TL;DR — Quick Answer
Mediation and moderation answer two different questions about a relationship between variables. Mediation asks how or why X affects Y — it identifies a third variable (M) that carries the effect, so X influences M, and M in turn influences Y. Moderation asks when or for whom X affects Y — it identifies a third variable (W) that changes the strength or direction of the relationship, tested through an interaction term. A mediator sits between cause and effect; a moderator sits alongside it. Mediation is now tested using bootstrapped confidence intervals for the indirect effect rather than the older Baron and Kenny causal-steps method, which has been widely criticised for low statistical power. Moderation is tested by centring your variables, creating an interaction term, and examining whether it is significant — then probing it with simple slopes. Both are readily run in SPSS using Hayes’s PROCESS macro (Model 4 for simple mediation, Model 1 for simple moderation) or within a structural equation model.
MK — Dr. Madhuri Kanojiya, Founder & Director · Empire Research Press
There is a moment in most quantitative research projects when a simple finding stops being satisfying. You have established that training influences employee performance, or that perceived usefulness influences technology adoption. The relationship is significant. And then your supervisor asks the question that turns a finding into a contribution: why does it work, and for whom does it work best?
Those two questions have precise statistical answers. The first is mediation. The second is moderation. Confusing them — or worse, testing one while describing the other — is among the most common analytical errors in submitted theses, and it is entirely avoidable once the underlying logic is clear.
This guide explains what each does, how to test them correctly using current methods, how to interpret and report the results, and where researchers most often go wrong.
The Core Distinction
Start with a basic relationship: X influences Y.
Mediation introduces a variable M that transmits the effect. X influences M, and M influences Y. The mediator is a link in the causal chain — remove it and the mechanism disappears.
An example: leadership style (X) influences employee performance (Y) through job satisfaction (M). The leader’s behaviour raises satisfaction, and satisfaction raises performance. Job satisfaction explains the mechanism.
Moderation introduces a variable W that changes the relationship without being part of it. X still influences Y, but the strength or direction of that influence depends on the level of W.
An example: training (X) influences performance (Y), but the effect is stronger for employees with high motivation and weaker for those with low motivation. Motivation (W) does not carry the effect — it conditions it.
A memory aid that holds up well: a mediator is a bridge — the effect travels across it. A moderator is a dial — it turns the effect up or down.
Mediation: The Vocabulary
Four terms recur throughout mediation reporting, and using them precisely will improve your methodology chapter immediately.
- Total effect (c) — the overall influence of X on Y before the mediator is introduced.
- Direct effect (c′) — the influence of X on Y that remains once M is in the model.
- Indirect effect (a × b) — the portion of the effect that travels through M. Path a is X→M, path b is M→Y, and their product is the indirect effect. This is the mediation.
- Total effect = direct effect + indirect effect.
Two further distinctions matter when writing up:
Full versus partial mediation. If the direct effect becomes non-significant once M is included, the older literature calls this full mediation; if it remains significant but reduced, partial mediation. Current methodological opinion increasingly discourages this language, because whether c′ reaches significance depends heavily on sample size rather than on the strength of the mechanism. It is more defensible to report the size and significance of the indirect effect and describe what it means substantively.
Competitive and complementary mediation. When the indirect and direct effects point in opposite directions, the mediation is competitive — sometimes called inconsistent mediation or a suppression effect. This is a genuine and interesting finding, not an error, and it deserves interpretation rather than being quietly omitted.
“A mediator answers the question your reader is already asking silently: yes, but how does that actually happen? A moderator answers a different one: does this hold for everyone? Researchers who can tell these two questions apart before they open their software write far better hypotheses — because the statistics are only ever as clear as the reasoning that preceded them.”
Dr. Madhuri Kanojiya — Empire Research Press
How to Test Mediation Properly
Mediation testing has changed substantially, and using the outdated approach is one of the fastest ways to attract reviewer criticism.
The older approach: Baron and Kenny’s causal steps
The classic method, from Baron and Kenny’s much-cited 1986 paper, required four conditions in sequence: X must significantly predict Y; X must significantly predict M; M must significantly predict Y while controlling for X; and the X→Y effect must weaken when M is included. A Sobel test was then used to assess the indirect effect.
This framework shaped decades of research, and it remains worth understanding because so much published work uses it. But it has two well-documented weaknesses. It has low statistical power, meaning genuine mediation frequently goes undetected. And the requirement that X must first significantly predict Y is now considered unnecessary — an indirect effect can exist even when the total effect is not significant, particularly with competing pathways or modest samples.
The current standard: bootstrapping the indirect effect
Contemporary practice tests the indirect effect (a × b) directly using bootstrapped confidence intervals. Bootstrapping repeatedly resamples your data — typically 5,000 times — to build an empirical distribution of the indirect effect without assuming that distribution is normal, which it generally is not.
The decision rule is straightforward: if the 95% bias-corrected confidence interval for the indirect effect does not include zero, mediation is supported. No separate significance test is required, and no preliminary condition about the total effect needs to be satisfied.
In SPSS this is done through Hayes’s PROCESS macro using Model 4 for simple mediation. In SEM software, the same logic applies with bootstrapping enabled — AMOS, lavaan in R, and SmartPLS all support it.
How to Test Moderation Properly
Moderation is tested through an interaction term, and the procedure has three stages.
Step one — centre your variables. Mean-centre (or standardise) both the independent variable and the moderator before creating the interaction term. This reduces multicollinearity between the interaction term and its components, and it makes the main effects interpretable at the mean of the moderator rather than at a meaningless zero point.
Step two — create and test the interaction. Multiply the centred X by the centred W to form the interaction term, then enter X, W, and X×W into the regression. If the interaction term is statistically significant, moderation is present. The change in R² attributable to the interaction indicates how much additional variance the moderation explains.
Step three — probe the interaction. A significant interaction tells you the relationship differs across levels of the moderator, but not how. Simple slopes analysis estimates the X→Y relationship at low (typically −1 SD), mean, and high (+1 SD) levels of W. Plotting these slopes is standard practice and makes the finding immediately legible to readers. The Johnson-Neyman technique goes further, identifying the exact values of the moderator at which the relationship becomes significant.
PROCESS Model 1 handles simple moderation, including simple slopes and Johnson-Neyman output.
When Both Appear in One Model
Many research models combine the two, and the terminology here is worth getting right.
Moderated mediation (conditional indirect effect) occurs when the strength of an indirect effect depends on a moderator — the mechanism works for some people and not others. It is assessed using an index of moderated mediation with a bootstrapped confidence interval; if that interval excludes zero, moderated mediation is supported.
Mediated moderation occurs when an interaction effect itself operates through a mediator. It is less commonly specified and requires careful theoretical justification.
PROCESS provides numbered templates for these more complex configurations — Model 7 and Model 14 are among the most frequently used for moderated mediation, depending on where the moderator enters the model. Choose the model number from your theoretical diagram, not by trying numbers until one produces significant results.
Reporting Your Results
For mediation, report: the path coefficients for a and b, the indirect effect with its bootstrapped 95% confidence interval, the direct effect, the total effect, the number of bootstrap samples used, and an effect size for the indirect effect where available. State the decision explicitly — that the confidence interval excluded zero, and what that means substantively.
For moderation, report: the main effects of X and W, the interaction coefficient with its significance, the R² change attributable to the interaction, the simple slopes at low, mean, and high levels of the moderator, and a plot of the interaction.
Throughout, describe results in terms of the relationships your theory proposed. A list of coefficients is a table; an interpretation is a contribution.
Common Mistakes
- Calling a moderator a mediator, or the reverse. The single most frequent error. Ask which question your variable answers — how (mediator) or when (moderator).
- Using only Baron and Kenny’s causal steps. Acceptable to cite for background, but bootstrapped confidence intervals should carry the analysis.
- Failing to centre variables before creating an interaction term. Produces inflated multicollinearity and uninterpretable main effects.
- Stopping at a significant interaction. Without simple slopes, you have not shown the nature of the moderation.
- Choosing a PROCESS model number by trial and error. The model must follow from your theoretical diagram.
- Claiming causal mediation from cross-sectional data. Mediation is a causal claim; a single-time-point survey cannot establish temporal order. State this as a limitation honestly.
- Adding mediators without theory. A variable is a mediator because your framework says it carries the effect — not because including it improves the model.
Quick-Reference Checklist
- Identify whether your question is how/why (mediation) or when/for whom (moderation)
- Justify the mediator or moderator theoretically before analysis
- Confirm your measures are reliable and valid — weak measurement distorts both
- For mediation: bootstrap the indirect effect with 5,000 resamples and bias-corrected CIs
- For moderation: centre variables, create the interaction, test it, then probe with simple slopes
- Select the PROCESS model number from your theoretical diagram
- Report indirect effects with confidence intervals, not just p-values
- Plot significant interactions
- Acknowledge design limitations on causal interpretation
Related Reading
This guide sits alongside the site’s statistics coverage: see Structural Equation Modelling Explained for testing these effects within a full latent-variable model, What Is Regression Analysis? for the underlying method, Correlation vs Causation for why mediation claims require careful design, What Is a Variable in Research? for the variable types involved, and What Is a Conceptual Framework in Research? for building the model you will test.
Sources
- 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.
- Hayes, A. F. Introduction to Mediation, Moderation, and Conditional Process Analysis. 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.
- 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.
Empire Research Press provides methodology consultation for researchers designing and testing mediation and moderation models — including model specification, PROCESS configuration, and reporting standards for thesis and journal submission.
About the Author
Dr. Madhuri Kanojiya is a researcher, author and educator with a PhD in Computer Science and Management. She is the Founder and Director of Empire Research Press — an independent international publisher and research consultancy based in Goa, India. She writes on research methodology, AI adoption, cloud computing, organisational systems and academic publishing.
- Google Scholar: https://scholar.google.com/citations?user=67en4MwAAAAJ&hl=en
- ORCID: https://orcid.org/0000-0003-1499-6112
- ResearchGate: https://www.researchgate.net/profile/Madhuri-Kanojiya