Hi,
after Dixon and Mood (1946) had originally proposed a sign test with ties
included, the suggestion to "discard" ties was then made by Dixon and
Massey (1951), Putter (1955) and, then, Pratt (1959). Recently, this
problem has been revisited and it has been found that ties can only be
discarded if they are "exact", meaning that there is no possibility that
differences have been merely overlooked. This often applies to genetics.
* Wittkowski KM, Liu X (2002) A statistically valid alternative to the
TDT. Hum Hered 54: 157-164
If, instead, ties are due to rounding or the use of discrete surrogate
variables for continuous phenomena, ties must not be excluded.
* Wittkowski, KM (1998) Versions of the sign test in the presence of
ties. Biometrics 54: 789�791
* Rayner JCW, Best DJ (1999) Modelling Ties in the Sign Test.
Biometrics 55: 663-665
* Randles HR (2001) On neutral responses (zeros) in the sign test and
ties in the Wilcoxon-Mann-Whitney Test. Am Statist 55: 96-101
* Fong DYT, Kwan CW, Lam KF et al. (2002) Use of the sign test for the
median in the presence of ties. Am Statist 57: 237-240
While the above papers address the sign test, rather than the Wilcoxon
signed rank test, the same rationale applies.
Note that if two tests, i.e., the sign test with ties discarded and the
sign test with ties not discarded, are not asymptotically equivalent, a
single "exact" tests cannot be appropriate for both.
I hope this helps
Knut
At 13:04 2003-12-01 +0100, you wrote:
>"Niels Steen Krogh" <[EMAIL PROTECTED]> writes:
>
> > a group of medical doctors ... is a little confused about there use of
> a test
> > named "willcoxon-pratt" for testing if the clinical and biochemical
> > markers has decreased significantly after the use of some drugs for a
> > group of patients.
> >
> > Looking into the R-functions I would in R recommand using a
> > matched-pairs Wilcoxon test with a formula like:
> > wilcox.test(pre,post,alternative='greater',paired=T)
> >
> > Looking deeper into the writings of Pratt I found some 1964-stuff
> > "Pratt JW. Remarks on zeros and ties in the Wilcoxon signed rank
> > procedures. J.Americ.Statistical Assoc. 1959; 54: 655-67. "
> >
> > Do'es any of you know what is the wilcoxon-pratt test compared with
> > the formula described above and how it should be implemented in R.
>
>wilcox.test does the original Wilcoxon procedure, discarding any tied
>pairs (zero difference). The Pratt procedure (rank all differences
>first, *then* discard tied pairs) is not immediately available, but
>would be fairly easy to implement, at least for the large-sample case.
>I wouldn't be surprised if the exact procedure could be performed
>using Torsten's ExactRankTests package (perm.test or wilcox.exact),
>but I haven't gone deeply into it.
>
>--
> O__ ---- Peter Dalgaard Blegdamsvej 3
> c/ /'_ --- Dept. of Biostatistics 2200 Cph. N
> (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918
>~~~~~~~~~~ - ([EMAIL PROTECTED]) FAX: (+45) 35327907
>
>______________________________________________
>[EMAIL PROTECTED] mailing list
>https://www.stat.math.ethz.ch/mailman/listinfo/r-help
Knut M. Wittkowski, PhD,DSc
------------------------------------------
The Rockefeller University, GCRC
Experimental Design and Biostatistics
1230 York Ave #121B, Box 322, NY,NY 10021
+1(212)327-7175, +1(212)327-8450 (Fax)
[EMAIL PROTECTED]
http://www.rucares.org/clinicalresearch/dept/biometry/
[[alternative HTML version deleted]]
______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
