The statistical power of this test may sometimes be improved by using a value other than the median to define the groups – that is, by using a value which divides the groups into more nearly equal groups than the median would.
This test is useful when the distributions of data differ markedly from normality, for example when raw scores have been arbitrarily classified into inappropriate ranges before being recorded. For this reason it is often useful as an initial step in exploratory analysis[?].
wikipedia.org dumped 2003-03-17 with terodump