How to report a t-test: templates for Welch, Student and paired tests

Quick answer. Report, in this order: what you compared and in how many independent units; each group’s mean and SD; the difference with its 95% confidence interval, saying which group was subtracted; then the test — which t-test (Welch, Student or paired; two-sided unless justified), t with its degrees of freedom, and the exact p-value. For example: mean 63.6 pg/mL (SD 13.0, n = 6) in treated mice and 50.9 pg/mL (SD 2.7, n = 12) in controls; difference (treated − control) 12.7 pg/mL, 95% CI −0.9 to 26.4; two-sided Welch’s t(5.22) = 2.37, p = 0.062.

The checklist reviewers apply

Report	Why	Typical gap
The exact test: Welch’s (no equal-variance assumption), Student’s (assumes equal variances) or paired; one- or two-sided	Different tests give different p-values for the same data	“Data were analysed by t-test”
n per group and what n counts (animals, cultures, patients — the unit of analysis)	Lets the reader judge independence and power	Wells or cells (technical replicates) counted as n
Mean and SD (standard deviation, the typical spread of individual values) for each group	SD describes the data; SEM (standard error of the mean) only describes the precision of the mean	SEM shown as if it were variability
Difference between means with 95% CI	Gives the size of the effect and its uncertainty	Only a p-value
t, degrees of freedom, exact p	Lets anyone check the numbers agree	“p < 0.05” or “p = 0.000”
Software and version	Defaults differ between programs	Missing

These follow the SAMPL guidelines for basic statistical reporting in biomedical journals, which ask authors to name the test, state whether it was one- or two-sided and paired or independent, give the actual p-value and report estimates with confidence intervals (Lang & Altman 2015); the APA’s reporting standards, which ask for effect-size estimates and confidence intervals for each inferential test (Appelbaum et al. 2018); and the t-test checklist of a review that found 53% of physiology papers using t-tests did not say whether any were paired (Weissgerber et al. 2018, Box 2). For p-values, APA style is to give exact values to two or three decimals and “p < .001” below that (APA Numbers and Statistics Guide); many biology journals keep the leading zero (p = 0.031).

Where to find each number

Number	R t.test output	SciPy result	GraphPad Prism
t	t =	.statistic	t
Degrees of freedom (df: how much information the variance estimate rests on)	df =	.df	df
p	p-value =	.pvalue	P value
95% CI of the difference (the range of differences compatible with the data)	95 percent confidence interval	.confidence_interval()	95% confidence interval
Which group was subtracted	First group listed minus second	First argument minus second	Shown in the results table

“Two-sided” means the test looks for a difference in either direction; it is the default in all three.

Templates

Replace the bracketed parts. Keep the order: estimate first, test second.

Welch’s t-test (two independent groups, the recommended default)

[Outcome] was [higher/lower] in [group A] (mean [x], SD [s], n = [n]) than in [group B] (mean [x], SD [s], n = [n]); difference ([A] − [B]) [d] [units], 95% CI [lower] to [upper]; two-sided Welch’s t([df, to two decimals]) = [t], p = [exact p].

Example: Mean serum marker was 63.6 pg/mL (SD 13.0, n = 6) in treated mice and 50.9 pg/mL (SD 2.7, n = 12) in controls; difference (treated − control) 12.7 pg/mL, 95% CI −0.9 to 26.4; two-sided Welch’s t(5.22) = 2.37, p = 0.062. (Worked through in Welch vs Student.)

Welch’s degrees of freedom are usually not a whole number; report them as calculated, rounded to two decimals.

Student’s t-test (equal variances assumed, with a stated reason)

… difference [d], 95% CI [lower] to [upper]; Student’s t([n₁ + n₂ − 2]) = [t], p = [p].

Paired t-test (the same subjects measured twice)

[Outcome] changed from [mean before] (SD [s]) to [mean after] (SD [s]) in [n] [subjects]; mean change [d] (SD of changes [s]), 95% CI [lower] to [upper]; paired t([n − 1]) = [t], p = [p].

Example: Fasting glucose fell from 9.19 mmol/L (SD 1.86) to 8.29 mmol/L (SD 1.67) in 8 mice; mean change −0.90 mmol/L (SD of changes 0.58), 95% CI −1.39 to −0.41; paired t(7) = −4.37, p = 0.003.

A non-significant result

… difference [d], 95% CI [lower] to [upper]; Welch’s t([df]) = [t], p = [p]. The interval is compatible with [no difference / an increase of up to (upper)].

Say what the interval allows; do not write “there was no difference”. A p-value above 0.05 is not evidence of no effect (ASA statement on p-values, principles 3 and 5).

Figure legend

Each point is one [mouse]; bars show mean ± SD (n = [n] per group). Groups were compared with a two-sided Welch’s t-test (p = [p]).

Show the individual points rather than a bar alone: bar graphs of small samples can hide very different distributions (Weissgerber et al. 2015).

Methods paragraph

Two groups were compared with a two-sided Welch’s t-test, which does not assume equal variances[, specified before the data were analysed]. Paired measurements were compared with a paired t-test on the within-subject differences. The unit of analysis was the [animal]; technical replicates were averaged within each [animal]. Results are reported as mean differences with 95% confidence intervals. Analyses used [R 4.3.3 / SciPy 1.18.1]; p < 0.05 was considered statistically significant.

Numbers that reviewers check for consistency

• df must match the design. Student’s: n₁ + n₂ − 2. Paired: number of pairs − 1. Welch’s: a decimal between the smaller group’s n − 1 and n₁ + n₂ − 2 — for 5 vs 7 mice, between 4 and 10.
• t, df and p must agree. Reviewers and automated checkers recompute p from t and df; a mismatch suggests a copying error.
• The CI and p must agree. A two-sided 95% CI excludes zero exactly when p < 0.05 for the same test.
• Means, SDs and n must reproduce the t. For Welch’s test, t = (mean₁ − mean₂) ÷ √(SD₁²/n₁ + SD₂²/n₂). Recompute from the unrounded data where you can: rounded summaries shift the last digit (the paired example’s t = −4.37 comes from the raw values; the rounded SD of 0.58 would give −4.39).
• Several t-tests in one paper need a plan. If you test many outcomes or many pairs of groups, either correct for multiple comparisons (Holm’s method, a step-down version of Bonferroni, is a safe default) or name one primary comparison and label the rest exploratory. For three or more groups, compare them together first (ANOVA) rather than with t-tests between every pair.

Effect sizes: which d?

Standardised effect sizes help comparisons across studies and power calculations, but different designs need different versions (Lakens 2013):

• Two independent groups: Cohen’s d_s = difference ÷ pooled SD (Hedges’ g corrects its small-sample bias).
• Paired: d_z = mean change ÷ SD of the changes — the one used for paired power calculations — or d_av, which divides by the average of the two SDs and is comparable with between-subject studies.

Name the version you report. The raw difference with its CI remains the primary result in most biology papers.

Rewrite these phrases

Each row turns a phrase reviewers query into one they accept.

Instead of	Write
“p = 0.000”	“p < 0.001”
“p < 0.05” alone	The exact p-value, e.g. “p = 0.031”
“There was a trend towards significance (p = 0.07)”	“difference 4.1, 95% CI −0.4 to 8.6; p = 0.07”
“No difference between groups (p = 0.4)”	“difference 1.2, 95% CI −2.0 to 4.4; p = 0.4”
“Data were analysed by t-test”	“two-sided Welch’s t-test”
“mean ± SEM” for describing data	“mean ± SD” (use SEM or CI only to show precision, and say which)

References

• Appelbaum M, Cooper H, Kline RB, et al. (2018). Journal article reporting standards for quantitative research in psychology: the APA Publications and Communications Board task force report. American Psychologist 73(1):3–25. doi:10.1037/amp0000191
• Lakens D (2013). Calculating and reporting effect sizes to facilitate cumulative science: a practical primer for t-tests and ANOVAs. Frontiers in Psychology 4:863. doi:10.3389/fpsyg.2013.00863
• American Psychological Association (2022). APA Style Numbers and Statistics Guide (7th ed.). PDF
• Lang TA, Altman DG (2015). Basic statistical reporting for articles published in biomedical journals: the “Statistical Analyses and Methods in the Published Literature” or the SAMPL Guidelines. International Journal of Nursing Studies 52(1):5–9. doi:10.1016/j.ijnurstu.2014.09.006
• Wasserstein RL, Lazar NA (2016). The ASA statement on p-values: context, process, and purpose. The American Statistician 70(2):129–133. doi:10.1080/00031305.2016.1154108
• Weissgerber TL, Milic NM, Winham SJ, Garovic VD (2015). Beyond bar and line graphs: time for a new data presentation paradigm. PLOS Biology 13(4):e1002128. doi:10.1371/journal.pbio.1002128
• Weissgerber TL, Garcia-Valencia O, Garovic VD, Milic NM, Winham SJ (2018). Why we need to report more than ‘Data were analyzed by t-tests or ANOVA’. eLife 7:e36163. doi:10.7554/eLife.36163