When is it unsafe to interpret a significant Kruskal–Wallis test as purely a difference in medians?
StatisticsLevel: IntermediateMean and paired inference
When is it unsafe to interpret a significant Kruskal–Wallis test as purely a difference in medians?
Hints
- Ranks depend on the whole distributions.
- Equal shapes enable a location interpretation.
Answer
When group distribution shapes or spreads differ. The rank-based test can respond to broader distribution differences; a median-location interpretation needs additional shape assumptions.
Source and licence: SciPy — statistical estimators and hypothesis tests · BSD-3-Clause · answer: Quantoros editorial