Part-Time Work and Women’s Careers: a Decomposition of the Gender Promotion Gap

Abstract

This paper studies the effect of working hours on vertical sex segregation using Belgian micro-data on promotions. Using Yun decompositions we find that more than 40% of the promotion gaps between men and women can be explained by gender differences in contract hours, overtime hours and occasional late work. The fact that women often work in sectors that offer less promotion possibilities is another important factor. The presence of children strongly affects the promotion chances of female employees, but not those of the male employees in our sample. This evidence supports theories that relate the availability of part-time work to the degree of vertical segregation in countries.

Appendix 1: Heckman Correction for Sample Selection Bias

We correct for potential bias resulting from non-random selection into employment using Heckman's (1979) two-step estimator. Assume a latent variable U _i that measures the utility of entering the labour market, so that U _i  = γW _i  + u _i and individuals decide to enter the labor market when U _i  > 0. Heckman showed that the expected outcome among the selected is E[Y _i |U _i  > 0] = βX _i  + β _λ λ _i  + ν _i , where λ _i is the ‘Inverse Mill’s Ratio’ (IMR). Table 5 presents the results of the first step probit selection equation for a binary dependent variable indicating employment. In order to reduce collinearity in the outcome equation (step 2), two additional variables – the number of children and whether a person lives together with a partner or spouse in the household – are included in the selection equation. All the significant interaction terms between these two instruments and the other independent variables were added as well in order to maximize the fit of the selection equation. In the literature on the returns to education, the number of children and the presence of a partner or spouse are often used as instruments for Heckman correction in wage equations (for example Arrazola and de Hevia 2008; Ben-Halima et al. 2014; Pastore and Verashchagina 2006; Ordine and Rose 2011). The identifying assumption of the Heckman strategy to correct for selection bias is that we assume both instruments to influence the outcome (promotions) only indirectly via their effect on selection (employment). This assumption requires there to be a causal effect in the first step, which is supported by the fact that the instruments are strongly related to the employment decision in Table 5. The assumption also requires that this is the only channel through which the instruments affect outcomes. These exclusion restrictions are difficult to test, but we can provide evidence using a strategy described by Angrist and Pischke (2015). They reason that in a subsample where the first-stage effects are small, the exclusion restriction implies that reduced-form estimates should also be small. If these estimates are not small then the exclusion restriction is falsified. We adopt this strategy by inspecting more closely a subsample of highly committed persons within our full sample. It can be expected that highly committed persons will always decide to work, regardless of their family characteristics. If in this subsample of highly committed persons the instruments do not affect the selection decision, then any effect of the instruments on the outcome (a reduced-form equation) would signal violations of the exclusion restrictions. We re-estimated the promotion functions from Table 2 by adding the two instruments and by restricting the sample to highly committed persons. Although the sample sizes in this exercise remain substantial (N = 1832 in the authority promotion function and N = 2165 in the management promotion function), none of the instruments show significant effects in the promotion functions (each of the four p-values is greater than 60%). These results support the exclusion restrictions. Moreover, and as should be expected, the IMR”s are negatively correlated to both authority promotions (r = −.04, p < .001) and management promotions (r = −.04, p < .001) and the estimated returns to education after Heckman correction are greater than the OLS estimates.

Table 5 Probit model for selection into employment

Table 6 Means of the explanatory variables in the decomposition, by children and part-time