Intra-family distribution of paid-work time

Abstract

This article analyses the intra-family distribution of paid-work time in five European countries (France, Germany, Italy, Spain and the UK). To that end, we formulate a collective model, which allows us to characterize the efficient labour supply decisions of each spouse. This two-equation model is then simultaneously estimated by using national panel data drawn from the European Community Household Panel-ECHP (1994–2001). Empirical results clearly show that, in all sample countries, the labour supply of wives is affected by own wages, household and own nonlabour incomes and the number of children, whereas evidence for husbands differs across countries.

Acknowledgements

This article was partially written while José Alberto Molina was Visiting Fellow at the Department of Economics of the University of Warwick (UK), to which he would like to express his thanks for the hospitality and facilities provided. An earlier version of this paper has been presented at the thirty-first Spanish Economic Analysis Meeting-2006 (Oviedo, Spain), with all the comments made by the participants being highly appreciated. Finally, the authors also thanks for the financial support provided by the Spanish Government (Grant SEC2005-06522), as well as by the Fundación BBVA. The usual disclaimer applies.

Notes

¹Samuelson (1956) and Becker (1974a, b) represent the early attempts to account for the fact that households may consist of different individuals with their own preferences. However, in both cases the authors finally accepted the traditional approach: in the first case, through an aggregate utility function, which is achieved by consensus among the individuals; in the second, by assuming the utility function of a benevolent head of the family, who takes into account the preferences of all household members.

²Other empirical restrictions, strongly rejected in the literature, include the income pooling hypothesis and the symmetry of the Slutsky matrix (see, among others, Kooreman and Kapteyn, 1987; Schultz, 1990; Thomas, 1990; Fortin and Lacroix, 1997; Tiefenthaler, 1999). With respect to the former, this implies that individual nonlabour incomes of the household members are pooled in a single household nonlabour income, which, in turn, implies that the source of this exogenous income plays no role in the household's distribution. As regards the latter, this requires that marginal compensated price changes of two individuals in a household have the same effect on each other's goods demands.

³See Vermeulen (2002) for an excellent recent survey on the collective household approach.

⁴Habitually, only the total household consumption is observed (and not the individual consumptions). We then talk of the Hicksian aggregate commodity.

⁵According to this view, the household optimum allocation can vary as a consequence of a change in the nonlabour income both, directly, via the usual income effect, and indirectly through a shift in the bargaining power. This means that the income pooling hypothesis no longer needs to be true.

⁶Empirical evidence on household labour supply for a number of countries has been found in, for example, Kawaguchi, 1994; Pradhan and Van Soest, 1997; Seaton, 1997; García and Molina, 1998; Barmby and Smith, 2001; Flood et al., 2004; Doiron and Guyonne, 2005; and Iyigun and Walsh, 2007. Additionally, particular studies on full-time/part-time work of spouses are, for example, Yamada and Yamada, 1987; Powel, 1998; and Connelly and Kimmel, 2003.

⁷Whenever α₄/β₄ is different from α₅/β₅.

⁸This supposes, we are not considering the possibility of nonparticipation in the labour market and then selectivity bias may arise. Rather, we focus on the case of the interior solutions. The main reasons why we have opted to ignore this bias in our analysis are the difficulties in both considering corner solutions in the collective model (Blundell et al., 2002; Donni, 2003; and Bloemen, 2004) and, similarly, dealing with selectivity bias in the case of panel data estimation (Wooldridge, 2002; Kyriazidou, 1997). The consideration of all these possibilities is beyond the scope of this article and is left for future research.

⁹As stated earlier, different distribution factors have been used in applied work. Although, we have tried a number of these, we finally chose this one given the inadequacy or nonsignificance of other possibilities. Thus, divorce laws are common within a country, the low regional disaggregation in the database used makes the construction of appropriate sex ratios more difficult, while differences in ages, in educational levels or in experiences between spouse are not applicable with the fixed effects estimation, which was ultimately selected, since they are time-invariant. Finally, this is a potentially continuous function and, thus, does not give rise to any incompatibility with the required assumption of continuously differentiability of the sharing rule, which usually fails elsewhere (Chiappori et al., 2002; Clark et al., 2002; Crespo, 2005).

¹⁰Since income variables refer to the period prior to the interview and the remaining data refer to the current period, the last year is lost for estimation. Thus, the time-series dimension reduces to 7 years.

¹¹These include experience and squared experience, three dummy variables indicating educational attainment, nine occupational dummies, seven firm size dummies, one dummy for the public sector and one for fixed-term contracts. Results of these estimations are not shown but are available from the authors upon request.

¹²All income variables have been deflated by the annual mean of the Consumer Price Index (CPI) base 1992 and are expressed in euros. The ECHP provides the nonlabour income of each of the spouses and the total household income. We have computed household nonlabour income as the difference between the total household income minus the total labour income.

¹³Initially, we also introduced the age and squared age into the equation. However, these were always nonsignificant and were eliminated from the final estimation.

¹⁴Considering only the couples that offer information about hours caring for children or adults severely reduces the samples.

¹⁵Since this is the specification finally chosen, a time-varying distribution factor is selected to avoid the elimination of constant regressors by mean-differencing, as well as the collinearity arising from variables whose differences are constant over time, as, for example, gender differences in age or in experience, which initially were also considered as possible alternative distribution factors.

¹⁶Note that the restrictions tested are if α₃/β₃ = α₅/β₅ for the semilog specification, or γ₃/δ₃ = γ₇/δ₇, for the quadratic.

¹⁷A recent study by Couprie (2003) shows that today, in the UK, the younger generation does not appear to have experienced a change of attitude regarding gender equality in the family.