A0673
Title: A Bayesian doubly robust estimation using entropic tilting method
Authors: Shunichiro Orihara - Tokyo Medical University (Japan) [presenting]
Tomotaka Momozaki - Tokyo University of Science (Japan)
Shonosuke Sugasawa - Keio University (Japan)
Abstract: In observational studies, propensity score methods are of central interest for estimating causal effects while adjusting for confounding variables. In particular, the doubly robust estimator has attracted considerable attention, as it provides a consistent estimator for causal effects when either the propensity score model or the outcome model is correctly specified. Like other propensity score approaches, the doubly robust estimator typically involves two-step estimation: first, estimating the propensity score and outcome models and then estimating the causal effects using the estimated values. However, such two-step inference does not naturally align with the Bayesian framework, where inference is understood as the updating of prior beliefs solely through the likelihood function. A novel Bayesian doubly robust estimation method is proposed using entropic tilting techniques, which avoids the need for two-step estimation. Specifically, the doubly robust property is achieved by modifying the posterior distribution through the imposition of certain moment conditions. As a result, the proposed procedure can be interpreted as a fully Bayesian approach to doubly robust estimation. A methodological explanation of the double robustness properties of the proposed method is provided. Its performance is also examined through simulation studies and real data applications, comparing it with other Bayesian approaches.
