### Department of Biostatistics Seminar/Workshop Series

# Theory and application of statistical methods for causal mediation analysis

## Eric Tchetgen Tchetgen, PhD

### Associate Professor of Epidemiology, Department of Epidemiology, Department of Biostatistics, Harvard University School of Public Health

### Wednesday, March 14, 1:30-2:30pm, MRBIII Room 1220

A formal causal theory of mediation analysis is presented, whereby a causal definition of direct and indirect effects of an exposure mediated by a variable on its causal pathway to the outcome is
formulated. Simple conditions for identification of such mediation effects are given, and nonparametric, semiparametric and parametric statistical methods of estimation are discussed. New multiply robust
estimators of marginal direct and indirected effects are presented that strike a balance between robustness to modeling assumptions and statistical efficiency. A multiply robust estimator is one that only
requires a subset of the working models indexing the data generating mechanism to be correct for valid inferences, without knowing which subset is in fact correct; the approach is contrasted in simulations and a data example, and shown to be superior to the maximum likelihood estimator in terms of robustness, with relatively low efficiency loss. Finally, an alternative and simple inverse odds ratio weighted (IORW) estimator is proposed for decomposing a total effect conditional on covariates into direct and indirect effects in the context of standard generalized linear models. IOWR is advantageous in its ease of implementation using standard statistical software, in addition IOWR scales appropriately in the context of multiple mediators, and can easily be used for quantile regression or in the context of survival
data, say in a Cox regression.

The talk is mainly based on the following two articles available online :

(1) Eric J. Tchetgen Tchetgen and Ilya Shpitser, "Semiparametric Theory for Causal Mediation Analysis: efficiency bounds, multiple robustness, and sensitivity analysis" (June 2011). Harvard University
Biostatistics Working Paper Series. Working Paper 130.

http://www.bepress.com/harvardbiostat/paper130
(2) Eric J. Tchetgen Tchetgen, "Inverse Odds Ratio-Weighted Estimation for Causal Mediation Analysis" (February 2012). Harvard University Biostatistics Working Paper Series. Working Paper 143.

http://www.bepress.com/harvardbiostat/paper143