The t-test and ANOVA are powerful but rest on assumptions — chiefly that data is roughly normal — that real data frequently violates. Non-parametric tests exist as the robust alternatives that make no assumption about the underlying distribution's shape, working instead on the ranks of the data. Without them, analysts facing skewed data, ordinal measurements, or small samples with outliers would be forced to either misapply parametric tests or abandon inference entirely. The Mann-Whitney U test and the Kruskal-Wallis test are the non-parametric counterparts to the two-sample t-test and one-way ANOVA, comparing distributions by their ranks rather than their means. They are the essential fallback whenever the comfortable normality assumption is untenable, trading a little power for substantial robustness.
25 minintermediate
Non-Parametric Tests — Mann-Whitney, Kruskal
Analogy🏏Cricket
🏏 Think of it like cricket: Imagine Virat Kohli has scored 45, 52, 38, 61, and 49 across five innings in a series, and a commentator wants to describe his form in one phrase. The commentator cannot read out all five scores every time, so they compress them into a single representative figure. Just as the commentator picks one number to stand in for the whole sequence of innings, a measure of central tendency picks one value to represent an entire dataset. Just as a misleading summary ('he averages 200') would distort how selectors judge Kohli, a wrongly chosen centre distorts how analysts judge data. And just as different summaries (best score versus typical score) tell different stories, the mean, median, and mode each emphasise a different aspect of the same innings. This reveals why central tendency is never one fixed number — it is a deliberate choice about which story the data should tell.
Lesson 28 of 35
0% complete