[R] using "cochran.test()" as a "mcnemar.test()" ?

[David Winsemius]

On Jun 1, 2009, at 6:01 PM, Tal Galili wrote:

> Hi Chuck,
> Thank you for the fast reply.
> I looked through that thread (that's how I came across the commands I
> wrote).
>
> 1) My question was if this command (cochran.test) could be used the  
> same way
> as (mcnemar.test). It obviously can't be, but maybe there is another  
> command
> I don't know about.

???. What is wrong with mcnemar.test()?

>
> 2) Why do you think I might not want this test ?

They test different hypotheses. The McNemar test tests for equal  
probabilities of cells (i,j) and (j,i), in essence a test of symmetry  
across the diagonal, whereas the Cochran test is for the similarity of  
distribution in k columns. Who can tell which test you really want?  
You have not described your design or research question at all. That  
Wikipedia article does not say they are equivalent in general, only  
that they are in the 2 x 2 case, which would seem readily apparent.

Thompson's accompaniment to Agresti's text says (p 177):
"Also, McNemar’s test is equivalent to a Cochran-Mantel-Haenszel test  
with subjects as strata. Thus, the CMH test is performed on a 2 x 2 x  
n table. This table is called the subject-specific table, and will be  
used in the next section."

https://home.comcast.net/~lthompson221/Splusdiscrete2.pdf

-- 
David

>
> Thanks,
> Tal
>
> On Tue, Jun 2, 2009 at 12:56 AM, Chuck Cleland  
> <ccleland at optonline.net>wrote:
>
>> On 6/1/2009 5:29 PM, Tal Galili wrote:
>>> Hello all
>>> I wish to perform a mcnemar.test() for a 5X5 matrix.
>>> Wikipedia tells me (http://en.wikipedia.org/wiki/Cochran_test) I  
>>> should
>> turn
>>> to cochran.test.
>>> The only place I found it was in the "outliers" package, but the  
>>> command
>>> cochran.test() acts differently then  mcnemar.test() , and doesn't  
>>> take a
>>> table as input.
>>>
>>> Any ideas on how to use it ?
>>>
>>> #Example code:
>>> aa =
>>>
>> as 
>> .table(matrix(c(0,0,0,0,1,0,2,2,1,0,1,2,2,1,0,1,1,3,2,0,0,0,1,2,3), 
>> 5,5))
>>> mcnemar.test(aa)
>>> #  p-value = NA
>>> install.packages("outliers")
>>> require(outliers)
>>> cochran.test(aa)
>>> # Error in x$terms : $ operator is invalid for atomic vectors
>>>
>>> Thanks,
>>> Tal
>>
>>  I'm not sure you actually want Cochran's Q Test, but if you do see
>> this thread:
>>
>> https://stat.ethz.ch/pipermail/r-help/2006-September/113139.html
>>
>> --
>> Chuck Cleland, Ph.D.
> ----------------------------------------------
>
>
> My contact information:
> Tal Galili

David Winsemius, MD
Heritage Laboratories
West Hartford, CT