Outliers in semi-parametric estimation of treatment effects

Average treatment effects estimands can present significant bias under the presence of outliers. Moreover, outliers can be particularly hard to detect, creating bias and inconsistency in the semi-parametric ATE estimads. In this paper, we use Monte Carlo simulations to demonstrate that semi-parametric methods, such as matching, are biased in the presence of outliers. Bad and good leverage points outliers are considered. The bias arises because bad leverage points completely change the distribution of the metrics used to define counterfactuals. Whereas good leverage points increase the chance of breaking the common support condition and distort the balance of the covariates and which may push practitioners to misspecify the propensity score. We provide some clues to diagnose the presence of outliers and propose a reweighting estimator that is robust against outliers based on the Stahel-Donoho multivariate estimator of scale and location. An application of this estimator to LaLonde's (1986) data allows us to explain the Dehejia and Wahba (2002) and Smith and Todd (2005) debate on the inability of matching estimators to deal with the evaluation problem.