Boosting and Double Robust Estimation

• Activity Number: 40
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Daniela Golinelli*+ and Greg Ridgeway
• Companies: Bureau of Justice Statistics and National Institute of Justice
• Keywords: Double robust estimation ; Propensity Score ; Boosting

Abstract

In the double robust estimation literature, authors tend to put more emphasis on the quality of the propensity score model stage than the outcome regression stage. Several studies have shown that generalized boosting, a non-parametric method, often outperforms other methods when estimating the propensity score. In this work, we investigate, via simulation, whether the use of boosting both for the propensity score stage and the outcome model stage produces treatment effect estimates with smaller mean square error than the estimates obtained using a more traditional regression approach for the outcome model.