Sign In
T-Tests: One-Sample, Independent, and Paired Explained
The t-test is the workhorse of inferential statistics. When a researcher wants to compare means, either a single sample against a known value or two groups against each other, a t-test is usually the first tool considered. Yet the choice between the three main types (one-sample, independent samples, and paired samples) causes frequent errors in published research, and using the wrong test can produce misleading results that peer reviewers catch.
This guide explains what t-tests are, when to use each type, the assumptions they require, and how to calculate, interpret, and report them. Worked examples appear throughout so the calculations are concrete, not abstract.
What Is a T-Test?
A t-test is a statistical procedure that determines whether the difference between means is likely to reflect a real effect in the population or could plausibly be attributed to sampling variability. All t-tests use the same underlying logic: calculate a t-statistic that measures the observed difference relative to the standard error of that difference, then compare the t-statistic to a t-distribution to obtain a p-value.
The t-distribution is similar to the normal distribution but has heavier tails, which accounts for the additional uncertainty that comes with estimating variability from a sample rather than knowing the population variance. As sample size increases, the t-distribution converges toward the normal distribution.
T-tests belong to the broader family of hypothesis testing procedures. Every t-test starts with a null hypothesis (typically that no difference exists) and an alternative hypothesis (that a difference does exist), and the test evaluates the evidence for rejecting the null.
The Three Types of T-Tests
The three t-tests differ in the structure of the data they compare, not in their underlying logic. Choosing the correct type is the single most important decision in any t-test analysis.
| T-Test Type | Data Structure | Example Research Question |
|---|---|---|
| One-sample t-test | One group of scores compared to a known or hypothesized value | Do the reading scores of students at this school differ from the national average of 500? |
| Independent samples t-test | Two separate groups of participants compared to each other | Do students taught with method A score higher than students taught with method B? |
| Paired samples t-test | Two measurements from the same participants (or matched pairs) | Do participants score higher on the posttest than on the pretest? |
The distinction between independent samples and paired samples depends on whether the two sets of scores come from separate groups or from linked observations. A pre-post design uses a paired test because each participant contributes two scores. A comparison of two intervention groups uses an independent test because each participant contributes one score to one group.
The One-Sample T-Test
The one-sample t-test compares the mean of a single sample to a specific value, usually a known population mean or a benchmark of theoretical interest. It answers the question: is our sample mean far enough from this reference value to suggest a real difference?
The formula for the t-statistic is:
t = (x̄ − μ) / (s / √n)
where x̄ is the sample mean, μ is the reference value, s is the sample standard deviation, and n is the sample size.
Example. A researcher wants to know whether the average reading score of 30 students at a particular school differs from the national mean of 500. The sample mean is 512 with a standard deviation of 40.
Step 1. Calculate the standard error: SE = 40 / √30 = 7.30
Step 2. Calculate the t-statistic: t = (512 − 500) / 7.30 = 1.64
Step 3. Identify degrees of freedom: df = n − 1 = 29
Step 4. Compare to the critical t-value or calculate a p-value. For a two-tailed test at α = 0.05 with 29 degrees of freedom, the critical value is 2.045. Because 1.64 is less than 2.045, we fail to reject the null hypothesis.
The p-value is approximately 0.112, meaning the school's mean is not significantly different from the national mean at the conventional 0.05 level.
The Independent Samples T-Test
The independent samples t-test compares the means of two separate groups. It is sometimes called a two-sample t-test or Student's t-test. The two groups must contain different participants, with no overlap or pairing between them.
The formula for the t-statistic when equal variances are assumed is:
t = (x̄₁ − x̄₂) / √(s²p (1/n₁ + 1/n₂))
where s²p is the pooled variance calculated from the two samples. Most statistical software calculates this automatically, so the formula matters less than knowing when the test is appropriate.
Example. A researcher compares final exam scores between two teaching methods. Method A has 25 students with a mean of 78 and standard deviation of 10. Method B has 28 students with a mean of 73 and standard deviation of 12.
Step 1. Calculate the pooled variance: s²p = [(24 × 100) + (27 × 144)] / (25 + 28 − 2) = (2400 + 3888) / 51 = 123.29
Step 2. Calculate the standard error: SE = √(123.29 × (1/25 + 1/28)) = √(123.29 × 0.0757) = √9.33 = 3.05
Step 3. Calculate the t-statistic: t = (78 − 73) / 3.05 = 1.64
Step 4. Identify degrees of freedom: df = n₁ + n₂ − 2 = 51
The critical value for a two-tailed test at α = 0.05 with 51 degrees of freedom is approximately 2.008. Because 1.64 is less than 2.008, we fail to reject the null hypothesis. The p-value is approximately 0.108.
When the assumption of equal variances is violated, most software offers Welch's t-test as an alternative. Welch's test adjusts the degrees of freedom to account for unequal variances and is generally recommended as the default in modern practice because it performs well even when variances are equal.
The Paired Samples T-Test
The paired samples t-test compares two measurements from the same participants or from matched pairs. Common applications include pre-post designs, repeated measures under two conditions, and studies with matched pairs of participants (such as twin studies).
The paired t-test works by calculating the difference score for each pair, then testing whether the mean difference is significantly different from zero. This structure makes the test more powerful than an independent samples t-test on the same data because it removes between-subject variability from the error term.
The formula for the t-statistic is:
t = d̄ / (sd / √n)
where d̄ is the mean of the difference scores, sd is the standard deviation of the difference scores, and n is the number of pairs.
Example. A researcher measures the writing scores of 20 students before and after a semester-long writing course. The mean difference (post minus pre) is 6.5 points with a standard deviation of 8 points.
Step 1. Calculate the standard error: SE = 8 / √20 = 1.79
Step 2. Calculate the t-statistic: t = 6.5 / 1.79 = 3.63
Step 3. Identify degrees of freedom: df = n − 1 = 19
Step 4. The critical value at α = 0.05 with 19 degrees of freedom is 2.093. Because 3.63 exceeds 2.093, we reject the null hypothesis. The p-value is approximately 0.002.
The writing course produced a statistically significant improvement in writing scores.
Assumptions of the T-Test
All three t-tests share a common set of assumptions. Violating these assumptions can inflate Type I error rates or reduce statistical power.
- Continuous dependent variable. The scores being compared should be measured on an interval or ratio scale, not ordinal or nominal.
- Independence of observations. Each observation should be independent of the others. The paired t-test relaxes this within pairs but requires independence between pairs.
- Normality. The data (or the difference scores in a paired test) should be approximately normally distributed. The t-test is robust to modest violations of normality when sample sizes are large (roughly n greater than 30 per group).
- Homogeneity of variance. For independent samples t-tests, the two groups should have similar variances. Welch's t-test handles violations of this assumption automatically.
When assumptions are seriously violated, non-parametric alternatives are available: the Wilcoxon signed-rank test replaces the paired t-test, the Mann-Whitney U test replaces the independent samples t-test, and the sign test can replace the one-sample t-test in some contexts.
Interpreting T-Test Results
The t-statistic itself is only part of the story. A complete interpretation requires four elements:
- The p-value indicates whether the observed difference is statistically significant at the chosen alpha level.
- The direction of the difference tells you which mean is higher and by how much. Report the means themselves, not only the p-value.
- The effect size (usually Cohen's d for t-tests) quantifies the magnitude of the difference in standardized units. A statistically significant result with a small effect size may not be practically meaningful.
- The confidence interval around the mean difference indicates the range of plausible values for the true difference in the population. Confidence intervals often communicate more than p-values because they show both the direction and the precision of the effect.
Reporting T-Tests in APA Format
APA style requires a specific format for t-test results. The general template is:
t(df) = t-value, p = p-value, d = effect size, 95% CI [lower, upper]
Complete examples for each t-test type:
One-sample t-test: "The mean reading score at the school (M = 512, SD = 40) did not differ significantly from the national mean of 500, t(29) = 1.64, p = .112, 95% CI [−3.06, 27.06]."
Independent samples t-test: "Students taught with Method A (M = 78, SD = 10) did not score significantly higher than students taught with Method B (M = 73, SD = 12), t(51) = 1.64, p = .108, d = 0.45, 95% CI [−1.12, 11.12]."
Paired samples t-test: "Writing scores increased significantly from pretest (M = 68.2, SD = 9.1) to posttest (M = 74.7, SD = 8.7), t(19) = 3.63, p = .002, d = 0.81, 95% CI [2.76, 10.24]."
Common Mistakes to Avoid
Using an independent samples t-test on paired data. This is the most common error. When each participant contributes two scores, the observations are not independent, and using an independent samples test inflates the standard error and reduces power.
Running multiple t-tests without correction. Comparing three or more groups pairwise using multiple t-tests inflates the family-wise error rate. Use ANOVA followed by post hoc tests instead.
Ignoring effect sizes. A statistically significant t-test with a small effect size may be practically trivial. Always report Cohen's d alongside the t-statistic and p-value.
Reporting only the p-value. APA style requires the t-statistic, degrees of freedom, p-value, and effect size at minimum. Include confidence intervals for a complete picture.
Assuming t-tests fix non-normal data. While t-tests are robust to modest non-normality with large samples, seriously skewed data or small samples call for non-parametric alternatives.
Before You Submit: A Self-Audit
Work through this checklist before submitting a paper reporting t-tests:
- Did you choose the correct t-test type for your data structure?
- Have you checked the assumptions (normality, independence, homogeneity of variance)?
- Are the means and standard deviations reported for each group?
- Is the t-statistic reported with degrees of freedom in parentheses?
- Is an effect size (Cohen's d) reported alongside the p-value?
- Are confidence intervals included for the mean difference?
- If multiple t-tests were run, is the family-wise error rate addressed?
Statistical reporting is one of the most heavily scrutinized parts of any manuscript. Peer reviewers catch t-test errors more often than almost any other statistical mistake because the tests are so widely used and the reporting conventions are so specific. If you have finished your analysis and want a professional editor to review your methods, results, and statistical reporting before submission, Editor World offers journal article editing by editors with subject-matter backgrounds. Clients browse editor profiles and choose the editor whose expertise best matches their field. A free sample edit on the first 300 words is available for every project, and a certificate of editing is available as an optional add-on for publishers or committees that require one.
Frequently Asked Questions
What is the difference between an independent samples and paired samples t-test?
An independent samples t-test compares two separate groups of participants, while a paired samples t-test compares two measurements from the same participants or matched pairs. The choice depends on whether the two sets of scores are linked at the individual level. Pre-post designs use a paired test; comparing two intervention groups uses an independent test.
When should I use a t-test versus ANOVA?
Use a t-test when comparing two means (or one mean to a reference value). Use ANOVA when comparing three or more means. Running multiple t-tests to compare more than two groups inflates the Type I error rate, so ANOVA followed by post hoc tests is the correct approach.
What are the assumptions of a t-test?
T-tests assume a continuous dependent variable, independence of observations, approximately normal distribution of the data or difference scores, and (for independent samples) homogeneity of variance. The test is robust to modest violations of normality with large samples but sensitive to serious violations or small samples.
Do I need to report effect sizes for t-tests?
Yes. APA style and most journal guidelines require effect sizes alongside t-statistics and p-values. Cohen's d is the standard effect size for t-tests. A statistically significant result with a very small effect size may not be practically meaningful, and effect sizes give readers the information needed to judge practical significance.
What should I do if my data are not normally distributed?
If sample sizes are large (roughly n greater than 30 per group), t-tests are robust to modest non-normality. For seriously skewed data or small samples, use a non-parametric alternative: the Wilcoxon signed-rank test for paired data or the Mann-Whitney U test for independent groups. Data transformations (such as log transformations) are another option when appropriate.