Contact
One address, read by a person. Corrections to the mathematics get answered first.
contact@onlinepvaluecalculator.comWhat to put in the message
A reproducible report gets fixed. A description of a feeling usually cannot be.
A number looks wrong
The highest-priority message we get. If a result here disagrees with R, SciPy, SAS, or Stata by more than floating-point noise, it is a bug and we want it.
Include: The exact inputs, the value this site returned, the value your software returned, and which software it was.
A calculator is missing
Tell us which test, and — this is the useful part — what you were trying to decide with it. The use case shapes what the page needs to show alongside the p-value.
Include: The test name, and a sentence on the situation that sent you looking for it.
Something is unclear or wrong in the writing
Interpretation prose is a load-bearing part of this site, so a sentence that misleads is as much a defect as a bad digit.
Include: The page URL and the sentence, plus what you read it as meaning.
Accessibility or display problems
Content clipped on a phone, a control unreachable by keyboard, poor contrast in one of the themes.
Include: Your browser, your device, and a screenshot if the problem is visual.
Reporting a numerical discrepancy
This site's central claim is that its output is provably accurate — every distribution is validated against SciPy to a relative error below 1 × 10⁻¹³, and the reference values are published on the methodology page. A confirmed disagreement with established software falsifies that claim, so it is treated as the most serious class of bug here.
Before writing, it is worth checking two things that account for most reports. The first is tail direction: chi-square and F tests are right-tailed, and a tool that doubles them for a "two-tailed" option is producing a quantity that does not mean anything. The second is degrees of freedom — an off-by-one in df moves the p-value in a way that looks like a calculation error. Beyond that, send it.
Reproducible discrepancies are fixed and recorded. Where the fix changes a published number, the methodology page changes with it.
What to expect
This is a small project, not a company with a support desk. Messages are read by a person and answered as time allows, with correctness reports going to the front of the queue. There is no ticket number, no autoresponder, and no service-level commitment.
What is not possible, whatever the queue looks like, is statistical consultation. We cannot tell you which test suits your design, whether your data satisfy its assumptions, or how to interpret a result in the context of your study — those questions need someone who can see your data and your protocol. The learn section covers the general versions, and your institution's statistician or a service like Cross Validated covers the specific ones.
Your message and your data
Email is not encrypted end to end, and it lands in an ordinary mailbox. Do not send patient records, personally identifying data, or anything else confidential. If you need to demonstrate a bug, a small synthetic dataset that reproduces it is better evidence than the real one, and carries no risk.
Messages are used to answer the message. They are not added to any mailing list. Nothing about this changes what happens in the calculators themselves — those still run entirely in your browser, as the privacy policy describes.