Rank Biserial Correlation

[stikpet]

Thanks for the great library.

I was looking at the rank-biserial correlation that comes with the Mann-Whitney U test results. It seems the result is the same as the one proposed by Glass (1965, p. 91; 1966, p. 626) which is actually also the same as Somers'd (1962, p. 804) and Cliff Delta (1993, p. 495). Also Kerby's formula seems to give this result. However, Glass clearly states the result is not to be used if there are ties present between the two categories.

Cureton actually addresses this in his 1968 article and sticks with his own version from 1956.

Wouldn't it be 'better' to use Cureton's version?

[raphaelvallat]

Thanks @stikpet — could you please create an issue with an example code implementation and comparison of all these different versions? Is there a reference implementation for the Cureton version (R, Matlab) that we can compare against ?

[stikpet]

created an issue :-)