Reporting Statistics in Biology Papers: What Journals Require and What Reviewers Catch

Reviewer comment 3 says “the statistical methods section is incomplete.” You re-read it, find nothing obviously wrong, and now you're guessing what they want. This is the most common style of revision request in biology because methods sections are written from a researcher's mental model of “what I did” rather than a reviewer's checklist of “what I need to evaluate.”

This guide is the checklist version. It covers the sentences a methods section must contain to survive review at Nature, Cell, JAMA, or PLOS journals, organized by test type, with concrete templates and the specific phrasing reviewers look for. The bar is reproducibility — a competent reader should be able to rerun your analysis from the methods alone.

The Five Required Elements for Every Statistical Analysis

Independent of test type, every statistical analysis reported in a biology paper needs five things. Missing any one of them is grounds for a methods-section revision request:

1. Software and version. “Analyses were performed in R 4.3.1 (R Core Team, 2023)” or “GraphPad Prism 10.2.3 (GraphPad Software, San Diego, CA).” Major version changes (R 4.x vs 3.x; Prism 10 vs 9) change default behaviors; reviewers can't reproduce without it.
2. The unit of analysis (n). State explicitly whether n is biological replicates, technical replicates, animals, cells, or wells. Biological replicates are independent samples (different animals, different patient donors); technical replicates measure the same sample multiple times and should not be averaged into the n.
3. The test name. “Unpaired t-test” not just “t-test.” “One-way ANOVA with Tukey HSD post-hoc” not just “ANOVA.” The exact name disambiguates parametric vs. non-parametric, paired vs. unpaired, and correction method.
4. Assumption checks. “Normality was assessed by Shapiro-Wilk; variance homogeneity by Levene's test.” If you used a non-parametric test, state why: “Because Shapiro-Wilk indicated non-normality (p < .05) in 2 of 3 groups, we used the Mann-Whitney U test instead.”
5. Significance threshold and effect size. “Significance was set at α = 0.05. Effect sizes are reported as Cohen's d for t-tests and partial η² for ANOVA.” APA 7th edition treats effect sizes as essentially required — reporting only p-values increasingly causes desk rejections.

The omission that triggers reviewer comment 3 is almost always one of these five. Audit your methods section against this list before submission.

APA Formatting Rules That Reviewers Check

APA 7th edition is the dominant biomedical convention even outside psychology journals. The formatting rules are mechanical, journals enforce them, and getting them wrong signals carelessness:

• Italicize statistical symbols: t, F, p, r, d, n, N, M, SD, df. Greek letters (α, β, χ, η) are NOT italicized.
• No leading zero on bounded statistics: p = .045, not p = 0.045. Same for r, R², proportions — any statistic that can't exceed 1. Statistics that can exceed 1 (t, F, M, SD) keep the leading zero.
• Exact p-values: report p = .032, not p < .05, when the value is > .001. Below .001, write p < .001. Never p = .000 — statistically impossible (probabilities can be arbitrarily small but never exactly zero).
• Degrees of freedom in parentheses: t(28) = 3.42, F(2, 27) = 4.81. The df identify which distribution the test statistic was compared against.
• Consistent decimal places: typically 2 decimals for descriptive statistics, 2 for test statistics, 3 for p-values. Don't mix 4-decimal p-values with 1-decimal means in the same table.

If you're auto-formatting from SPSS output, strip the leading zeros and convert “Sig. = .000” to p < .001 manually. Tools that emit APA-formatted strings (JASP, GraphHelix, R packages like apaTables) handle this in one step.

Templates by Test Type

t-test

Methods sentence: “Differences between treatment and control groups were assessed using an unpaired two-tailed Student's t-test, with normality verified by Shapiro-Wilk (p > .05 in both groups) and equal variance by Levene's test.”

Results sentence: “Treated cells showed elevated expression compared to controls, t(28) = 3.42, p = .002, d = 1.25 (95% CI [0.47, 2.03]).”

The 95% confidence interval on the effect size is what reviewers increasingly expect — it shows the precision of the effect, not just whether it cleared a threshold. For more on choosing between SEM, SD, and 95% CI in figures, see error bars in scientific figures.

One-way ANOVA

Methods sentence: “Differences across the three treatment conditions were tested by one-way ANOVA, with significant omnibus effects followed by Tukey HSD pairwise comparisons (family-wise α = .05).”

Results sentence: “Treatment had a significant effect on outcome, F(2, 27) = 7.84, p = .002, η_p² = .37. Post-hoc Tukey HSD showed condition B exceeded condition A (p = .003) and condition C (p = .014); A and C did not differ (p = .68).”

Always state which post-hoc correction you used. Tukey, Bonferroni, and Holm-Šidák give different adjusted p-values, and reviewers ask. See Tukey, Bonferroni, and FDR compared for which to pick.

Non-parametric tests

Methods sentence: “Because Shapiro-Wilk indicated non-normality in 2 of 3 groups (p < .05), differences across conditions were tested by Kruskal-Wallis H with Dunn's post-hoc adjustment.”

Results sentence: “Conditions differed significantly, H(2) = 12.4, p = .002, ε² = .42. Dunn's post-hoc showed condition B exceeded condition A (adj. p = .008).”

Stating the reason for using a non-parametric test — not just “we used Mann-Whitney” — signals that the choice was principled rather than convenient. The non-parametric test should always come with a non-parametric effect size: rank-biserial r for Mann-Whitney, ε² for Kruskal-Wallis. The decision tree for which test belongs to which design is covered in choosing the right statistical test.

Linear regression

Methods sentence: “The relationship between dose and response was modeled by linear regression, with assumptions assessed via residual Q-Q plots, residuals-vs-fitted scatter, and variance inflation factors for multicollinearity (all VIF < 5).”

Results sentence: “Dose predicted response, β = 0.42 (SE = 0.11), t(28) = 3.82, p < .001, R² = .34, adjusted R² = .31.”

Survival analysis

Methods sentence: “Survival curves were estimated by Kaplan-Meier and compared by log-rank test. Hazard ratios were estimated from Cox proportional hazards regression adjusted for sex and baseline age; the proportional hazards assumption was verified by Schoenfeld residuals (global p = .42).”

Results sentence: “Median survival was 18.2 weeks (treatment) vs. 11.4 weeks (control), log-rank χ²(1) = 7.93, p = .005. Adjusted hazard ratio for treatment was 0.52 (95% CI [0.31, 0.87], p = .013).”

Field-Specific Reporting Guidelines

Beyond APA formatting, certain study types have field-specific guidelines that journals require. Submission systems increasingly demand a completed checklist:

• Animal studies: ARRIVE 2.0 (Animal Research: Reporting of In Vivo Experiments) — required by Nature, BMJ, and ~1,000 other journals. The Essential 10 items cover study design, sample size, inclusion/exclusion, randomization, blinding, outcome measures, and statistical methods.
• Randomized clinical trials: CONSORT 2010 — 25-item checklist for parallel-group RCTs. Variants exist for cluster trials, non-inferiority, and pragmatic designs.
• Observational studies: STROBE — for cohort, case-control, and cross-sectional studies. Item 12 covers statistical methods, item 16 covers main results with confidence intervals.
• Systematic reviews / meta-analyses: PRISMA 2020 — 27-item checklist; submission systems often require the populated checklist as a supplementary file.
• Statistical methods generally: SAMPL (Statistical Analyses and Methods in the Published Literature) — broad guidelines by Lang & Altman (2015) covering all biomedical statistical reporting. Less formally enforced than ARRIVE or CONSORT but widely cited.

The EQUATOR Network hosts a searchable directory of all reporting guidelines; check it before drafting the methods section for any study type that isn't a standard bench experiment.

What Reviewers Catch

A Complete Methods Section Template

Pulling it together, here's a paragraph that hits every required element for a typical bench experiment:

Statistical analyses were performed in R 4.3.1 (R Core Team, 2023). Data are presented as mean ± SEM unless otherwise specified, with n indicating biological replicates. Normality was assessed by Shapiro-Wilk and variance homogeneity by Levene's test. Differences between two groups were compared by unpaired Student's t-test or Mann-Whitney U where normality was violated. Differences across three or more groups were assessed by one-way ANOVA with Tukey HSD post-hoc, or by Kruskal-Wallis with Dunn's post-hoc where appropriate. Significance was set at α = 0.05. Effect sizes are reported as Cohen's d for t-tests and partial η² for ANOVA. Exact p-values are reported to three decimal places; p-values below .001 are reported as p < .001.

This paragraph is reusable across most bench-experiment papers with minor edits. Drop it into a template, swap test names and effect sizes when your design differs, and the “methods section incomplete” revision request goes away.

For automated APA formatting of test statistics — the italics, leading-zero handling, and effect size with magnitude labels — tools like free biostatistics software handle the mechanics so the methods section becomes a writing task rather than a transcription task.