Paired T-Test Calculator

Analyze before-after and matched-pair studies. Test whether the mean difference between paired observations is statistically significant, with formulas, effect sizes, and confidence intervals.

Data entry:

Example: 10 students' quiz scores before and after a tutoring intervention. Replace with your own data.

T 0.8486
DF 9
P 0.4190
COHEN'S D 0.2684
95% CI [3.8852, 8.9148]
There is not enough evidence to reject H₀ at α = 0.05.
T distribution with observed test statistic
Fig. 1 — Observed t statistic (red dashed) on the t distribution (df = 9).
  1. d̄ = 6.4, s_d = 2.632, n = 10 Difference statistics.
  2. t = 6.4 / (2.632 / √10) = 6.4 / 0.8326 = 7.688 t-statistic.
  3. df = 10 − 1 = 9 Degrees of freedom.
  4. p = 0.0001 Two-tailed p-value.
  5. d = 7.688 / √10 = 2.431 Cohen's d effect size.

§1 The Formula

The paired t-test compares two related samples by analyzing the mean difference between paired observations.

t = d̄ / (s_d / √n)
Mean of the differences (after − before)
s_d
Standard deviation of the differences
n
Number of pairs
df
Degrees of freedom = n − 1

§2 Pretest/Posttest Worked Example

Ten students take a quiz before and after a tutoring session. Do their scores improve significantly?

Student Before After Difference (d)
172786
268757
375827
470766
573818
671798
774839
869778
976859
1072808

Step 1: Calculate the differences for each student. The differences are: 6, 7, 7, 6, 8, 8, 9, 8, 9, 8.

Step 2: Compute the mean and SD of differences. d̄ = (6 + 7 + 7 + 6 + 8 + 8 + 9 + 8 + 9 + 8) / 10 = 76 / 10 = 7.6

Step 3: Compute the t-statistic. s_d = 1.075 (sample SD of differences); SE = 1.075 / √10 = 0.340

Step 4: Calculate t. t = 7.6 / 0.340 = 22.35, df = 9, p < 0.0001 (highly significant).

Step 5: Report effect size and CI. Cohen's d = 22.35 / √10 = 7.07 (very large); 95% CI [6.76, 8.44].

Why Paired Designs Need Fewer Participants & Have More Power

Paired designs control for individual differences by comparing each person to themselves. This reduces variability in the differences compared to the variability between two independent groups. Because paired designs remove person-level "noise," they need fewer participants to detect the same effect size.

Example: Suppose you want to detect a medium effect (Cohen's d = 0.5) at α = .05 with 80% power. An independent-samples t-test requires roughly 128 participants per group (256 total). A paired t-test requires roughly 34 participants (one sample of 34 measured twice). The reduction occurs because pairing eliminates between-subject variance.

This efficiency gain comes with a cost: pairing assumes observations within a pair are related, and the test is sensitive to violations of independence among the differences. Always report your design (paired or independent) clearly in methods.

§3 Assumptions & When This Test Is Wrong

Pairing Structure

The two observations must be genuinely paired (before-after on the same person, matched pairs, etc.). The differences must have a clear definition. If pairing is arbitrary or weak, use an independent-samples t-test instead.

Independence of Differences

The differences between pairs must be independent of each other. If pairs are clustered (e.g., multiple measurements per day, students in the same class), consider a multilevel or mixed-effects model.

Approximate Normality of Differences

The differences should be approximately normally distributed. With small samples (n < 15), check a histogram or Q–Q plot of the differences. The test is fairly robust to mild non-normality. If distributions are severely skewed or have outliers, use the Wilcoxon signed-rank test instead.

Nonparametric Alternative

If the differences are non-normal or ordinal, use the Wilcoxon signed-rank test. It tests whether the median difference is zero and makes no assumption about normality, though it is slightly less powerful if normality holds.

§4 How to Report a Paired t-Test in APA Style

Use this format in results sections and tables:

t(df) = t-value, p = p-value, d = effect size, 95% CI [lower, upper]

Example: Scores increased significantly from pretest (M = 72.0, SD = 2.1) to posttest (M = 79.0, SD = 3.0), t(9) = 22.35, p < .001, d = 7.07, 95% CI [6.76, 8.44].

Formatting notes:

  • Always include df in parentheses.
  • Report p-values with a leading zero (p = .045) unless p < .001.
  • Always use a 95% two-sided CI, even if your test is one-tailed.
  • Cohen's d, t, and p are italicized.
  • State the pairing in the results (e.g., "paired samples" or "matched pairs").

§5 FAQ

What if a participant is missing either their pre- or post-measurement?

They must be excluded from the analysis entirely. Pairing requires both values to compute the difference. Do not skip them or pair them with someone else. Report the number of participants excluded and why in your methods. Missing data reduces your sample size and statistical power, so try to minimize it during data collection.

Can I use a paired t-test for unequal time intervals between measurements?

Yes. The test does not require equal time intervals between before and after measurements. It only requires two measurements per participant. However, if your time intervals vary widely, describe them in your methods. Some researchers also use regression or mixed models to account for varying intervals, but paired t-test results are still valid.

Should I use a one-tailed or two-tailed test?

Use two-tailed by default. A two-tailed test is more conservative and avoids the temptation to "hunt" for significance in the direction that looks good. Only use one-tailed if your hypothesis predicted the direction of change before seeing the data. Always report which you used in methods.

How do I report Cohen's d for a paired t-test?

Divide the mean difference by the standard deviation of the differences: d = d̄ / s_d. This is sometimes called the standardized mean change. Cohen's conventions apply: d ≈ 0.2 (small), 0.5 (medium), 0.8 (large). Always report d alongside p to convey effect magnitude.

What is the difference between paired and independent t-tests?

Paired t-tests compare two related measurements on the same individuals (before-after or matched pairs). Independent t-tests compare two separate groups of different individuals. Paired designs are more powerful for the same sample size because they control for individual differences. However, paired designs require the pairing structure to be meaningful. Choose based on your study design, not on which p-value looks better.

Should I correct for multiple comparisons if I run paired t-tests on many outcomes?

Yes. If you test multiple outcomes (e.g., 10 different measures), apply a Bonferroni correction or false discovery rate (FDR) adjustment to control error rates. For example, with Bonferroni correction, divide your α (e.g., 0.05) by the number of tests. Reporting unadjusted p-values without mentioning multiple comparisons risks overstating significance.

§6 Sources

§7 Related Tools

Explore statistical tests and related calculators: