Mediation analysis
Investigate an explicit proposed pathway
Mediation analysis decomposes a proposed relationship into direct and indirect paths through a mediator. Regression, SEM and causal-mediation frameworks make different assumptions. Bootstrap intervals address the sampling uncertainty of an indirect-effect estimate; they do not establish a mechanism. PROCESS implements common regression-based models, but the causal argument must come from design and identification.
Use when a theory proposes how an exposure affects an outcome through an intermediate variable. Plan temporal ordering, relevant confounder measurements and identification assumptions before deciding how to estimate the indirect effect.
Strengths
- Makes mechanism hypotheses explicit
- Bootstrap inference handles non-normal indirect-effect sampling distributions
Limitations
- Cross-sectional associations rarely identify a causal mechanism
- Mediator-outcome confounding can persist after exposure randomisation
Know the boundary
A significant indirect effect or a nonsignificant direct effect does not prove complete causal mediation.