Outliers in Semi-Parametric Estimation of Treatment Effects

Outliers can be particularly hard to detect, creating bias and inconsistency in the semi-parametric estimates. In this paper, we use Monte Carlo simulations demonstrate that methods, such as matching, are biased presence of outliers. Bad good leverage point outliers considered. Bias arises case bad points because they completely change distribution metrics used define counterfactuals; points, on other hand, increase chance breaking common support condition distort balance covariates, which may push practitioners misspecify propensity score or distance measures. We provide some clues identify correct for effects following a reweighting strategy spirit Stahel-Donoho (SD) multivariate estimator scale location, S-estimator location (Smultiv). An application experimental data is also implemented.