Dear _all;
I remember few years ago, there was a discussion on
the list about whether the Mann-whitney-wilcoxon test
in Stata was testing the null that the medians from
two distributions where equal, or it was really
testing that the the degree of overlap between the two
distributions was equal. i.e. equality of
distributions.
I recall that it was concluded that test was not
medians but of distributions. I looked through the
archives but could not find the tread. Does any one
recall this discussion or can offer an opinion on this
topic? Textbooks and other sources refer to this test
a a test of medians, are they wrong?
The Wilcoxon test can be seen as a test of a null hypothesis proposing a
zero value for 2 parameters, namely the Hodges-Lehmann median difference
and Somers' D of the outcome with respect to membership of the first group
(instead of the second group). Both of these parameters can be estimated
with confidence limits, and most scientists these days consider a
confidence interval to be more informative than a P-value for a hypothesis
test.
