[R] kruskal test p value way too low.

[Prof Brian Ripley]

On 26/10/2014 05:37, Rolf Turner wrote:
> On 26/10/14 16:40, David Doyle wrote:
>> Hello,
>>
>> I'm trying to run kruskal test on some data but the p values seemed
>> way too
>> low.  So I tried it on some similar data and still got p-value =
>> 1.611e-09.  I'm sure it is a simple mistake but I can't figure it out.
>>
>> Below is my data and code.  Could it be because there are some miss
>> data /
>> NAs in the data set??  If so, could some one point me towards a
>> solution??
>>
>> Thank you for your time.
>> David
>>
>> mydata <-read.csv("http://doylesdartden.com/R/test.csv", sep=",")
>> kruskal.test(mydata, AMMONIA~Well)
>>
>>          Kruskal-Wallis rank sum test
>>
>> data:  mydata
>> Kruskal-Wallis chi-squared = 36.3952, df = 1, p-value = 1.611e-09
>
> Thank you for providing a clear question and an easily reproducible
> example.
>
> The problem is your syntax for the call to the kruskal.test() function.
>
> It should be:
>
>      kruskal.test(AMMONIA ~ Well, data=mydata)
>
> This gives a p-value of 1 (which agrees with wilcox.test); t.test()
> gives a p-value of 0.7958) so harmony is restored to the universe.
>
> IMHO there is a bit of a design flaw in kruskal.test(); it should have
> thrown an error, given your syntax.  The wilcox.test() function *does*

 From the help page

      ## Default S3 method:
      kruskal.test(x, g, ...)

Arguments:

        x: a numeric vector of data values, or a list of numeric data
           vectors.

        g: a vector or factor object giving the group for the
           corresponding elements of ‘x’.  Ignored if ‘x’ is a list.

x was a list, hence g is ignored.  The only thing you can quibble about 
is if the elements of x were 'numeric data vectors': one is a factor and 
it is conventional in R to coerce to the specification, as happened here.

I have added a warning if such coercion is done.


> throw an error.
>
> cheers,
>
> Rolf Turner
>


-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford
1 South Parks Road, Oxford OX1 3TG, UK