Accounting for Selection Bias in RIF Decomposition

Abstract

This paper suggests an extension of the RIF Decomposition for taking into account the average selectivity bias. Wages regressions are estimated using Reimers recentered influence functions in order to obtain unconditional quantiles partial effects controlled for sample selection. The extension used is compatible with the reweighting strategy needed for estimating a counterfactual. I illustrate the methodology by examining the effect of covariates on the French gender wage gap from full-time workers. I show that gender disparities are larger along the wage distribution when accounting for full-time selection. The wage structure is highly underestimated without selection control and drives the extension results. Then, occupational status follows by experience, workplace, industry and education are the main covariates affecting the gender wage gap.

Keywords: Gender wage gap, Decomposition, Sample Selection, RIF Regression

JEL Classification: C18, J31, J7