Causal Mediation Analysis with Multiple Mediators: A Simulation Approach

• In Event: 20858 - Reimagining How We Measure and Model the Social World: Causation, Flows, and Artificial Intelligence

Abstract

Analyses of causal mediation often involve exposure-induced confounders or, relatedly, multiple mediators. In such applications, researchers aim to estimate a variety of different quantities, including interventional direct and indirect effects, multivariate natural direct and indirect effects, and/or path-specific effects. Each of these estimands capture—in different ways—how the influence of a focal exposure on an outcome is transmitted through intermediate variables. This study introduces a general approach to estimating all these effects by simulating potential outcomes from a series of distribution models for each mediator and the outcome. Building on similar methods developed for analyses with only a single mediator (Imai et al. 2010a,b, 2011), we first outline how to implement this approach with parametric models. The parametric approach can accommodate linear and nonlinear relationships, both continuous and discrete mediators, and many different types of outcomes. However, it depends on the correct specification of each model used to simulate the potential outcomes, which is a stringent requirement, especially when modeling multiple mediators. If violated, it can lead to biased estimates, even when the target estimands are nonparametrically identifiable. To address the risk of misspecification bias, we introduce an alternative implementation using a novel class of nonparametric models, which leverage deep neural networks to approximate the relevant distributions without relying on strict assumptions about functional form. We illustrate both methods with a reanalysis of the effects of prenatal care on preterm birth, as mediated by maternal smoking and preeclampsia risk VanderWeele et al. (2014). Open-source software is available in R to facilitate implementation.

Authors

• Geoffrey Thomas Wodtke, University of Chicago

• Jesse Zhou, University of Chicago