Causal Inference
Contrasting Natural Effects and Separable Effects: Insights into Mediation Analysis Etsuji Suzuki* Etsuji Suzuki Tomohiro Shinozaki Eiji Yamamoto
As has been well appreciated in the causal mediation literature, the total effect can be decomposed into the natural direct effect and the natural indirect effect in the counterfactual framework. If certain assumptions about confounding are met, we can use the so-called mediation formula to identify these effects under the nonparametric structural equation models with independent errors (NPSEM-IE). However, the natural effects have been criticized because they are cross-world quantities, and the so-called cross-world independence assumption cannot be empirically verified. Furthermore, interventions on the mediator may sometimes be challenging even to conceive. As an alternative approach, separable effects were more recently proposed and applied in mediation analysis, often under the finest fully randomized causally interpretable structured tree graph (FFRCISTG) models. In this approach, the exposure is assumed to be separated into two (or more) components, one having a direct effect only on the mediator and the other one having a direct effect only on the outcome. Furthermore, each separable component can in principle be intervened on separately, and the total effect can be decomposed into the separable direct effect and the separable indirect effect. The separable effects are not defined as cross-world quantities and are claimed to be identifiable under assumptions that are, in principle, testable, thereby addressing some of the challenges associated with natural effects. In this presentation, we contrast natural effects and separable effects under the NPSEM-IE, thus highlighting their similarities and differences. Additionally, we illustrate these two approaches graphically by using causal directed acyclic graphs, incorporating potential outcomes determined by the NPSEM-IE. By examining their required properties/assumptions and sufficient conditions for identification, we aim to provide deeper insights into mediation analysis.