Online t test calculator
Paste two columns of numbers. The calculator picks the right variant for you — Welch when the variances differ, paired when the rows are the same people — and gives you the effect size and a sentence you can paste into a manuscript.
Which t test is the right one
The question that decides it is not statistics, it is sampling: are the two sets of numbers from the same people?
- Paired. Before and after on one group. Left hand versus right hand. The same samples run twice.
- Independent. A treatment group and a control group. Two different hospitals. Men versus women.
Choosing the wrong one is the most common error in this test. It changes the degrees of freedom, and pairing data that should not be paired usually shrinks the p value unfairly.
Why the variant matters
The textbook version pools the two variances and assumes they are equal. That assumption is routinely violated in real data, and when it is, the pooled test can be liberal — it reports significance too readily. The Welch correction does not pool, and adjusts the degrees of freedom downward to compensate.
There is no good reason to use the pooled version when the variances differ, which is why this calculator runs Levene's test first and switches automatically. If you need the pooled result for a course requirement, the page shows both.
What to report
A t test on its own is an incomplete result. Report the test statistic and degrees of freedom, the p value, the mean difference with a 95% confidence interval, and an effect size. The sentence this page produces includes all of them, in APA style, so you do not have to assemble it by hand.
Frequently asked
Should I use a paired or an independent t test?
Same people measured twice — paired. Two separate groups — independent. If you are unsure, the pairing is defined by how the data was collected, not by the numbers themselves.
My p value came out much higher than expected. Why?
Usually one of three things: the wrong variant was selected for the design, the sample is smaller than you thought because of missing values, or the difference genuinely is not detectable at this sample size.
Is my data sent anywhere?
No. The calculation runs in JavaScript on this page. Nothing is uploaded.
What if my data is clearly not normal?
With small samples and strong skew, a Mann–Whitney U test (independent) or a Wilcoxon signed-rank test (paired) is the better choice. The full StatsToTable app makes that switch for you when a file is analysed.
When you have a file rather than two columns
If your data is already a .sav, .xlsx or .csv, the
StatsToTable app does this and more: it reads the file, works out the variable
types, picks the procedure and exports a three-line table. One report export each month is free, and nothing is uploaded there either.
Open a .sav file without SPSS · StatsToTable compared with SPSS · method notes · which test should I use · verification record