A Comparison of Propensity Score Models and Covariate Selections for Propensity Score in Multilevel Data

• In Event: Advanced Methods With Practical Implications

Abstract

This simulation study was designed to compare propensity score methods and covariates selections for propensity score in multilevel data under a variety of manipulated conditions including: ratios of treatment to control group members (1:3, 1:1), the number of groups (50, 100) and the average group sizes (20, 40, 50). Two propensity score models were compared: traditional logistic model verse multilevel logistic model. In addition, the accuracy of parameter estimates was compared among the seven types of covariate selection. These results showed that including variables related to both treatment and the outcome decreased the estimated treatment effect. In general, the degrees of relative bias tended to be larger when the logistic model was applied as compared to the multilevel logistic model.

Authors

• Hyeonjeong Jeon, National Youth Policy Institute

• Hyewon Chung, Chungnam National University