Mediation vs Moderation Analysis Explained With Dissertation Examples

By Dr. David C., PhD in Biostatistics. Reviewed by Dr. Lisa H.. 2026-05-18.

Get expert help with this chapter | University guides | Dissertation writing services

Key takeaways

• Mediation asks 'how' (mechanism). Moderation asks 'when' (boundary condition).
• Use PROCESS Model 4 for simple mediation, Model 1 for simple moderation.
• Bootstrap confidence intervals are the modern standard, not Baron and Kenny.
• Mean-center continuous predictors before creating interaction terms.
• Cross-sectional mediation cannot establish causality; acknowledge this limitation.

The plain-English distinction

Mediation asks 'how does X affect Y?' by inserting a third variable M that explains the mechanism. X causes M, which causes Y.

Moderation asks 'when, or for whom, does X affect Y?' by inserting a third variable W that changes the strength or direction of the X to Y relationship.

Mediation is about mechanism. Moderation is about boundary conditions.

Choosing the right model

Use mediation when your theory predicts a causal chain: X causes M causes Y.

Use moderation when your theory predicts that the relationship between X and Y depends on a third factor: X x W predicts Y.

Use moderated mediation when both apply: X causes M causes Y, and the strength of one path depends on W.

Mediation example (dissertation)

Research question: Does academic self-efficacy (M) mediate the relationship between parental involvement (X) and student GPA (Y)?

Run in PROCESS Model 4. Report the total effect (c), direct effect (c'), indirect effect (a*b), bootstrapped 95 percent confidence interval for the indirect effect, and effect size (kappa squared or proportion mediated).

APA 7 template: 'Bootstrap analysis (5,000 resamples) revealed a significant indirect effect of parental involvement on GPA through academic self-efficacy, b = .15, 95% CI [.08, .23]. The direct effect of parental involvement on GPA remained significant after accounting for the mediator, b = .22, p = .003, indicating partial mediation.'

Moderation example (dissertation)

Research question: Does gender (W) moderate the relationship between workplace stress (X) and burnout (Y)?

Run in PROCESS Model 1. Mean-center continuous predictors before creating interaction terms. Report the main effects, the interaction term, and the simple slopes at high and low values of the moderator.

APA 7 template: 'The interaction between workplace stress and gender was significant, b = .18, t(196) = 2.45, p = .015, R-squared change = .02. Simple slope analyses indicated that workplace stress predicted burnout more strongly for women (b = .42, p < .001) than for men (b = .24, p = .003).'

Tools that handle both

PROCESS Macro (Hayes): runs 92 models for mediation, moderation, and moderated mediation. Free SPSS and SAS plug-in. Industry standard for US dissertations.

lavaan in R: best for complex mediation and structural equation modeling. Free.

Mplus: gold standard for latent variable mediation and longitudinal models. Paid license.

Sample size requirements

Simple mediation (PROCESS Model 4): minimum 80 to 100 cases for detection of medium effects with bootstrapping.

Simple moderation: minimum 80 to 100; interaction effects need more power than main effects.

Moderated mediation: minimum 150 to 200 depending on model complexity.

Always run an a priori power analysis (G*Power or simr in R) before defending the design at your proposal stage.

Common mistakes

Confusing mediation with moderation in the writeup.

Reporting Baron and Kenny steps without bootstrapping (outdated since Hayes, 2018).

Forgetting to mean-center continuous predictors before creating interaction terms.

Drawing causal language from cross-sectional mediation data (mediation implies causality but cross-sectional design cannot establish it).

How DissertlyPro handles mediation and moderation

Our biostatisticians design the model, run PROCESS or lavaan, produce APA 7 tables and figures, and write the methodology and results narrative in your voice. Every engagement includes a methodology memo defending the model choice and bootstrap settings.

Need a statistician on your committee timeline? Book a free 20-minute call with a PhD biostatistician on the DissertlyPro team.

FAQ

Can I run both in the same dissertation?

Yes, often called moderated mediation. PROCESS Models 7, 14, and 15 handle the most common configurations.

Is PROCESS free?

Yes, free from Andrew Hayes' website for SPSS and SAS.

Do I need a longitudinal design for mediation?

Strictly, yes, to make causal claims. Cross-sectional mediation is common in dissertations but must be acknowledged as a limitation.

What sample size do I need for mediation?

Minimum 80 to 100 for simple mediation; 150 to 200 for moderated mediation. Run an a priori power analysis.