Nonparametric Tests

The t-test and ANOVA compare means, and they assume roughly normal data. When your data is skewed, has strong outliers, is measured on a rating scale, or the sample is tiny, the mean stops being a fair summary — and the test can mislead. Nonparametric tests step in.

The Rank Idea

Sort every value, replace each with its rank, and compare the ranks across groups — as in the chart above. A single extreme value only moves one rank, so it can't drag the whole result the way it drags a mean.

The Common Tests

Each replaces a mean-based comparison with a rank-based one, answering the same question without the normality assumption.

The Trade-off

• More robust: skew and outliers barely move the result
• Fewer assumptions: no normality required
• Slightly less powerful when the data really is normal
• Tests a shift in distribution, not strictly a difference in means

Nonparametric Tests in the SKARI Statistical Lab

The Statistical Lab checks normality first and, when it fails, offers the matching rank-based test — Mann-Whitney, Wilcoxon, or Kruskal-Wallis — with the same plain-language read as its parametric cousin.

• Automatic normality check that points you to the right test
• Mann-Whitney, Wilcoxon, Kruskal-Wallis, and Spearman
• Interpretation alongside the p-value, not just a number

Frequently Asked Questions

When should I use a nonparametric test?

When the data is skewed, has strong outliers, is ordinal, or the sample is too small to trust normality.

Is it always safer?

No — when the data is genuinely normal, the parametric test detects real differences more reliably.

What exactly does it compare?

A shift between distributions (often described via the median), not strictly the difference in means.

Conclusion

Nonparametric tests are the honest fallback when normality fails. Match each to its parametric cousin, use it when the assumptions actually break, and your comparison stays trustworthy.