Key points at a glance
- Justify X, mediator and Y.
- Check timing and confounders.
- Report the indirect effect and uncertainty.
Specify the mechanism you propose
Mediation asks whether an exposure X is related to outcome Y through an intermediate variable M. That is a possible mechanism, not merely three correlations. Draw the proposed paths before fitting software and justify them with theory or prior evidence. Define how and when each variable is measured.
Example: Study support X might improve self-efficacy M, which could relate to student satisfaction Y. If all three are measured in one self-report survey, you can examine an indirect statistical association. You cannot establish from those observations alone that support caused self-efficacy first and satisfaction afterwards.
Consider design before calculation
A causal interpretation needs a defensible temporal order and assumptions about confounding. An unmeasured variable may influence both mediator and outcome. Think through variables arising before X, between X and M, and between M and Y. Adding every available covariate is not automatically safer; controlling a wrongly placed variable can introduce bias.
Discuss whether the design supports mediation with your supervisor. For cross-sectional data, an exploratory indirect association is usually a more defensible description. Plan sample size and measurement quality in advance; a poorly measured mediator makes the estimated pathway difficult to interpret.
Estimate and report the indirect path
In a simple linear model, the path from X to M is often labelled a and the conditional path from M to Y labelled b. Their product a×b estimates an indirect effect. The product need not follow a convenient normal distribution, so suitable confidence intervals, often using a bootstrap, are common.
State the model, scale of each variable, analysed sample, handling of missing data and the interval. Check modelling assumptions and whether a reverse direction is plausible. An insignificant total effect does not automatically rule out an indirect effect; a significant indirect estimate does not make an observational design causal.
Use language your data support
Present estimates and uncertainty before the substantive explanation. With cross-sectional observations, “the data are consistent with an indirect association” is safer than “M causally mediates”. Identify potential shared causes and the uncertainty around timing.
Mediation is not mandatory in a strong thesis. If theory, measurement sequence or sample size are weak, a clear descriptive or simple association analysis may answer the question better. A well-justified simple model is preferable to a complex diagram whose arrows were invented after seeing the results.
Final submission context
Formal details can feel like a separate writing task, but they matter just as much for a printed submission. Anything missing, misplaced or inconsistently formatted in the document will also appear in the bound copy.
Use this guide together with your cover page, table of contents, page numbers, source notes and appendices. Prepare the final PDF for printing and binding only after the complete file has been checked.
Practical check before PDF export
- Do headings, chapter structure and page numbers match?
- Are sources, figures and tables included completely?
- Are required elements such as declarations, appendices or the cover page included where required?
- Did you open and check the final PDF after exporting it?
Checklist
- Grounded the path model in theory.
- Checked timing and confounders.
- Reported indirect effect and interval.
- Matched causal language to design.
Common mistakes
- Calling three correlations causal proof.
- Choosing covariates without a model.
- Reporting only a p-value.
Prepare the finished thesis for submission
Once the content and university requirements are checked, review your PDF and the available printing and binding options.
Review printing optionsRelated content
Explore related guidance on thesis structure, sources, formatting and final submission.
Frequently asked questions
Is a mediator just a control variable?
No, it represents a proposed intermediate path requiring justification.
Can cross-sectional data support mediation?
A statistical model is possible, but causal claims are strongly limited.
Must the total effect be significant?
Not necessarily; examine the indirect effect and the full model.