How to Report Your Statistics in a Methods Section Reviewers Won't Question
- 7 minutes ago
- 5 min read
When reviewers examined 100 clinical medicine papers, 65% needed revisions just to explain what their statistical tests were for, and 64% had omitted effect sizes entirely. These aren't obscure edge cases. They're the basic contents of a statistics section, and most manuscripts go out incomplete.
The fix costs almost nothing. Statistical reporting problems are rarely about doing the wrong analysis; they're about not writing down enough for a reader to judge it. A reviewer who can't tell which test you ran, on how many independent units, with what assumption checks, has to ask, and that question becomes a revision round.
Here's what belongs in the section, in the order it reads best, plus the omissions that generate the most reviewer queries.
What "enough detail" means in statistical reporting
There's a standard for this: the SAMPL guidelines (Statistical Analyses and Methods in the Published Literature), written by Lang and Altman and hosted by the EQUATOR Network. They're the closest thing to a universal checklist, and they're badly underused, referenced by only about 5% of the World Association of Medical Editors' members.
Two principles run through them, and they're worth internalizing because everything else follows:
Describe your methods in enough detail that a knowledgeable reader could reproduce the analysis with access to your data.
Report results in enough detail that they could be incorporated into another analysis, such as a meta-analysis. This is why a p-value alone is never sufficient: re-analysis needs the descriptive statistics behind it, the group sizes, the effect estimate, and a measure of precision.
If you write with the second principle in mind, you'll pass the first automatically.
Seven things to state in your statistics methods section
Most reviewer queries come from one of these being absent:
The test, and why you chose it. Name the test and tie it to the design ("a one-way ANOVA compared the three treatment groups"). The most-flagged omission is failing to explain what a test was for.
What your n refers to. State the number of independent experimental units per group, and distinguish them from technical replicates. "n = 3" is ambiguous; "n = 3 independent cultures, each measured in triplicate" is not.
How assumptions were checked. Name the check, not just the assumption: normality by Q-Q plot or Shapiro-Wilk, equal variance by Levene's test, and what you did if one failed. Inadequate assumption assessment is a recurring reviewer complaint.
The multiple-comparison correction, if you ran more than one test, and which method (Bonferroni, Holm, Tukey, false-discovery-rate).
How outliers and missing data were handled, decided in advance. Silence here reads as "we decided after seeing the results."
The software and version, including the package where relevant. SAMPL asks for this explicitly, and it's a one-line fix that reviewers notice when missing.
Your significance threshold, stated once, and whether tests were one- or two-tailed.
Tip: write this section before you run the analysis, not after. Pre-specifying forces the decisions that are hardest to defend retrospectively, and the text is nearly identical either way.
What to report in results, and in what format
Formatting conventions are where small errors accumulate:
Exact p-values, not thresholds. Report p = 0.03, not p < 0.05. Use three decimal places, and reserve p < 0.001 for values genuinely below it. A bare "p < 0.05" discards information and is increasingly rejected outright.
The effect size, every time. This is the single most common omission, missing from roughly two-thirds of papers in that review. A p-value tells the reader whether an effect exists; only the effect size tells them whether it matters.
A confidence interval alongside every effect estimate, giving both the plausible range and the precision.
The test statistic with its degrees of freedom: F(2, 42) = 6.81, or t(28) = 2.14.
Descriptive statistics with a named spread. Give the mean with SD, or median with IQR, and say which. Never write "mean ± error" without naming what the error bar is.
Consistent values across text, tables, and figures, rounded consistently. Mismatches between the abstract and the results table are a routine production flag.
Caution: every figure legend needs its own statement of what the error bars represent and the n behind them. A figure that says only "error bars shown" is incomplete, and it's one of the fastest ways to draw a query.
A methods section template you can adapt
A complete, unremarkable methods paragraph looks like this:
"Data are presented as mean ± SD from n = 6 independent cultures per group, each measured in technical triplicate and averaged before analysis. Normality was assessed by Shapiro-Wilk and equal variance by Levene's test. Group differences were compared by one-way ANOVA with Dunnett's correction for multiple comparisons against the vehicle control. Two-tailed p < 0.05 was considered significant. Analyses were performed in [software, version]."
And the matching results sentence:
"Treatment increased migration relative to vehicle (mean difference 18.2 µm, 95% CI 7.4–29.0; F(2, 15) = 9.34, p = 0.002; partial η² = 0.55)."
Nothing in either is elaborate. They're just complete, and completeness is the whole trick. Drafting the paragraph from your actual design and having the gaps flagged before submission takes minutes (Sophie's analytics reports the effect size, interval, and assumption checks alongside the test, which is most of what the paragraph needs).
Write it so nobody has to ask
A methods section that survives review isn't more sophisticated than one that doesn't, it's more complete. State the test and why you chose it, define what your n counts, name the assumption checks you ran, and report exact p-values with effect sizes and confidence intervals rather than thresholds alone. Say which software produced the numbers, and make sure the figure legends stand on their own. Every one of those is a sentence, and each is a question a reviewer now doesn't have to ask.
FAQ
What should I include in the statistical methods section? Name the test and why it suits the design, define what n refers to (independent units, not technical replicates), describe how assumptions were checked, state any multiple-comparison correction, explain how outliers and missing data were handled, give your significance threshold, and specify the software and version.
Should I report exact p-values or p < 0.05? Report exact values, typically to three decimal places (p = 0.03), reserving p < 0.001 for values genuinely below that. Reporting only a threshold discards information the reader needs and is increasingly refused by journals.
Why do reviewers ask for effect sizes? Because a p-value indicates only whether an effect is distinguishable from chance, not how large it is. Effect sizes are the most commonly omitted element in statistical reporting, missing from roughly two-thirds of papers in one review, and they are also what makes a study usable in later meta-analyses.
What are the SAMPL guidelines? SAMPL (Statistical Analyses and Methods in the Published Literature), by Lang and Altman, is a set of guidelines for reporting statistical methods and results in biomedical journals, hosted by the EQUATOR Network. They are designed to prevent the most common reporting deficiencies and can be applied to almost any empirical paper.
How should I describe my sample size? State the number of independent experimental units per group and distinguish them from repeated measurements of the same unit. "n = 3 independent cultures, each measured in triplicate" is unambiguous; "n = 3" alone is not.
