Covariate Balancing Propensity Score for General Treatment Regimes

• Activity Number: 629
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Kosuke Imai*+ and Marc Ratkovic and Christian Fong
• Companies: Princeton University and Princeton University and Princeton University
• Keywords: causal inference ; instrumental variables ; propensity scores ; covariate balance ; matching ; weighting

Abstract

We extend the recently proposed methodology, the Covariate Balancing Propensity Score (CBPS; Imai and Ratkovic, 2014, JRSSB), to general treatment regimes. The original CBPS methodology improves the propensity score estimation by optimizing the resulting balance of observed covariates between the treatment and control groups. We first extend the CBPS to a multi-valued treatment and then to a continuous treatment. The CBPS is particularly useful in these settings with many treatment values because standard diagnostics of covariate balance are less applicable. Finally, we also apply the CBPS to generalize an instrumental variable estimate to the average treatment effect. Empirical and simulation studies show that the CBPS, often dramatically, improve the performance of propensity score methods.