Causal Mediation Analysis with a Binary Mediator: The Influence of the Estimation Approach and Causal Contrast

Abstract

Although causal mediation analysis clarifies causal effect estimation, little attention has been devoted to the differences between causal estimation approaches. This paper illustrates the difference between the causal estimation approaches for mediation models with a binary mediator. Using a Monte Carlo simulation study and an empirical data example we show that the regression- and simulation-based approaches provide indirect and total effect estimates that are dependent on the chosen causal contrast, while the imputation- and weighting-based approaches provide overall effect estimates. The results underline the importance of choosing an estimation approach that provides estimates of the causal effect of interest.