Skip to content

Chi-squared test (PR#9350)

3 messages · McGehee, Robert, Kasper Daniel Hansen

Original
McGehee, Robert
# 
Is Sahotra's abuse sending a wishlist item to R-bugs, or is his error
phrasing a suggestion in the form of a question? It seems others have
commented on the inability to specify df for chisq.test (see:
http://tolstoy.newcastle.edu.au/R/help/05/01/10539.html), and adding an
option certainly seems like a reasonable suggestion to me.

And of course, R-FAQ 9.2 does explicitly give a standard for reporting
wishlist or non-bug items to the R-bugs:
"There is a section of the bug repository for suggestions for
enhancements for R labelled `wishlist'. Suggestions can be submitted in
the same ways as bugs, but please ensure that the subject line makes
clear that this is for the wishlist and not a bug report, for example by
starting with `Wishlist:'."

So if your qualm is with how onerous and manual the R bugs tracking
system is, rather than if the sender sent his question/comment to the
wrong list or in the wrong form, then perhaps an upgrade to the R bug
tracker is in order rather than discouraging sending wishlist items to
the bugs archiving system. 

Robert

-----Original Message-----
From: r-devel-bounces at r-project.org
[mailto:r-devel-bounces at r-project.org] On Behalf Of Kasper Daniel Hansen
Sent: Tuesday, November 07, 2006 10:24 PM
To: sarkar at mail.utexas.edu
Cc: R-devel
Subject: Re: [Rd] Chi-squared test (PR#9350)

(not cc'ed to R-bugs)

If this is not a bug why do you abuse the bug report system to ask a  
simple question? Now a member of R-core has to use valuable time to  
handle this bug report, time that could be far more profitable spent  
making improvements to R (or spend it otherwise). This is incredible  
selfish behavior.

Kasper
On Nov 7, 2006, at 5:41 PM, sarkar at mail.utexas.edu wrote:

            

      
      
      
    

        

      
      
    

        
______________________________________________
R-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

  
  


  


    

    

      
      

        


  

        

      Kasper Daniel Hansen
    

    

      
      #
    

  


  

      
Upon further reflection on my part I have realized that my original  
email was way too harsh - even for the sometimes harsh tone on R-devel.

Having spent a few more minutes reflecting on the original question  
and reading the old thread Robert linked to, it seems that there is  
indeed some questions about the non-existence of a df argument.

I guess it can be interesting in the case where you are comparing to  
an estimated probability vector and you want to take that estimation  
into account (which does not seem straightforward, see Peter  
Dalgaard's old email). In that case a user might be interested in  
explicitly passing along a df argument.

On the other hand, there are several circumstances where I would say  
a df is unnecessary. One case is when the p value is computed by  
simulation and another is the case where the argument is a matrix and  
the function tests for independence (at least it would be highly non- 
standard to have any other degrees of freedom that the usual one).

Since the function returns the statistic it is very easy to directly  
compute the df by
   statistic = chisq.test( ... )$statistic
   p.value = 1 - pchisq(statistic, df = df)
(or p.value = pchisq(statistic, df = df, lower.tail = FALSE) ).

Altogether I would personally be slightly inclined to say it is  
better not to provide a df argument - I would think that the  
possibility of misuse is to high.

Kasper

        
On Nov 8, 2006, at 7:40 AM, McGehee, Robert wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Kasper Daniel Hansen
    

    

      
      #
    

  


  

      
On Nov 8, 2006, at 9:44 PM, Kasper Daniel Hansen wrote:

        

            

      
      
      
    

        
Ok, I am writing pretty bad here, this paragraph should fo course read

Since the function returns the statistic it is very easy to directly
compute the p-value for a given degrees of freedom by
    statistic = chisq.test( ... )$statistic
    p.value = 1 - pchisq(statistic, df = df)
(or p.value = pchisq(statistic, df = df, lower.tail = FALSE) ).