[R] wilcox.test returning 'NA' p-value

[Peter Ehlers]

Ambar Amarelo wrote:
>> Did you read the help page?
> 
> Hi  Ista Zahn and Peter, thanks to response  me.
> Well, some times on my data I have two samples with exactly values.
> (not small samples)
> 
> If the mann-whitney test (wilcox.test) return a p-value = NA
> Can I consider that my two sample have a p-value = 1 ???
> 
You probably can, but so what? I realize that p-value obsession
is rampant, but sometimes it's actually best to study the data.

You can always check for the sort of pathological case you're
interested in before passing data to wilcox.test(), using, say,
diff() or rle() on your vectors.

You might also find the error messages in coin::wilcox_test()
more informative.

> Or there's another's cases that this test will return NA and the samples are
> not the same ???
> 
> BTW, on the help ?wilcox.test
> 
>     "...Otherwise, if both 'x' and 'y' are given and 'paired' is 'FALSE',
>      a Wilcoxon rank sum test (equivalent to the Mann-Whitney test: see
>      the Note) is carried out..."
> 
> This is why i set the parameter PAIRED=FALSE on wicox.test.
> 
Yes, but my point was that R is case-sensitive;
"paired" != "PAIRED"

  -Peter Ehlers

> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 2009/11/5 Peter Ehlers <ehlers at ucalgary.ca>
> 
> 
>> Ambar Amarelo wrote:
>>
>>> Hi folks, sorry for this beginner question but what means a p-value = NA
>>> on
>>> a menn-whitney test?
>>>
>>> v1 <- c(0.022, 0.022, 0.022, 0.022, 0.022, 0.022)
>>> v2 <- c(0.022, 0.022, 0.022, 0.022, 0.022)
>>> wilcox.test(v1, v2, PAIRED=FALSE)
>>>
>>> W = 15, p-value = NA
>>>
>>> I know that there's no statistical difference between v1 and v2, so why my
>>> p-value is not ONE ??
>>>
>>> Can I consider p-value= NA  as ONE ??
>>> what i do when get this type of value on wilcox.test ???
>>>
>>>
>>> PS: sorry for my bad english.
>>>
>>>        [[alternative HTML version deleted]]
>>>
>>>
>> Did you read the help page? If there are ties, then a
>> Normal approximation is used. So how would you do a
>> Normal approximation? Try it "by hand" - you have a
>> small enough amount of data.
>>
>> Try also a t.test() and see what message it gives you.
>>
>> BTW, there is no PAIRED argument to wilcox.test().
>>
>> --
>> Peter Ehlers
>> University of Calgary
>>
> 
> 
>