A Data-Adaptive Approach to Modeling Propensity Scores for Inverse Weighted Estimation of Causal Effects

• Activity Number: 111
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: ASA
• Author(s): Yeying Zhu*+
• Companies: Pennsylvania State University
• Keywords: Boosting algorithms ; Causal inference ; Logistic regression ; Observational data ; Random forests

Abstract

In observational studies, differences in the outcome may arise not only due to the treatment but also because of the effect of confounders. Based on the potential outcome framework, propensity-score-based methods are often used to estimate the causal treatment effect. In much of the literature, propensity scores are estimated by logistic regression. However, it is challenging to determine the interaction and nonlinear terms when the covariates are high-dimensional. We develop more flexible approaches to modeling propensity scores from a model combining/averaging perspective. The proposed estimator is a weighted average of parametric and nonparametric machine learning estimates. When there are multiple candidate models, we propose using a cross-validation criterion to select the optimal subset of the candidate models for combining. In the two-stage modeling framework with propensity scores as nuisance parameters, the criterion achieves a balance between the number of models we combine and the variability of the estimated treatment effect. The proposed method can be extended to the estimation of causal treatment effects in longitudinal studies with time-varying treatments.