Écarts de salaires entre femmes et hommes : effets de sélection et décompositions

Persistent gender disparities in labor market outcomes continue to attract significant aca- demic and policy attention. In France, Insee publishes annual indicators of gender gaps in annual wages, full-time equivalent wages (EQTP), and wages for comparable jobs in the private sector Gerardin (2025). These measures, based on observed wages among em- ployees, show a long-term decline in gender wage differentials. These indicators rely however on the population of wage earners, a selected subset of the potential labor force. This selection bias may distort the interpretation of observed wage gaps. A decline in average observed wage differences does not necessarily imply a reduc- tion in the underlying differences in offered wages between women and men. For instance, job losses concentrated among low-paid male workers can artificially increase the observed gap, while the entry of more highly educated women into the labor market tends to reduce it. These mechanisms illustrate the importance of compositional effects and the interac- tion between economic conditions, demographic changes, and social norms in interpreting observed wage gaps. The economic literature proposes several methods to correct for this selection bias. The Heckman model Heckman (1979) remains the benchmark: it jointly estimates the probabi- lity of employment and conditional wages, using an instrument that affects employment but not wages directly (such as the presence of young children in the household). Other ap- proaches restrict the analysis to subpopulations less affected by selection, though such samples may not be representative of the employed population. Blau et al. (2024) re- cently proposed a method to correct for selection bias in comparisons of mean and median wages by imputing wages for the non-employed using data from the Panel Study of Income Dynamics (PSID). Their findings suggest that, once selection is accounted for, the decline in gender wage gaps is even larger than that observed among the employed population. This paper applies the methodology of Blau et al. (2024) to French data from the panel tous Salariés(PTS) and the Échantillon Démographique Permanent (EDP) between 1988 and 2022. It also develops a complementary approach inspired by Chiquiar and Hanson (2005), based on the reweighting method of DiNardo, Fortin and Lemieux DiNardo et al. (1996), which allows for the estimation of selection-corrected wage gaps conditional on observable characteristics. The study examines the impact of selection on both average wage gaps and their distribution, using unconditional quantile regressions Firpo et al. (2009). Results show that women and men are differently selected into wage employment, in ways that are partly but not fully explained by observable characteristics. As a result, measured gender wage gaps are biased by selection effects whose magnitude varies over time and across methods. Once corrected for selection, gender wage gaps in annual wages exhibit a substantial and consistent decline between 1988 and 2022, regardless of the correction method applied.