METHOD GUIDE
Moderation Analysis in SPSS
Product Term, Centering and Conditional Effects
A moderator changes how strongly, or in which direction, a predictor relates to an outcome, and the evidence sits in a single product term. Centering, the interaction row and the effects at chosen moderator values are what a reader needs to see.
IN SHORT
What a Moderation Model Tests and What It Reports
A moderation model asks whether the relationship between a predictor and an outcome depends on a third variable, the moderator, and it tests this as the product of predictor and moderator in a regression. The product term is the evidence, and the effects of the predictor at particular moderator values describe what it means. To run moderation analysis in SPSS, install the PROCESS add-on by Andrew Hayes, choose model 1, assign the outcome, the predictor and the moderator, and add the Johnson-Neyman option where the moderator is continuous (labels and defaults vary by PROCESS version). OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, carries out and reviews this analysis through an in-house PhD-level statistician.
ANALYSIS SUPPORT
Analysis Support for This Method
When the question concerns a mechanism and not a single effect, the model changes, and the second card describes the extension that combines the two ideas.
Moderation Analysis
The interaction model fitted within your study design, with the scaling of the variables, the probing of the interaction and the plot agreed for your data.
Moderated Mediation
A moderator that changes the strength of an indirect route, for when the question is about a pathway that differs by group or by level.
BEFORE YOU RUN IT
Theory, Scaling and Power for the Product Term
A moderation test is an interaction in a regression, so its theory, its scaling and its power are decided before PROCESS is opened.
A Reason for the Moderator State beforehand why the effect should differ across the moderator and in which direction. Without a reason, the interaction is exploratory, and the report labels it that way.
Scale of the Moderator A continuous moderator keeps its information, while splitting it at the median discards information. A grouped moderator needs a coding that makes the reference group meaningful.
Power for the Interaction Interactions are usually harder to detect than main effects and often need larger samples, so the plan states how large an interaction the sample could reasonably find.
Range of the Moderator The product term says nothing about moderator values the sample never reached, so conditional effects are reported only within the observed range of the data.
IN SPSS
How to Run Moderation Analysis in SPSS
Most of these steps need the PROCESS add-on in IBM SPSS, though the same interaction can be fitted as a product term in an ordinary regression, and labels and defaults differ by version.
- Decide which variable is the focal predictor and which the moderator, since the product term is symmetric but the story told about it is not.
- Decide on mean centering in SPSS for the continuous predictor and moderator, by saving centered variables or through the option your PROCESS version offers.
- Open PROCESS from the menu where it was installed, and assign the outcome, the focal predictor and the moderator.
- Select model number 1, and request the conditional effects and the plotting data that your version provides.
- Check which moderator values the output uses for probing, and change them if they do not suit the variable.
- Paste the syntax so the settings stay on record, and read the interaction row before the conditional effects.
READING THE OUTPUT
Reading the Interaction Row and the Conditional Effects
PROCESS model 1 in SPSS prints one regression that contains a product term and then probes that term, so the rows that matter are the product term and the effects at chosen moderator values.
| Output element | What it tells you | What to check |
|---|---|---|
| Model summary | The R-squared and overall F test of the regression with predictor, moderator and product term | That a significant overall model is not mistaken for evidence of moderation |
| Coefficients for the predictor and the moderator | Each variable’s effect when the other equals zero | That zero is a meaningful value, which is why centering changes these rows but not the product term |
| Product term (labeled Int_1 in most versions) | The change in the predictor’s slope for a one-unit change in the moderator | That its interval is read with its p-value, and that its sign is put into words |
| Test(s) of highest order unconditional interaction(s) | The increase in R-squared due to the interaction, with an F test | That the size of the increase is reported and not only its significance |
| Conditional effects of the focal predictor at values of the moderator | The predictor’s effect at low, middle and high moderator values, with intervals | That the values used are named and lie inside the observed range |
| Johnson-Neyman output (on request, for a continuous moderator) | The moderator values at which the predictor’s effect crosses from nonsignificant to significant | That the region is described as a feature of this sample and not a fixed threshold |
| Data for visualizing the effect (on request) | Predicted outcome values at the chosen moderator levels | That the plot carries intervals and shows how much of the moderator’s range the sample covers |
REPORTING
Reporting the Interaction Before Its Probes
The order matters here: the product term and its interval come first, and the probed effects follow only when the interaction is supported.
Report the interaction first: product term b = ___, 95% CI [___, ___], with the increase in R-squared = ___ that it adds. Say whether the variables were centered and how, because the coefficients for the predictor and the moderator are conditional on that choice. If the interaction is supported, give the effect of the predictor at the stated moderator values as b = ___, 95% CI [___, ___], and describe the pattern in words, naming the values and the range of the data. Add a plot drawn with intervals. If the interaction is not supported, still give the product term with its interval, because a wide interval means the data could not settle the question, and avoid saying that the moderator has no effect.
COMMON MISTAKES
Interaction Results That Reviewers Query
Questions about a moderation result usually concern what the lower-order coefficients mean, how the plot was probed and how far the pattern extends.
Main Effects Read as Averages Without centering, the predictor's coefficient is its effect when the moderator is zero, which may be a value no participant has. It is not the average effect unless zero marks the mean.
Probing Values Chosen by Result Selecting the moderator values that give a significant effect after seeing the output is a form of searching. The values are set by the theory or by a stated rule, such as the mean and one standard deviation either side.
Lower-Order Terms Dropped Removing the predictor or the moderator because its coefficient is nonsignificant, while keeping the product term, changes what the product term tests. Both lower-order terms stay in the model beside the product.
Standardized Coefficient for the Product Term In an ordinary SPSS regression, the Beta printed for a product term standardizes the product variable itself, not the product of standardized variables, so it is not a standardized interaction. The unstandardized product term is reported, or the variables are standardized before the product is formed.
Several Moderators Tried Testing a list of possible moderators and reporting the one with a significant product term inflates the chance of a false finding. The moderators are named before the analysis or the search is disclosed.
Moderation Read as Causation A product term shows that the association differs across levels, and it does not show why. Causal wording needs a manipulated predictor and an argument about the moderator.
Interaction figure drafted? Independent Statistical Review checks the centering, the probing values and the wording of the pattern before you submit.
FREQUENTLY ASKED QUESTIONS
Questions About Moderation Output in PROCESS
The interaction appears in several places in the PROCESS output, and the first three answers explain how those places relate to each other.
Is the R-squared change test the same as the test of Int_1?
With one product term, yes: the F in the R-squared change row is the square of the t for Int_1, and the two p-values match. The R-squared change adds the size of the interaction in variance terms, which the coefficient alone does not show. When the moderator has more than two categories, PROCESS builds several product terms, and the F test then becomes the joint test of all of them, which no single row can give.
What does the Johnson-Neyman result add to the conditional effects?
It shows where along the moderator the predictor’s effect changes from nonsignificant to significant, which a handful of chosen values cannot show. The boundaries are estimates with uncertainty of their own and belong to this sample. They are reported with the range of the moderator observed, and a boundary that falls outside that range is not interpreted.
Does it matter which variable I call the predictor in PROCESS?
The product term and its test are the same whichever variable is named the predictor, and only the probing and the plot change. The conditional effects are computed for the focal predictor at values of the moderator, so swapping the roles gives different effects and a different figure. The role that matches your theory is chosen before the output is seen.
Would OnlineSPSS review the centering and plot of my interaction model?
Yes. An in-house PhD-level statistician can review the scaling of the variables, the product term, the probing values and the plot against your design, or fit the interaction model from the start. Where the model is settled and the reporting is what remains, Results Reporting & Interpretation picks up from there. Why the moderator should matter is your argument to make, and the conclusions are yours.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the dataset or the output you have, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Which Moderator Values
Belong in Your Plot?
Centering, probing values and the observed range decide what an interaction plot can honestly show, and they are easier to settle with a statistician before the figure reaches a draft.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014