When to use it
- •You expect an effect to hold more strongly for some people, or in some contexts, than others.
- •You have a theoretical reason for the interaction — fishing for significant interactions is a well-known route to unreplicable findings.
- •Your predictor, moderator and outcome are all measured continuously (or the moderator is categorical).
Assumptions to check first
- •The same assumptions as linear regression, since that is what is running underneath.
- •Predictor and moderator mean-centred (or standardised) before forming the product term, to keep the lower-order coefficients interpretable.
- •Enough power — interaction effects are typically much smaller than main effects and need larger samples to detect.
What to report
- •The interaction term's coefficient, with its standard error, t and p.
- •The change in R² attributable to the interaction.
- •Simple slopes at meaningful levels of the moderator — conventionally ±1 SD — each with its own significance test.
- •A plot of the interaction, which communicates the finding far better than the coefficient does.
APA 7 example
The interaction between workload and support was significant, B = −0.18, SE = 0.06, t = −3.02, p = .003, ΔR² = .04. Simple slopes showed workload predicted burnout strongly at low support (B = 0.44, p < .001) but not at high support (B = 0.08, p = .34).
Numbers are illustrative — the Advisor generates this sentence with your own results.
The mistake reviewers catch
Without centring, the "main effects" in a moderation model are conditional effects at zero on the other variable — which is often a value that cannot exist, like zero years of age.
Run it on your own data
Paste or upload your dataset — nothing leaves your browser. The Analysis Advisor checks the assumptions above, runs moderation, and drafts the APA Methods and Results text with your numbers in it.
Open the Analysis Advisor →Related analyses
Frequently asked questions
When should I use moderation?
Does the effect of X on Y depend on W? You expect an effect to hold more strongly for some people, or in some contexts, than others.
What do I need to report for moderation in APA 7?
The interaction term's coefficient, with its standard error, t and p. The change in R² attributable to the interaction. Simple slopes at meaningful levels of the moderator — conventionally ±1 SD — each with its own significance test. A plot of the interaction, which communicates the finding far better than the coefficient does.
What is the most common mistake with moderation?
Without centring, the "main effects" in a moderation model are conditional effects at zero on the other variable — which is often a value that cannot exist, like zero years of age.