Non-parametric comparison with Kruskal-Wallis

Learning objectives

• Recognise when Kruskal-Wallis is appropriate.
• Interpret a rank-based result.
• State its limitations.

What it tests

Kruskal-Wallis compares distributions across independent groups using ranks. Similar group shapes let it be read as a median comparison.

It does not test means and it is not assumption-free — independence and similar shapes still matter.

Garment-factory example

Compare highly skewed repair minutes across three defect categories where ANOVA residual assumptions fail.

Method

1. Plot groups and confirm independence.
2. Run the test in a free spreadsheet or tool.
3. Follow up with adjusted-error pairwise comparisons.
4. Report effect and its practical meaning.

Common mistakes

• Using it for paired data.
• Claiming it tests means.