Statistics Glossary → Moderator variable
Moderator variableconcept
A third variable that changes the direction or strength of the relationship between two variables; its effect is usually modelled with an interaction term.
A moderator W explains 'when' or 'for whom' the effect of X on Y is stronger. In the regression model Y = b₀ + b₁X + b₂W + b₃(X·W), a significant b₃ means that the effect of X depends on the value of W: the conditional effect of X is b₁ + b₃W. To ease interpretation, conditional effects are reported at chosen values of W (e.g. the mean and ± 1 standard deviation, or the 16th, 50th and 84th percentiles); the Johnson-Neyman technique finds the range of W in which the effect is significant. In meta-analysis, study characteristics are examined as moderators of the effect size.
A moderator should not be confused with a mediator: a mediator explains 'how' X affects Y (the path X → M → Y), while a moderator explains 'under which conditions' that effect changes. The two can appear together in moderated mediation models, where one of the mediation paths depends on a moderator (such as Hayes's Model 7 and Model 14). Testing interaction terms usually needs a much larger sample than testing main effects, and when an interaction is present the main-effect coefficients must be read as conditional effects at the value zero of the other variable.
An everyday example
The effect of coffee on attention is not the same for everyone: it may give a clear boost to someone who slept little and make almost no difference to someone well rested. Here, hours of sleep is a moderator: it changes the strength of the relationship between coffee and attention. If coffee affected sleep and changed attention through it, sleep would instead be a mediator.
In MerQur
In Mediation Analysis the 'Moderator (W)' and 'Moderation model (Model 7 / Model 14)' parameters compute the conditional indirect effect and the index of moderated mediation. In Meta-Analysis numeric or categorical moderators are used through the 'Meta-regression moderators' and 'Subgroup (moderator)' parameters.
Analyses where this concept appears
Meta-AnalysisMediation Analysis
Related concepts
Interaction effectIndirect effect (mediation)Meta-regression