Assessment of Propensity Score Performance in Small Samples

Abstract

BIOSTATISTICS

ASSESSMENT OF PROPENSITY SCORE PERFORMANCE IN SMALL SAMPLES

EMILY PETERSON

Thesis under the direction of Professor Tatsuki Koyama

In observational studies, treatment selection is determined by the characteristics of the subject, and therefore, cannot be randomized. One must account for systematic differences in baseline characteristics between treatment groups. Propensity score is a subject’s probability of receiving a specific-treatment, which is conditioned on the observed baseline covariates, and is a method to account for differences in baseline characteristics between treatment groups. There has been little research on variable selection for propensity score models when dealing with restrictive sample sizes. The purpose of this study is to assess performance of propensity score models that use data reduction to allow for inclusion of all baseline covariates. The results of this simulation study showed that inclusion of baseline covariates related to both treatment and outcome yield the optimal propensity score model in small samples. In addition, penalized maximum likelihood methods in conjunction with propensity score models yield optimal type I error.

Approved ________________________________________________________ Date ____________________