Worker-firm matching and the parenthood pay gap: Evidence from linked employer-employee data

Abstract

The parenthood pay gap is not fully explained by human capital depreciation and unobserved heterogeneity. Endogenous worker-firm matching could also account for such wage differences. This hypothesis is tested thanks to linked employer-employee data on the French private sector between 1995 and 2011. Childbirth penalties are estimated for women and for men from hourly wage equations including firm- and worker-fixed effects on top of usual measures of human capital. Though worker-firm matching explains none of the motherhood wage penalty, it plays a role in the case of fathers who do not experience any wage loss after childbirth, but do not enjoy any premium either; there is evidence of an erosion of this premium since the end of the 1990s. In a counterfactual where women do not incur any penalty after childbirth, the gender gap still amounts to 2/3 of the one that currently prevails.

Appendix A : data

1.1 Cleaning

I proceed to some cleaning of the DADS panel. First, I recode the age variable as the difference between the current year and the year of birth. The former age variable exhibits some errors due to scan problems before the numerical DADS was introduced. Second, département codes are sometimes one-digit instead of being two-digit; other département or region codes are missing. In that case, I rely on other observations in the whole database in order to recover that information.

In the EDP database, I eliminate observations for which days or months of marriage or birth are equal either to 00 or 99, as well as observations for which the year of birth is 0000.

1.2 Selection

I restrict my attention to individuals born on October of even-numbered years: careers of individuals born on October of odd-numbered years is unknown before 2002. The most important selection is dictated by the necessity of measuring experience properly (see infra): I focus on individuals who entered the panel after 1995, which leaves me with 46,280 individuals (338,879 observations at the individual-year level and 489,852 observations at the individual-firm-year level). We eliminate further individuals whose net annual earnings are missing or less than 10 euros in 2011 terms. I also restrict my sample to individuals aged 16 to 65, working at least 10 h a year, whose job duration is consistent with worked hours (for instance, the ratio of the latter over the former must be less than 24), which leaves me with 45,483 individuals (317,476 individual-year observations). After trimming observations with a hourly wage that is smaller than 80 % of the legal minimum wage,²⁰ and after dropping years 2003 to 2005, my estimation sample is composed of 41,531 individuals (212,189 individual-year observations and 301,079 individual-firm-year observations). Among those individuals, 19,932 are women while 21,599 are men. Last but not least, I define time-varying variables for marriage (parenthood) as the fact of being married (experiencing a childbirth) before time t for individual i.

In this sample, 62 % (69.5 %) of women (men) do not have any child yet, against 36.3 % (43.9 %) in the DADS-EDP database, the difference being mainly due to the youth of individuals in the estimation sample, and being at the source of the main empirical limit of the current analysis. Among the women in the working sample,²¹ 29.2 % leave after first birth; among the remaining mothers, 34.5 % leave after second birth, this rate being roughly the same for subsequent births. The estimation sample does not seem gender-biased with respect to the DADS-EDP database: for instance, 54.6 % of non-parents are women in our sample, against 57.5 % in the DADS-EDP; about 47 % of parents of either one child or two children are women in our sample, against about 50 % in the DADS-EDP. Finally, individuals who work in small firms are relatively less likely in the estimation sample: in 2011, 42 % of individuals present in the DADS-EDP and working in the private sector belong to a firm with less than 20 employees, against 33 % in the estimation sample.

1.3 Definition of main employment

Aggregating data at the individual-year level requires to define for each individual her main employment in the year. I select the employment with (in successive order) the highest number of working days, the highest wage, a full-time position (if any), and the highest number of worked hours. If there are still ties after applying those criteria, I choose the job with the last SIREN in lexicographical order—to keep the code deterministic. Finally, if several observations resisted to the last iteration, I would consider them as authentic doubles and eliminate them—which does not happen here. We define job characteristics (private/public sector, industry, geographic location, firm’s size, full-time/part-time, but also seniority) at the individual-year level as being related to the main employment. I sum wages and working hours, and define working days as the minimum of 360 (the annual number of working days in the DADS by convention) and the sum of working days over the whole year.

1.4 Computation of experience

Mincer (1958) demonstrated how important it is to control properly for experience and seniority in wage equations. I devote much attention to compute these variables as precisely as possible. Seniority is defined as the difference between the current date and the first appearance of a pair (individual, firm). Thanks to the comprehensive nature of the DADS panel, it is possible to reconstitute the whole salaried career of an individual, hence to compute his experience from observed working times. Experience will thus be defined as closely as possible as the amount of salaried time spent on the labor market. Since worked hours have been available from 1995 onwards only, I restrict my attention to individuals who entered the panel after 1995. I consider that workers increase their full-time/part-time experience variable every year by their share of working hours expressed in full-time units (FTU).