9.4 Take-Home Points

• An interaction variable represents moderation in a regression model also if the moderator is numerical.

• An interaction variable is the product of the predictor and moderator.

• The effect of the predictor in a model with an interaction variable does not represent a main or average effect. It is a conditional effect: The effect for cases that score zero on the moderator. The same applies to the effect of the moderator, which is the conditional effect for cases scoring zero on the predictor.

• The unstandardized regression coefficient of the interaction variable specifies the predicted change in the effect of the predictor on the dependent variable for a one unit increase in the moderator variable.

• We recommend to mean-center a numerical moderator and a numerical predictor that are involved in an interaction effect. Observations with a mean score on the moderator are a substantively interesting reference group.

• To interpret moderation, describe the effects (slopes, unstandardized regression coefficients) and visualize the regression lines for some interesting levels of the moderator, such as the mean and one standard deviation below or above the mean.