Tracing Causal Paths from Experimental and Observational Data
Xiang Zhou, Teppei Yamamoto
Abstract
The study of causal mechanisms abounds in political science, and causal mediation analysis has grown rapidly across different subfields. Yet, conventional methods for analyzing causal mechanisms are difficult to use when the causal effect of interest involves multiple mediators that are potentially causally dependent—a common scenario in political science applications. This article introduces a general framework for tracing causal paths with multiple mediators. In this framework, the total effect of a treatment on an outcome is decomposed into a set of path-specific effects (PSEs). We propose an imputation approach for estimating these PSEs from experimental and observational data, along with a set of bias formulas for conducting sensitivity analysis. We illustrate this approach using an experimental study on issue-framing effects and an observational study on the legacy of political violence. An open-source R package, paths, is available for implementing the proposed methods.