Seminar

Nonseparable sample selection models

Blaise Melly (Brown University)

June 21, 2011, 15:30–17:00

Toulouse

Room Amphi S

Econometrics Seminar

Abstract

This paper analyzes the nonparametric identification of nonseparable models in the presence of sample selection. Most existing sample selection models severely restrict the effect heterogeneity of the observed variables on the outcome distribution. In contrast, we allow for essential heterogeneity by assuming the outcome to be a nonparametric and nonseparable function of the explanatory variables and the disturbance term. We impose a quantile restriction on the disturbance term to derive sharp bounds on the functions of interest such as the partial effects. The identified set shrinks to a single point if separability holds or if some observations are observed with probability one. We also provide a simple estimator for the identi.ed set in the linear quantile regression model and apply it to female wage data.