Steve Albert asked about a stratified Wilcoxon test that is different from the
Wilcoxon-Breslow test in -sts test-. Richard Goldstein confirmed that Cytel has the
stratified rank-sum test in its repertoire.
I believe that what Steve's looking for is also called van Elteren's test. SAS can
perform this, too, using its PROC FREQ with the Cochran-Mantel-Haenszel option.
See ftp.sas.com/techsup/download/stat/vanelter.html for how. Stata cannot perform
this test with any built-in command or currently available user-written command, as
far as I know.
Steve was surprised that Stata doesn't have much in the way of stratified rank-sum
tests. I believe that Stata Corp. had intended to implement more along this line
(see the last paragraph of www.stata.com/support/faqs/stat/trend.html).
I understand that Steve's firm has committed to using the method in this case, but
keep in mind that the Cochran-Mantel-Haenszel method has foibles when it comes
to stratified analysis (see, for example, www.faqchest.com/appl/sas-l/sas-02/sas-
0208/sas-020830/sas02082705_09296.html).
In another context--I'm using Stata to go through M. E. Stokes, C. S. Davis & G. G.
Koch, _Categorical Data Analysis Using the SAS System, 2nd, (Cary, North
Carolina: SAS Institute, 2000)--I've found that a corresponding parametric model
with an intraction term will typically best the Cochran-Mantel-Haenszel technique. If
the sample sizes are small, the parametric model using Stata's -perm- or -permute-
has worked nicely. Perhaps this is one reason why Stata hasn't had much incentive
to implement the technique more vigorously.
Joseph Coveney
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
