Title: Flexible models for time-dependent causal mediation
Authors: Jason Roy - Rutgers University (United States) [presenting]
Abstract: Suppose interest is in estimating time-varying direct and indirect effects of exposure from observational data. We propose a Bayesian nonparametric approach for modeling the observed data. We then develop causal identifying assumptions and computational methods. We assess the performance of the method and compare with alternative approaches using simulation studies. Finally, we implement the method on longitudinal air pollution data.