关键词: 处置效应, 后定变量, 倾向得分, LASSO

Abstract: This article studies the influence of post-treatment variable on propensity score and LASSO/Post-LASSO estimations in treatment effect model.We introduce post-treatment variable into Rubin Causal Model,and prove it will lead to the failure of the unconfoundedness assumption and further failures of several identification equations,and bias in the estimation of propensity score, which leads to inconsistency of propensity score-based methods. Besides,for the structural equation model,endogeneity will cause inconsistency of both LASSO and Post-LASSO methods.Monte Carlo simulations verify our theoretical results.Finally, we take Dickson et al.(2015)as an application to demonstrate the influence of post-treatment variablebias in empirical studies.

Key words: Treatment Effect, Post-treatment Variable, Propensity Score, LASSO