Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Insights · Study guide · Formatting a data-analysis write-up

How should a data-analysis write-up for a manuscript be formatted?

The test is whether a knowledgeable reader could verify your results. Here is what that requires, item by item, from the guidelines that actually apply.

Pubrica AcademyWritten by a research writing specialist Reviewed bya research methodology reviewer Published15 October 2026 Last checked15 October 2026 Reading time13 minutes Sources5 primary

A data-analysis write-up has two jobs: say what you did precisely enough that someone could check it, and report what you found in a form that can be read, compared and pooled. This page sets out what the reporting standards actually require for each, and which standard applies to your design.

What this section has to do

The governing standard is the same in SAMPL and in the ICMJE Recommendations, and it is a usefully concrete test: “Describe statistical methods with enough detail to enable a knowledgeable reader”1 — the ICMJE adds the purpose, “to verify the reported results.”2

That test settles most arguments about how much detail to include. If a reader could not tell which model you fitted, which participants were in it, or what you did with the missing values, the section has not met the standard, however fluent it reads.

The statistical methods section

For a randomised trial, CONSORT 2025 breaks the requirement into four parts3:

Table 1. CONSORT 2025 item 21, statistical methods

ItemWhat it requires
21aStatistical methods used to compare groups for primary and secondary outcomes, including harms
21bDefinition of who is included in each analysis (eg, all randomised participants), and in which group
21cHow missing data were handled in the analysis
21dMethods for any additional analyses (eg, subgroup and sensitivity analyses), distinguishing prespecified from post hoc

Those four generalise well beyond trials, because each addresses a decision a reader cannot reconstruct from the results: what you compared and how, who was in the analysis, what happened to missing data, and what was planned as against decided later.

For an observational study, STROBE item 12 asks for a comparable set, with confounding and loss to follow-up named explicitly4:

Table 2. STROBE item 12, statistical methods

ItemWhat it requires
12aDescribe all statistical methods, including those used to control for confounding
12bDescribe any methods used to examine subgroups and interactions
12cExplain how missing data were addressed
12dIf applicable, explain how loss to follow-up was addressed
12eDescribe any sensitivity analyses

Item 12 is quoted here from the cohort-study checklist. Item 12(d) is worded differently in the case-control and cross-sectional variants, so quote the variant that matches your design.

Beyond the checklists, two things belong in this section that neither names explicitly: the software and its version, because defaults change between releases and a reader reproducing your analysis needs to know which one you ran, and the significance level and whether it was adjusted for multiple comparisons, with the method if it was.

The results section

The single most useful requirement in any of these documents is CONSORT 2025 item 26, because it specifies what a result is. For each primary and secondary outcome, by group3:

  • the number of participants included in the analysis
  • the number of participants with available data at the outcome time point
  • result for each group, and the estimated effect size and its precision (such as 95% confidence interval)
  • for binary outcomes, presentation of both absolute and relative effect size

Read the third of those carefully: the estimated effect size and its precision. A result is an estimate plus an interval. A p-value on its own is not a result, and a table of p-values is not a results section.

Item 28 then covers everything else you ran: “Any other analyses performed, including subgroup and sensitivity analyses, distinguishing pre-specified from post hoc”3 The phrase to notice is “distinguishing pre-specified from post hoc”, which is the difference between a finding and a hypothesis.

How to report a number

Table 3. Reporting conventions, with the source for each

ConventionWriteNotWhy
An estimate, always with its precisionA difference of 4.2 mmHg (95% CI 1.8 to 6.6)A difference of 4.2 mmHg, p = 0.01SAMPL asks you to “report a measure of precision, such as the 95% confidence interval” The interval tells a reader the range the data are consistent with; the p-value does not.
A p-value as an equalityp = 0.03p < 0.05SAMPL: “P values should be reported as equalities when possible” A closed-form p-value discards information and cannot be used in later synthesis.
A floor, not a zerop < 0.001p = 0.000SAMPL sets the floor: “The smallest P value that need be reported is P <0.001, save in studies of genetic associations.” And p = 0.000 is a software display artefact, not a probability — a p-value is never zero.
The actual value, not a labelp = 0.21NSSAMPL: “Do NOT report 'NS'; give the actual P value.”
Denominators with percentages18/204 (8.8%)8.8%A percentage without its denominator cannot be checked, and small denominators make percentages misleading.
Sensible precisionMean 24.3 kg/m² (SD 3.1)Mean 24.31724 kg/m²Report to the precision the measurement supports, not to the precision the software prints.

The reporting guideline that applies to you

Table 4. Which guideline, by design

Your studyGuidelineStatistical items
Randomised trialCONSORT 2025Item 21 for methods, item 26 for results, item 28 for additional analyses
Observational studySTROBEItem 12, in the variant for your design
Systematic reviewPRISMA 2020Its synthesis and results items; see our page on review tables
Diagnostic accuracy studySTARDIts statistical methods items
Any of the aboveSAMPLSits across all designs for the statistical content specifically

Note that the guideline follows from your design, not from your journal or your discipline. And SAMPL is additive rather than alternative: it covers how to report the statistics within whichever design-specific checklist applies1.

The errors that get flagged

These are not hypothetical. An audit of published radiology articles quantified them5:

Table 5. Error frequencies in published articles

Error classRateNote
Errors in summarising data66%“The most frequently encountered error was 'errors in summarizing data' with a rate of 66%.”
Incorrect representation of P values42%Including “incorrect representation of P values (e.g., P = 0.000, P < 0.0005).”
P values not reported at all25.48%The second most frequent P-value-specific error.

Two of those deserve naming individually. P = 0.000 is listed explicitly among incorrect representations5; write p < 0.001. And closed-form p-values — “P values given in closed form (e.g., P < 0.01, P < 0.05, P > 0.05).”5 — discard information that a later meta-analysis would need.

The error that audit does not cover, and which is arguably the most consequential, is reporting a p-value with no effect estimate or interval beside it. For that the authority is the ICMJE, which asks authors to “Avoid relying solely on statistical hypothesis testing, such as P values,” because they “fail to convey important information about effect size and precision of estimates.”2 — and CONSORT 2025 item 26, which requires the effect size and its precision for every outcome3.

A working checklist

How to apply this to your own work

  1. Name the reporting guideline for your design and download its checklist before you write the section, not after a reviewer asks.
  2. In methods, state: the model or test per outcome; who was included in each analysis and in which group; how missing data were handled; which analyses were pre-specified; the significance level and any multiplicity adjustment; and the software with its version.
  3. In results, give every outcome an estimate and an interval. If a number appears without its precision, it is not finished.
  4. Report p-values as equalities, floor them at p < 0.001, and never write p = 0.000.
  5. Give denominators with every percentage, and round to the precision the measurement actually supports.
  6. Label post hoc analyses as post hoc, in both the methods and the results.
  7. Reconcile every number in the text against the table it came from, and against the abstract, before submission.

Take-home message

The standard for a statistical methods section is that a knowledgeable reader could verify your results — so name the model, who was in it, what happened to missing data, and what was pre-specified. The standard for results is CONSORT 2025 item 26: an effect estimate and its precision for every outcome. A p-value alone is not a result, and the ICMJE says so. Report p-values as equalities, floor them at p < 0.001, never write p = 0.000, and give a denominator with every percentage.

No, and the ICMJE says so directly. Its recommendations ask authors to “quantify findings and present them with appropriate indicators of measurement error or uncertainty,” with confidence intervals as the example, and to “Avoid relying solely on statistical hypothesis testing, such as P values,” because these “fail to convey important information about effect size and precision of estimates.” Report the effect and its interval; the p-value is additional information, not the result.

You should not — write p < 0.001. A p-value is a probability and is never exactly zero; “0.000” is what statistical software prints when the value rounds below its display precision. An audit of radiology journals lists it explicitly among incorrect representations of P values, and that category was the second most common error class in the papers examined, at 42%.

Enough that a knowledgeable reader could verify your results — that is the ICMJE’s standard and SAMPL’s. Concretely: what test or model, for which outcome; which participants were included in each analysis and in which group; how missing data were handled; what software and version; and which analyses were pre-specified as against decided after seeing the data. If you are reporting a trial, CONSORT 2025 item 21 asks for exactly those four things.

Yes, and it is the item most often skipped. CONSORT 2025 asks for the methods of additional analyses “distinguishing prespecified from post hoc”, and again in the results for any other analyses performed. The distinction is what lets a reader judge a subgroup finding, and a page of subgroup results with no indication of which were planned is not interpretable.

It depends on the design, not on the journal. A randomised trial is reported against CONSORT 2025; an observational study against STROBE; a systematic review against PRISMA 2020; a diagnostic accuracy study against STARD. SAMPL sits across all of them for the statistical content specifically. Most journals ask for the completed checklist at submission, and filling it in honestly while you write surfaces the gaps while you can still fix them.

The test is whether a knowledgeable reader could reproduce what you did. That usually means naming the model and its specification rather than just the test family — a mixed-effects linear model with site as a random effect, rather than “regression analysis”. Software version matters because defaults change between versions. What is not needed is a textbook explanation of why the test exists.

References and sources

  1. Lang TA, Altman DG. Basic Statistical Reporting for Articles Published in Biomedical Journals: the ‘Statistical Analyses and Methods in the Published Literature’ or the SAMPL Guidelines. In: Smart P, Maisonneuve H, Polderman A, editors. Science Editors’ Handbook. European Association of Science Editors, 2013. read 15 October 2026. View on EQUATOR NetworkThe reporting guideline for statistical methods and results, indexed by EQUATOR. Volume and page numbers are not shown on the copy we read.
  2. International Committee of Medical Journal Editors. Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals. Section IV.A.3.d.iii, Statistics. Updated January 2026. read 15 October 2026. View on ICMJEThe authoritative statement that a p-value alone is not adequate reporting, from the body whose recommendations most medical journals adopt.
  3. Hopewell S, Chan A-W, Collins GS, Hróbjartsson A, Moher D, Schulz KF, Tunn R, et al., Boutron I. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ. 2025;389:e081123. doi:10.1136/bmj-2024-081123 read 15 October 2026. View on The BMJCited here for its statistical items. Item 21 covers the methods, item 26 the results, and item 26 is the one that requires an effect size and its precision for every outcome.
  4. STROBE Initiative. STROBE Statement — checklist of items that should be included in reports of cohort studies. read 15 October 2026. View on Duke UniversityThe cohort-study variant of the STROBE checklist. Read from a university-hosted copy, which prints no version date; the official checklists are at strobe-statement.org.
  5. Günel Karadeniz P, Uzabaçı E, Atış Kuyuk S, Kaskır Kesin F, Can FE, Seçil M, Ercan İ. Statistical errors in articles published in radiology journals. Diagnostic and Interventional Radiology. 2019;25:102–108. doi:10.5152/dir.2018.18148 read 15 October 2026. View on Diagnostic and Interventional RadiologyAn audit of statistical reporting in published radiology articles. The source for the frequency of reporting errors and for naming P = 0.000 as an incorrect representation. It does not address p-values reported without an effect estimate.

Pubrica AcademyWritten and maintained by the subject matter experts of Pubrica who do this work. Articles are revised when the standard underneath them moves, and each one carries the date it was last checked.

How Pubrica helps

Our biostatistics and medical writing teams draft and review statistical methods and results sections against the applicable reporting checklist, reconcile every reported figure against the analysis output, and complete the submission checklist with page references. If a reviewer has asked for the statistics to be rewritten, that is a contained piece of work.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.