Propensity Score Estimation with Boosted Regression

• Activity Number: 632
• Type: Topic Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Claude Setodji* and Daniel F. McCaffrey and Lane Burgette and Beth Ann Griffin and Daniel Almirall
• Companies: RAND Corporation and Educational Testing Service and RAND Corporation and RAND Corporation and University of Michigan
• Keywords: Causal inference ; Generalized boosting model ; propensity score ; variable selection ; Covariate balance

Abstract

In this paper, we use the generalized boosting model (GBM) as an alternative to logistic regression for estimating propensity scores. GBM combines many simple regression trees to provide a smooth and flexible propensity score model. We present methods for using covariate balance to guide fitting the GBM. Using a simulation study, we compare GBM to alternative methods for propensity score estimation. We find that in terms of mean squared error, GBM appears to be advantageous in the commonly encountered situation of propensity score model building in the presence of many candidate confounders, some of which may not actually be related to the outcomes of interest. We also discuss a generalization of propensity score analysis to three or more treatments and the use of GBM in this context.