Glossary
Mediation
What is Mediation?
Mediation describes a proposed pathway through which one variable affects another via an intermediate variable. The mediator represents part of the explanation of how an effect arises.
Interpretation
A statistical association involving an intermediate measure does not, by itself, establish that measure as the causal mechanism. Imai, Keele, and Tingley’s causal framework separates effect definition, identification assumptions, estimation, and sensitivity analysis.
Example and limits
For example, a study might propose training → confidence → later performance. Training is the intervention or independent variable, confidence measured after training is the proposed mediator, and subsequent performance is the outcome or dependent variable. The indirect effect concerns the part of the training effect that operates through confidence; a direct effect concerns pathways not through that mediator.
This is a hypothesis about a pathway, not proof that any intervention that raises confidence will reproduce the training effect. Prior ability, for example, could influence both confidence and performance, complicating a causal reading of their association.
How mediation is analyzed
The historical Baron–Kenny approach checks a sequence of relationships among the intervention, mediator, and outcome, including whether the intervention–outcome relationship decreases after accounting for the mediator. An alternative is to estimate the indirect effect directly and report its uncertainty, rather than rely only on that sequence of significance tests.
Resampling methods such as bootstrapping can help estimate uncertainty under a chosen statistical model. They do not establish the causal assumptions. A causal mediation analysis must address timing, confounding of the mediator–outcome relationship, and how conclusions change when assumptions are weakened. Randomizing training alone does not randomize confidence.