Re: Differences between Mann-Whitney U Test / Wilcoxon and Spearman Correlation?

From: Kelly (nilotna2001_at_yahoo.com)
Date: 12/28/04 

Date: Tue, 28 Dec 2004 14:00:18 +0000 (UTC)

On Mon, 27 Dec 2004 05:28:36 +0200, J. Mitchell wrote:
>I have a function Y=f(A,B,C,...). The function is of course not
known;
>A,B,C,... are not necessarily distributed Normally.
>
>I want to know which of the independent variables affect Y and to
what
>degree.
>
>What is the difference between: (the 7th percentile is just an
example
>threshold)
>(1) Doing Spearman correlation in the 0th-7th percentile and also in
the
>7th-10th percentile between Y and A,B,C,...
>(2) Doing Mann-Whitney (having divided A,B,C,... into groups
according to
>the 0th-7th percentiles and 7th-10th percentiles of Y).
>
>I realize that obviously Spearman and M-W are not the same, but it
appears
>that the end result (i.e. the final conclusion which variables affect
Y)
>will be very similar.
>
>And what happens if A,B,C,... are correlated (doing Wilcoxon instead
of
>M-W)?
>
>Help will be most appreciated.

It is simple... the Spearman seems to be the most appropriate... but
with the right groupings on data the result will be most of the time
the same with M-W... it is for the reason that the group that yields
the higher observations seem to be significant(through ranks).. the
same thing will happen if you use wilcoxon... what you should thake in
mind the appropriateness of the test... there is no mathematical proof
to prove your claim but I experienced it...

Relevant Pages