Mediation Analysis in SPSS with PROCESS Model 4

To test mediation in SPSS, the standard tool today is Andrew Hayes' PROCESS macro, Model 4. You enter X, Y and the mediator M, ask for 5,000 bootstrap samples, and look at one number: the indirect effect and its confidence interval. If the 95% bootstrap interval does not include zero, mediation is supported. This tutorial shows the full procedure, how to read each part of the output and how to write it up.

What mediation means

A mediator explains how or why X affects Y. For example, perceived value (X) may increase purchase intention (Y) because it improves attitude toward the product (M). The model has three paths:

• a: effect of X on M
• b: effect of M on Y, controlling for X
• c': direct effect of X on Y, controlling for M

The indirect effect is the product a × b. The total effect is c = c' + ab.

Why not use Baron and Kenny's steps?

The old causal steps approach required a significant total effect before testing mediation and relied on the Sobel test. Both are now considered outdated. An indirect effect can exist even when the total effect is not significant, and bootstrap intervals are more accurate than the Sobel test because the product ab is rarely normally distributed (Preacher & Hayes, 2008; Zhao et al., 2010).

Install PROCESS in SPSS

1. Download PROCESS for free from processmacro.org and unzip the file.
2. Open SPSS as administrator (on Windows, right-click and choose Run as administrator).
3. Go to Extensions > Utilities > Install Custom Dialog (Compatibility mode) and select the process.spd file.
4. PROCESS now appears under Analyze > Regression.

Run Model 4 step by step

1. Go to Analyze > Regression > PROCESS.
2. Move purchase intention (PI) to Y variable, perceived value (PV) to X variable and attitude (ATT) to Mediator(s) M.
3. Set Model number to 4.
4. Set Number of bootstrap samples to 5000 and keep the confidence level at 95%.
5. Click Options and tick Show total effect model. Tick Standardized coefficients if you also want standardised effects.
6. Add control variables (age, gender) in Covariate(s) if your model needs them.
7. Click OK.

Read the output

PROCESS prints several blocks. Here is what to look for in each one, with example values (N = 300).

Outcome variable: ATT (path a)

This regression shows the effect of X on M. In our example: a = 0.52, SE = 0.05, p < .001. Perceived value increases attitude.

Outcome variable: PI (paths b and c')

This regression includes both X and M as predictors of Y.

Total effect model (path c)

If you asked for it: c = 0.46, p < .001. Note that 0.46 = 0.21 + (0.52 × 0.48), which is a quick way to check the output.

Indirect effect(s) of X on Y

The interval [0.17, 0.34] does not include zero, so the indirect effect is significant. PROCESS does not give a p-value for the indirect effect, and you do not need one.

Full or partial mediation?

Zhao et al. (2010) propose clearer labels than "full" and "partial":

In our example, both ab and c' are significant and positive: complementary mediation.

How to report the results

Mediation was tested with PROCESS Model 4 (Hayes, 2022) using 5,000 bootstrap samples. Perceived value had a positive effect on attitude (a = 0.52, p < .001), and attitude had a positive effect on purchase intention (b = 0.48, p < .001). The indirect effect of perceived value on purchase intention through attitude was significant (ab = 0.25, 95% CI [0.17, 0.34]). The direct effect remained significant (c' = 0.21, p < .001), indicating complementary mediation (Zhao et al., 2010).

A path diagram with a, b and c' values makes the result easier to follow for reviewers.

FAQ

Can I test several mediators at once?

Yes. Add them all in the Mediator(s) box of Model 4 to test parallel mediators. For mediators in a chain, use Model 6.

Should I report unstandardised or standardised coefficients?

PROCESS reports unstandardised coefficients by default, and they are fine to report. If all variables use the same scale, many authors add the standardised values too.

Why are my bootstrap intervals slightly different each time?

Bootstrapping uses random resamples. Set a seed in the Options (for example 12345) to get identical results every time.

Does mediation prove causality?

No. With cross-sectional survey data, mediation is consistent with your causal model but does not prove it. Say so in your limitations.

References

• Hayes, A. F. (2022). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach (3rd ed.). Guilford Press.
• 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
• 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