Comparing Different Methods of Estimating the Variance of Propensity Score Matching Estimator

Abstract

Propensity score matching is extensively utilized in estimating the effects of policy interventions and programs for data observations. This method compares two treatment and control groups to make statistical inferences about the significance of the effects of these policies on target variables. Therefore, when using propensity score matching, it is significant to obtain the standard error to estimate the treatment effect. The precise estimations of variance and standard deviation facilitate more efficient statistical testing and more accurate confidence intervals. However, there is no agreement in the literature on the estimation method of standard error; some methods rely on resampling, while others do not. This study compares these methods using Monte Carlo simulation and calculating the Mean Squared Errors (MSE) of these estimators. Our results indicate that Jackknife and standard methods are superior to Abadie and Imbens (2006) bootstrap, and subsampling ones in terms of accuracy. Finally, reviewing Tayyebi et al. (2019) indicated that different methods of estimating variance in the matching estimator led to different statistical inferences in terms of statistical significance.