ATLASResearch
methods
Quantitative/ Analysis

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.

WHEN IT FITS

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.

USED ACROSS
Banking & financeBusiness & MBAPsychologyComputer scienceEducation