MerQuR+Statistics Glossary

Statistics Glossary → Rank-biserial correlation

Rank-biserial correlationconcept

A rank-based effect size ranging from -1 to +1 that accompanies the Mann-Whitney U or Wilcoxon signed-rank test.

For two independent groups, r_rb is the difference between the probability that a randomly chosen observation from one group is larger than one from the other group and the probability that it is smaller. It is computed from U as r_rb = 1 - 2U/(n₁n₂) (the sign depends on which group U is computed for). +1 means every value in one group exceeds every value in the other, and 0 that the groups overlap completely. For paired data Kerby's simple difference formula is used: r = (T+ - T-)/(T+ + T-), the difference between the rank sums of positive and negative differences divided by the total rank sum.

r_rb is directly related to the common-language effect size (CLES, probability of superiority A): r_rb = 2A - 1. It is not the same as the 'r' effect size computed as Z/√N, which is derived from the test statistic and is on a different scale. It needs no distributional assumption and is robust to outliers, but it reflects only the ordering advantage, not the size of the difference.

An everyday example

Ten loaves each are taken from two bakeries and rated for freshness. In 75 of the 100 possible pairs of loaves bakery A's loaf is fresher, and in 25 bakery B's is. The rank-biserial correlation is (75 - 25)/100 = 0.50: a random loaf from A has a 75% chance of being fresher than a random loaf from B.

In MerQur

In the Mann-Whitney U test the rank-biserial r is given in every run and the 'Extra outputs: Rank-biserial CI' checkbox adds a confidence interval based on Fisher's z transformation; in the Wilcoxon signed-rank test the 'Paired rank-biserial and CI' checkbox gives r = (T+ - T-)/(n(n + 1)/2) with a Fisher z interval. It is also among the rank-based measures under 'Analysis type' in the effect size calculator.

Analyses where this concept appears

Mann-Whitney UWilcoxon Signed-RankEffect Size Calculator

Related concepts

Common-language effect size (CLES)r effect size from Z/√NHodges-Lehmann estimator (pseudomedian)