On the Power of Stratified Analysis Using Propensity Score Matching
Jeff Jonkman, PhD.
Thursday, February 17, 2000 at 2:00-3:00 p.m.
14th Floor WARF
ABSTRACT
Propensity score matching is used to reduce bias in observational studies, where subjects may not be randomly assigned to treatments. For example, treatment for early-stage breast cancer is usually based on physician advice and the patient's decision. A common use of propensity scores is to stratify the data into quintiles, based on the estimated propensity score, and perform a Mantel-Haenszel test. While the bias-reducing properties of propensity score matching are well documented, less work has been done regarding power when propensity score matching is used. We give a brief introduction to propensity score methods, investigate the power of the test described above using an extensive simulation study, and consider the question of an approximation for power to be used in study planning.
