Variable Selection and Inference in Modeling for the Semiparametric Treatment Effect Estimator

• Activity Number: 528
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Education
• Author(s): Shuai Yuan*+
• Companies: North Carolina State University
• Keywords: Semiparametric treatment effect estimation ; Semiparametric treatment effect estimation ; Covariates adjustment ; Shrinkage variable selector ; Oracle property ; Multiple-split method

Abstract

Inference on the effects of treatment on the basis of a primary outcome is the objective of randomized clinical trials. Zhang et al. (2008, Biometrics) propose a semiparametric method to estimate the treatment effect using a covariate adjustment to improve efficiency, which involves modeling the regression of outcome on covariates separately by treatment group. We study the use of modern variable selection approaches, including shrinkage-type and false selection rate methods, to identify important covariates to be included in the treatment-specific regression models. We show that the standard error formula for the treatment effect proposed by Zhang et al. underestimates the true sampling variation in finite samples under these conditions. To correct for this underestimation we consider use of the empirical sandwich formula and demostrate its superior performance both through simulations and theoretically. Our simulation results show that shrinkage type methods improve the efficiency of treatment effect estimation in many practical settings.