On Tue, 1 Oct 2002, Thom Baguley wrote:

> Donald Burrill wrote:
> >  And, to address your last question, a Mann-Whitney test (aka a
> > Wilcoxon rank-sum test) is exactly equivalent to a t-test on the mean
> > ranks;  so yes, you can, but use alpha/2 for your tabled significance
> > level.
>
> Pedantry alert! Is it _exactly_ equivalent - don't the results
> differ very slightly (though not to a degree that would make a
> difference in practice)?

Pedantic reply:  My proximate source is quoted below.  If you want more
details, consult Donald Zimmerman or Bruno Zumbo.   (Bruno, I believe,
used to subscribe to this edstat list;  I don't know if he still does.)
 As I read their prose, the equivalence is exact for the large-sample
normal approximation.  If the "differ very slightly" Thom has in mind
reflects whatever difference there is between the true distribution and
the normal approximation thereto, he's probably right;  and he's certainly
right, I think, that it would not affect practice.   -- DFB.

>From Donald W. Zimmerman & Bruno D. Zumbo, "The relative
power of parametric and nonparametric statistical methods", Chapter 19 in
_A Handbook for Data Analysis in the Behavioral Sciences_, edited by
Gideon Keren and Charles Lewis (Erlbaum, 1993), pages 487-488:

"It has been known in mathematical statistics and recently emphasized by
Conover (1980), Conover and Iman (1981), and others, that the
Wilcoxon-Mann-Whitney test in standard form, that is the large sample
normal approximation, is equivalent to an ordinary Student _t_ test
performed on the ranks of measures instead of the measures themselves.
 [They then give an equation expressing _t_ in terms of _W_ and _N_.]
...
  "This means that, apart from details of computation, it makes no
difference whether a researcher performs a Wilcoxon test based on rank
sums, or alternatively, pays no attention to _W_ and simply performs the
usual Student _t_ test on the ranks ...
  "[Similarly], the Wilcoxon matched-pairs signed-ranks test is equivalent
to a paired-samples Student _t_ test on signed ranks instead of original
signed differences.  The Kruskal-Wallis test is equivalent to an ordinary
_F_ test performed on the ranks of the measures in an ANOVA design instead
of the measures themselves, and so on.  These findings have not received a
great deal of attention in introductory statistics textbooks and are not
widely known among researchers in psychology."

The references are
  Conover, W.J. (1980).  _Practical nonparametric statistics_ (2nd ed.).
New York:  Wiley.
  Conover, W.J. & Iman, R.L. (1981).  Rank transformations as a bridge
between parametric and nonparametric statistics.  _The American
Statistician, 35,_ 124-129.

 -----------------------------------------------------------------------
 Donald F. Burrill                                            [EMAIL PROTECTED]
 56 Sebbins Pond Drive, Bedford, NH 03110                 (603) 626-0816
 [Old address:  184 Nashua Road, Bedford, NH 03110       (603) 471-7128]

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to