A Bootstrap Procedure of Propensity Score Estimation

Propensity score estimation plays a fundamental role in propensity score matching for reducing group selection bias in observational data. To increase the accuracy of propensity score estimation, the author developed a bootstrap propensity score. The commonly used propensity score matching methods: nearest neighbor matching, caliper matching, and Mahalanobis metric matching, are used for evaluating the quality of bias reduction results using propensity scores estimated with or without the bootstrap. The study reveals that the matching results with bootstrap propensity score are better than or comparable to those of matching without bootstrap procedures. With regard to the merit of the bootstrap method in reducing sampling errors and, therefore, improving the accuracy of propensity score estimation, the results of this study imply a great potential to use the bootstrap method in estimating propensity scores. This study provides a preliminary framework for further research in this area.