On the Difficulty of Establishing Causal Mechanisms without Mediation

• In Event: Causal Complexity: Effect Heterogeneity, Mediation, and Generalization

Abstract

Political scientists are increasingly attuned to the promises and pitfalls of establishing the existence or nonexistence of a causal effect. But for many researchers, the most vital question is not if a causal effect exists but why an effect exists. Causal mediation analysis has been a popular way to assess these types of causal mechanism questions, but many researchers have been skeptical of mediation due to the assumptions it requires. This paper investigates two popular alternative techniques for exploring causal mechanisms. First, we explore “effect on mediator” tests in which scholars establish the effect of treatment on a series of potential mediators to establish the possibility of a causal mechanism through those mediators. We show that without further assumptions these tests are not necessary nor sufficient for a causal mechanism to exist and thus are not well-suited to the task of establishing causal mechanisms. Second, we investigate the possibility of a “mechanism-canceling moderator” design where the researcher assumes no treatment effect for a particular group of units. We show that if direct effects are stable across levels of the moderator, this design can identify indirect effects from the interaction between the treatment and the moderator. The paper shows how being explicit about our causal assumptions can clarify how our empirical evidence connects to our theoretical arguments.

Authors

• Matthew Blackwell, Harvard University

• Ruofan Ma, Harvard University

• Aleksei Opacic