A Bayesian Semiparametric Approach to Causal Inference with Intermediate Variables

• Activity Number: 296
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Fan Li*+ and Scott Schwartz and Fabrizia Mealli
• Companies: Duke University and Duke University and University of Florence
• Keywords: Bayesian nonparametrics ; causal inference ; Dirichlet process mixture ; intermediate variables ; principal stratification ; compliance

Abstract

In causal inference, treatment comparisons often need to adjust for post-treatment (intermediate) variables. Principal stratification (PS) is a popular framework to deal with such variables. Continuous intermediate variables introduce inferential challenges to the PS analysis. Existing methods usually rely on a fully parametric model for the joint distribution of the potential intermediate variables, which is often inadequate to capture complex density shapes. To provide the necessary flexibility in modeling, we propose a Bayesian semiparametric approach that combines a parametric model for the response variables with a Bayesian nonparametric model for the intermediate variables using the Dirichlet process mixture. The method is illustrated by two examples: one concerning partial compliance in randomized clinical trial, another concerning the lock-in effect in job training programs.