Mediation analysis assesses whether the relationship between an independent variable and a dependent variable operates through a third variable, called the mediator. It helps researchers understand how or why an effect occurs. In mediation, the independent variable influences the mediator, which in turn influences the dependent variable. Mediation is different from an interaction effect: an interaction describes how the strength or direction of an effect changes depending on another variable, whereas mediation describes a causal pathway through another variable.
Researchers often test for mediation using a series of regression analyses. First, they show that the independent variable predicts the dependent variable. Next, they demonstrate that the independent variable predicts the mediator. Then, they show that the mediator predicts the dependent variable while controlling for the independent variable. If the effect of the independent variable on the dependent variable shrinks (or disappears) after accounting for the mediator, it suggests a mediation effect.
For example, a study might find that a new exercise program leads to lower stress levels. Mediation analysis could show that the exercise improves sleep quality (the mediator), and better sleep then reduces stress. The effect of exercise on stress would thus be explained partly or fully by its effect on sleep.
« Back to Glossary Index