How to Write the Methods Section of a Research Paper by Study Design: Cohort, Trial, Diagnostic, Qualitative and Prediction-Model Studies

A research paper methods section lets readers assess bias, follow the analysis and judge reproducibility. Because requirements differ by design, the most reliable approach to how to write the methods section of a research paper is a consistent framework plus design-specific detail.

A practical way to organise research methodology writing is a five-part methods framework: design, data, decisions, analysis and credibility. It is an editorial aid, not a validated reporting standard.

• Design: research question, study design, setting and period.
• Data: participant selection, eligibility, recruitment, data sources and measurements.
• Decisions: outcomes, exposures, predictors, exclusions and why key choices were made.
• Analysis: statistical models or qualitative procedures, assumptions and missing-data methods.
• Credibility: how bias, uncertainty, validation and reproducibility were addressed.

Report what was done, separate prespecified from exploratory analyses and keep results out of the methods.

A cohort study methods section should say whether the design was prospective or retrospective and explain cohort entry, eligibility, data sources, exposure and outcome measurement, and follow-up. Define outcomes explicitly and report losses to follow-up and censoring.

Name the effect measure and model (risk, rate, odds or hazard ratio). Select confounders from subject-matter knowledge and a causal framework rather than statistical significance, describe how survival-model assumptions were checked, and report subgroup and sensitivity analyses.

What a reviewer will ask: could loss to follow-up, misclassification or inappropriate adjustment have distorted the association? STROBE supports reporting of cohort, case-control and cross-sectional studies; it improves completeness but does not prescribe the right analysis [2].

A clinical trial methods section must make the intervention, comparator, allocation and planned comparison transparent. State the design (parallel-group, crossover, cluster-randomised), setting, eligibility, recruitment period, and the intervention and comparator with dose, frequency and duration; incomplete treatment descriptions are common and block replication [3]. For randomised trials, report sequence generation, allocation concealment and blinding, because inadequate concealment is associated with exaggerated treatment effects [4].

Report the sample-size calculation and its assumptions (target effect, significance level, power and attrition). Specify the primary outcome, analysis population and, where relevant, the estimand, including how intercurrent events and missing outcomes were handled, plus multiplicity, interim analyses and stopping rules.

What a reviewer will ask: does the sample-size calculation match the primary outcome, and is allocation transparent? CONSORT 2025 and its extensions guide trial reporting [5].

Diagnostic accuracy studies evaluate how well an index test identifies a target condition against a reference standard. Define all three, then explain participant selection, test procedures, thresholds and interpretation criteria, whether assessors were blinded, the interval between tests and how indeterminate or missing results were handled. Explain the sampling strategy: recruiting clearly diseased and clearly healthy people (a case-control design) inflates accuracy beyond what the intended clinical population would show [6].

Name the planned measures: sensitivity and specificity, positive and negative predictive values, likelihood ratios and the area under the ROC curve (AUC). State how confidence intervals were calculated and whether thresholds were prespecified (predictive values depend on prevalence).

What a reviewer will ask: could participant selection or an unsuitable reference standard have biased the estimates? STARD 2015 supports transparent reporting [7].

Name the methodological approach (phenomenology, grounded theory, ethnography or qualitative description) and why it fits. Describe recruitment, sampling strategy and participant characteristics, and justify sample adequacy against the question and methodology rather than a universal number.

Specify the data source (interviews, focus groups, observation or documents), who collected it and how recordings were transcribed. Describe coding, theme development and how analysts handled contradictory accounts. Address reflexivity: how the researchers’ backgrounds and relationships with participants may have shaped the findings.

What a reviewer will ask: is the analysis transparent, and is reflexivity addressed? COREQ suits interviews and focus groups [8]; SRQR covers qualitative approaches more broadly [9].

A prediction model methods section must show how predictions were generated and that evaluation did not reuse development data inappropriately. Describe the population, data source, outcome, prediction horizon and predictor selection. Report the sample-size rationale, missing-data handling, feature selection and model type (regression, survival or machine learning).

Explain the validation strategy and how model selection and hyperparameter tuning were done. Reusing the same observations for fitting, tuning and evaluation gives optimistic estimates, and bootstrapping generally gives more stable internal validation than split samples [10]. Then report three distinct dimensions:

• Discrimination: how well the model separates people who have the outcome from those who do not.
• Calibration: how closely predicted risks match observed outcomes.
• Clinical utility: whether model-guided decisions could improve on alternatives, where relevant.

Give performance measures with uncertainty, calibration plots and the final model specification. What a reviewer will ask: was evaluation independent of development and tuning? TRIPOD+AI replaces TRIPOD 2015 for prediction models built with regression or machine learning [11].

When structuring a methods section by study design, choose the reporting guideline that matches the design and objective, then check relevant extensions and the target journal’s instructions. The EQUATOR Network library lists them all [12].

Reporting quality is not methodological quality: a checklist shows what is missing, not that the design or analysis is valid.

Use this checklist to find omissions, adapted to your design: randomisation matters for trials, calibration for prediction models.