Publication Date

Summer 7-31-2021

Abstract

In this work, we developed ten-way decompositions of the total causal effect on continuous outcomes in the presence of two mediators that have mediation and interaction components. We considered a natural counterfactual interaction effect framework that facilitates the decomposition of the total casual effects of the exposure on the outcome attributable to the effects of mediators and their interaction effects with the exposure. We discussed the identification assumptions and provide empirical formulas for estimating the components of the ten-way decompositions. We also developed multiplicative total causal effect decompositions based on risk ratios for binary outcomes, hazard ratios and logarithm survival time differences for time-to-event outcomes, respectively. We conducted one simulation study and two real data applications to demonstrate the proposed decompositions. The detailed effect decompositions provide a deeper understanding of the total casual effect and the components that are attributable to the direct effects or indirect effects due to mediation and/or interaction, and hence provide valuable information for the development of interventional strategies based on the relative importance of different pathways.

Degree Name

Statistics

Level of Degree

Doctoral

Department Name

Mathematics & Statistics

First Committee Member (Chair)

Li Li

Second Committee Member

Li Luo

Third Committee Member

Yan Lu

Fourth Committee Member

Huining Kang

Fifth Committee Member

Fletcher Christensen

Language

English

Keywords

causal inference, causally sequential mediators, binary outcome, survival outcome, risk ratio, hazard ratio

Document Type

Dissertation

Download

Share

COinS