SERtalks – Ohio

Causal mediation analysis is used for decomposing total causal effects into pathway specific causal effects in epidemiologic research. There is a long tradition of path analysis and structural equations modeling in social sciences, but these techniques were not traditionally applied in epidemiology for various reasons, including their reliance on linearity, additivity and other model assumptions that were uncommon in the epidemiologic setting.  Instead, epidemiology has developed over the last 2 decades a set of mediation concepts and techniques derived from counterfactual and non-parametric structural equations models.  Within the last decade these have crystallized into two main types of effects: controlled effects and natural (pure) effects.  In this 3-hour seminar, Dr. Kaufman will retrace this history, define these effects formally, and explain their identification, estimation, and interpretation.  In the final hour, he will demonstrate estimation and interpretation of the quantities using Stata Statistical Software.