Anderson-Darling Test
What you'll learn: How the Anderson-Darling test provides a more sensitive check for normality, especially in the tails of your data distribution.
What Makes It Different?
You've already learned two normality tests: Shapiro-Wilk and Kolmogorov-Smirnov. The Anderson-Darling test is another statistical test that checks whether your data follows a normal distribution, but with a special feature: it gives more weight to the tails (the extreme values at both ends) than the K-S test does.
Think of it like a quality inspector who pays extra attention to the edges of a product, where defects are more likely to show up. While the Kolmogorov-Smirnov test treats all parts of the distribution equally, Anderson-Darling is particularly strict about whether the extreme values match what we'd expect from a normal distribution.
Why Care About the Tails?
In many real-world scenarios—financial risk, rare events, outlier detection—the tails matter most. If your data has unusual behavior at the extremes (too many or too few outliers), the Anderson-Darling test is more likely to catch it than K-S.
How to Interpret
Like other hypothesis tests:
- Null hypothesis (H₀): The data comes from a normal distribution
- Test statistic: A calculated value measuring deviation from normality (larger = less normal)
- p-value: If p-value < 0.05, reject normality; the data likely isn't normal
The test is particularly powerful for detecting departures from normality in small-to-medium sample sizes.
Key Takeaway: The Anderson-Darling test is a normality test that's more sensitive to deviations in the tails of your distribution than the Kolmogorov-Smirnov test, making it valuable when extreme values matter most.