molecalculator

Moles & Molar Mass

How to Read a Normality Test Results

Guide 01-8615 · Updated · Step by step

This guide explains how to interpret the output of a normality test, such as the Shapiro–Wilk, Kolmogorov–Smirnov, Anderson–Darling, or Jarque–Bera test. You will learn what the test statistic and p-value mean, how to decide whether your data is normally distributed, and what to do when the test fails. No prior statistical software knowledge is needed, but having the test output (statistic and p-value) in front of you will make the steps concrete.

What You’ll Need

  • The output from a normality test: the test statistic (often labeled W, D, A, or JB) and the p-value.
  • The sample size (n) used in the test.
  • The significance level (alpha) you plan to use, commonly 0.05.
  • Optional: a histogram or Q-Q plot of the data to visually confirm the result.

Step-by-Step

  1. Locate the p-value. Most statistical software (R, Python, SPSS, Minitab, Excel) reports a p-value for the normality test. It is usually labeled “p”, “Sig.”, “p-value”, or “Pr(>W)”. Write it down.
  2. Set your significance level. The default is almost always 0.05 (5%). This is the probability of rejecting normality when the data actually is normal. If you have a reason to be stricter, you might use 0.01, but 0.05 is standard.
  3. Compare p-value to alpha.
    • If p-value > alpha (e.g., p = 0.20 > 0.05): fail to reject the null hypothesis. The data is consistent with a normal distribution.
    • If p-value < alpha (e.g., p = 0.003 < 0.05): reject the null hypothesis. The data is not normally distributed.
  4. Interpret the test statistic. The statistic itself is less intuitive than the p-value, but here is what it means for each common test:
    • Shapiro–Wilk (W): W ranges from 0 to 1. Values close to 1 indicate normality. A very low W (e.g., 0.85) suggests deviation from normality.
    • Kolmogorov–Smirnov (D): D measures the maximum distance between the empirical cumulative distribution and the theoretical normal CDF. Smaller D (closer to 0) indicates normality. Larger D means more deviation.
    • Anderson–Darling (A²): A² is a weighted squared distance. Smaller values indicate normality. The p-value is what you use for the decision.
    • Jarque–Bera (JB): JB combines skewness and kurtosis. Values close to 0 indicate normality. Larger values indicate non-normality.
  5. Consider sample size. For small samples (n < 30), the Shapiro–Wilk test is preferred because it has more power. For large samples (n > 1000), even trivial deviations from normality will produce a significant p-value. In that case, rely on visual checks and the magnitude of the deviation, not just the p-value.
  6. Check the histogram or Q-Q plot. A normality test alone can be misleading. Look at the data: if the histogram is roughly bell-shaped and the Q-Q plot points fall close to the diagonal line, the data is approximately normal even if the p-value is slightly below 0.05.
  7. Make a decision for your analysis. If the data is normal, you can use parametric tests (t-tests, ANOVA, Pearson correlation). If not, either transform the data (log, square root, Box–Cox) or use non-parametric alternatives (Mann–Whitney U, Kruskal–Wallis, Spearman correlation).

Common Mistakes

  • Using the wrong test for small samples. The Kolmogorov–Smirnov test is less powerful for small samples; use Shapiro–Wilk when n < 50.
  • Ignoring sample size. With large samples, the test may reject normality for minor, practically irrelevant deviations. Always look at the plot.
  • Confusing the null hypothesis. The null hypothesis is that the data IS normal. A high p-value means you do NOT reject normality, not that you prove it.
  • Using a one-tailed p-value. Normality tests are two-tailed; make sure your software reports the two-tailed p-value.
  • Not reporting the test name. When you write up results, always state which test you used (e.g., Shapiro–Wilk, W = 0.97, p = 0.32).

Tips

  • Always run a normality test before choosing between parametric and non-parametric tests, especially for small samples.
  • Combine the test with a Q-Q plot and histogram. The visual is often more informative than the p-value alone.
  • If your data has ties (repeated values) or is discrete, note that some tests (like Shapiro–Wilk) handle ties poorly. Consider the Anderson–Darling test instead.
  • For data with outliers, normality tests are sensitive. Check for outliers first; they can cause a false rejection of normality.
  • Remember that normality tests assume the data is a random sample from a continuous distribution. If your data is collected in a non-random way, the test is not valid.

Troubleshooting

  • Software gives an error or no p-value. This often happens when the sample size is too small (n < 3) or when all values are identical. Ensure you have at least 3 distinct data points.
  • p-value is exactly 0.000. This means the p-value is less than 0.001; report it as p < 0.001. The data is strongly non-normal.
  • Test says non-normal but Q-Q plot looks fine. This can occur with large samples. Look at the skewness and kurtosis values; if they are within ±1, the deviation is likely not practically significant.
  • Different tests give different results. This is common. Shapiro–Wilk is more sensitive to deviations in the tails, while Kolmogorov–Smirnov is more sensitive to the center. If they disagree, look at the plots and consider the sample size.
  • Data has many zeros or is heavily skewed. No normality test will pass. Consider a transformation or use a non-parametric test.

Frequently Asked Questions

What is the null hypothesis of a normality test?

The null hypothesis (H0) is that the data comes from a normal distribution. The alternative hypothesis (Ha) is that it does not.

What p-value indicates normality?

A p-value greater than your chosen alpha (usually 0.05) indicates that you cannot reject normality. It does not prove normality, but it suggests the data is consistent with a normal distribution.

Which normality test is best for small samples?

The Shapiro–Wilk test is generally recommended for small samples (n < 50) because it has better statistical power than the Kolmogorov–Smirnov test.

Can I use a normality test on ordinal data?

No. Normality tests assume continuous data. For ordinal data, use non-parametric tests without checking normality.

What if my data is not normal? Can I still use t-tests?

If the sample size is large (n > 30), the Central Limit Theorem allows t-tests to be robust to non-normality. For small samples, use non-parametric alternatives like the Mann–Whitney U test.

How do I report a normality test result?

Report the test name, the test statistic, and the p-value. For example: “The Shapiro–Wilk test indicated that the data was not normally distributed (W = 0.89, p = 0.02).”

Related

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this section

The whole Moles & Molar Mass section →