The Impact of Social Interventions: Nonparametric Identification from Choice-Based Samples

Abstract

This paper presents conditions for the identification of the AVERAGE effect of training CONSTRAINED to the population of trainees. All of the identification conditions we consider are based on a variety of conditional independence assumptions. More specifically, they assume the participation status and post-training latent earnings are conditionally (mean) independent. The identification conditions differ according to the conditioning set they use. Nowhere are any functional-form imposed. In either words, a nonparametric approach has been taken. Moreover, the conditions are robust to choice-based sampling.