Writing the Statistical Methods Section

Overview

The statistical Methods section should allow a knowledgeable reader with access to the data to understand, assess and reproduce the analysis.

It should explain:

• what was estimated;
• why each method was selected;
• how variables were coded;
• how dependence and design were handled;
• how uncertainty was quantified;
• how missing data and multiplicity were addressed;
• which analyses were prespecified;
• which software and versions were used.

ICMJE states that statistical methods should be described in enough detail for knowledgeable readers to judge appropriateness and verify the results [1]. SAMPL provides more specific biomedical statistical-reporting guidance [2].

Why this matters

A sentence such as:

Data were analysed using appropriate statistical tests in SPSS.

is not reproducible.

It omits:

• outcome;
• groups;
• estimand;
• test version;
• assumptions;
• effect measure;
• confidence level;
• missing-data approach;
• adjustment variables;
• software version.

The Methods section should document decisions made before or without reference to the observed results. Outcome-driven changes belong in deviations or exploratory analyses.

When to use this framework

Use the framework when writing:

• protocols;
• statistical analysis plans;
• trial reports;
• observational manuscripts;
• diagnostic studies;
• prediction-model reports;
• theses and dissertations.

The reporting depth should match the study complexity and journal format. Technical details can be placed in a supplement, but they must remain accessible and peer reviewed.

Core elements

Study design and analytical unit

State:

• design;
• sampling or allocation;
• repeated or clustered structure;
• analysis population;
• independent unit.

Example:

Participants were individually randomised in a 1:1 ratio. The participant was the analytical unit.

Outcomes and estimands

Define:

• primary outcome;
• time point;
• effect measure;
• contrast direction;
• analysis population.

Example:

The primary estimand was the telemonitoring-minus-usual-care difference in mean clinic systolic blood pressure at 12 weeks among all randomised participants.

Descriptive methods

State how each variable type was summarised.

Avoid:

Continuous data were reported as mean or median as appropriate.

Better:

Numerical variables were summarised using mean and SD when the arithmetic mean was clinically meaningful; skewed duration variables were summarised using median and first and third quartiles. Categorical variables were reported using counts and percentages with variable-specific denominators.

Primary analysis

Name the exact model, outcome, predictor and covariates.

Example:

The primary outcome was analysed using linear regression adjusted for baseline systolic blood pressure and clinic, with usual care as the reference group.

Effect measures

State what was reported:

• adjusted mean difference;
• risk difference;
• risk ratio;
• odds ratio;
• hazard ratio;
• median difference;
• correlation.

Specify 95% confidence intervals and test sidedness where relevant.

Assumption assessment

Describe diagnostic approaches, not only formal tests.

Example:

Linearity, residual variance and influential observations were assessed graphically. The treatment-by-baseline interaction was examined to evaluate homogeneity of regression slopes.

Missing data

State:

• amount expected;
• primary strategy;
• variables in imputation;
• number of imputations;
• pooling method;
• sensitivity analyses.

“Missing data were excluded” is incomplete.

Multiplicity

Identify:

• primary and secondary outcomes;
• confirmatory comparisons;
• adjustment method;
• exploratory analyses.

Software

Report:

• package;
• version;
• important functions or packages;
• operating environment where relevant.

Current ICMJE recommendations also state that AI technology used to conduct the study should be described sufficiently to enable replication, including tool, version and prompts where applicable [1].

Data example

Methods information map

simulated data

Question	Required detail
What is estimated?	Adjusted mean difference at 12 weeks
Which population?	All randomised participants
Which model?	Linear regression
Which covariates?	Baseline SBP and clinic
How is uncertainty reported?	Two-sided 95% CI
How are missing outcomes handled?	Multiple imputation
Which secondary measure?	Risk difference and risk ratio for control
Which software?	R 4.6.1 with named functions/packages

Worked example

A structured writing template

Analysis population:
Descriptive summaries:
Primary outcome and estimand:
Primary statistical model:
Adjustment variables and reference categories:
Effect measure and confidence interval:
Assumption and diagnostic assessment:
Missing-data method:
Multiplicity and subgroup analyses:
Sensitivity analyses:
Software and version:
Protocol or SAP deviations:

Model-specific details

Linear regression

Report:

• outcome;
• predictors;
• coding;
• functional forms;
• interactions;
• variance estimator;
• diagnostics.

Logistic regression

Report:

• event definition;
• reference categories;
• odds-ratio interpretation;
• continuous-variable forms;
• separation handling;
• predicted probabilities where relevant.

Survival analysis

Report:

• time origin;
• event definition;
• censoring;
• competing risks;
• proportional-hazards assessment;
• effect measure.

Prediction models

Report:

• candidate parameters;
• model-building strategy;
• shrinkage;
• internal validation;
• performance measures;
• external validation.

Prespecified and exploratory analyses

Analyses should be labelled as:

• primary confirmatory;
• secondary;
• sensitivity;
• exploratory;
• post hoc.

A post hoc analysis is not invalid merely because it was unplanned, but its evidential status is different and must be transparent.

Interpreting the Methods section

Presenting the methods

Suggested reporting

Use the relevant design-specific reporting guideline alongside SAMPL. For clinical-trial protocols, SPIRIT 2025 provides the minimum items needed for transparent planning and reporting [3]. Complex regression specifications should also be documented with enough detail to reconstruct coding, transformations and interactions [4]. Statistical reporting guidance complements rather than replaces CONSORT, STROBE, PRISMA, STARD or TRIPOD+AI.

Common mistakes

Knowledge check

What is the guiding principle for statistical Methods?

Answer and reasoning

Provide enough detail for a knowledgeable reader with the data to judge appropriateness and reproduce the analysis.

Why should the estimand be stated?

Answer and reasoning

It identifies the population quantity the model is intended to estimate and aligns the method with the research question.

Is software name alone sufficient?

Answer and reasoning

No. Include version and relevant packages, functions or settings.

Where should an unplanned analysis be reported?

Answer and reasoning

Label it transparently as exploratory or post hoc and explain the rationale.

What should be reported about missing data?

Answer and reasoning

Amount and pattern, primary handling method, model variables and assumptions, and relevant sensitivity analyses.

Key take-home messages

References

1. International Committee of Medical Journal Editors. Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals: preparing a manuscript for submission. Updated 2026. https://www.icmje.org/recommendations/browse/manuscript-preparation/preparing-for-submission.html
2. Lang TA, Altman DG. Basic statistical reporting for articles published in biomedical journals: the SAMPL guidelines. Int J Nurs Stud. 2015;52(1):5–9. https://doi.org/10.1016/j.ijnurstu.2014.09.006
3. Chan AW, Boutron I, Hopewell S, Moher D, Schulz KF, Collins GS, et al. SPIRIT 2025 statement: updated guideline for protocols of randomised trials. BMJ. 2025;389:e081477. https://doi.org/10.1136/bmj-2024-081477
4. Harrell FE Jr. Regression modeling strategies. 2nd ed. Cham: Springer; 2015. https://doi.org/10.1007/978-3-319-19425-7