[R] mann-whitney test with more groups

[Peter Dalgaard]

Peter Ehlers wrote:
> Sorry, correction below.
> 
> Peter Ehlers wrote:
>>
>> Kim Vanselow wrote:
>>> Dear r-helpers,
>>> I want to test groups of samples for significant differences.
>>> Question: Does Group1 differ significantly from group2.
>>> This is a question to be answered by mann-whitney-u-test.
>>>
>>> I know that I can use wilcox.test with 2 samples.
>>>
>>> My problem: How can r perform the test automatically if there are 
>>> more than 2 groups in my data frame.
>>> Test group1 vs. 2, 1 vs. 3, 1 vs. 4, etc.
>>>
>>>
>>> This is my skript:
>>> Deckung <- read.table("Gesamtdeckung.csv", sep=";", header=TRUE, 
>>> dec=",", row.names=1)
>>>
>>> x <- Deckung$Gesamtdeckung
>>> y <- Deckung$Klasse
>>>
>>> #U-Test
>>> wilcox.test(x ~ y, paired = FALSE, subset = y %in% c("1", "2"))
>>> wilcox.test(x ~ y, paired = FALSE, subset = y %in% c("1", "3"))
>>> wilcox.test(x ~ y, paired = FALSE, subset = y %in% c("2", "3"))
>>>
>>> Any help would be greatly appreciated!
>>>
>>> Thanks
>>> Kim 
>>
>> This sounds like serious data dredging, but if you're
>> sure that it's what you want, try the combn() function:
>>
>>  y <- gl(4, 5)
>>  x <- rnorm(20)
>>  m <- cbind(t(combn(4, 2)), NA)
>>  for(i in 1:nrow(idx))
>>     m[i, 3] <-
>>       wilcox.test(x ~ y, subset = y %in% idx[i,])$p.value
>>  m
>>
>>  -Peter Ehlers
>>
>   y <- gl(4, 5)
>   x <- rnorm(20)
>   m <- cbind(t(combn(4, 2)), NA)
>   for(i in 1:nrow(m))  # change 'idx' to 'm'
>      m[i, 3] <-
>        wilcox.test(x ~ y, subset = y %in% m[i,])$p.value  # ditto

There's also pairwise.wilcox.test, with multiple testing correction and 
all. (But someone called Lumley may chime in and remind you of the lack 
of guaranteed transitivity of rank tests.)


-- 
    O__  ---- Peter Dalgaard             Øster Farimagsgade 5, Entr.B
   c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
  (*) \(*) -- University of Copenhagen   Denmark      Ph:  (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)              FAX: (+45) 35327907