A Data Adaptive Estimator for the Average Treatment Effect Based on Inverse Probability Weighting

Abstract

In observational studies, the propensity score method is often used as an approach to handle the confounding by indication bias. Among other methods, inverse probability of treatment weighting uses weights based on the propensity score to create a synthetic sample in which the distribution of measured baseline covariates is independent of treatment assignment. However, methods that use inverse probabilities as weights are sensitive to misspecification of the propensity score model when some estimated propensities are small. The recently developed doubly robust method applies the propensity score model and the outcome model simultaneously and it remains asymptotically unbiased of the parameter even if one of the models is misspecified. In this study, we propose a data adaptive estimator based on the usual doubly robust estimator with the introduction of a tuning parameter. We demonstrate through simulations that the proposed data adaptive robust estimator is a more efficient estimator of the standard error of the treatment-control mean difference.