How to Report Independent t-Test Results

The independent-samples t-test is one of the most familiar comparisons in clinical papers, and its write-up is one of the easiest for a reviewer to pick apart. The two failures are the same every time: a bare P value with nothing to size the difference, and "no significant difference" written as if it meant "no difference". Both are fixable with one sentence, provided the sentence carries the right six pieces. A separate guide covers when the t-test is the right choice; this one picks up after you have run it and need to write it down.

The six elements of a complete t-test result

The SAMPL guidelines, the general-purpose statistical reporting checklist for biomedical journals, ask that a comparison report the summary statistic for each group, the difference between groups with a confidence interval, the test statistic, and the exact P value.³ For a t-test that works out as follows.

The mean difference and its interval are the elements most often missing, and they are the only two that say anything about clinical size. A t value and a P value between them tell the reader how surprising the data would be if the groups were identical; they do not tell the reader how far apart the groups are.

A real result: age at surgery and tumour ulceration

The melanoma cohort from Odense University Hospital records 205 patients who had radical surgery for malignant melanoma, with age at surgery and whether the tumour was ulcerated.¹ Age is roughly normal in both groups (Shapiro-Wilk P = .27 with ulceration, P = .33 without) and the variances are similar (Levene P = .69), so the t-test is the correctly routed comparison rather than a rank test. Ninety patients had ulcerated tumours and 115 did not.

Example data: the melanoma cohort of Andersen, Borgan, Gill, and Keiding (1993), 205 patients treated at Odense University Hospital, Denmark; a public teaching dataset. Computed with scipy, not generated.

Two things about that card are worth noticing before you write the sentence. First, the engine reports Welch's t-test by default, so the degrees of freedom are fractional (182.5) rather than the 203 a pooled Student's test would give; for this dataset Student's version returns t(203) = 1.81, P = .072, the same conclusion by a slightly different route. Second, the effect size is printed as Hedges' g, the small-sample corrected form of Cohen's d; at n = 205 the two agree to two decimals (0.25), so either label is defensible as long as you name the one you use.⁷

Writing the AMA sentence

The AMA Manual of Style, the format most clinical and surgical journals follow, uses a capital italic P with no leading zero, keeps the leading zero on effect sizes and coefficients, writes the interval as "95% CI, lower to upper", and puts the group descriptives in the form mean (SD) rather than mean ± SD.² Assembled from the card above:

"Patients with ulcerated tumours were older at surgery than those without (mean [SD], 54.8 [17.4] vs 50.6 [15.9] years), but the difference was not statistically significant (mean difference, 4.2 years; 95% CI, −0.4 to 8.9; Welch t_182.5 = 1.79; P = .08; Cohen d = 0.25)."

One detail differs between the card and the sentence. StatsPlease prints P to three decimals (P = .075). The AMA Manual's convention is two digits for P values of .01 or more, three digits between .001 and .01, and P < .001 below that, keeping a third digit only when rounding would carry a value across .05.² So .075 becomes P = .08 in the manuscript. Trim the digit when you write the sentence; do not round the interval or the effect size, which keep their own precision.

The Methods section carries the rest: "Age was compared between groups with Welch's independent-samples t-test, which does not assume equal variances. Normality was assessed with the Shapiro-Wilk test and homogeneity of variance with Levene's test. Cohen d is reported as the effect size, and all tests were two-sided with α = .05."

This is also the sentence ChatGPT will write for you fluently and wrongly. Ask an LLM to report an independent t-test and it will produce a perfectly formatted AMA sentence around a t, a P, and an interval it never computed from your data. The numbers above are computed with scipy, not generated; upload the same two columns to R or SPSS and the t statistic, the degrees of freedom, and the interval match to the last decimal.

How to report a result that is not significant

The example above was chosen because it is not significant, and that is where most t-test reporting fails. "No significant difference in age between groups" is the sentence reviewers expect to see, and it is wrong in a specific way: the confidence interval runs from a 0.4-year deficit to an 8.9-year excess. The data are compatible with the ulcerated group being nearly nine years older, a gap any clinician would call meaningful, and equally compatible with no gap at all. What the study has shown is that it could not tell; absence of evidence is not evidence of absence.⁶

Two phrases to strike from the draft. "NS" and "P > .05" both discard information the reader needs; the AMA Manual asks for the exact value wherever it is available.² And "a trend toward significance" describes a P value of .08 as if it were on its way somewhere. It is not; a P value is a single number from a single sample. If the interval spans clinically important differences, say that in words: "the study was not powered to exclude an age difference of up to 9 years."

Welch or Student: name the one you ran

Student's t-test assumes the two groups have equal variances; Welch's version does not, and it costs almost nothing in power when the variances happen to be equal. Delacre, Lakens, and Leys made the case in 2017 that Welch should be the default rather than the fallback, because the two-step routine of "run Levene, then pick" inflates error rates and the equal-variance assumption is rarely worth defending.⁴ StatsPlease reports Welch by default for that reason. Whichever you use, name it: "Welch t" or "Student t" in the sentence, and the fractional degrees of freedom of the Welch test are not a typo to be rounded away.

If a reviewer asks why the degrees of freedom are 182.5 and not 203, the answer is one line: Welch's approximation adjusts the degrees of freedom for the two groups' separate variances and sample sizes, and SPSS prints the same figure on its "Equal variances not assumed" row. The Levene and Welch guide works through the mechanism.

Where APA differs

If the target journal uses APA style, the same six elements are reported with different typography: a lowercase italic p (still with no leading zero), M and SD as labelled abbreviations, and the interval in square brackets, 95% CI [−0.4, 8.9]. The numbers do not change. StatsPlease's deterministic output is AMA by default; the Reporting Style picker in the Analyze flow switches the rendering to APA for journals that ask for it.

Frequently asked questions

Should I report Welch's or Student's t-test?

Report the one you ran, and name it in the sentence. Welch's t-test does not assume equal variances and loses almost nothing when the variances are equal, which is why several methodologists recommend it as the default and why StatsPlease reports it. If you ran Student's test after a Levene test, say so in the Methods; the two versions give different degrees of freedom and slightly different P values.

How do I report a t-test result that is not statistically significant?

Exactly as you would report a significant one: both group means with SD and n, the mean difference with its 95% confidence interval, the t statistic with degrees of freedom, the exact P value, and an effect size. Never write NS or P greater than .05. The interval is the part that matters, because it shows which differences the data have and have not ruled out.

Do I need an effect size for an independent t-test?

Yes. Reporting guidelines for biomedical journals ask for an estimate of the size of the effect alongside the test, and a t statistic with a P value does not give one. Report the raw mean difference with its confidence interval, and add Cohen's d or Hedges' g if the journal or the reader needs a standardised figure that can be compared across studies.

References

1. Andersen PK, Borgan Ø, Gill RD, Keiding N. Statistical Models Based on Counting Processes. New York, NY: Springer-Verlag; 1993.
2. Christiansen SL, Iverson C, Flanagin A, et al. AMA Manual of Style: A Guide for Authors and Editors. 11th ed. New York, NY: Oxford University Press; 2020. https://doi.org/10.1093/jama/9780190246556.001.0001
3. 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. Int J Nurs Stud. 2015;52(1):5-9. https://doi.org/10.1016/j.ijnurstu.2014.09.006
4. Delacre M, Lakens D, Leys C. Why psychologists should by default use Welch's t-test instead of Student's t-test. Int Rev Soc Psychol. 2017;30(1):92-101. https://doi.org/10.5334/irsp.82
5. Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates; 1988.
6. Altman DG, Bland JM. Absence of evidence is not evidence of absence. BMJ. 1995;311(7003):485. https://doi.org/10.1136/bmj.311.7003.485
7. Lakens D. Calculating and reporting effect sizes to facilitate cumulative science: a practical primer for t-tests and ANOVAs. Front Psychol. 2013;4:863. https://doi.org/10.3389/fpsyg.2013.00863