Improving the Propensity Score Equal Frequency Adjustment Estimator Using an Alternative Weight

• Activity Number: 191
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Daniel K. Yang*+ and Alix I. Gitelman and Virginia M. Lesser and David S. Birkes
• Companies: Oregon State University and Oregon State University and Oregon State University and Oregon State University
• Keywords: propensity scores ; adjustment ; sub-classification ; equal frequency ; equal variance ; inverse variance weights

Abstract

The bias of the treatment effect estimator in an observational study can be reduced by the propensity score (PS) adjustment. Rosenbaum and Rubin (1983, 1984) provide a theoretical framework to use equal frequency subclassification (EFS) on PS with equal weights (EW) that was introduced in Cochran (1968). Hullsiek and Louis (2002) propose equal variance subclassification (EVS) using inverse variance weights (IVW). We develop a theoretical framework to illustrate if higher variation occurs with larger bias among the subclasses in EFS, then an IVW estimator has smaller bias than the EW estimator; and it always has smaller variance. We show if the equal variance in the EVS approach is larger than the harmonic mean of the subclass-specific variances under EFS, then the IVW estimator using EFS has lower variance. Numerical verification of these results and a data application are also given.