Robert Dawson wrote:

[a long description of an instransitivity problem with WMW]

This is very interesting!

I'm interested to know what happens in these cases with 
Kruskal-Wallis - presumably it will reject.

It does make the point (which I always try to make clear
to people) that unless you have a shift-alternative (*or*
what would be a shift-alternative after a monotonic 
transformation), you probably need to think about the
question of interest more carefully. (i.e. what is it
you're really interested in?) It often turns out in those
cases that any difference in distribution is of interest, 
but good power against location shift is desired. This
can be done without pounding WMW's square peg into that
particular round hole.


Reply via email to