Model-Averaged Double Robust Estimation

• Activity Number: 526
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Matthew Cefalu*+ and Francesca Dominici and Giovanni Parmigiani
• Companies: Harvard University and Harvard School of Public Health and Dana-Farber Cancer Institute
• Keywords: causal inference ; doubly robust ; confounding ; model uncertainty ; model averaging ; propensity score

Abstract

We propose a new class of estimators for the average causal effect, the model averaged double robust (MA-DR) estimators, that account for model uncertainty in both the propensity score and outcome model through the use of model averaging. The MA-DR estimator is defined as a weighted average of double robust estimators, where each double robust estimator corresponds to a specific choice for the outcome model and the propensity score, respectively. The MA-DR estimators extend the desirable double robustness property by achieving consistency under the much weaker assumption that either the true propensity score model or the true outcome model be within a specified, possibly large, class of models. We provide asymptotic results and conduct a large scale simulation study that indicates the MA-DR estimator has better finite sample behavior than the usual double robust estimator. We show that the MA-DR estimator that a priori links the propensity score and the outcome model can have 90% less variance than a double robust estimator constructed via model selection for the propensity score and the outcome model separately.