Data Normality Test screenshot

Data Normality Test

Shapiro-Wilk W and its p-value by Royston's AS R94 algorithm, for 3 to 5000 values, computed in your browser.

Privacy First
0 Runs
Updated 2026-11-06
No upload
Runs in this tab

Shapiro-Wilk, not a substitute

Paste a sample, one value per entry or separated by commas, spaces or new lines. The page sorts it, builds Royston's published weights for the W statistic, and prints W, the lower-tail p-value and the normalising transform it used. n is held to the range the algorithm is published for, 3 to 5000, and a sample with no spread is refused because the denominator of W would be zero.

The p-value is the probability of a W as small as yours if the sample came from a normal distribution, so a small p means the sample is unusual for it. That is the whole claim: at a five percent threshold the mechanical decision is to reject the null hypothesis, and nothing follows from it about what the data are. No Jarque-Bera, Anderson-Darling or Kolmogorov-Smirnov substitution is made anywhere on the page.

The constants are literals in the page and the arithmetic runs in this tab. No sample is uploaded, nothing is stored, and pasting the same numbers prints the same W and the same p.

Run a Shapiro-Wilk test on a pasted sample and read W, its lower-tail p value, the sample size and the transform the algorithm used.

It is one test of one assumption: the statistic is the published W, the p value is a lower-tail probability, and nothing is uploaded or stored.

How to use Data Normality Test

Paste the sample into One value per entry, or separated by commas, spaces or new lines; between 3 and 5,000 values is accepted.

Read W and its tail for the statistic and its lower-tail p, with Values read and Transform used beside them.

Read Mean and Standard deviation too, and weigh Values read before calling a departure meaningful.

What the numbers mean

W close to 1 is what a normal sample tends to produce, and the printed p is P(W ≤ w) under the null hypothesis. The statistic has no upper tail to report, which is why only one p is printed.

The three forms

The algorithm has separate coefficient sets for n = 3, for n = 4 to 11, and for n = 12 and above, and the answer names which one ran. A sample of identical values is refused because the denominator vanishes and W has no value.

Where it stops

Between 3 and 5,000 values, one variable, no grouping and no alternative distributions. A large sample can make a trivial departure significant, so the page reports W and n together rather than a verdict.

What to Expect

Best for

  • checking a small sample before choosing a test
  • seeing whether a residual set is plausibly normal

What this tool keeps

  • The sample stays in the tab, the coefficients ship with the page, and the values are not stored after a reload.

Frequently asked questions

Which test is this?

Shapiro-Wilk. The page prints W, the lower-tail p, the number of values read, the mean, the standard deviation and the coefficient form the algorithm used for that sample size.

Why is only one p value shown?

Because W is a one-sided statistic: the published probability is P(W ≤ w). There is no upper tail, so a second p would be an invention.

What are the limits?

Between 3 and 5,000 values, all numeric, and not all identical. A zero-spread sample is refused because the statistic's denominator is zero.

Developer program

Build custom tools for your store

Join the SDK program to create useful ecommerce utilities for your store.