Academic Statistics · Module 06 of 11

Non-parametric tests

This module covers rank-based alternatives to parametric tests: the Mann-Whitney U, Wilcoxon signed-rank, Kruskal-Wallis with Dunn's post-hoc test and Friedman tests, and Spearman's rank correlation. You will learn when these tests are genuinely needed, what hypotheses they test and how to report them with effect sizes.

Watch the first lesson of every module free. The remaining lessons, exercises and live sessions open when you enrol.

What you will be able to do

  • Match each parametric test to its non-parametric equivalent
  • Decide when a non-parametric test is appropriate, for example with ordinal outcomes or small samples with clear non-normality
  • Run Kruskal-Wallis with Dunn's post-hoc tests and an adjustment for multiple comparisons
  • Report effect sizes such as r = Z / sqrt(N) or rank-biserial correlation

Key terms

  • Mann-Whitney U
  • Wilcoxon signed-rank test
  • Kruskal-Wallis test
  • Friedman test
  • Dunn's test
  • Rank-biserial correlation
  • Ordinal data

Hands-on practice

Run a Mann-Whitney U test and a Kruskal-Wallis test on a provided ordinal dataset in SPSS or jamovi and report each with an effect size.

Common pitfalls this module fixes

  • Describing the Mann-Whitney U test as a comparison of medians without qualification; that interpretation holds only when the distributions have similar shapes.
  • Switching to non-parametric tests at the first sign of a significant Shapiro-Wilk result in a large sample; inspect plots and remember that t-tests and ANOVA are fairly robust to moderate non-normality.
  • Reporting means and SDs alongside a rank-based test; report medians and IQRs to match the analysis.