The male-female wage gap in France: an analysis using non-parametric methods

Using the French data set “2002 Employment Survey�, this paper aims to shed light on the nature of the gender wage differential in France, exploring the added-value of a non-parametric analysis over previous knowledge based on parametric estimates. The parametric decomposition of the gender wage gap is run following various assumptions on the “reference� wage structure and on sample selection. It is applied to different sub-populations (wage earners, full time workers only, private or public sector workers only). Furthermore, we analyze the contribution of each productivity characteristic in explaining the wage gap. One can highlight two main limitations in this standard Oaxaca’s decomposition methodology. First, it is based on parametric assumptions about the form of the conditional expected earnings function, which can induce specification errors. Secondly, the gender earnings gap is measured at the mean, thereby ignoring the differences in the form of the entire earnings distribution. Hence, we implement a non-parametric procedure to analyze the influence of workers’ productive characteristics on gender differences in the distribution of wages. The main estimation problem is thus to construct a counterfactual wage density that would prevail for men (women) if they had women (men) ’s distribution of characteristics. This counterfactual density can be estimated by applying standard nonparametric Kernel density estimation techniques to a re-weighted sample of men (women). We are then able to bring out the role of male-female variations in each productive characteristic in explaining the differences in densities.