Housework share between partners: Experimental evidence on gender-speci ﬁ c preferences

This paper uses a novel vignette-based experimental design to investigate the reasons underlying the gendered division of housework. We are particularly interested in the role of gender-speci ﬁ c preferences: are there differences in the utility that men and women derive from housework, and might these be responsible for the fact that women continue to do more housework than men? It is dif ﬁ cult to address these questions with conven- tional survey data, because of inherent problems with endogeneity and ex-post ration-alization; our experimental design circumvents these problems. We ﬁ nd remarkably little evidence of any systematic gender differences in preferences, and a general inclination towards an equal distribution of housework; this suggests that the reasons for the gendered division of housework do not derive from gender differences in preferences, and must lie elsewhere.


Introduction
This paper reports the results of a novel experiment which investigates whether men and women have different preferences over the allocation of housework within couples, and asks whether such differences may provide an explanation for the fact that women do more housework than men.
Gender disparities in the allocation of housework have attracted interest from across the social sciences (Becker, 1965;Oakley, 1974;Hakim, 1996Hakim, , 2000Akerlof and Kranton, 2000;Baker and Jacobsen, 2007;Lachance-Grzela and Bouchard, 2010;Stratton, 2012;Thompson, 1991;and many others). This issue is becoming ever more relevant in the contemporary Western context: women are now educated at least as well as men, the gender gap in labor market participation and earnings continues to narrow, but women continue to do much more housework than their male partners (Brines, 1993(Brines, , 1994Alvarez and Miles, 2003;Bitman et al., 2003;Washbrook, 2007;Kan et al., 2011).
A range of theories have been advanced to explain what Hochschild (1989) refers to as the "stalled revolution." These theories will be discussed in Section 2; the debate essentially boils down to whether women do more housework because their capabilities, characteristics or standards are systematically different from those of men; or because they are responding to pressure arising from power dynamics within the partner relationship or from society at large; or because men's and women's preferences over housework differ systematically, with women liking housework more (or disliking it less) than men.
It is this hypothesis of systematically different gender preferences which this paper sets out to explore. There currently exists little empirical evidence on the gendered nature of preferences over the allocation of time, largely because of the difficulty of using survey data to obtain meaningful estimates of the relevant preferences. Although several household surveys carry questions on the allocation of paid work and housework between partners, and on individuals' satisfaction with these arrangements, the fact that these data relate only to people's actual arrangements (rather than what people would experience under alternative arrangements) leads to three main problems.
First, some distributions of housework and paid work are rarely observed in surveys (for example, surveys typically contain very few households where the woman does more paid work, earns more, and does less housework, than her male partner). This means that it is not possible to estimate preferences over the entire range of potential distributions of housework and paid work, because there are simply too few observations in some parts of the full space.
Second, people's satisfaction with the situation in which they actually find themselves may be affected by a process of expost rationalization, and may be a poor reflection of what their preferences would be, given a range of possible alternatives. Pedulla and Th ebaud (2015) show that preferences are sensitive to the institutional context, with people's preferences in the presence of institutional constraints such as a gender pay gap differing systematically from their "true" preferences in the absence of such constraints.
Third, people's hours of domestic and market work, as well as related factors such as their wages, are largely determined by their own characteristics and those of their partners e and these may be the same characteristics which drive their preferences over housework arrangements. Empirical analyses are thus subject to problems of endogeneity, meaning that it is difficult to draw causal inferences from survey data as to whether women's greater contribution to housework arises as the result of gender-specific preferences, or as the result of a process of specialization triggered by partners' differences in productivity in the market and in the home. 1 In many contexts where behavior is endogenously determined, a randomized experiment would address the problem. However, the difficulties in carrying out a real-world randomized experiment in this context are obvious and insurmountable: it would not be possible to randomly allocate paid work, earnings or housework among a sample of couples.
An alternative empirical approach is the use of laboratory, field or survey experimental designs (see Croson and Gneezy, 2009;Bertrand, 2011). Experimental studies on gender identity or gender-specific preferences include the laboratory experiment of Cadsby et al. (2013) investigating the effect of gender identity on attitudes to risk and competition; the experiment of G€ orges (2015) testing gender specific patterns in couples' work specialization decisions; the factorial survey experiment adopted by Abraham et al. (2010) testing the effect of gender role attitudes on migration decisions within dualearner partners; and the survey-experiments run by Pedulla and Th ebaud (2015) examining the extent to which institutional constraints, such as workplace policies, influence young, unmarried men's and women's preferences for their future workfamily arrangements.
However, to best of our knowledge, no experimental research has yet been conducted to assess gender differences in preferences over housework arrangements. In our experiment people are invited to imagine themselves and their partners in different hypothetical domestic scenarios ("vignettes"), and to tell us how satisfied they would be with each set of arrangements. These hypothetical scenarios are generated using a multi-factorial experimental survey design (Auspurg and Hinz, 2015) 2 which, as well as varying the distribution of housework between scenarios, also varies a range of other factors: the share of paid work done by each partner; the level of respondents' own earnings and their partners' earnings; the presence and age of children; and whether the household employs paid help (i.e. whether there is some market substitution of domestic work). Factorial survey methods are a proven method in the study of intragroup differences in judgement rules, such as cross-gender differences (Jasso, 1994) and have been shown to be a useful tool for researching how individuals' preferences would change over a range of scenarios (e.g., Shlay, 2010). In our design vignettes are randomly allocated between households, with male and female members of couples receiving sets of vignettes which are identical but "reflected" (that is, the same housework and paid work arrangements, but with the roles of the male and female partners exchanged). This design allows us to assess directly whether preferences over work arrangements differ systematically between men and women, free from the problems of endogeneity and post-hoc rationalization. A finding of systematic differences in preferences would lead us to conclude that gender identity, i.e. the internalization of social gender norms, is a factor contributing to the unequal distribution of housework; conversely, a finding of few or no differences would lead us to conclude that the gendered division of housework arises not because of gendered preferences, but must be due to some other factor: women's comparative advantage in domestic activities, as 1 Hwang et al. (1998)