Guide clusterResearch Methodology
The methodology diagram mistakes that draw objections most often are missing exclusion counts, unlabeled decision branches, mixing protocol steps with analysis steps, and diagrams that contradict the methods text. Each is verifiable, so reviewers can raise them as concrete objections rather than general concerns.
What this page covers
- Missing counts and unbalanced numbersA methodology diagram without numbers is a protocol sketch.
- Unlabeled branches and hidden decisionsDecision diamonds without labels are a common fault…
- Mixing levels of abstractionMethodology diagrams frequently collapse protocol detail and analysis detail into one figure.
- Contradictions between the figure and the textThe most avoidable fault is disagreement between the diagram and the methods section.
Missing counts and unbalanced numbers
A methodology diagram without numbers is a protocol sketch. Once counts are added at each transition, the figure becomes a statement of what evidence the study actually rests on. Reviewers use these numbers to judge statistical power, to check for differential attrition, and to see whether the analysis set is well defined.
Numbers must also balance. The count entering the study should equal the sum of those excluded and those continuing, at every stage, and the final analysed count should reconcile with what the results tables report. When they do not reconcile, reviewers rightly suspect that the exclusions were not pre-specified or that the analysis set was chosen after seeing results.
Add a count to every transition, including each exit point.
Confirm that the numbers balance at every junction in the diagram.
Cross-check the final analysed count against the first results table.
Note where the analysis set differs from the allocated sample, and why.
Unlabeled branches and hidden decisions
Decision diamonds without labels are a common fault, and they are costly because they leave the reader unable to judge whether a choice was pre-specified. Every branch should carry the condition that routes a unit: meets criteria or does not, consented or declined, complete data or incomplete. Where a decision was made during analysis rather than at design, say so openly rather than drawing it as though it had always been planned.
Related to this is the silent switch in analysis. A study that changes its model because an assumption was violated should show the check and the branch, with the reason recorded. Presenting only the final model hides a decision that affects how the results should be interpreted.
An unlabeled branch is read as an unexamined decision.
Mixing levels of abstraction
Methodology diagrams frequently collapse protocol detail and analysis detail into one figure. A box describing how a questionnaire was administered sits beside a box describing a mixed-effects model, and the reader cannot tell at what level the diagram operates. Splitting the figure into a data collection panel and an analysis panel solves this neatly without losing content.
A related fault is the merged mega-box. When a step contains several distinct procedures with different outputs, it should be split, because the reader cannot see where the process could fail or where an intermediate artifact is produced. The rule of thumb is one action and one output per process box.
| Fault | Symptom a reviewer sees | Fix |
|---|---|---|
| Mixed abstraction | Protocol steps beside statistical models | Split into collection and analysis panels |
| Merged processes | One box containing several distinct actions | Split so each box has one action and one output |
| Silent analysis switch | Only the final model is shown | Show the assumption check and the branch with its reason |
| Unclear unit | Reader cannot tell what a box represents | State the unit of study in the caption and label each process accordingly |
Contradictions between the figure and the text
The most avoidable fault is disagreement between the diagram and the methods section. The figure says four sites and the text says five. The diagram shows a follow-up at twelve months that the text never mentions. These mismatches are easy to create during revision and easy for a reviewer to find, and they undermine confidence in the whole methods description because they suggest the study was not carefully documented.
Build a small comparison list while revising: one column for each element in the figure, one for the corresponding sentence in the text, and a note where they diverge. Update both from that list rather than editing either in isolation, and do the final check after the last text revision rather than before it.
- Compare counts, time points, sites and instruments between figure and text.
- Check that every procedure in the text appears in the figure and vice versa.
- Confirm terminology is identical, not merely equivalent.
- Rerun the comparison after any late change to either document.
Accessibility and reproduction quality
Methodology figures are often reproduced at small size in print or viewed on mobile devices by readers of online articles. Thin line weights, low-contrast fills and text below roughly seven points all fail in those conditions. Use sufficient contrast, avoid relying on color alone to distinguish paths, and check the figure in grayscale before submission.
Alt text is also worth writing, even where it is not mandatory, because methodology figures carry information that some readers will access through a screen reader. A short description naming the entry point, the number of exclusion stages and the final analysis set is enough to convey the structure.
Worked example
Worked example: correcting a methods figure before resubmission
A team submitted a paper with a flow diagram showing recruitment, allocation, measurement and analysis as four unlabeled boxes connected by arrows. Reviewers asked for the effective sample size at each stage, the reasons for withdrawal, and clarification of whether the analysis included participants who had switched conditions.
The revision began by inserting counts. Entry, exclusions by reason, allocation to each arm, withdrawals at two follow-up points and the final analysed numbers were added at each transition. The counts immediately revealed a discrepancy: the analysed sample in the diagram was larger than the number reported in the results table, because the table had excluded participants with incomplete baseline data and the diagram had not.
The next step was labeling the branches. Two decision diamonds gained conditions, one for eligibility and one for complete follow-up data. A third diamond, previously absent, was added at the analysis stage to show the pre-specified handling of participants who switched arms, with a footnote explaining the rationale and pointing to the sensitivity analysis.
Finally, the merged measurement box was split into three, one per instrument, because the original single box concealed the fact that the three measures had different completion rates. The figure became taller but substantially clearer, and the resubmission noted the changes to the diagram alongside the corrections to the results table.
Common mistakes
Frequently asked questions
Do methodology figures need a caption?
Yes, and the caption should state the unit of study, the number entering and the number analysed. Many readers read captions before figures, so a caption that only names the figure wastes the most-read sentence on the page.
How much detail can a methods figure carry before it becomes unreadable?
Aim for the level at which a reader can audit the sample and see the design's safeguards. If a figure needs more than about fifteen process elements, split it into panels rather than shrinking the type. Two readable panels beat one dense one.
Will an AI tool catch inconsistencies between my figure and my methods text?
It can compare counts and terms if you provide both, and it is reasonably good at spotting mismatched labels. It cannot judge whether an exclusion was appropriate, so use it for the mechanical comparison and make the methodological judgments yourself.
Should sensitivity analyses appear in the diagram?
Show them as a branch from the main analysis, labeled with what they test. This signals that robustness was considered without cluttering the primary path. The results of those analyses belong in the text or supplementary material.
Sources and further reading
- ARRIVE - reporting guidance for animal researchARRIVE Guidelines
- JATS - Journal Article Tag Suite, the standard article structureNational Center for Biotechnology Information, U.S. National Library of Medicine
- EQUATOR Network - library of reporting guidelinesUK EQUATOR Centre, University of Oxford
- PRISMA - reporting guidance for systematic reviewsPRISMA Executive
- CONSORT - reporting guidance for randomised trialsCONSORT Group
These are the organisations whose published guidance the recommendations on this page follow. Listing them does not imply that they endorse Pathform AI, and this page is not affiliated with them.