T Statistic Calculator
The t-statistic measures how far a sample mean sits from a hypothesized population mean, in units of standard error. It is the workhorse of the one-sample t-test: the test you run when you want to know whether your sample really differs from a claimed value, or whether the gap could just be sampling noise. This calculator runs all three common t-tests – one-sample, Welch two-sample, and paired – computing the t-statistic, degrees of freedom, standard error, and the exact two-tailed p-value, then stating the verdict in plain English.
The p-value is computed from the exact t-distribution via the regularized incomplete beta function, not read off a printed table. Very small p-values display in scientific notation (e.g. 1.914e-98).
Welch’s test does not assume equal variances, so the degrees of freedom come from the Welch–Satterthwaite equation and are usually fractional.
The paired test is a one-sample t-test on the differences (A − B) against a null of zero mean difference.
Quick Answer
Four oil-change wait times (46, 58, 40, 71 minutes): mean 53.75, standard deviation 13.72, tested against a claimed mean of 30. The standard error is 6.86, the t-statistic is 3.4621 with 3 degrees of freedom, and the two-tailed p-value is 0.0406 – below 0.05, so the test rejects the null hypothesis. These values are pre-filled in the calculator above; it also runs Welch two-sample and paired t-tests with the same step-by-step arithmetic.
Key takeaways
- t = (sample mean − hypothesized mean) ÷ (s ÷ √n). The t-statistic is a distance measured in standard errors, not in the original units.
- The p-value comes from the exact t-distribution with n − 1 degrees of freedom, computed directly – no printed tables involved.
- Use one-sample for a sample vs a claimed value, Welch two-sample for two independent groups, paired for before/after or matched pairs.
- A significant result means the gap is unlikely to be sampling noise at your chosen α; it says nothing about whether the gap matters in practice.
How to use the t-statistic calculator
Pick the tab that matches your data, fill in the summary numbers, and run the test. The results stay hidden until you calculate.
- One-sample: enter the sample mean (53.75), the hypothesized mean (30), the sample standard deviation (13.72), the sample size (4) and the significance level (0.05). Click Calculate. You get t = 3.4621, df = 3, SE = 6.8600, two-tailed p = 0.0406, one-tailed p = 0.0203, and a verdict: significant at α = 0.05.
- Two-sample (Welch): enter each sample’s mean, SD and size (5, 2, 10 vs 7, 2.5, 12). Click Run test. You get t = −2.0842, df = 19.98, SE = 0.9596, p = 0.0502: just above 0.05, so not significant – a near miss worth reading honestly.
- Paired: paste both samples with the same count, any separator (10, 12, 9, 11, 13 vs 9.5, 12.3, 8.2, 10.8, 13.1). Click Run test. You get t = 1.1083, df = 4, mean difference 0.2200, p = 0.3299: not significant.
- Read the steps: every mode shows the full arithmetic as a numbered list, so you can follow or check each number. Copy or share the result from the buttons in the results box; Reset restores the worked-example values.
Invalid inputs are rejected with plain messages: a non-numeric mean, a zero or negative standard deviation, a sample size below 2, mismatched paired counts, non-numeric paste values, and the special case where all paired differences are identical (the t-statistic is then undefined, and the tool says so).
The t-test formulas behind the calculator
All three tests share one idea: divide the observed gap by the standard error of the gap, then ask the t-distribution how surprising that ratio is. The differences are in what “the gap” and “its standard error” mean. The statistics calculator can produce the mean and standard deviation inputs from raw data if you have the dataset rather than the summary.
Welch two-sample: t = (x̄₁ − x̄₂) ÷ √(s₁²/n₁ + s₂²/n₂), df from Welch–Satterthwaite
Paired: t = d̄ ÷ (sd ÷ √n), df = n − 1, on the differences d = A − B
The gap in standard-error units. t = 3.46 means the sample mean sits 3.46 standard errors from the claim – surprising under the null.
How much independent information the estimate used. One-sample and paired spend one degree on the mean, so df = n − 1. Welch blends both samples, so df is fractional (19.98).
How much sample means typically wobble: s ÷ √n. Larger samples shrink it, which is why the same 2-point gap becomes significant at n = 100 but not at n = 10.
The probability of a t at least this extreme if the null were true, read off the t-distribution. The one-tailed value is half the two-tailed one, for the observed direction.
The p-value here is exact, computed from the t-distribution through the regularized incomplete beta function – the same numerical route the t-statistic p-value calculator uses when you already have a t and only need the p-value. If instead you want the cutoff t for your chosen α and df, the t critical value calculator reads the table in the other direction.
Worked examples: three tests, three verdicts
Each example below is pre-filled in its tab, and every figure is exactly what the calculator produces.
| Test | Inputs | t | df | p (two-tailed) | Verdict at α = 0.05 |
|---|---|---|---|---|---|
| One-sample | mean 53.75, μ₀ 30, s 13.72, n 4 | 3.4621 | 3 | 0.0406 | Significant: reject the null |
| Welch two-sample | 5 ± 2 (n 10) vs 7 ± 2.5 (n 12) | −2.0842 | 19.98 | 0.0502 | Not significant (a near miss) |
| Paired | A: 10, 12, 9, 11, 13; B: 9.5, 12.3, 8.2, 10.8, 13.1 | 1.1083 | 4 | 0.3299 | Not significant |
The middle row deserves a second look. A p of 0.0502 against a 0.05 line is not significant – the threshold is a line, not a neighborhood, and “almost significant” is not a verdict the tool will give you. If your study pre-registered α = 0.10, the one-sample tab’s dropdown covers that choice and the same p = 0.0406 would clear it comfortably.
Which t-test do you need?
Three questions, one answer. Run through them in order before you touch a tab.
Sample against a known or claimed mean μ₀. Use the One-sample tab. Example: is our average wait really 30 minutes?
Two separate samples, possibly different sizes and spreads. Use Two-sample (Welch) – it does not assume equal variances, the safer default.
The same subjects measured twice, or matched pairs. Use Paired – it tests the mean of the differences and is more powerful than treating the samples as independent.
If the population standard deviation is genuinely known (rare outside textbooks and calibrated instruments), you want a z-test, not any of these tabs. If you want the estimate form of the same information, the confidence interval calculator turns the standard error into a range.
What competing calculators miss
Most t-test tools on the web stop early, and the ones that go far go far in the wrong direction:
- t-statistic-only pages: several ranking tools compute the t value and the degrees of freedom and stop, leaving you to hunt a printed t-table for the verdict. This page carries the comparison through: exact p-value plus a significance verdict at your chosen α.
- One-mode tools: the common pattern is one-sample only, with no Welch and no paired option. The FIRGELLI-style multi-mode tools exist, but they split modes across separate pages or calculators rather than one tabbed tool with shared step-by-step output.
- Table-lookup p-values: tools that interpolate printed t-tables can be off in the third decimal on small df. The p-value here is computed from the t-distribution through the regularized incomplete beta function, exact to the displayed digits.
- Equal-variance defaults: many two-sample tools quietly run Student’s pooled-variance test, which misbehaves when the spreads differ. Welch’s version is the safer default and the only one offered here, with the fractional df shown honestly rather than rounded.
- Verdict without arithmetic: several tools print “reject” or “fail to reject” with no working shown. Every mode here lists the numbered steps – SE, t, df, p – so a student can check each line against a textbook.
- The gap this page fills: three tests in one tabbed tool, exact p-values, the one-tailed value reported alongside the two-tailed one, fractional Welch df displayed, and the full step-by-step arithmetic on every result.
Assumptions and honest limits
A t-test is a model, and the verdict is only as good as the fit. The assumptions, stated without hedging:
- Approximately normal data (or a large enough sample that the Central Limit Theorem takes over – roughly n above 30). With n = 4, as in the worked example, you are trusting the normality assumption.
- Independent observations. The paired tab handles dependence within pairs; dependence across pairs still breaks the test.
- Random sampling. A t-test on a convenience sample tests the sample, not the population you wish you had.
- The one-tailed p-value is half the two-tailed value, for the observed direction. That is only the correct one-tailed p if the direction was predicted before seeing the data. Halving after the fact to chase significance is p-hacking.
- Significance is not importance. With n = 100,000, a t of 21 gives p = 1.914e−98 for a 1-unit gap – overwhelmingly “significant” and possibly meaningless. Always report the effect size (the actual mean difference) alongside the p-value.
- Not significant is not “no difference.” It means the data cannot rule out sampling noise at your chosen α – nothing more.
Frequently asked questions
The distance between a sample mean and a hypothesized mean, measured in standard errors. In the worked example the sample mean sits 3.4621 standard errors from the claimed value of 30. Larger absolute values mean the sample is more surprising under the null hypothesis.
Two-tailed counts surprises in both directions (mean higher or lower than claimed); one-tailed counts only one direction and is half the two-tailed value for the observed direction. The calculator reports both, but use the two-tailed value unless you predicted the direction before seeing the data.
Use the t-test when the population standard deviation is unknown and estimated from the sample, which is nearly always in practice. The z-test needs a known population σ, which usually only happens in textbook problems or calibrated processes. With large samples the two give nearly identical answers.
Only that the observed difference is unlikely to be pure sampling noise at your chosen α. It says nothing about whether the difference is large enough to matter in practice – with 100,000 observations even a 1-unit gap gives p = 1.914e−98. Always report the effect size alongside the p-value.
Roughly, how many independent pieces of information went into the estimate. For the one-sample and paired t-tests, df = n − 1, because one degree is spent estimating the mean itself. Smaller df means fatter tails and a higher bar for significance. Welch’s df blends both samples and is usually fractional, like 19.98.
When comparing two independent samples whose variances may differ. It is the safer default over the classic Student version, which assumes equal variances and can misbehave when that assumption fails. Use the Two-sample tab here; it is Welch’s test.
Before/after measurements on the same subjects, or matched pairs. It tests whether the mean difference is zero, which is more powerful than treating the samples as independent when pairing exists. Paste both samples with the same count; the tool runs a one-sample t-test on the differences.
Yes, free for as many tests as you run. The math happens entirely in your browser: no account, nothing stored, and your pasted samples never leave the page. The one-sample, Welch and paired modes and the numbered step-by-step workings are all included, and “Copy my results” or “Share results” moves the t-statistic, p-values and verdict wherever you need them.
Add This Calculator to Your Website
Embed this calculator on your own site with the code below. Pick a size to preview it live, then copy the code.
Powered by Calculator Zilo
References
- National Institute of Standards and Technology. T Test (NIST/SEMATECH e-Handbook reference: one-sample, unpaired two-sample and paired two-sample t-tests, test statistics, critical regions). Accessed October 24, 2026.
- Penn State Eberly College of Science, STAT 200. One Sample Mean t Test, Formulas (hypotheses, assumptions, t = (x̄ − μ₀) ÷ (s ÷ √n) with df = n − 1). Accessed October 24, 2026.
- Penn State Eberly College of Science, STAT 800. Hypothesis Testing for One-Sample Mean (worked one-sample t example: t = −1.3 with n = 61, decision logic). Accessed October 24, 2026.
- Minitab. Hypotheses for 1-Sample t (null and alternative hypothesis forms for one-sample t-tests). Accessed October 24, 2026.
- LibreTexts Statistics. The Summary of Hypothesis Testing for One Parameter (t-test assumptions, test statistic, reject/fail-to-reject decision rules). Accessed October 24, 2026.