Propensity Score Matching to Recover Latent Experiments

• Activity Number: 392
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Ben B. Hansen*+
• Companies: University of Michigan
• Keywords: covariate balance ; matched sampling ; observational study ; confounding ; causality ; common support

Abstract

Propensity score matching aims to gain for an observational study various benefits characteristic of experiments. Some of these have observable manifestations, others not. When it succeeds in ways that can be seen, this suggests it has succeeded in the remaining ones; so we cite its successes in visible domains, such as covariate balance, as evidence of likely success in others, namely removing bias from estimates of treatment effects.

Yet existing theory licenses such inference only vaguely, if at all. Much of it assumes exact matching on the true propensity score, whereas in practice the best one can do is to match approximately, and on an estimate of the score. I'll present a novel large-sample account of causal inferences with propensity-matched data that relaxes these limitations and clarifies the contribution of propensity matching diagnostics.