Mediation and Moderation with PROCESS in SPSS: A Guide for Your Thesis
The difference between a mediator and a moderator, how to run Model 4 and Model 1 in Hayes' PROCESS, how to read the indirect effect and the interaction, and how to report them in APA.
Part of the series: Statistical Analysis for a Seminar Paper or Thesis: The Complete Guide
"Test whether X affects Y through M" or "whether the X-Y relationship depends on W" - these two questions appear in almost every thesis and many advanced seminar papers. The first is mediation, the second moderation. The standard tool for both is Andrew Hayes' PROCESS, a free SPSS add-on.
Mediator vs moderator in one line
- Mediator (M): the mechanism - why or how X affects Y. "Social support reduces burnout by increasing self-efficacy."
- Moderator (W): the condition - when or for whom the relationship is stronger. "Workload predicts burnout more strongly among employees with low support."
| Mediator | Moderator | |
|---|---|---|
| Question | how? why? | when? for whom? |
| PROCESS model | Model 4 | Model 1 |
| What you test | indirect effect (a × b) | interaction (X × W) |
Installing
Download PROCESS from Hayes' site (processmacro.org); in SPSS: Extensions → Utilities → Install Custom Dialog, and pick the .spd file. It then appears under Analyze → Regression.
Mediation - Model 4
Dialog: Y, X, the mediator under Mediator(s), Model number = 4. Under Options tick Show total effect model and Effect size. Keep the default 5,000 bootstrap samples.
Reading the output: 1. Path a (X → M): the coefficient in the first table, where M is the Outcome. 2. Path b (M → Y, controlling for X) and the direct effect c' (X → Y): in the table where Y is the Outcome. 3. Indirect effect at the end: a × b with its bootstrap confidence interval (BootLLCI to BootULCI).
The key rule: mediation is significant if the confidence interval does not include 0. There's no p-value for the indirect effect. If c' is also significant - partial mediation; if not - full mediation (many supervisors no longer insist on the distinction, but say which).
Moderation - Model 1
Dialog: Y, X, the moderator under Moderator variable W, Model number = 1. Under Options: Mean center for construction of products (easier interpretation) and Generate code for visualizing interactions.
Reading the output: 1. Int_1 (X × W): if significant, there's moderation. 2. R²-chng on the interaction row: how much variance the interaction adds. 3. Conditional effects of X at values of W: the effect of X on Y at low, mean and high W (mean ± SD). This is where you see the direction - where the relationship is strong and where it fades.
Plot: the code PROCESS generates draws one line per level of W. A plot is close to mandatory for a moderation finding.
Reporting in APA
Mediation: "There was a significant indirect effect of social support on burnout through self-efficacy, b = −0.14, 95% CI [−0.23, −0.07]. Support predicted higher self-efficacy (a = 0.41, p < .001), which predicted lower burnout (b = −0.35, p < .001). The direct effect remained significant (c' = −0.19, p = .012), indicating partial mediation."
Moderation: "Workload and support interacted significantly in predicting burnout, b = −0.18, p = .004, ΔR² = .04. Workload was strongly related to burnout at low support (b = 0.52, p < .001) and more weakly at high support (b = 0.17, p = .048)."
Common mistakes
- Mixing the terms - calling it "mediation" when the hypothesis is about a condition.
- Using Baron and Kenny's four steps as the only approach - outdated; the bootstrapped indirect effect is the standard.
- Looking for a p-value for the indirect effect instead of the confidence interval.
- Reporting moderation without conditional effects or a plot.
- Reading mediation as causation in a one-time correlational study. You may say the findings are consistent with a mediation model.
FAQ
Both a mediator and a moderator? That's moderated mediation - Model 7 or 14 in PROCESS, for example. More complex; plan it with your supervisor.
How many participants? For bootstrapped mediation, 100+ is usually enough for medium effects. See sample size and power.
Where does it fit? It extends regression - see Statistical analysis for a seminar paper: the complete guide. Supervisor asked for a PROCESS model you never learned? Tell us - we'll explain every line of the output.
Tell us what you need, and we'll come back with an expert in Statistical analysis and a quote.
Last updated: