Use a paired t-test when the same people (or matched pairs) are measured twice, and use an independent t-test when you're comparing two separate, unrelated groups. The single question that decides it: are your two sets of scores linked to the same individuals, or do they come from completely different people? Pick the wrong one and your p-value, degrees of freedom, and conclusions can all be wrong — a surprisingly common mistake in theses and journal submissions.
Key Takeaways
- A paired t-test compares two measurements from the same subjects (e.g., before vs after treatment) and tests whether the average difference within each person is zero.
- An independent t-test compares the means of two separate groups (e.g., treatment vs control) made up of different people.
- The deciding question is always: "Are these two scores from the same subject, or from different subjects?" Same subject → paired. Different subjects → independent.
- Using a paired test when you should use an independent one (or vice versa) changes your degrees of freedom and p-value, and can flip a non-significant result into a "significant" one.
- Both assume roughly normally distributed data; if that fails, use the Wilcoxon signed-rank test (paired) or the Mann-Whitney U test (independent).
What is the difference between a paired and independent t-test?
The difference is whether your two columns of data are connected to the same subjects. A paired t-test (also called a dependent or repeated-measures t-test) looks at two measurements taken from each individual and analyses the difference within each person. An independent t-test (also called a two-sample t-test) compares the average of one group of people against the average of a completely different group of people.
Think of it this way: in a paired design, every data point in column A has a partner in column B that belongs to the same person. In an independent design, there is no pairing — row 3 in group 1 has nothing to do with row 3 in group 2.
This isn't a cosmetic distinction. The paired t-test exploits the fact that each person is their own control, which cancels out individual differences and usually gives you more statistical power with fewer participants.
When should I use a paired t-test?
Use a paired t-test whenever each participant contributes two related scores, or when you have naturally matched pairs. Classic situations include:
- Before-and-after designs: anxiety scores before therapy and after therapy, for the same patients.
- Repeated measures: reaction time under caffeine vs placebo, with every participant doing both conditions.
- Matched pairs: twins, or participants matched on age and sex, where one of each pair gets the treatment.
The test takes the difference for each pair, then checks whether the average of those differences is significantly different from zero. Because it works on within-person change, it strips out stable individual traits (one person being naturally faster, another naturally more anxious).
When should I use an independent t-test?
Use an independent t-test when you compare two groups made up of different, unrelated people and each person provides only one score. Typical cases:
- Men vs women on a memory task.
- A group given a new drug vs a separate control group given a placebo.
- Students taught with Method A vs a different cohort taught with Method B.
Here there's no meaningful way to pair up individual scores, so the test compares the two group means and asks whether the gap between them is bigger than you'd expect from random sampling variation alone. If you're unsure your groups are normally distributed, see our guide on the Mann-Whitney U vs the t-test.
Paired vs independent t-test: key differences
| Feature | Paired t-test | Independent t-test |
|---|---|---|
| Also known as | Dependent / repeated-measures t-test | Two-sample t-test |
| Data structure | Two scores per subject | One score per subject, two groups |
| Example | Before vs after, same people | Treatment group vs control group |
| What it tests | Mean difference = 0 | Difference between two group means = 0 |
| Degrees of freedom | n − 1 (pairs) | n₁ + n₂ − 2 |
| Key assumption | Differences are normally distributed | Each group normal; equal variances (or use Welch's) |
| Non-parametric backup | Wilcoxon signed-rank test | Mann-Whitney U test |
A worked example with real numbers
Imagine a study of 20 participants testing whether a mindfulness app lowers self-reported stress.
If every participant used the app for four weeks and you measured stress before and after, this is a paired design. Suppose the mean stress dropped from 6.8 to 5.4, with a mean within-person difference of 1.4 points. A paired t-test gives:
t(19) = 3.41, p = .003, d = 0.76, 95% CI [0.52, 2.28]
The t(19) tells you there were 20 pairs (df = n − 1 = 19). The p = .003 means a drop this large would be very unlikely if the app did nothing. Cohen's d = 0.76 is a medium-to-large effect, and the confidence interval not crossing zero confirms a real reduction.
Now imagine instead you had two different groups — 10 people using the app and 10 using nothing — each measured once at the end. That's an independent design. You might get:
t(18) = 2.10, p = .050, d = 0.94, 95% CI [0.00, 2.80]
Notice the degrees of freedom are now 18 (n₁ + n₂ − 2), and even with a similar effect size the p-value is weaker — because the independent design can't remove individual differences the way the paired design does. Same research question, different design, different test, different conclusion.
What happens if I pick the wrong test?
Choosing the wrong t-test produces an incorrect p-value and can lead to a false conclusion. If you run an independent t-test on paired data, you throw away the pairing information, inflate your error variance, and usually lose power — meaning you may miss a real effect. Run a paired t-test on genuinely independent groups and you'll compute meaningless "differences" between unrelated people, giving a result that's simply invalid.
In practice, reviewers and supervisors catch this often. A 2016 methods review in psychology estimated that a meaningful share of published t-tests used degrees of freedom inconsistent with their described design — a red flag that the wrong test was applied. Getting the design-to-test match right is the first thing a careful reader checks.
How do I run and report this correctly?
Both tests require only seconds to compute, but the reporting is where many students lose marks. APA 7 expects the test statistic, degrees of freedom, exact p (no leading zero), an effect size (Cohen's d), and ideally a confidence interval — exactly as shown in the examples above.
This is where a tool helps. StatRyx is an AI-powered statistical analysis tool that replaces manual SPSS workflows with automated, APA 7-formatted reporting. Instead of guessing between a paired and independent test, you upload your data, StatRyx detects the structure, checks normality assumptions, selects the correct test (including the Welch correction or a non-parametric alternative if needed), and writes the APA 7 results sentence for you. Unlike R (powerful but code-heavy) or desktop installs like JASP and jamovi, it runs in your browser with no setup, and it won't cost you the roughly $99/month that an SPSS subscription typically runs.
Stop calculating this by hand — run it free in StatRyx → Try StatRyx
Frequently Asked Questions
How do I know if my data is paired or independent?
Ask whether each score in one column belongs to the same person (or matched pair) as a score in the other column. If yes — such as a before-and-after measurement on