METHOD GUIDE
Mediation Analysis in SPSS
The Indirect Effect, Its Interval and Its Limits
Mediation asks whether the effect of one variable on an outcome runs through a third. The design questions come first, then the PROCESS output that carries the answer and the causal wording your design can support.
IN SHORT
What a Mediation Model Estimates and What It Cannot Show
A mediation model asks whether part of the association between a predictor and an outcome is transmitted through an intermediate variable, the mediator, and it estimates that route as the product of two paths, called the indirect effect. The word mediation also implies an ordering in time and a causal story, which the data alone cannot supply. To run mediation analysis in SPSS, install the PROCESS add-on by Andrew Hayes, choose model 4, assign the outcome, the predictor and the mediator, and check the bootstrap settings that PROCESS applies to the indirect effect (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
Because the interval for an indirect effect rests on resampling, Bootstrapping is worth reading alongside the mediation and moderated mediation support described here.
Mediation Analysis
The mediation model fitted within your study design, with the ordering, the confounding questions, the indirect effect and its interval agreed for your data.
Moderated Mediation
An indirect route whose strength changes with a third variable, for when the mechanism is expected to differ across groups or contexts.
BEFORE YOU RUN IT
What Turns Three Variables Into a Mediation Model
PROCESS will estimate any three variables it is given, so whether the result deserves the name mediation is decided by the study and not by the output.
Order in Time The mediator should plausibly follow the predictor and precede the outcome. When all three are measured once, the order rests on theory, and the model describes association along a proposed path.
Confounding of the Second Path Randomizing the predictor does not randomize the mediator. Variables that influence both the mediator and the outcome bias the path between them, so measured confounders enter as covariates.
Quality of the Mediator Score Measurement error in the mediator weakens its path to the outcome and can distort the split between direct and indirect effects, so each variable is a sound score with reliability reported.
Precision for a Product of Paths A product of two coefficients has a wide interval in a small sample. The plan states the precision expected and does not count on resampling to repair a sample that is too small.
IN SPSS
How to Run Mediation Analysis in SPSS
Each step assumes the PROCESS add-on has been installed in IBM SPSS, since it does not ship with the program, and dialog labels and defaults differ by PROCESS and SPSS version.
- Name the roles: one outcome, one predictor, one mediator and any covariates, each scored in the direction the theory assumes.
- Open PROCESS from the menu where its installation placed it, and assign the outcome, the predictor and the mediator.
- Select model number 4, and add the covariates that apply to both equations.
- In the options, confirm the number of resamples and the interval type, and request a fixed seed if your version offers one.
- Paste the syntax instead of only pressing OK, so that every setting stays on record.
- Read the path from predictor to mediator and the path from mediator to outcome before the indirect effect.
READING THE OUTPUT
Which PROCESS Blocks Carry the Indirect Effect
PROCESS model 4 in SPSS prints a regression for the mediator and one for the outcome, then the direct and indirect effects (the total effect on request), and only the indirect effect is tested by resampling.
| Output element | What it tells you | What to check |
|---|---|---|
| Outcome variable: the mediator | The regression of the mediator on the predictor and covariates, whose predictor coefficient is the first path | That its sign agrees with the theory and that the interval is reported with it |
| Outcome variable: the final outcome | The regression of the outcome on the predictor and the mediator, which gives the second path and the direct effect | That covariates match those in the first equation unless a reason is stated |
| Total effect model (when requested in the options) | The effect of the predictor on the outcome with the mediator left out | That a nonsignificant total effect is not treated as a bar to examining the indirect effect |
| Direct effect of X on Y | The predictor’s effect once the mediator is in the model | That its significance alone is not used to call the mediation full or partial |
| Indirect effect(s) of X on Y | The product of the two paths, with a bootstrap standard error and the lower and upper limits (BootLLCI and BootULCI) | That an interval containing zero is read as imprecision about the effect and not as proof that none exists |
| Completely standardized indirect effect (on request, where your version offers it) | The indirect effect rescaled by the standard deviations of predictor and outcome | That the raw and standardized values are not mixed in one sentence |
| Notes on resamples and confidence level | The number of resamples and the level used for every interval | That both are copied into the methods text, with the seed where one was set |
REPORTING
Reporting the Paths and the Indirect Effect
Each blank is filled from your own PROCESS output, and the closing sentence is set by the design, not by the width of the interval.
Report the path from predictor to mediator (a = ___, 95% CI [___, ___]), the path from mediator to outcome (b = ___, 95% CI [___, ___]) and the indirect effect (ab = ___, bootstrap 95% CI [___, ___]) as separate estimates. An indirect effect bootstrap in SPSS gives an interval from resamples of your own data, and the interval, not a p-value, carries the claim. State the number of resamples, the interval type and the covariates, and say whether the coefficients are unstandardized, which is what PROCESS prints by default. Add the direct and total effects with their intervals. Then match the wording to the design: with the three variables measured at one time point, the data are consistent with an indirect association through the mediator, and the text does not say that the predictor causes the outcome through it.
COMMON MISTAKES
Claims a Mediation Table Cannot Support
Most reviewer queries on a mediation result concern the causal reading of the arrows and the test of the indirect effect, not the arithmetic inside PROCESS.
Path P-Values as the Test Reading each path's p-value in turn is not a test of the indirect effect. A significant first path beside a nonsignificant second leaves the product undecided, so its own interval is reported.
Mediators Chosen After the Fact Trying candidate mediators one by one until an interval excludes zero makes the reported route a finding by search. The mediators are named in advance, and any added later are labeled exploratory.
Causal Verbs From One Wave Measuring all three variables at once supports association along a proposed path. Phrases such as leads to or works through need a design and an argument that justify them.
Unmeasured Confounders Ignored Covariates adjust only for what was measured. The remaining risk to the second path is stated, and a sensitivity analysis is run where one is available for the model.
Full Mediation Declared A nonsignificant direct effect can reflect limited power and does not show that the mediator explains everything. The size of the remaining direct effect is given with its interval.
Resampling Settings Unreported Intervals from different numbers of resamples or different interval types can differ slightly, so the methods text gives the settings that were used.
PROCESS tables ready for a second reading? Independent Statistical Review reads the paths, the interval and the causal claims against your design before you submit.
FREQUENTLY ASKED QUESTIONS
Questions About the Mediation Output in PROCESS
After a first model 4 run, questions usually turn to the resampling, the interval columns and whether a second mediator fits.
Why does my indirect effect change slightly each time I rerun PROCESS?
Each run draws new random resamples of your data, so the interval limits move slightly unless a seed is fixed. The estimate of the indirect effect itself does not change, because it comes from the original data. More resamples narrow the run-to-run differences but do not make a small sample more informative. If a limit sits very close to zero and the conclusion flips between runs, the result is borderline, and the text should say so.
What do BootLLCI and BootULCI mean in the PROCESS output?
They are the lower and upper limits of the interval for the indirect effect, obtained by resampling your data. If the interval excludes zero, the data support an indirect effect in that direction at the stated confidence level. If it includes zero, the effect is imprecise or absent, and the data cannot say which. A percentile interval is the default in recent versions and other interval types may be offered, so the type is named in the report.
Can I enter more than one mediator in model 4?
Yes, model 4 accepts several mediators in parallel and prints a specific indirect effect for each, plus the total indirect effect. Mediators that correlate share variance, so each specific effect is unique to that mediator given the others. If theory says one mediator influences the next, a serial model (PROCESS model 6) matches it better. Every added mediator adds paths and asks more of the sample.
Can OnlineSPSS check the resampling settings and wording of my indirect effect?
Yes. An in-house PhD-level statistician can review the specification, the covariates, the resampling settings and the causal wording against your design, or fit the model on your data from the start. When the estimates are final and only their presentation is open, Results Reporting & Interpretation is the stage that fits. The argument for the mediator, and the conclusions drawn from it, 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
Can Your Design Carry
a Mediation Claim?
A mediation claim rests on when and how the three variables were measured as much as on the bootstrap interval, so the design and any PROCESS output are considered together from the first exchange.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014