Estimating causal mediation effects from a single regression model
Christina Tripp Saunders, PhD Post Doctoral Fellow Vanderbilt University School of Medicine
We describe a classical regression framework for estimating causal mediation effects and their variance from the fit of a single regression model, rather than from a system of equations. We present this new approach in the context of two widely used frameworks: the traditional Baron-Kenny approach and the potential outcomes approach. We show how our approach is used to estimate the portion eliminated, which measures the maximum preventive effect of an intervention on the mediated pathways. Estimating the portion eliminated has the potential to enhance understanding of disease prevention and intervention research, for which treatments are designed to change mediators which are hypothesized to change important health outcomes. Requiring the fit of only one model to estimate mediation effects yields an analytical formula for the variance of mediation effects and increases computational efficiency. It also permits the use of a rich suite of regression tools that are not easily implemented on a system of equations. We show how to apply the single-model approach to more complex research hypotheses, including models with moderated mediation and exposure-mediator interactions.