statcompute

Statistics, computed and explained

Sixty-two statistics calculators that show their work

Paste a column or drop a CSV, get the answer — plus the formula it came from, the arithmetic behind it, and a plain reading of what it means.

  • Nothing is uploaded. Every calculation runs in your browser.
  • Verified against R. Expected values kept in the repository.
  • The working is shown. Formula, arithmetic, and what it means.
02

Browse by category

Hypothesis tests

Every test here answers the same question in a different setting: could this pattern have come from chance alone? Start with the decision guide if you are not sure which one you need.

All 14 in Hypothesis tests →

Descriptive statistics

Before any test, describe the data. These tools summarise a column of numbers and show you its shape, which is what decides which test is appropriate in the first place.

All 17 in Descriptive statistics →

Distributions

A distribution is a model of how a quantity varies. Pick the one whose assumptions match your situation and everything else — probabilities, critical values, p-values — follows from it.

All 9 in Distributions →

Regression & correlation

These tools quantify how two or more variables move together: how strongly, in which direction, and how much of one is explained by the others.

All 7 in Regression & correlation →

Charts

Paste a column, get a chart you can download — plus the numbers behind it, because a picture nobody can read the values off is only half a tool.

All 8 in Charts →

A/B testing

Decide the sample size before you start, read the result after you stop, and know which differences your test was ever capable of detecting.

All 8 in A/B testing →

03

How to read a statistical result

Every calculator here does the same three things: it computes the number correctly, it shows the arithmetic that produced it, and it tells you what the number does and does not mean. That last part is the one most tools skip, and it is where results get misread.

Nothing you type is transmitted. The statistics run in your browser — there is no server to send data to, which is why a column of salaries or patient measurements is safe to paste in.

What a p-value is, and what it is not

Most of the tests on this site end in a p-value, and it is the most misread number in statistics. A p-value is the probability of seeing data at least as extreme as yours if the null hypothesis were true. It is not the probability that the null hypothesis is true, it is not the probability that your result is a fluke, and 1 − p is not the probability that your hypothesis is correct.

Two consequences follow. A small p-value with a tiny effect is common in large samples and rarely worth acting on — check the effect size and the confidence interval. And a large p-value is not evidence of no effect; it often means the study was too small to detect one, which is what statistical power measures.

Choosing a test in thirty seconds

Comparing two group means, use a t-test. Three or more, use ANOVA. Comparing counts in categories, use chi-square — or Fisher's exact test when the counts are small. Comparing two conversion rates, use the A/B test calculator. Looking at how two measurements move together, use correlation or linear regression. If the data are skewed or full of outliers, take the rank-based route: Mann-Whitney and Kruskal-Wallis assume far less.

The decision guide walks through the same choice properly, with the assumptions spelled out.

Where the numbers come from

Every calculator is checked against R, the reference implementation used across academia and industry, with the fixtures and the expected values kept in the repository rather than in someone's head. Distribution functions are computed with Lanczos log-gamma and the continued-fraction incomplete beta and gamma functions; tail probabilities are computed directly rather than as one minus the CDF, which is the difference between a correct p-value of 6.2 × 10⁻¹⁶ and a wrong one. Variances use Welford's algorithm rather than the textbook shortcut, which fails outright on data with a large mean. The methodology page sets out exactly what is verified and how.

0.0010.010.050.100.501.00α = 0.05strong evidence against H₀nothing to see
p = 0.049 and p = 0.051 are the same evidence. The 0.05 line is a convention borrowed from Fisher in 1925, not a property of the world — report the number, not just which side of the line it fell.