This paper assesses the performance of common estimators adjusting for differences in covariates, such as matching and regression, when faced with so-called common support problems. It also shows how different procedures suggested in the literature affect the properties of such estimators. Based on an Empirical Monte Carlo simulation design, a lack of common support is found to increase the root mean squared error (RMSE) of all investigated parametric and semiparametric estimators. Dropping observations that are off support usually improves their performance, although the magnitude of the improvement depends on the particular method used.
Empirical Monte Carlo Study matching estimation regression common support outlier small sample performance