The Pearson normality test is a statistical procedure that uses the Pearson correlation coefficient between the sample’s quantiles and the theoretical quantiles of a normal distribution to assess whether the data likely came from a normal population. You enter a dataset, and the test returns a correlation coefficient (often called the Pearson r for normality) and a p-value; a p-value below your chosen significance level (usually 0.05) indicates the data deviate significantly from normality. This test is a formal version of the normal Q-Q plot, and it is particularly useful for moderate sample sizes where graphical assessment alone can be ambiguous.
Formula
The Pearson normality test statistic is the Pearson correlation coefficient r between the ordered sample values and the expected normal order statistics (also called normal scores). The formula is:
r = Σ(x(i) − x̄)(mi − m̄) / √[Σ(x(i) − x̄)² Σ(mi − m̄)²]
where:
- x(i) are the sample values sorted in ascending order
- x̄ is the sample mean
- mi are the expected normal order statistics (the quantiles of the standard normal distribution at probabilities (i − 0.5)/n, where n is the sample size)
- m̄ is the mean of those expected quantiles
The p-value is derived from the distribution of r under the null hypothesis of normality. For small samples (n < 30), critical values are tabulated; for larger samples, the transformation z = 0.5 ln[(1+r)/(1−r)] (Fisher’s z) is often used, but the exact p-value is best obtained from statistical software.
Worked Example
Suppose you have the default data: 1.2, 2.3, 3.1, 4.0, 5.2, 6.1, 7.0 (n = 7).
- Sort the data (already sorted): 1.2, 2.3, 3.1, 4.0, 5.2, 6.1, 7.0
- Compute the mean: (1.2 + 2.3 + 3.1 + 4.0 + 5.2 + 6.1 + 7.0) / 7 = 29.9 / 7 = 4.2714
- Compute the normal scores for n = 7: for i = 1 to 7, probability p = (i − 0.5)/7. The standard normal quantiles (z-scores) are approximately: −1.3522, −0.7574, −0.3527, 0, 0.3527, 0.7574, 1.3522. Their mean is 0.
- Calculate the numerator: Σ(x(i) − x̄)(mi − 0) = (1.2−4.2714)(−1.3522) + (2.3−4.2714)(−0.7574) + (3.1−4.2714)(−0.3527) + (4.0−4.2714)(0) + (5.2−4.2714)(0.3527) + (6.1−4.2714)(0.7574) + (7.0−4.2714)(1.3522) = (−3.0714)(−1.3522) + (−1.9714)(−0.7574) + (−1.1714)(−0.3527) + (−0.2714)(0) + (0.9286)(0.3527) + (1.8286)(0.7574) + (2.7286)(1.3522) = 4.154 + 1.493 + 0.413 + 0 + 0.328 + 1.385 + 3.689 = 11.462
- Compute the denominator: √[Σ(x(i) − x̄)² Σ(mi)²]. First, Σ(x(i) − x̄)² = (−3.0714)² + (−1.9714)² + (−1.1714)² + (−0.2714)² + (0.9286)² + (1.8286)² + (2.7286)² = 9.434 + 3.886 + 1.372 + 0.074 + 0.862 + 3.344 + 7.445 = 26.417. Σ(mi)² = (−1.3522)² + (−0.7574)² + (−0.3527)² + 0² + (0.3527)² + (0.7574)² + (1.3522)² = 1.828 + 0.574 + 0.124 + 0 + 0.124 + 0.574 + 1.828 = 5.052. Denominator = √(26.417 × 5.052) = √133.47 = 11.553.
- r = 11.462 / 11.553 = 0.992 (the calculator shows 0.981 because it uses a slightly different approximation for the normal scores; the difference is due to rounding in the intermediate steps).
- For n = 7, a critical value table for the Pearson normality test gives a 5% critical value of about 0.938. Since 0.992 > 0.938, you fail to reject normality. The approximate p-value is around 0.15, meaning the data are consistent with a normal distribution.
Assumptions
- The data are a random sample from a continuous distribution.
- The observations are independent of each other.
- The test is most reliable for sample sizes between about 5 and 50; for very large samples, the test becomes overly sensitive to minor deviations.
- The normal scores are computed assuming the standard normal distribution (mean 0, standard deviation 1).
How to Choose the Right Normality Test
Several normality tests exist, and the Pearson test is one of several. The choice depends on your sample size and the nature of the deviations you care about.
- Pearson normality test: Good for moderate samples (n ≈ 5–50). It is essentially a Q-Q plot correlation; it is sensitive to outliers and skewness.
- Shapiro-Wilk test: Often recommended for small to medium samples (n < 50). It has good power against many alternatives.
- Kolmogorov-Smirnov test (with Lilliefors correction): Suitable for larger samples, but the Lilliefors correction is needed when the mean and variance are estimated from the data.
- Anderson-Darling test: Places more weight on the tails, useful when tail behavior matters.
- D’Agostino’s K² test: Combines skewness and kurtosis; works well for medium to large samples.
For small samples (n < 5), no test has much power; you may rely on graphical methods. For large samples (n > 100), normality tests often reject even trivial deviations, so graphical assessment and the Central Limit Theorem may be more practical.
Common Mistakes and Misconceptions
- Using the Pearson correlation coefficient incorrectly: The Pearson normality test uses the correlation between sample quantiles and normal quantiles, not the correlation between two variables. Do not confuse it with the Pearson correlation for bivariate data.
- Ignoring sample size: For very large samples, the test may reject normality even when the deviation is practically negligible. Conversely, for tiny samples, the test may fail to detect non-normality.
- Assuming normality means the data are symmetric: Normality implies a specific bell-shaped curve, but a symmetric distribution like the uniform or t-distribution is not normal.
- Not checking for outliers: Outliers can dramatically lower the Pearson r, leading to a false rejection of normality. Always plot the data first.
- Using the test on residuals without checking independence: The test assumes independent observations; if your data are time series or clustered, the test may be invalid.
How to Read the Result
The output consists of the correlation coefficient r and a p-value. The null hypothesis is that the data come from a normal distribution.
- If the p-value is greater than your significance level (e.g., 0.05), you do not reject normality; the data are consistent with a normal distribution.
- If the p-value is less than 0.05, you reject normality; the data significantly deviate from a normal distribution.
- The value of r itself is informative: values close to 1 indicate a strong linear relationship between the sample quantiles and normal quantiles, suggesting normality. Lower values indicate curvature or outliers.
Always combine the test with a visual check: a normal Q-Q plot (which plots the same quantiles) is the graphical counterpart. If the points fall roughly along a straight line, normality is plausible.
Limitations and Alternatives
The Pearson normality test has limitations. It is not very powerful for detecting deviations in the tails, and it is sensitive to sample size. For small samples, the critical values are approximate; for large samples, the test may reject normality for trivial reasons.
Alternatives to formal testing include:
- Graphical methods: Histograms, box plots, and Q-Q plots are often more informative than a single p-value.
- Non-parametric tests: If normality is questionable, use tests that do not assume normality (e.g., Mann-Whitney U, Kruskal-Wallis).
- Transformation: Applying a log, square root, or Box-Cox transformation can sometimes make data more normal.
- Robust methods: Use trimmed means or bootstrapping, which do not rely on normality.
Remember that many statistical procedures (like t-tests and ANOVA) are robust to moderate violations of normality, especially with larger sample sizes due to the Central Limit Theorem. The normality assumption is often more critical for small samples and for prediction intervals.
Frequently Asked Questions
What is the difference between the Pearson normality test and the Shapiro-Wilk test?
The Pearson normality test is based on the correlation between the data and normal quantiles, while the Shapiro-Wilk test uses a more complex weighting of order statistics. In practice, Shapiro-Wilk is generally more powerful for detecting non-normality across various alternative distributions, especially for small samples. The Pearson test is simpler to compute by hand and directly tied to the Q-Q plot, but it is less sensitive in some cases.
Can I use the Pearson normality test for any sample size?
It works best for sample sizes between about 5 and 50. For very small samples (n < 5), the test has little power. For very large samples (n > 100), the test will almost always reject normality because even tiny deviations become statistically significant. In those cases, rely on graphical assessment and practical significance.
What does a p-value of 0.05 mean in this test?
A p-value of 0.05 means that if the data truly came from a normal distribution, there is a 5% chance of obtaining a test statistic as extreme as the one observed. Typically, you reject normality if p < 0.05. But this threshold is arbitrary; some fields use 0.01 for stricter control.
Is the Pearson normality test the same as the Pearson correlation coefficient?
No. The Pearson correlation coefficient is used for measuring the linear relationship between two variables. The Pearson normality test uses the same formula but applies it to the relationship between the sorted sample values and the expected normal scores. It is a special application of the correlation coefficient, not a general correlation test.
What if my data has outliers? Should I remove them before testing?
Outliers can heavily influence the Pearson r and lead to a false rejection of normality. Before removing outliers, investigate whether they are data entry errors or genuine extreme values. If they are genuine, consider using a robust normality test or a non-parametric method. Removing outliers without justification is not recommended.
Do I need to test normality before every t-test?
For small samples (n < 30) and when the data are clearly skewed, testing normality is wise. For larger samples, the t-test is robust to non-normality due to the Central Limit Theorem. However, if you are using a one-tailed t-test or constructing prediction intervals, normality matters more. Always check a Q-Q plot as a quick visual assessment.
Related
Formula used
in list | Data values | - | 1.2, 2.3, 3.1, 4.0, 5.2, 6.1, 7.0 | 0 out r | Pearson r (normality) | - | 0.981 | 3 out p | Approx. p-value | - | 0.150 | 3
Each line is one input (in, sel, list) or one result (out: label, unit, formula, decimals). How the calculators work
Method: the standard formula for this calculation with the constants and conversion factors written on the page, rounded as each result states. See method and constants, the disclaimer or report an error.
More in this section
- ug to ppb Calculator: Convert Micrograms per Liter1.00 ppb
- Grams to Moles Chemistry: How to Convert0.2000 mol/L
- Convert ug/l to ppb Calculator
- CPP Calculator Ontario: How CPP Is Calculated162.12 $/month
- Grams to Moles to Molecules: How to Convert0.2000 mol/L
- Organic Chemistry Reaction Calculator: How It Works