Covariate Balancing Propensity Score

• Activity Number: 670
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Kosuke Imai*+ and Marc Thomas Ratkovic
• Companies: Princeton University and Princeton University
• Keywords: instrumental variables ; marginal structural models ; matching ; weighting ; observational studies ; randomized experiments

Abstract

Propensity score plays an essential role in a variety of settings for causal inference. In observational studies, for example, matching and weighting methods based on the estimated propensity score have been employed to adjust for confounders. Despite their popularity and theoretical appeal, the practical difficulty of these methods is that propensity score is unknown and must be estimated. The misspecification of propensity score model may result in covariate imbalance and bias the estimation of treatment effects. In this paper, we show how to estimate propensity score while simultaneously maximizing the covariate balance. The key insight is to derive the moment conditions for achieving the optimal covariate balance (e.g., mean independence between treatment and covariates) as well as for maximizing the predictive power of the propensity score model (e.g., likelihood function). These moment conditions are then combined within the generalized method of moments or empirical likelihood frameworks. We apply the proposed methodology to a variety of situations where the estimation of propensity score is required for causal inference.