To test whether the relationship between an independent variable (IV or X) and a dependent variable (DV or Y) changes depending on the level of a third variable, the moderator (W). In other words, moderation answers the question: "Does the effect of X on Y depend on W?"
This is also called an interaction effect. The moderator does not explain how X affects Y (that's mediation) , but it tells you under what conditions or for whom the effect is stronger, weaker, or reversed.
Note: Mediation and moderation are often confused. A mediator (M) lies in the causal chain between X and Y. A moderator (W) sits outside that chain and changes its strength.
You have ONE continuous or categorical IV (X), ONE continuous DV (Y), and ONE moderator (W) that is continuous or categorical.
You have a theoretical reason to expect that the effect of X on Y is not the same across all levels of W.
Do NOT use moderation if: You do not have a theory-driven reason to expect the effect of X on Y to differ by W, your sample size is small (aim for N ≥ 200), you want to know why the effect of X on Y occurs, not when (use mediation instead), you have more than one DV (use MANOVA or SEM).
Before running a test, four assumptions should be met:
Linearity: The relationship between each predictor and the DV should be linear.
Independence of residuals: Observations are independent of each other.
Homoscedasticity: Residual variance is constant across all levels of the predictors.
Multicollinearity: The IV, moderator, and their interaction term should not be excessively correlated. Note: Some correlation between X and the interaction term (X × W) is expected and normal — this is why centering variables before computing the interaction term is recommended (see below).
Before running the analysis, you should mean-center your continuous IV and moderator to reduce multicollinearity and make the intercept interpretable. This is done by clicking "data" → "Compute" then create a new variable: X_c = X - MEAN(X). Repeat for W: W_c = W - MEAN(W)
Click "Modules" in the upper right-hand corner
Click "jamovi library"
Search for "medmod" under the "Available" tab and click "Install"
Click "medmod"
Click "Moderation"
Move the DV into the "Dependent Variable" box
Move the IV (X or X_c) into the "Predictor" box
Move the moderator (W or W_c) into the "Moderator" box
Under "Estimates," check "Labels," "Test statistics," and "Confidence interval"
Under "Simple Slopes," check "Simple slopes plot"
Interaction term (X × W): Tests whether the slope of X predicting Y significantly differs across levels of W.
If p < .05 for the interaction term → the moderation is significant. The effect of X on Y does depend on W.
If p > .05 for the interaction term → the moderation is not significant. The relationship between X and Y does not significantly change across levels of W.
Simple slopes: When the interaction is significant, you need to interpret the effect of X on Y at specific levels of W (typically: low W = −1 SD, mean W, high W = +1 SD). These are called simple slopes.
If the simple slope is significant at high W but not low W → X predicts Y only when W is high
If the simple slope reverses direction across levels of W → this is a crossover interaction, the strongest form of moderation
Simple slopes plot: This is the most intuitive way to interpret and communicate moderation results. The plot shows the regression line of X predicting Y drawn separately for low, medium, and high levels of W. Non-parallel lines indicate an interaction. Always include this plot in your results.
β (standardized estimate): The standardized slope for each predictor and the interaction term. Larger absolute values = stronger effect.
R²: The proportion of variance in Y explained by the full model (X + W + X×W). Interpret the same as in regression.
Appropriate data visualization: Simple slopes plot
Sample table: Report the full regression model (including X, W, and X×W) in a regression table format.
Sample write-up: A moderated regression analysis examined whether social support (W) moderated the relationship between perceived stress (X) and depressive symptoms (Y). After mean-centering the predictor and moderator, the interaction term (stress × social support) was entered into the regression model alongside the main effects. The overall model was significant, F(3, 196) = 12.43, p < .001, R² = .16. The interaction term was significant, b = −0.31, t(196) = −2.87, p = .004, 95% CI [−0.52, −0.10], indicating that social support moderated the association between stress and depression. Simple slopes analysis revealed that the effect of stress on depression was significant at low social support (b = 0.54, p < .001) and at mean social support (b = 0.29, p = .012), but not at high social support (b = 0.04, p = .731). These results suggest that the effect of stress on depressive symptoms weakens as social support increases.
Note: Replace bracketed values with your own output. Always report b (unstandardized), t, df, p, and 95% CI for the interaction term, plus simple slopes at all three levels of W.