Does job satisfaction explain why supportive leadership lowers turnover intention? Does that effect weaken for employees with long tenure? The first question asks about mediation and the second about moderation. Mediation analysis in SPSS used to mean running three regressions and applying the Sobel test by hand. Today most theses use the PROCESS macro by Andrew Hayes, which estimates the whole model at once and tests the indirect effect with bootstrap confidence intervals.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.
This guide explains the difference between the two ideas, which PROCESS model fits which question, the steps in SPSS, how to read the output, and how to write the results.
What are mediation and moderation? Mediation asks how or why X affects Y: part of the effect passes through a third variable M (X → M → Y), and the quantity tested in mediation analysis in SPSS is the indirect effect, the product of path a (X → M) and path b (M → Y). Moderation asks when or for whom X affects Y: the strength or direction of the X → Y relationship changes with the level of a moderator W, tested through the X × W interaction term.
Mediation or moderation: which one is your question?
The wording of the hypothesis decides. “Satisfaction mediates the relationship between leadership and turnover intention” is mediation. “Tenure moderates the relationship between leadership and turnover intention” is moderation. Both can be combined in one model, called moderated mediation, when the indirect effect itself depends on a moderator.
| Feature | Mediation | Moderation |
|---|---|---|
| Question | How or why does X affect Y? | When, or for whom, does X affect Y? |
| Third variable | Mediator M, caused by X | Moderator W, not caused by X |
| Key test | Indirect effect a × b with a bootstrap CI | X × W interaction coefficient |
| PROCESS model | Model 4 (simple mediation) | Model 1 (simple moderation) |
| Typical figure | Path diagram with a, b, c’ | Simple slopes plot |
Two other models appear often in theses: Model 7, where W moderates the X → M path (first-stage moderated mediation), and Model 14, where W moderates the M → Y path (second-stage moderated mediation). PROCESS numbers its models according to the templates in Hayes (2022), and the model number is what you report in the methodology chapter.
Why bootstrapping replaced the Baron and Kenny steps
Baron and Kenny (1986) proposed a sequence of conditions: X must predict Y, X must predict M, M must predict Y controlling for X, and the X effect must shrink. Later work showed that this sequence has low power and that the first condition is not necessary: an indirect effect can exist even when the total effect is not significant (Hayes, 2009; Zhao, Lynch, & Chen, 2010). The Sobel test, which assumes the product a × b is normally distributed, is also weaker than resampling methods (Preacher & Hayes, 2008).
Current practice for mediation analysis in SPSS is therefore to estimate the indirect effect directly and judge it with a percentile bootstrap confidence interval. If the 95% interval does not include zero, the indirect effect is considered statistically different from zero.
📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.
How do you install and run the PROCESS macro?
The steps below are the standard route for mediation analysis in SPSS and for moderation alike.
- Download PROCESS from Hayes’s site (afhayes.com). The current release requires SPSS 26 or later and runs on Windows and Mac. The archive contains the custom dialog file and documentation.
- Install the custom dialog as the documentation describes. In recent SPSS versions this is done through Extensions, then Utilities, then Install Custom Dialog (Compatibility mode). Installation may need administrator rights.
- Open the dialog under Analyze, then Regression, then PROCESS.
- Move the outcome to Y, the predictor to X, and the mediator to the Mediator(s) M box, or the moderator to the Moderator W box.
- Choose the model number: 4 for mediation, 1 for moderation.
- Set the number of bootstrap samples to at least 5,000 and keep the confidence level at 95%.
- Under Options, request the total effect model for mediation, and for moderation tick mean-centering for the product terms if you want interpretable lower-order coefficients, plus the Johnson-Neyman output if W is continuous.
- Click OK and save the output. Bootstrap results change slightly between runs unless you set a seed, so record the seed you used.
How do you read Model 4 output for mediation analysis in SPSS?
The output is printed as a series of regression models.
- Outcome variable: M. The coefficient of X here is path a.
- Outcome variable: Y. The coefficient of M is path b, and the coefficient of X is c’, the direct effect controlling for M.
- Total effect model (if requested). The coefficient of X is c, the total effect. In ordinary least squares c = c’ + a × b.
- Total, direct and indirect effects of X on Y. The indirect effect row gives the point estimate (Effect), its bootstrap standard error (BootSE) and the interval (BootLLCI to BootULCI). This is the line that answers the mediation hypothesis.
The terms “full” and “partial” mediation are discouraged by Hayes because they depend on sample size; report the indirect and direct effects with their intervals instead and describe them.
A worked scenario of mediation analysis in SPSS
The numbers below are an illustrative scenario, not data from a real study. A sample of 250 employees; X = supportive leadership, M = job satisfaction, Y = turnover intention, all scale means from 1 to 5.
| Path | Coefficient | SE | p |
|---|---|---|---|
| a: leadership → satisfaction | 0.52 | 0.06 | < .001 |
| b: satisfaction → turnover intention | −0.41 | 0.07 | < .001 |
| c’: direct effect | −0.12 | 0.07 | .086 |
| c: total effect | −0.33 | 0.06 | < .001 |
| Indirect effect a × b | −0.21 | Boot SE 0.05 | 95% CI [−0.31, −0.12] |
APA sentence: “A simple mediation analysis (PROCESS Model 4, 5,000 bootstrap samples) showed that supportive leadership was indirectly related to turnover intention through job satisfaction, ab = −0.21, 95% CI [−0.31, −0.12]. Leadership predicted higher satisfaction (a = 0.52, p < .001), which predicted lower turnover intention (b = −0.41, p < .001). The direct effect was not significant, c’ = −0.12, p = .086.”
How do you read Model 1 output for moderation?
The Y model contains X, W and the product term Int_1 (X × W). Its coefficient and p value test the moderation hypothesis, and the line “Test(s) of highest order unconditional interaction(s)” gives the R² change due to the interaction.
If the interaction is significant, probe it:
- Conditional effects of X at values of W. For a continuous moderator PROCESS uses the 16th, 50th and 84th percentiles by default; you can request the mean and plus or minus one standard deviation instead.
- Johnson-Neyman output, which gives the value of W where the effect of X crosses the significance threshold.
- A simple slopes plot. PROCESS can print the data for the plot, which you then chart in SPSS or Excel.
Mean-centering does not change the interaction test. It only makes the coefficients of X and W refer to the average level of the other variable, which is easier to explain.
Reporting mediation and moderation in a thesis
When you report mediation analysis in SPSS, the methodology chapter names the macro, its version, the model number, the bootstrap method and number of samples, and the treatment of covariates. The results chapter gives a path diagram with unstandardized coefficients for mediation, a coefficient table for moderation, and a plot of simple slopes. Report unstandardized effects with their intervals; add the completely standardized indirect effect if your field expects a standardized size.
Mediation analysis in SPSS with cross-sectional data does not prove causation. The model assumes a causal order that comes from theory and design, so the discussion chapter should state this limitation plainly.
Common mistakes in mediation analysis in SPSS
- Claiming no mediation because the total effect c is not significant.
- Reporting the Sobel test as the main evidence instead of the bootstrap interval.
- Using a variable as a moderator when it is clearly caused by X, or as a mediator when it is measured before X.
- Interpreting X and W coefficients in Model 1 as “main effects” without centering.
- Running mediation on single items instead of scale scores whose reliability you have reported.
- Forgetting that bootstrap intervals vary slightly between runs and not recording the seed.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
Is the PROCESS macro free?
Yes. Hayes distributes PROCESS for SPSS, SAS and R at no cost from his website. The documentation is in Appendix A of his book, so most theses cite the book together with the macro version.
What sample size does mediation analysis in SPSS need?
It depends on the expected sizes of paths *a* and *b*. Simulation work by Fritz and MacKinnon (2007) showed that small paths can require several hundred cases for adequate power with bootstrap tests, while medium paths need far fewer. A Monte Carlo power analysis is the most defensible justification.
Can I include covariates in PROCESS?
Yes. In mediation analysis in SPSS with PROCESS, variables entered in the Covariate(s) box are included in all the models by default. Report them in the methodology and explain why each was controlled.
Can PROCESS handle two mediators?
Model 4 accepts several parallel mediators and reports each specific indirect effect and their contrasts. Model 6 handles serial mediators, where M1 affects M2.
Should I use PROCESS or structural equation modeling?
PROCESS works with observed scale scores and suits most master’s theses. If your model has latent constructs with measurement error, many mediators or a full measurement model, structural equation modeling in AMOS or SmartPLS is more appropriate.
Sources
- Hayes, A. F. (2022). Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach (3rd ed.). Guilford Press.
- Hayes, A. F. The PROCESS macro for SPSS, SAS, and R: download page. https://www.afhayes.com/download.html
- Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research. Journal of Personality and Social Psychology, 51(6), 1173–1182. https://doi.org/10.1037/0022-3514.51.6.1173
- Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40(3), 879–891. https://doi.org/10.3758/BRM.40.3.879
- Hayes, A. F. (2009). Beyond Baron and Kenny: Statistical mediation analysis in the new millennium. Communication Monographs, 76(4), 408–420. https://doi.org/10.1080/03637750903310360
- Zhao, X., Lynch, J. G., & Chen, Q. (2010). Reconsidering Baron and Kenny: Myths and truths about mediation analysis. Journal of Consumer Research, 37(2), 197–206. https://doi.org/10.1086/651257
- Fritz, M. S., & MacKinnon, D. P. (2007). Required sample size to detect the mediated effect. Psychological Science, 18(3), 233–239. https://doi.org/10.1111/j.1467-9280.2007.01882.x
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.