A0482
Title: Bayesian nonparametric methods for causal mediation analysis
Authors: Jason Roy - Rutgers University (United States) [presenting]
Abstract: In many settings, interest is not just in the effect of an exposure on an outcome, but also on possible mechanisms. Causal mediation analysis aims to estimate how much of the impact of the exposure on the outcome is due to the exposure's impact on intermediate variable(s). However, the inference is challenging due to both the need for strong identifying assumptions and the need for multiple models. We describe some Bayesian nonparametric models that were motivated by the desire to avoid the risk of bias that comes with parametric modeling. We illustrate the methods with real examples for both the single mediator and multiple mediator scenarios.
