Potential Outcomes Effects in Mediation Designs with Three Experimental Conditions

Causal mediation analysis, based on the potential outcomes framework, is an important new tool for prevention research. Causal mediation analysis is a non-parametric, general technique for estimating causal effects in mediation analysis with explicit emphasis on the identification of assumptions necessary for causal inference. Causal mediation analysis is generally described for the simplest case of two experimental conditions (e.g., treatment and control), which limits researchers to splitting their study into pairwise contrasts when they have more than two experimental conditions. This paper expands causal mediation analysis based on the potential outcomes framework to include the definition, identification, and estimation of causal effects when there are three experimental conditions. Interpretations of the causal effects are illustrated with synthetic data, based on a three-arm mindfulness study. The analytical approach for calculating the causal effects with three experimental conditions is described and implemented in various software, including SAS, Mplus and the R package. The implications of this new approach to mediation analysis are discussed, including the method's clear focus on meeting the assumptions for estimating causal effects and making causal claims.