Thanks to Kit Baum, a Stata package -vanelteren- has been posted on SSC for
installation.
This package implements van Elteren's test, an extension of the Wilcoxon-Mann-
Whitney rank sum test that allows for blocking or stratification on a third
variable. It can be thought of as a nonparametric analogue of randomized
blocks ANOVA, with the exception that, like the Wilcoxon-Mann-Whitney test, van
Elteren's test allows only two levels of the -by- variable.
The response variable may be ordered categorical or continuous. It imposes
fewer assumptions than, say, -ologit- or -anova-. The test appears to be
popular in clinical trials, in both academic and industrial arenas.
Like the rank sum test and the corresponding Hodges-Lehmann estimate (-search
npshift cendif somersd Newson, or-), van Elteren's test can be inverted to
produce effect estimates and confidence intervals. Michael O'Kelly is giving a
talk on this topic tomorrow at the annual conference of Statisticians in the
Pharmaceutical Industry; see www.statistikersamfundet.se/fms/psiinvit.pdf .
Although the implementation in -vanelteren- is for the asymptotic test
statistic, it is a weighted rank sum test at heart, and an example of how to
calculate an exact p-value is given by Lehmann (cited in the help file). There
is some finagling in -vanelteren- to allow string variables (e.g., ID
variables) to be used in the -strata()- option, but -vanelteren- remains
reasonably fast, and you can match, for example, Lehmann's exact p-value to any
degree of precision desired using -vanelteren- with -permute-.
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/
