standarddeviationcalculator.net

Updated

Hypothesis test calculators

Every significance test on the site, grouped by what you are comparing: means, proportions, variances or ranks. Alongside them are the pieces a test is built from, such as test statistics, degrees of freedom, p-values and critical values, and the tools for planning a study and reporting its effect size.

Which calculator do I need?

You have or wantUse
Two group means from a sample, population SD unknown (the usual case)T-test
Three or more group means at onceANOVA
Conversion rates from two versions of a page or emailA/B test
A 2×2 table with small counts (any expected count below 5)Fisher's exact test
The same people measured yes/no before and afterMcNemar's test
Two groups of skewed data or ratings, where a t-test is doubtfulMann–Whitney U test
A test statistic already in hand and you need its p-valueP-value calculator
How many participants you need before collecting dataPower analysis

Compare means

Is the difference between a sample mean and a target, or between group means, larger than chance?

Compare proportions and counts

Tests for categorical outcomes: observed counts, 2×2 tables, paired yes/no data and conversion rates.

Compare variances and ranks

Equal-variance checks, and rank-based tests that do not assume normal data.

Test statistics, p-values and corrections

The building blocks: compute a statistic, its degrees of freedom, the cut-off and the p-value, then adjust for multiple tests.

Plan the study and report the effect

Sample size before the data, effect size after it.

How the pieces fit together

Every test here follows the same four steps. Compute a test statistic (t, z, F or χ²) that measures how far the data sit from the null hypothesis. Find its degrees of freedom. Convert the statistic to a p-value, or compare it with a critical value. Then report an effect size, because a p-value says whether a difference is likely real, not whether it is large. The full test calculators do all four at once; the statistic and p-value pages are for when a textbook or paper gives you one step and asks for the next.

A worked comparison

Group A: n = 10, mean 52, SD 8. Group B: n = 10, mean 45, SD 6. The pooled SD is √[(9 × 64 + 9 × 36) / 18] = √50 = 7.07, so the standard error of the difference is 7.07 × √(1/10 + 1/10) = 3.16 and t = 7 / 3.16 = 2.21 on 18 degrees of freedom. The two-sided p-value is 0.040, just under 0.05. Welch's test, which does not assume equal variances, gives the same t here (the groups are the same size) but 16.7 degrees of freedom and p = 0.041. Cohen's d = 7 / 7.07 = 0.99, a large effect.

Now suppose this was one of five comparisons in the same study. The Bonferroni threshold becomes 0.05 / 5 = 0.01, and p = 0.040 no longer counts as significant. With only ten per group, a power analysis would have warned that even an effect this large is easy to miss.

Common mix-ups

  • z or t? Use a z-test only when the population SD is known or you are testing a proportion with a large sample. For means with a sample SD, use the t-test.
  • Several t-tests instead of ANOVA. Running every pairwise t-test inflates the false-positive rate. Run ANOVA first, then correct any follow-up comparisons.
  • Paired data tested as independent. Before-and-after measurements on the same people need the paired t-test, or McNemar's test for yes/no outcomes.
  • Mann–Whitney and Wilcoxon rank-sum are the same test. U and W differ by a constant and give the same p-value; pick whichever your course or journal expects.

Guides to read alongside

Common questions

What does a p-value of 0.03 actually mean?

If the null hypothesis were true, a result at least as extreme as yours would turn up about 3% of the time. It is not the probability that the null hypothesis is true, and it says nothing about how large or important the effect is.

Should I use a one-tailed or two-tailed test?

Use two-tailed unless you decided before seeing the data that only one direction matters and a result in the other direction would be treated the same as no effect. A one-tailed test halves the p-value, so choosing it after looking at the data overstates the evidence.

My data are not normal. Can I still use a t-test?

With moderate samples (about 30 or more per group) the t-test is fairly robust, because the sampling distribution of the mean is close to normal. With small, skewed samples or ordinal ratings, a rank test such as the Mann–Whitney U test is safer.