Re: [S] nonparametric pairwise comparisons

Andrew M Kuhn (
Thu, 5 Mar 1998 14:43:28 -0500 (EST)


For #2, a multivariate Wilcoxon statistics may be what you are looking
for. At each time point, a two sample Wilcoxon test is computed as well
as the covariance matrix between all of these (correlated) tests. Two
reference papers are:

Wei and Lachin ('Two-Sample Asymptotically
Distribution-Free Tests for Incomplete Multivariate Observations',
Journal of the American Statistical Association 79:653-661, 1984)


Wei and Johnson ('Combining Dependent Tests with
Incomplete Repeated Measurements', Biometrika 72:359-364, 1985).

As far as Splus implementations, a paper by

Davis ('A Computer Program
for Nonparametric Analysis of Incomplete Repeated Measures from Two
Samples', Computer Methods and Programs in Biomedicine 42:39-52, 1994)

provides Fortran implementations of these two methods. I would recommend
the statistic from Wei and Lachin for Wilcoxon test since it is more
computationally efficient than Wei and Johnson's test.

Max Kuhn

On Thu, 5 Mar 1998 wrote:

> Hi folks,
> I have a time series for several study sites. I'm interested
> in the differences in the mean response among sites, not necessarily
> among times, so I've done a Friedman's test and verified the existence
> of significant differences among sites. I'd now like to do pairwise
> comparisons among my sites, and I have two S-PLUS related questions in
> this regard:
> 1. The method of which I'm aware for these pairwise comparisons is
> to calculate the U-statistic for each pair of sites, substituting the
> number of times at which samples were taken (i.e. blocks) for sample
> size. Does anyone have an Splus function that will do this? As near
> as I can tell, I can't do a Mann-Whitney in S-PLUS at all.
> 2. Does anyone know of a (nonparametric) method for pairwise
> comparisons that takes advantage of the temporal structure of the
> data-- something along the lines of a post-hoc version of the Wilcoxon
> test for paired comparisons? This would avoid losing a lot of the
> structure in the data.
> Thanks in advance for any suggestions.
> Cheers,
> Sean
> *************************************
> Sean Connolly
> Department of Biological Sciences
> Stanford University
> Stanford, CA 94305-5020
> (650) 723-4365
> *************************************
> -----------------------------------------------------------------------
> This message was distributed by To unsubscribe
> send e-mail to with the BODY of the
> message: unsubscribe s-news

This message was distributed by To unsubscribe
send e-mail to with the BODY of the
message: unsubscribe s-news