Normality tests at small sample sizes may lead to inaccurate medical reference intervals.
Normality tests may not accurately identify samples from a Gaussian population at small sizes. The study evaluated 4 normality tests on samples of 30 and 60 values from different populations. Shapiro-Wilk and D'Agostino-Pearson tests performed best, but had poor specificity at n = 30. Using parametric methods on samples falsely identified as Gaussian led to inaccurate reference intervals. Nonparametric methods or adjusting significance levels based on sample size can reduce errors in constructing reference intervals.
