Propensity Score Adjustments Using Covariates in Observational Studies

• Activity Number: 665
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Government Statistics
• Author(s): Daniel Yang*+ and Alix I. Gitelman and Virginia Lesser and David D. Birkes
• Companies: Bureau of Labor Statistics and Oregon State University and Oregon State University and Oregon State University
• Keywords: propensity scores ; equal frequency weights ; inverse variance weights

Abstract

In observational studies, propensity score (PS) methods have been used to reduce the bias of the treatment effect estimator. The equal frequency (EF) subclassification method, which has been widely applied, equally divides the sample space into subsets using the PS percentiles and assigns equal weights (EW) to subclasses. Some researchers have used an equal variance (EV) method, which divides the samples by equalizing the estimated variances of the treatment effect estimator among subclasses and assigns inverse variance (IV) weights. We conduct simulations to indicate that under quadratic term misspecification, the EF-IV estimator provides the lowest bias and root mean square error as compared to the ordinary least square estimator and other propensity score estimators. Our theoretical results demonstrates that if higher variation occurs with larger bias for within subclass treatment effect estimates then the EF-IV estimator has a smaller overall bias than the EF-EW estimator. We show that if the variance in the EV approach is larger than the harmonic mean of the within subclass variances of the EF approach, then the EF-IV estimator has a lower variance than the EV-IV estimator.