Mediation Analysis to Uncover Causal Mechanisms: How Specifically Designed Experiments Can Help for Identification

Understanding the mechanisms through which causal effects come about is crucial for designing effective policy interventions. Causal mediation analysis aims at identifying causal mechanisms which is challenging and typically requires strong assumptions. In this project, we explore identification and estimation of causal mechanisms in so-called double randomization designs that combine two experiments. The first and main experiment randomizes the treatment and measures its effect on the mediator and the outcome of interest. A second auxiliary experiment randomizes the mediator of interest and measures its effect on the outcome. We analyze different experimental designs that randomize different combinations of treatment and mediator values. We show that such designs allow for identification based on an assumption that is weaker than the assumption of sequential ignorability which is typically made in the literature. We also show that for certain research questions it is possible to design an experiment that identifies the causal mechanism of interest without strong assumptions.