A1276
Title: Bayesian causal synthesis for meta-inference on heterogeneous treatment effects
Authors: Kenichiro McAlinn - Temple University (United States) [presenting]
Kosaku Takanashi - Riken (Japan)
Shonosuke Sugasawa - Keio University (Japan)
Abstract: A novel Bayesian methodology is proposed to mitigate misspecification and improve estimating treatment effects. A plethora of methods to estimate- particularly the heterogeneous- treatment effect have been proposed with varying success. It is recognized, however, that the underlying data-generating mechanism, or even the model specification, can drastically affect each method's performance without comparing its performance in real-world applications. Using a foundational Bayesian framework, Bayesian causal synthesis is developed, a method that synthesizes several causal estimates to improve overall inference. This process is called meta-inference, as the inference of BCS is occurring above any individual estimate, treating each estimate as data to be updated. A fast posterior computation algorithm is provided, and the proposed method is shown to provide consistent estimates of the heterogeneous treatment effect. Several simulations and an empirical study highlight the efficacy of the proposed approach compared to existing methodologies, providing improved point and density estimation of the heterogeneous treatment effect.