A process in which part or all of the effect of an exposure on an outcome is transmitted through an intermediate variable (the mediator) that is caused by the exposure and in turn affects the outcome.

Principle

The total effect of an exposure can be decomposed into direct effects (not through the mediator) and indirect effects (through the mediator); adjusting for a mediator removes the indirect component from the estimated effect of the exposure.

Demonstration

Illustrative scenario → Exposure: increased physical activity. Recognition → Physical activity reduces body weight (mediator), and lower weight reduces blood pressure. Action → Mediation analysis estimates indirect effect through weight and direct effect independent of weight. Consequence → Findings identify whether interventions should target the mediator, the exposure, or both.

Misapplication

Treating a mediator as a confounder and adjusting for it when the research question concerns the total effect. This removes the indirect effect and can falsely suggest the exposure has no effect.

Consequence

Mediation analysis clarifies mechanisms and can identify intermediate targets for intervention; misapplied adjustment can obscure total causal impact or introduce bias when mediator–outcome confounding is unaccounted for.

Reversal

If the mediator–outcome relationship is confounded by an unmeasured pre‑ or post‑exposure variable, simple adjustment is invalid; methods for causal mediation require assumptions (no unmeasured mediator–outcome confounding, correct temporal ordering) that, if violated, invalidate mediation estimates.

Boundary

Clearly within: a variable caused by exposure that causally influences outcome. Boundary case: a variable partially caused by exposure and partially preexisting—classification depends on the causal model. Clearly outside: a pre‑exposure confounder or an effect modifier that does not lie on the causal path.

Semantic Tension

Interest in decomposing effects (mechanism) versus interest in total effect for policy; adjusting for mediators helps mechanism understanding but can conflict with estimating total impact.

Synthesis

Mediation locates where effects occur in a causal chain: it informs mechanism and intervention design but requires careful causal assumptions and methods to avoid biased decomposition.