Modern management and social-science theses rarely stop at 'X affects Y'. Committees and journals expect you to explain the mechanism or the boundary conditions — which is exactly what mediation and moderation do. They're routinely confused; the distinction is one sentence each.
Mediation — the how
A mediator (M) carries the effect: X → M → Y. Training (X) improves job performance (Y) because it raises self-efficacy (M). Testing asks how much of X's effect on Y flows through M — the indirect effect. The modern standard is bootstrapping the indirect effect (thousands of resamples, confidence interval); if the CI excludes zero, mediation is supported. The old Baron & Kenny causal-steps method and the Sobel test alone are now considered outdated — expect reviewers to say so.
Moderation — the when
A moderator (W) changes the strength or direction of X → Y. Stress hurts performance more for employees with low social support: support moderates the stress–performance link. Statistically it's an interaction term (X×W) in regression; a significant interaction means the effect of X depends on W. You then probe it with simple-slopes analysis and plot it — the plot is usually the most convincing exhibit in the chapter.
How to actually run them
- PROCESS macro (Hayes) in SPSS — the workhorse: Model 4 for simple mediation, Model 1 for moderation, Models 7/14 for moderated mediation. Free, bootstrap-native, hugely documented.
- SmartPLS — mediation via bootstrapped specific indirect effects and moderation via interaction terms, inside your full PLS-SEM model; the usual choice in management research with latent constructs.
- AMOS — covariance-based SEM with bootstrapped indirect effects; standard when you're doing CB-SEM anyway.
Reporting that satisfies examiners
For mediation: report total, direct and indirect effects with bootstrap CIs, and say plainly whether mediation is full, partial or absent. For moderation: report the interaction coefficient, then simple slopes at low/high moderator levels, with the interaction plot. In both cases, interpret in theory language — the statistics support the story; they aren't the story.
Mediation and moderation hypotheses need to be theorised, not bolted on after data collection — 'we also tested…' invites the question why?. If your model is still forming, get the framework right first; our methodology mentoring and data analysis help cover exactly this.
Frequently asked
Can a variable be both a mediator and a moderator?+
In different models, yes — but not in the same relationship at once. Which role it plays must come from theory: does it transmit the effect (mediator) or condition it (moderator)?
What sample size do I need for mediation analysis?+
Bootstrap mediation gains power with larger samples; 150–200+ is a common comfort zone for simple models, more for moderated mediation. Run a power check rather than relying on rules of thumb — our sample size calculator helps.
What is moderated mediation?+
A model where the strength of an indirect (mediated) effect depends on a moderator — e.g. training raises performance via self-efficacy, but more strongly for junior staff. PROCESS Models 7 and 14 are the standard implementations.
phdguide's mentors are senior academics, former supervisors, statisticians and publication specialists with 25+ years of combined experience guiding MBA, MPhil and PhD scholars from topic to viva.
Ethical, compliant guidance: We provide academic support, mentoring, analysis, editing and structuring — not authorship. Your work stays compliant with university policies.