Paired T-Test (Dependent Samples)
Compare two related measurements on the same subjects, such as before and after a change. The paired t-test tests whether the mean of the differences is zero, giving the t-statistic, p-value and confidence interval for the mean difference.
Run Paired T-Test →What is a Paired T-Test?
A paired t-test, also called a dependent-samples t-test, compares two related measurements taken on the same subjects or matched pairs. Common examples are before-and-after measurements on the same units, or the same characteristic measured by two methods on the same items.
Rather than comparing two group means directly, the paired t-test works on the differences within each pair. It calculates the difference for every pair, then tests whether the mean of those differences is significantly different from zero. This is mathematically a one-sample t-test applied to the difference scores.
Pairing is powerful because it removes the variation between subjects. By comparing each subject to itself, the test cancels out person-to-person or unit-to-unit differences, leaving a more sensitive comparison of the effect than an independent two-sample test on the same data would provide.
In plain terms: When you measure the same things twice, before and after, comparing each item to itself is far more sensitive than comparing two separate groups. The paired t-test looks at the change within each pair and asks whether the average change is real or just noise.
Why Pairing Helps
Works on Differences
The test computes each pair's difference, then tests whether the mean difference is zero. It is a one-sample t-test on the differences.
Removes Between-Subject Variation
By comparing each subject to itself, pairing cancels individual differences, making the test more sensitive than an independent-samples test.
Related Measurements
The two measurements must be genuinely linked (same subject, matched pair). If they are independent, use the two-sample t-test.
Key Formulas
Interpreting the Result
The null hypothesis is that the mean difference is zero (no change). A small p-value means the average change across pairs is unlikely to be due to chance, so a real effect exists.
The confidence interval for the mean difference shows the plausible size of the effect. If it excludes zero, the test is significant, and its distance from zero indicates how large the change is, which matters more than significance alone for practical decisions.
Assumptions & Validation
Paired Observations
Each observation in one condition is meaningfully paired with one in the other.
If violated: If observations are independent, use the two-sample t-test.
Normality of Differences
The differences (not the raw values) are approximately normal.
If violated: Use the Wilcoxon signed-rank test for non-normal differences.
Random Sampling
Pairs are a random sample from the population of interest.
If violated: Address sampling bias.
⚠️ Check assumptions first
The paired t-test assumes the differences, not the original measurements, are approximately normal. A common error is checking normality of the raw before and after values instead of the difference scores. If the differences are strongly skewed or contain outliers, especially with a small number of pairs, use the Wilcoxon signed-rank test, the non-parametric paired alternative.
When NOT to Use Paired T-Test
Independent Groups
If the two sets of measurements are unrelated, use the two-sample t-test, not the paired test.
Non-normal Differences
For small samples with clearly non-normal differences, use the Wilcoxon signed-rank test.
Three or More Conditions
For repeated measurements across three or more conditions, use repeated-measures ANOVA.
Industry Applications
Before-and-After Studies
Test whether an intervention changed a measurement taken on the same subjects before and after.
Method Comparison
Compare two measurement methods applied to the same items to detect systematic bias.
Matched Pairs
Compare treatment and control when subjects are matched on key characteristics.
Process Change Verification
Confirm whether a process change shifted a characteristic measured on the same units.
Frequently Asked Questions
When should I use a paired t-test?
Use a paired t-test when you have two related measurements on the same subjects or matched pairs, such as before-and-after readings or the same items measured by two methods. The pairing removes between-subject variation, making the test more sensitive than an independent two-sample t-test. If the two sets of measurements are unrelated, the two-sample t-test is the correct choice instead.
How does a paired t-test differ from a two-sample t-test?
A two-sample t-test compares the means of two independent groups. A paired t-test compares two related measurements by working on the differences within each pair and testing whether their mean is zero. Because pairing cancels out individual subject differences, the paired test is more powerful for related data, but it is only valid when the observations are genuinely linked.
Why does the paired t-test use differences?
By reducing each pair to a single difference score, the test removes the variation that comes from differences between subjects, isolating the effect of the change or treatment. This is why it is equivalent to a one-sample t-test performed on the difference scores, testing whether their mean differs from zero. The approach is what gives the paired design its extra sensitivity.
Which values need to be normally distributed?
The paired t-test assumes the differences between the paired measurements are approximately normal, not the raw before and after values. A frequent mistake is to check normality of the original measurements. If the differences are strongly skewed or have outliers, particularly with few pairs, the Wilcoxon signed-rank test is the appropriate non-parametric alternative.
What is the non-parametric alternative to the paired t-test?
The Wilcoxon signed-rank test is the non-parametric counterpart. It ranks the absolute differences within pairs and uses the signs to test whether the differences are centered at zero, without assuming the differences are normally distributed. It is the recommended choice when the difference scores are clearly non-normal or the number of pairs is small.
How do I interpret the confidence interval for the mean difference?
The confidence interval gives the plausible range for the true average difference between the paired conditions. If it excludes zero, the change is statistically significant at the corresponding level. Its distance from zero indicates the magnitude of the effect, which is often more informative than the p-value for deciding whether the change is practically meaningful.
Test Before-and-After Change
Run a paired t-test on the differences with full statistics. Free during Beta.
Run Paired T-Test →