About this calculator

Why build another one?

There is no shortage of p-value calculators. Most of them do the same narrow thing: take a test statistic, return a tail probability, stop. They are usually correct in the middle of the distribution and they are usually where the story ends.

Two problems follow from that. The first is technical — a surprising number of them report p = 0.000 for extreme statistics, which is never a valid p-value and is a straightforward consequence of computing the tail as 1 − CDF. The second is more consequential: handing someone a bare number is exactly the condition under which p-values get misread, and the research on that is unambiguous.

So this site is built around a different premise. The number is table stakes. The job is to return it accurately and make sure you can say what it means.

Principles

Correctness is not negotiable

Every distribution is implemented from published numerical methods and validated against SciPy to a relative error below 1 × 10⁻¹³. The algorithms, the reference values, and the limits are all published on the methodology page. If you are putting a number from a free web tool into a paper, you should be able to check where it came from.

Interpretation is part of the product

Every result comes with a plain-English reading and an explicit note about the specific misinterpretation that result invites — which differs depending on whether it was significant. Data-driven calculators report an effect size and a confidence interval at equal visual weight with the p-value, because the interval usually answers the question you actually had.

The tool comes first

The calculator is above the fold, interactive on load, and updates as you type. There is no wizard, no submit button, and no wall of prose between you and the input box. The explanatory content sits below, where it belongs.

Your data stays yours

Every calculation runs in your browser. Nothing you type is transmitted, logged, or stored — there is no server to send it to. That matters if you work with clinical, HR, or commercial data, and it means the site keeps working on a plane. See our privacy page.

Honest about limits

Where a method stops working, the calculator says so rather than returning a number anyway. Chi-square and F tests are locked to right-tailed because a two-tailed version is not a meaningful quantity. Small expected counts trigger a warning naming the right alternative. Unequal variances in a pooled t-test produce an advisory. A tool that never warns you is not being confident, it is being quiet.

What this is not

This is a calculator, not a statistician. It cannot tell you whether your study design was sound, whether your data meet the assumptions of the test you picked, whether you have controlled for the right things, or whether the question you are asking is the right one. It cannot know how many other tests you ran.

For consequential work — clinical research, regulatory submissions, anything where a wrong answer causes real harm — a calculator is a check on your arithmetic, not a substitute for statistical expertise.

Corrections

If any result here disagrees with R, SciPy, SAS, or Stata by more than floating-point noise, that is a bug. Send the exact inputs and both outputs via the contact page. Reproducible discrepancies get fixed and recorded on the methodology page.

The terms and conditions cover what you may do with the output — in short, use it for anything, including commercially, with no attribution required.