A Comparison of three Propensity Score Methods’ Average Treatment Effect on the Treat by Simulation

Abstract

This study compared the average treatment effect on the treated(ATT) of three propensity score methods- logistic regression model, generalized boosted model, and Bayesian model– by simulation. The simulated data were generated under two sample sizes, four covariates models, and four model intercepts conditions. The results shaw that generalized boosted model and Bayesian model also provide smaller bias than logistic regression model when the sample size was small(N=200). And, generalized boosted model and Bayesian model provide small bias than logistic regression model. It was interpreted that the propensity score method which takes into account the distribution of covariates produce more adequate estimation of causal effect.