Moderation (Interaction) Analysis
Also known as: interaction analysis, moderated regression, simple moderation, Düzenleyici Değişken Analizi (Moderation / İnteraksiyon)
Moderation analysis tests whether the effect of a predictor X on an outcome Y changes with the level of a third variable W, the moderator. It is estimated within a regression framework through an interaction term X×W, popularised by Aiken & West (1991) and Hayes's PROCESS macro (2018).
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use moderation analysis when you suspect that the strength or direction of an X→Y relationship is conditional on a third variable, and you have a reasonably large sample (a rule of thumb of at least about 80 observations, more with several predictors). The moderator should be measured with little error, the interaction term must be formed from mean-centred variables, and multicollinearity (VIF) needs to stay under control. With too few observations the interaction test is underpowered and Type II error risk is high.
Strengths & limitations
- Directly tests the 'it depends' hypothesis, revealing for whom or under what conditions an effect holds.
- Fits naturally inside the familiar regression framework and is interpretable via simple slopes.
- The Johnson-Neyman technique pinpoints the exact moderator range where the effect of X is significant, rather than only probing a few arbitrary points.
- Detecting an interaction term needs more statistical power than a main effect, so small samples (n < 80) carry a high Type II error risk.
- Adding the product term can sharply inflate multicollinearity (VIF), distorting the coefficients unless variables are centred.
- Measurement error in the moderator attenuates and biases the estimated interaction.
Frequently asked
Why should I mean-centre X and W before building the interaction?
The product X×W is highly correlated with its components on the raw scale, which inflates the variance inflation factor (VIF) and destabilises the coefficients. Mean-centring X and W reduces this collinearity and keeps the lower-order coefficients interpretable at the variables' means.
What does the Johnson-Neyman technique add?
Instead of only checking the effect of X at a few moderator values (-1SD, mean, +1SD), the Johnson-Neyman technique identifies the entire range of the moderator over which the effect of X is statistically significant, giving a continuous picture of the conditional relationship.
How large a sample do I need?
Interaction terms are underpowered relative to main effects. A common rule of thumb is at least about 80 observations, growing with the number of predictors (roughly n ≥ 80 + 8k). Below that, a non-significant interaction may simply reflect low power rather than a true absence of moderation.
How is moderation different from mediation?
Moderation asks whether an effect changes with a third variable (when or for whom it holds), tested via an interaction term. Mediation asks how an effect is transmitted through an intervening variable. When both are combined in one model, the analysis becomes conditional process analysis (moderated mediation).
Sources
- Hayes, A. F. (2018). Introduction to Mediation, Moderation, and Conditional Process Analysis (2nd ed.). Guilford Press. ISBN: 978-1462534654
- Aiken, L. S. & West, S. G. (1991). Multiple Regression: Testing and Interpreting Interactions. Sage Publications. ISBN: 978-0761907121
How to cite this page
ScholarGate. (2026, June 1). Moderation (Interaction) Analysis. ScholarGate. https://scholargate.app/en/causal-inference/moderation-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Causal Mediation AnalysisCausal inference↔ compare
- Conditional Process AnalysisCausal inference↔ compare
- Logistic RegressionResearch Statistics↔ compare
- OLS RegressionEconometrics↔ compare
- Panel Fixed EffectsEconometrics↔ compare