Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Nemenyi test


From   Célia Dechavanne <[email protected]>
To   [email protected]
Subject   Re: st: Nemenyi test
Date   Thu, 29 Aug 2013 08:22:19 +0200

Thank you for your help.
My first mail was not enough clear, sorry for that. The five groups are small (minimum n=11 and maximum n=26). The KW test gives a significant difference between the groups, now I would like to compare 2 by 2. But I can't perform pairwise comparison, I would like to compare the shift between the five groups. I hope is better clear now.
Thank you so much for the time spent to answer.
Best
Celia

Le mercredi 28 août 18:29, Marta García-Granero a écrit :
I have Hochberg&Tamhane "Multiple Comparisons Procedures", and there are a couple of references for a Nemenyi procedure, one for pairwise signed ranks tests (after a non-parametric K related samples omnibus test) and another for pairwise median tests, after a k-samples median test (not Kruskal-Wallis). This last method is also known as Levy test.

Unless there is another "Nemenyi procedure" not mentioned in H&T book, I think Celia is a bit confused, since she has performed a Kruskal-Wallis test, not a k-medians test. Anyway, I have found a web page that described multiple comparisons method named Nemenyi after a KW test...

Maybe the simplest pairwise comparison method she can use is pairwise rank-sums (AKA: Mann-Whitney's U tests) with Sidak adjustment.

This program is rather unpolished (absolutely no error trapping or handling), but it does the task (quick&dirty solution):

program multimw,nclass
version 12
args x y
confirm var `x'
confirm var `y'
sum `y',meanonly
local comp=r(max)*(r(max)-1)/2
qui levelsof `y',local(levels)
foreach i of local levels{
    foreach j of local levels{
        if `i'<`j'{
        quietly ranksum `x' if (`y'==`i')|(`y'==`j'), by(`y')
        display `j'  _c
        display " vs " `i'  _c
        display "p = " %6.4f 2*(1-normal(`r(z)')) "    " _c
        display "p’= " %6.4f 1-(1-2*(1-normal(`r(z)')))^comp
        }
    }
}
end

It gives both unadjusted and adjusted p-values.

HTH.
Marta GG

El 27/08/2013 13:22, David Hoaglin escribió:
Celia,

I'm not sure what you mean by the "Nemenyi test." Kruskal-Wallis and
similar tests based on ranks do not compare distributions.  They focus
on shifts (e.g., differences in medians among distributions that have
the same shape).

How do your continuous distributions behave?  What is the reason for
comparing the whole distributions?  How large are the samples?

David Hoaglin

On Tue, Aug 27, 2013 at 2:16 AM, Célia Dechavanne <[email protected]> wrote:
I am attempting to compare the distributions of a continuous variable
between five groups after a Kruskal Wallis in Stata without success.

I think that Nemenyi test is the one I need, but I don't find anything in
Stata. Could you help me to perform this test?

Thanks for your help,

--
Célia Dechavanne
INSERM - UMR S665
Equipe 3 - Pathogénèse du paludisme sévère
6, rue Alexandre Cabanel
75739 Paris Cedex 15


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index