Which non-parametric test is appropriate for comparing three or more independent groups on a non-normal distribution?

Which non-parametric test is appropriate for comparing three or more independent groups on a non-normal distribution?

• A. Kruskal-Wallis
• B. Friedman's ANOVA
• C. Wilcoxon
• D. Mann-Whitney U

Answer: (A)

When you’re comparing three or more independent groups and the data aren’t normally distributed, you want a test that uses ranks rather than raw values. Kruskal-Wallis fits this need perfectly because it generalizes the idea of rank-based comparison to any number of groups. It pools all observations, assigns ranks, and then looks at whether the mean ranks differ across groups. The null hypothesis is that the distributions (or medians) are the same across all groups. Because it relies on ranks, it doesn’t assume normality and works well with skewed data or ordinal measurements. If you find a significant result, you’d typically follow up with post-hoc pairwise comparisons (like Dunn’s test) with a correction for multiple testing to see which groups differ. The other options aren’t appropriate here: Friedman’s ANOVA is for related (repeated) measures; the Wilcoxon family tests are for paired data or two independent groups (Wilcoxon signed-rank for paired data, Wilcoxon rank-sum for two groups); Mann-Whitney U is the two-group version. So Kruskal-Wallis is the right choice.

When you’re comparing three or more independent groups and the data aren’t normally distributed, you want a test that uses ranks rather than raw values. Kruskal-Wallis fits this need perfectly because it generalizes the idea of rank-based comparison to any number of groups. It pools all observations, assigns ranks, and then looks at whether the mean ranks differ across groups. The null hypothesis is that the distributions (or medians) are the same across all groups. Because it relies on ranks, it doesn’t assume normality and works well with skewed data or ordinal measurements. If you find a significant result, you’d typically follow up with post-hoc pairwise comparisons (like Dunn’s test) with a correction for multiple testing to see which groups differ. The other options aren’t appropriate here: Friedman’s ANOVA is for related (repeated) measures; the Wilcoxon family tests are for paired data or two independent groups (Wilcoxon signed-rank for paired data, Wilcoxon rank-sum for two groups); Mann-Whitney U is the two-group version. So Kruskal-Wallis is the right choice.