Statistics Glossary → Kendall's W (coefficient of concordance)
Kendall's W (coefficient of concordance)concept
A coefficient between 0 and 1 that shows how far several raters agree with one another when ranking the same set of objects.
When m raters rank n objects, the sum of the ranks each object receives is computed. If the raters agree perfectly these sums are very different from one another (one object always first, another always last); if there is no agreement they stay close together. In W = 12·S / (m²(n³ - n)), S is the sum of squared deviations of the rank sums from their mean. W = 1 indicates perfect agreement and W = 0 no agreement at all; with ties a corrected formula is used.
Kendall's W is directly linked to the Friedman test: W = χ²_F / (m(n - 1)). It is therefore also reported as the effect size for the Friedman test; the Friedman test answers 'do the objects differ?', while W expresses how strong that difference and the consistency among raters are. W is also related to the average Spearman correlation between raters. It is suitable only for ranked data; for agreement on continuous measurements, the intraclass correlation coefficient is used.
An everyday example
Five food critics rank the same six restaurants from best to worst. If they all put the same restaurant first and the same one last, W approaches 1. If every critic produces a completely different ranking, the restaurants' rank sums end up close together and W approaches zero: the critics cannot be said to share a common taste.
In MerQur
It is given as the effect size in the Friedman test; the 'Kendall's W CI' checkbox adds a percentile bootstrap confidence interval obtained by resampling blocks.