Test for normality

Choose Stat > Basic Statistics > Normality Test. The test results indicate whether you should reject or fail to reject the null hypothesis that the data come from a normally distributed population. You can do a normality test and produce a normal probability plot in the same analysis. The normality test and probability plot are usually the best tools for judging normality.

The following are types of normality tests that you can use to assess normality.

Anderson-Darling test

This test compares the ECDF (empirical cumulative distribution function) of your sample data with the distribution expected if the data were normal. If the observed difference is adequately large, you will reject the null hypothesis of population normality.

Ryan-Joiner normality test

This test assesses normality by calculating the correlation between your data and the normal scores of your data. If the correlation coefficient is near 1, the population is likely to be normal. The Ryan-Joiner statistic assesses the strength of this correlation; if it is less than the appropriate critical value, you will reject the null hypothesis of population normality. This test is similar to the Shapiro-Wilk normality test.

Kolmogorov-Smirnov normality test

This test compares the ECDF (empirical cumulative distribution function) of your sample data with the distribution expected if the data were normal. If this observed difference is adequately large, the test will reject the null hypothesis of population normality. If the p-value of this test is less than your chosen α, you can reject your null hypothesis and conclude that the population is nonnormal.

Anderson-Darling and Kolmogorov-Smirnov tests are based on the empirical distribution function. Ryan-Joiner (similar to Shapiro-Wilk) is based on regression and correlation.

All three tests tend to work well in identifying a distribution as not normal when the distribution is skewed. All three tests are less distinguishing when the underlying distribution is a t-distribution and nonnormality is due to kurtosis. Usually, between the tests based on the empirical distribution function, Anderson-Darling tends to be more effective in detecting departures in the tails of the distribution. Usually, if departure from normality at the tails is the major problem, many statisticians would use Anderson-Darling as the first choice.