Written and maintained by CASRAI Editorial Board
Published
On this page
This guide is for researchers who have two groups of measurements and want to compare their means in GraphPad Prism. After reading it you can choose between an unpaired, Welch, paired or other analysis, run it from the menus, and read each line of the results table without over-reading it. Menu names follow the Prism 11 Statistics Guide (read 2026-11-08); wording in your installed version may differ slightly.
What a t test in Prism is used for
Prism’s t test compares the means of two groups. In Prism the same analysis, t tests (and nonparametric tests), covers unpaired, Welch, paired and lognormal variants, and the nonparametric alternatives. The main results to focus on are the 95% confidence interval of the difference and the P value (confidence interval, P value). For the statistics behind the choice of test, see conditions for a t test and the comparison of the z test and t test.
Which Prism t test should you choose?
The Prism checklist for the unpaired t test asks eight questions before you accept a result. Mapped to a decision, they give this table.
| Your situation | What Prism’s guide says to use |
|---|---|
| Two separate groups, variances not known to be equal | Unpaired design with Welch’s t test (the default choice in the dialog) |
| Two separate groups, certain the population variances are equal | Unpaired t test (assumes equal variances) |
| Measurements matched, such as before and after in the same subject | Paired t test (Experimental design = Paired) |
| Data not sampled from Gaussian populations | Transform the data, or use the Mann-Whitney test instead |
| Three or more groups | One-way ANOVA with multiple comparisons, not repeated t tests (see one-way vs two-way ANOVA) |
| One column of data compared with a theoretical value | One-sample t test |
The checklist also requires that errors are independent. For example, triplicate measurements taken from the same animal are not independent observations of the group.
Where to get Prism
GraphPad Prism is licensed commercial software from GraphPad (Dotmatics). Get it, or a trial, only from the official GraphPad website, graphpad.com. This guide does not cover licensing terms.
Step by step: unpaired t test
- In the Welcome (or New Table and graph) dialog, choose the Column tab and enter replicate values stacked in columns. Put each group in its own column. Groups may differ in size and cells may be empty. To practise first, choose the sample data t test – unpaired.
- From the data table, click Analyze in the Analysis section of the toolbar and choose t tests (and nonparametric tests) from the column analyses.
- On the Experimental Design tab set Experimental design to Unpaired. Choose the distribution: Normal if the difference between groups matters, Lognormal if the ratio matters.
- Under Choose test pick Welch’s t test or lognormal Welch’s t test (the default), or Unpaired t test or lognormal t test if you are certain the population variances are equal.
- On the Options tab choose a one-sided or two-sided P value (if in doubt, choose two-tail), the direction of the differences, and the confidence level (95% by default).
- Review the results, then work through the checklist before accepting them.
Step by step: paired t test
- On the Column tab enter each group in its own column, with matched values on the same row. Rows with missing values are ignored. Optional row labels (such as subject initials) can be added.
- Click the analysis button on the toolbar and choose t tests from the column analyses.
- On the Experimental Design tab set Experimental design to Paired and choose the paired t test under Choose test.
- On the Options tab set the P value type, direction of differences and confidence level as above.
- Focus on the 95% confidence interval of the difference and the P value. For the graph, Prism’s guide recommends a before-after graph that shows every subject, and advises against a bar graph because it shows only the mean and SD, not individual changes.
Reading the output
| Output line | What it means | What not to over-read |
|---|---|---|
| Difference between means (with SE and 95% CI) | The size of the difference and how precisely it is known | The CI is the main result; the sign changes if you flip the direction in Options |
| t ratio | Difference divided by its standard error | Reported mainly for comparison with other programs and textbooks; the sign only shows which mean is larger |
| df | Total sample size minus 2 for the standard test; Welch’s test uses a more complex equation | The two tests do not calculate df the same way |
| F test for equal variances | Larger variance divided by smaller, so always at least 1 | Its P value tests the variances, not the means. The checklist says not to base your conclusion solely on the F test |
| R squared | Fraction of the variation among all values explained by the difference between group means | 0 when the means are equal, approaching 1 when differences are large relative to scatter within groups |
One-tailed P values: the checklist says the prediction of direction must precede the data. If you did not decide in advance, choose the two-tail P value. For wider guidance on interpretation see how to interpret t test results; for a standalone P value calculation see the GraphPad P value calculator guide.
Troubleshooting
The groups are matched but Prism treated them as unpaired
Check that the Experimental design on the first tab is set to Paired and that matched values sit on the same row.
The F test is significant. Should I switch tests?
Prism’s checklist cautions against basing the decision only on the F test. Welch’s t test, the default, does not assume equal variances.
I have three groups
The checklist says to use one-way ANOVA with multiple comparisons rather than several t tests.
My data do not look Gaussian
Transform the data or use the Mann-Whitney test. The nonparametric tests are in the same t tests (and nonparametric tests) analysis list.
Free alternative: R
R’s t.test function in the stats package (documentation read 2026-11-08) offers the same choices. Its default is the Welch test, as in Prism.
| Prism choice | R t.test argument |
|---|---|
| Welch’s t test (default) | var.equal = FALSE (the default) |
| Unpaired t test assuming equal variances | var.equal = TRUE |
| Paired design | paired = TRUE (x and y must be the same length) |
| Two-sided P value | alternative = "two.sided" (the default); "greater" or "less" for one-sided |
| Confidence level 95% | conf.level = 0.95 (the default) |
Frequently asked questions
Should I use Welch’s t test or the standard unpaired t test in Prism?
Prism’s guide calls Welch’s test generally reasonable unless you are certain the population variances are equal. It is the default choice in the dialog.
Should I pick a one-tailed or two-tailed P value?
If in doubt, choose two-tail. A one-tailed P value is justified only if you predicted the direction before collecting data.
What is the alternative if my data are not normally distributed?
The checklist names transformation or the Mann-Whitney test, available from the same analysis dialog.
What if I have more than two groups?
Use one-way ANOVA with multiple comparisons.
Does changing the direction of the differences change the P value?
No. It changes only the sign of the difference and its confidence interval.
Sources
- GraphPad Prism 11 Statistics Guide, How to perform an unpaired t test (read 2026-11-08).
- GraphPad Prism 11 Statistics Guide, How to perform a paired t test (read 2026-11-08).
- GraphPad Prism 11 Statistics Guide, Checklist: Is an unpaired t test the right test? (read 2026-11-08).
- GraphPad Prism 11 Statistics Guide, How the unpaired t test works (read 2026-11-08).
- R Documentation, Student’s t-Test (stats::t.test) (read 2026-11-08).







