The paired samples t-test (also called the dependent samples t-test, repeated measures t-test, or matched pairs t-test) compares the means of two related measurements. Use it when the same subjects are measured twice — for example, before and after an intervention — or when two subjects are deliberately matched on important characteristics. By computing the difference for each pair, the test removes between-subject variability, making it more statistically powerful than an independent t-test for matched designs.

Common applications include pre-test/post-test educational studies, crossover clinical trials, left-eye vs right-eye medical measurements, and paired ecological transects.

The p-value is the probability of observing a mean difference as large as yours (or larger) if there were truly no difference between the two conditions — in other words, if chance alone were operating. A p-value of 0.03 means there is a 3% chance of seeing a difference this large by chance.

Important: the p-value is NOT the probability that the null hypothesis is true, and it is NOT the probability of making an error. A p-value below your chosen α (e.g., .05) means you reject the null hypothesis — but always look at the effect size to judge whether the difference is practically meaningful.

Statistical significance (p < α) tells you that the observed difference is unlikely to be due to chance at your chosen threshold. It does NOT tell you whether the difference is large enough to matter in practice.

With a very large sample (e.g., n = 5,000 pairs), even a trivially small mean difference (e.g., 0.1 mmHg blood pressure change) will be statistically significant. That's why you must always report and interpret the effect size (Cohen's d) alongside the p-value. A difference can be statistically significant but practically negligible — or practically important even when p is slightly above .05 (consider the power of your study).

Cohen's d for the paired t-test is calculated as the mean difference divided by the standard deviation of the differences (d̄ / s_d). It tells you how large the mean change is relative to the variability in change scores.

Cohen's (1988) benchmarks: d = 0.2 = small (small but real effect, roughly the difference between heights of 15- and 16-year-old girls); d = 0.5 = medium (noticeable to careful observers, e.g. a moderate intervention effect); d = 0.8 = large (obvious to any observer, e.g. a highly effective treatment). Note that these benchmarks were designed for social sciences — in medical research, a d of 0.2 may still be clinically meaningful if the outcome is mortality.

The paired t-test requires: (1) Paired / related observations — each value in Group 1 must correspond to one specific value in Group 2; (2) Normality of differences — the d_i = x1_i − x2_i values should be approximately normally distributed (verified by Shapiro-Wilk test or Q-Q plot); (3) No extreme outliers in the differences — a single outlier can dramatically inflate the SD and distort the result; (4) Continuous or interval measurement — the test is not valid for ordinal or nominal data.

If normality is violated and n < 30: use the Wilcoxon Signed-Rank Test. If you have only ordinal data: use Wilcoxon Signed-Rank Test. If n > 30: the Central Limit Theorem means non-normality is less critical, and you can usually proceed.

The minimum number of pairs depends on the size of the effect you want to detect. For 80% power at α = .05 (two-tailed): detecting a large effect (d = 0.8) requires approximately 15 pairs; detecting a medium effect (d = 0.5) requires approximately 34 pairs; detecting a small effect (d = 0.2) requires approximately 198 pairs.

With fewer than 10 pairs, the normality assumption becomes almost impossible to verify, and the test can be unreliable even if p < .05. Samples with 5–9 pairs should be treated with caution and supplemented with a Wilcoxon Signed-Rank test for verification.

A two-tailed test detects a mean difference in either direction (Group 1 could be higher OR lower than Group 2). A one-tailed test tests only one direction — either that Group 1 is higher (right-tailed) or that Group 1 is lower (left-tailed). One-tailed tests are more powerful (easier to reach significance) but must be justified by a strong, pre-specified directional hypothesis made before data collection.

Recommendation: use a two-tailed test by default. Applying a one-tailed test after seeing your data direction is considered p-hacking and can inflate the Type I error rate. Only switch to one-tailed if your pre-registration or study protocol explicitly committed to a directional hypothesis.

APA 7th edition requires the test name, means and SDs of both groups, the t-statistic, degrees of freedom, exact p-value (or "p < .001"), effect size (Cohen's d), and confidence interval.

Example: "A paired samples t-test was conducted to assess whether blood pressure changed from pre-treatment (M = 148.3, SD = 9.2) to post-treatment (M = 135.7, SD = 8.6). Results indicated a statistically significant decrease, t(11) = 4.23, p = .001, d = 1.22, 95% CI [6.1, 19.2] mmHg." Note: use italics for all statistical symbols in actual manuscripts. See the "How to Write Your Results" section above for all 5 reporting templates.

This calculator is designed for education, learning, and exploratory data analysis. The computational formulas used are standard and accurate. For formal publication, always verify results using peer-reviewed statistical software such as R (base::t.test), Python (scipy.stats.ttest_rel), SPSS, SAS, or Stata.

To cite this tool in your work: StatsUnlock. (2025). Paired samples t-test calculator. Retrieved from https://statsunlock.com/tools/paired-samples-t-test-calculator/

No. A non-significant result (p > α) does not prove there is no effect — it means your data do not provide sufficient evidence to reject the null hypothesis at your chosen significance level. This could be because: (a) there genuinely is no meaningful effect, (b) your sample size was too small to detect the effect (low statistical power), or (c) the effect size is smaller than your study was designed to detect.

Always check the confidence interval for the mean difference. If the entire CI is within your "practically unimportant" zone, you can more confidently conclude the effect is negligible. If the CI is wide or overlaps meaningful values, the result is inconclusive — not negative. Consider a Bayesian analysis (Bayes Factor) to quantify evidence in favor of the null hypothesis.