cohabiting couples, while others have introduced civil unions like those in France. More research is needed to evaluate the impact of such institutional changes on gender equality.
AcknowledgmentsWe thank Francine Deutsch, Marta Domínguez-Folgueras, Sophie Ponthieux, Amandine Schreiber, and the anonymous referees, all of whom provided relevant comments and suggestions.
Estimating the wage equation
50Our estimation of the wage equations uses Heckman’s method (1979) to take into account the selection effect on the labor market by simultaneously estimating the equations for wage (1) and for participation in the labor market (Heckman, 1979):
52where w is the hourly wage, the index i designates the individual, and xi1 is the vector of the control variables: the obtained diploma (less than high school, high school, above high school); union type (marriage, civil partnership, cohabitation); residence in a rural area; and potential experience  and its square. As we use potential experience, this variable probably overestimates actual experience—particularly for women, due to the birth of children likely affecting career continuity. Therefore, we account for career breaks in estimating the woman’s wage by multiplying both potential experience and its square by the number of children in the household. β1 is the vector of the corresponding coefficients, and εi1 is the error term.
53For the selection equation (2), the latent variable si* determines selection (employment) in the labor market but is not observed; so we use an observable variable that is defined as follows:
55Hence, the probability of working versus the fact of being unemployed or inactive is Pr (si = 1|xi2) = Pr (si* > 0); xi2 is the vector of control variables containing the variables used in the wage equation; and xi1 and the exclusion restriction variables zi indicate the existence of non-labor income (interest, savings income, dividends). For women, variables are introduced for the presence of children under age 3, ages 3 to 6, and for their partner’s education diploma. β2 defines the vector of corresponding coefficients, and εi2 is the error term. The εi and ε2 error terms of the two equations follow a normal joint distribution, with zero mean and a variance–covariance matrix ∑. For identification purposes, the variance of ε2 is normalized to 1. The reference person is a married individual, with less than a baccalauréat (high school diploma) and living in an urban area. The results are interpreted in relation to this reference.
Table A.4: Estimates of a woman’s share of domestic work (results of OLS estimations)
Table A.4: Estimates of a woman’s share of domestic work (results of OLS estimations)* p < 10%.
Coverage: Couples in which both members filled in diaries and at least one is active.
Table A.5: Individual opinions in France on men’s and women’s roles in a couple, by union status (2002 and 2012), results of logit estimate(a)(a) The logit estimate with the probability of having answered “strongly disagree” to the following statement, “A man’s job is to earn money; a woman’s job is to look after the home and family,” as a dependent variable.
*** p < 1%. ** p < 5%. * p < 10%.
Note: n/a = not available. The 2002 cohabitation category contains PACS (information undocumented that year).
Coverage: Individuals aged between 25 and 55 years, living in a couple, with at least one active partner, and neither is retired, a student, in training, or disabled.
The obligations toward children are the same for each union type.
The days reported by both partners are not chosen by the respondents but instead randomized and fixed by the investigators.
Only the 2009 time-use survey has a variable indicating whether partners lived together for more than 1 year. While this variable is available for all households in the general questionnaire, it was not well filled in. Therefore, not retaining the couples with this missing value would have reduced the sample size by 1,108 couples. Another question specifies the exact length of union, but it covers a subsample of couples in only the supplemental Decision-making Within Couples module.
For couples reporting that total time spent on domestic work is null, we excluded 28 and 20 couples in, respectively, the 2009 and 1998 surveys. We also removed 1 and 46 couples with missing information about household income in, respectively, the 2009 and 1998 surveys, and another couple reporting inconsistent diary days in the 2009 survey.
Maintenance of heating and water, other household maintenance work, and non-professional moving.
The bargaining power of members in the couple can be influenced by other parameters, such as the state of the labour market and the legal framework. The latter partly determines the financial terms governing a couple’s breakup and thus alters each partner’s bargaining power.
Two variants were tested: one that defines bargaining power as the ratio between the woman’s hourly wage and the sum of the couple’s hourly wages; and another using a dummy variable that equals 1 if the man’s wage is more than twice that of his partner. The results of the analysis were not profoundly altered by integrating these two alternatives.
This restriction is needed to find individuals with similar propensity scores. In the tails of distributions, it is difficult to find married couples with low propensity scores or, inversely, non-married couples with high propensity scores. If a given interval of the propensity score includes only married or non-married couples, we are unable to match them with another couple.
To respect both assumptions, we drop about 25% and 21%, respectively, when matching married and civil partners in 2009 (and 27% and 34% when matching married and cohabiting couples). In 1998, we drop 29% of married couples and about half of cohabiting couples, and in 1985 12% and 25% of married and cohabiting couples.
See Heckman et al. (1997), Heckman et al. (1998), Greene (2002), and Imbens and Woolridge (2009). Robustness tests based on other matching methods (nearest neighbor, radius, and stratification) were also carried out but not presented here.
In 1998, only one couple shared this status.
For the 2009 survey, the woman’s education has a weaker and less significant effect than for the other two surveys. However, the effect of the bargaining power variable (calculated from the relative wages of both partners) is significant in 2009 but not in 1998 (and not included in 1985). The partner’s education and bargaining power variables capture some of the same effect.
Estimates that exclude activities related directly to children indicate that this negative effect persists (impact of about 2 percentage points, an effect significant at the 5% threshold).
For results, see online supplementary material at: https://www.cairn.info/docs/Kandil_Perivier_Supplementary_Table.xlsx
ISSP 2002 is the first Family and Changing Gender Roles survey that includes France.
Individuals aged between 25 and 55 years, living in a couple, with at least one active partner, and neither is retired, a student, in training, or disabled.
In 2002 (resp. 2012), 21,683 PACS contracts (153,715) were signed compared to 286,169 marriages (245,930) (INSEE, 2017).
Being a female respondent increases the probability of strongly disagreeing, and more educated respondents more often disagree with the statement. Having children decreases the probability of strongly disagreeing, which can be explained by gender roles being reinforced by parenthood.
Difference between current age and age upon completion of studies.