## Department of Biostatistics Seminar/Workshop Series

## Biostatistics Student Research Forum:

## Mediation Analysis: A Review and Some Musings

## Christina Maria Tripp, Ph.D. Candidate, Department of Biostatistics

## Vanderbilt University, School of Medicine

Mediators are behavioral, biological, psychological, or social constructs that transmit the effect of one variable to another. Mediation analysis seeks to understand how the effect of a predictor on an outcome passes through a mediator by estimating the predictor’s

*indirect* effect. The indirect effect is how much of the predictor’s effect on the response is transmitted through a mediator, and it can be estimated from the difference between the

*total* and

*direct* effects.

In this talk, I provide a review of the classical regression framework for mediation analysis, as well as some ideas for a more general framework that is part of my dissertation research with Jeffrey Blume,

PhD. State of the art mediation analysis requires fitting multiple regression models and aggregating model effect estimates in order to measure mediation. We derived an analytical formula for the indirect effect and show how it can be computed from just a single regression model. Our approach reduces computation time and extends to multi-dimensional effects, such as when dealing with multiple predictors and mediators.