Systematic Evaluation and Comparison of Entropy Balancing and Covariate Balancing Propensity Score Methods

Evgeniya Reshetnyak, Fordham University


Propensity scores methodology is a popular approach to treatment effect estimation in observational research. Correctly specified propensity scores guarantee the balance of pre-treatment covariates between treatment and control groups, and lead to unbiased estimated treatment effect. The major challenge of propensity scores methodology is that the true propensity scores are unknown and need to be estimated using correctly specified model. Typically, researchers go through several rounds of manual specification of propensity scores model, followed by covariate balance check and model’s correction. In this dissertation a series of simulation studies were conducted to evaluate and compare Covariate Balancing Propensity Scores (CBPS) and entropy balancing methods, that automatically equate groups on pre-treatment measures, and do not require manual specification of propensity scores model or covariate balance check. The first part of this dissertation assessed balancing properties of these methods and compared with the traditional weighting techniques. According to the results, both CBPS and entropy balancing produced well-balanced data; CBPS showed slightly worse results when true propensity score model contained cubic effects. The second part of this work evaluated accuracy of treatment effect estimation (ATT) on the data that were preprocessed by these methods. When an outcome was continuous, all techniques including reference methods produced similar unbiased results. In the models with binary outcome CBPS and entropy balancing performed equally well with an exception of the scenarios with true propensity scores models containing cubic effects, when CBPS produced less biased results. In conditions with count outcome the difference between CBPS and entropy balancing was observed only in the scenarios of true propensity scores model with cubic effects, where entropy balancing produced less biased results.

Subject Area


Recommended Citation

Reshetnyak, Evgeniya, "Systematic Evaluation and Comparison of Entropy Balancing and Covariate Balancing Propensity Score Methods" (2017). ETD Collection for Fordham University. AAI10278790.