Improving causal inference of mediation analysis with multiple mediators using interventional indirect effects

Mediation analysis is indispensable for investigating how a treatment causally affects an outcome via intervening variables. Existing discussions on the validity of causal inference drawn from mediation analysis have prioritized single mediator settings. In this article, we focus on improving causal inference when investigating multiple mediators. We pay particular attention to the prevalent practice of exploring mediated effects along various paths linking several mediators. Such approaches are fraught with stringent-yet often overlooked-causal assumptions that predicate valid inference of indirect or mediated effects. To mitigate the risk of incorrect inference, we introduce a conceptually and substantively novel approach from the causal inference literature: interventional indirect effects. Interventional indirect effects focus on the contributions of each distinct mediator to the treatment effect on the outcome. An appealing feature of this approach is that valid causal inference of mediation analysis with multiple mediators can be attained without assuming a (correct) causal structure among the mediators. Using a social psychology experiment as a running example, we illustrate how researchers can readily estimate and interpret the proposed interventional effects in practice. We hope this article will encourage explaining and substantiating the causal assumptions underpinning mediation analysis with multiple mediators to fortify causal inference in psychology research.