F Test Calculator
Two-sample F test for equal variances, with the F statistic, degrees of freedom and p-value.
The F test asks whether two populations plausibly share the same variance, by comparing the ratio of the two sample variances against the F distribution. A ratio near 1 says the spreads look alike; the p-value then asks how surprising a ratio at least this far from 1 would be if the population variances really were equal. The worked pair on this page — (2, 4, 6, 8) against (1, 2, 3, 4) — has sample variances 6.667 and 1.667, so F = 4.0 on 3 and 3 degrees of freedom, with a two-tailed p of about 0.285: a fourfold variance ratio that is still unremarkable at these tiny sample sizes.
The F test for equal variances is notoriously sensitive to non-normal data: with skewed or heavy-tailed samples it can flag a “difference” that is really just the shape of the data. Treat the p-value as one piece of evidence, check what your data looks like, and consider a robust alternative such as Levene’s test when normality is in doubt.
Quick answers
How do you calculate an F test for two variances?
Divide one sample variance by the other and compare the ratio with the F distribution on (n₁ − 1, n₂ − 1) degrees of freedom. Variances 6.667 and 1.667 give F = 4.0, df 3 and 3, two-tailed p ≈ 0.285.
The test assumes both samples come from normal populations and is unusually sensitive when they do not — check the shape of your data, or use Levene’s test instead.
What does a significant F test mean?
That a variance ratio this far from 1 would be unlikely (at your chosen level, conventionally p < 0.05) if the two population variances were truly equal — evidence, not proof, that the spreads differ.
It does not say the means differ, does not say by how much the variances differ in practical terms, and with non-normal data it may be reporting the shape of the data rather than its spread.
How to use
- Paste the first sample into the top box and the second sample into the second box, values separated by commas, spaces or new lines.
- Make sure each sample has at least two values and is not constant — the page refuses either case with a plain explanation.
- Read F (the ratio of the first sample variance to the second), the two degrees of freedom and the two-tailed p-value.
- Weigh the p-value together with the normality warning below the result before drawing any conclusion.
About this tool
The assumption that decides whether to trust this test is normality — and the F test is famous for failing that trust. Unlike the t test, which shrugs off moderate non-normality, the variance-ratio F test can produce badly wrong p-values for skewed or heavy-tailed data, mistaking shape for spread. Its legitimate home is as a preliminary check (historically, before a pooled t test) on data you have reason to believe is roughly normal. For anything else, a robust alternative such as Levene’s or Brown–Forsythe’s test is the defensible choice. This page computes the textbook statistic accurately; it does not vouch for any scientific conclusion drawn from it, and a small p-value here is a prompt to inspect the data, not a verdict.
Frequently asked questions
What does the F test test, exactly?
The null hypothesis that two populations have equal variances. F is the ratio of the two sample variances; the further F sits from 1 in either direction, the smaller the two-tailed p-value becomes. It says nothing directly about the means — two groups can share a variance and have very different means, or the reverse.
Why is my F less than 1?
Because the first sample happens to have the smaller variance. This page keeps the ratio in the order you entered the samples (sample 1 ÷ sample 2) and uses a two-tailed p-value, so F = 0.25 and F = 4 are treated as equally strong evidence, as they should be.
How many values does each sample need?
At least two — a single value has no variance to estimate. In practice two or three values per group give a test with almost no power: the worked example shows F = 4.0 failing to reach significance at n = 4 per group. Double-digit samples are where variance comparisons start to mean something.
When should I not use an F test?
When the data are visibly skewed, heavy-tailed or full of outliers, or when they are counts, scores or other clearly non-normal measures. The F test for variances is unusually sensitive to non-normality; Levene’s test is the standard safer substitute in those cases.
First time embedding? Read the step-by-step guide →
More free tools
Percentage Calculator
Work out percentages, increases and discounts in one click.
Age Calculator
Find your exact age in years, months and days.
BMI Calculator
Check your Body Mass Index in seconds.
Tip Calculator
Split the bill and work out the tip in seconds.
Discount Calculator
See the sale price and how much you save, instantly.
Scientific Calculator
Full scientific calculator: trig, logs, powers, factorial. Free, private, runs in your browser.