Skip to content

kruskal wallis post hoc?

7 messages · Mario Garrido, Tal Galili, Iasonas Lamprianou +2 more

#
Try to use "kruskalmc" in the package pgirmess

I give one of my results
Kruskal-Wallis chi-squared = 132.3091, df = 3, p-value < 2.2e-16
Multiple comparison test after Kruskal-Wallis 
p.value: 0.05 
Comparisons
              obs.dif critical.dif difference
abril-julio 134.52815     56.84018       TRUE
abril-junio  53.42185     58.73584      FALSE
abril-mayo   58.96383     59.73167      FALSE
julio-junio 187.95000     44.40539       TRUE
julio-mayo   75.56432     45.71446       TRUE
junio-mayo  112.38568     48.05106       TRUE



-----
Mario Garrido Escudero
PhD student
Dpto. de Biolog?a Animal, Ecolog?a, Parasitolog?a, Edafolog?a y Qca. Agr?cola
Universidad de Salamanca
--
View this message in context: http://r.789695.n4.nabble.com/kruskal-wallis-post-hoc-tp4288008p4288263.html
Sent from the R help mailing list archive at Nabble.com.
#
The Kruskal-Wallis test is a special case of the proportional odds ordinal
logistic model.  You can get any contrast you want by testing regression
coefficients.   In a couple of weeks the rms package's contrast function
will allow for individual confidence intervals of effects that together have
a 0.05 type I error, by using the multcomp package (called automatically
from contrast.rms).
Frank

Iasonas Lamprianou wrote
the answer is yes:
I would say that the post hoc correction wouldn't help you (it could be that
the reason for this significance is based on some weird contrast...)
Details:-------------------------------------------------------
www.r-statistics.com (English)
wrote:
conducted all pairwise comparisons and found no significant results. Could
anyone please give me a hint as to why this happens or redirect me towards a
specific web page where I can find more info? (I used alpha=5% and made no
bonferroni or other correction for the pairwise comparisons)
http://www.R-project.org/posting-guide.html
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/kruskal-wallis-post-hoc-tp4288008p4288894.html
Sent from the R help mailing list archive at Nabble.com.
#
On Jan 12, 2012, at 14:11 , Frank Harrell wrote:

            
Eh? Can you elaborate on that?

I would expect that at best it is equivalent to some _test_ in a polr-type model. It is never really clear what the model is when some groups are different and others not.

  
    
#
Peter,

The score test from the P.O. model for the global null hypothesis (k-1
degrees of freedom for comparing k groups) is almost exactly the
Kruskal-Wallis test statistic.  For the case where k=2 (Wilcoxon test) the
numerator of the score test is exactly the numerator of the
Wilcoxon-Mann-Whitney U test.

Frank

Peter Dalgaard-2 wrote
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/kruskal-wallis-post-hoc-tp4288008p4290363.html
Sent from the R help mailing list archive at Nabble.com.