Priors and Propensity Scores in Bayesian Causal Inference
Observational Studies · 2025-03 · 11 (1) · 47-59
DOI: 10.1353/obs.2025.a956841 ↗Abstract
Abstract: Using synthetic examples, Aronow et al. (2025) provide a convincing case for the special status of randomized controlled trials (RCTs) in which the propensity scores are known and can be used to make causal inferences. We provide a Bayesian perspective on the propensity score, which has been a controversial topic for some time. We review the Bayesian approach for population-level causal inference with an emphasis on role of the prior. Under the assumption of prior independence of the propensity score and outcome model parameters, the propensity score plays no role in Bayesian inference for causal effects. However, recent work has shown why assuming prior independence can be problematic in high-dimensional models. We discuss how various approaches that have been proposed for incorporating the propensity score in Bayesian causal inference correspond to different ways of relaxing prior independence. Given these considerations, we illustrate how a Bayesian might approach the synthetic examples of Aronow et al. (2025) in RCTs.
EUVIMED