Presentation: Mediation Analysis with Complex Intermediate Causal Structure
Candidate: Fan Xia, Graduate Student, UW Biostatistics
Committee Members: Gary Chan (chair), James Hughes, Thomas Richardson, Carey Farquhar, Walter Kukull (GSR)
Abstract: In this talk, we focus on the third project on the decomposition, identification, and multiply robust estimation of natural mediation effects with multiple mediators.
We consider a decomposition of the total indirect effect through multiple mediators, with an unspecified causal ordering, into individual components termed exit indirect effects and a remainder interaction term. We provide a set of identification assumptions for estimating all components. The identified expressions, which are closely related to the interventional indirect effects, continue to have causal interpretations when some identification assumptions are violated, as long as the total indirect effect is identified.
We provide four moment-type estimators for each decomposed effect based on different parametrizations and derive the semiparametric efficiency bounds for the effects. The efficient influence functions contain conditional densities that are variational dependent, which is uncommon in existing problems, and we consider a reparameterization based on copulas to avoid model incompatibility and proposed a quadruply robust estimator for each of the decomposed effects that remains consistent and asymptotically normal under four types of possible misspecification and is also locally semiparametric efficient.