Mediation
Also known as: Mediator, Indirect effect
The mechanism in between — how a cause produces its effect.
What it means
Mediation describes a causal chain in which an independent variable influences an outcome through an intervening variable, the mediator, that transmits part or all of the effect. Analyzing mediation decomposes a total effect into a direct path and an indirect path running through the mediator, addressing not just whether a cause works but how. Modern treatments stress that valid mediation analysis is fundamentally causal, requiring strong assumptions of no unmeasured confounding for both the cause-mediator and mediator-outcome links. Because the mediator is rarely randomized, naive mediation estimates from observational data are easily biased, a caution often underappreciated by the technique's many users.
Examples
Education raises income partly by increasing skills, so skills mediate the education-income relationship.
A wellbeing app lifts mood mainly by improving sleep, so sleep is the mediator — and an app that lifts mood without touching sleep must be working some other way.
A price cut raises sales partly because it pulls more shoppers to the shelf, not only because the product is cheaper: footfall carries part of the effect.
First described in Baron & Kenny (1986); causal framework by Imai, Keele & Tingley (2010).