[R] Difference Between R: wilcox.test and STATA: signrank

[peter dalgaard]

On Aug 9, 2010, at 3:03 PM, Alain Guillet wrote:

> Hi,
> 
> Look at the output of the test made in R and you can see it is a Wilcoxon rank sum test and not a Wilcoxon signed rank test.

It might be helpful to add that paired=TRUE is needed in the call to get the signed-rank test.

> If there are ties, I know I prefer wilcox.exact from the exactRankTests.
> 

(Not that much of an issue in larger sample sizes, I'd say. Even with binary data, the normal approximation works reasonably well under the usual assumptions of expected counts > 5, since the tie-adjustment for the variance is exact for the distribution of the ranks. The continuity correction doesn't quite work though. Anyways, wilcox.exact is of course a nice thing to have.)


> Alain
> 
> On 09-Aug-10 12:43, Capasia wrote:
>> This is my first post to the mailing list and I guess it's a pretty stupid
>> question but I can't figure it out. I hope this is the right forum for these
>> kind of questions.
>> 
>> Before I started using R I was using STATA to run a Wilcoxon signed-rank
>> test on two variables. See data below:
>> 
>> https://spreadsheets.google.com/pub?key=0ApodAA2GAEP_dDZkdzZHSFBqX1JHOWJBX1dMQUZCVkE&hl=en&output=html<%20%20https://spreadsheets.google.com/pub?key=0ApodAA2GAEP_dDZkdzZHSFBqX1JHOWJBX1dMQUZCVkE&hl=en&output=html>
>> 
>> STATA Output:
>> . signrank x=y
>> 
>> Wilcoxon signed-rank test
>> 
>>        sign |      obs   sum ranks    expected
>> -------------+---------------------------------
>>    positive |       41        3101      2330.5
>>    negative |       18        1560      2330.5
>>        zero |       49        1225        1225
>> -------------+---------------------------------
>>         all |      108        5886        5886
>> 
>> unadjusted variance   106438.50
>> adjustment for ties     -282.38
>> adjustment for zeros  -10106.25
>>                     ----------
>> adjusted variance      96049.88
>> 
>> Ho: transfer_2_a = transfer_2_b
>>             z =   2.486
>>    Prob>  |z| =   *0.0129*
>> 
>> When running a Wilcoxon signed-rank test
>> 
>> 
>>> wilcox.test(datablatt$x, datablatt$y)
>> Wilcoxon rank sum test with continuity correction
>> 
>> data:  datablatt$x and datablatt$y
>> W = 7059.5, p-value = *0.09197*
>> alternative hypothesis: true location shift is not equal to 0
>> 
>> As you can see the p Values are different (one with H0 rejection and the
>> other one not). I tested whether it could be that the STATA one isn't paired
>> but this doesn't seem to be the problem.
>> 
>> I'm dumbfound what could lead to such a difference. I couldn't find any
>> seetings I have missed but I somehow I guess I'm using the function in the
>> wrong way...
>> Any ideas?
>> Thanks a lot in advance!
>> 
>> 	[[alternative HTML version deleted]]
>> 
>> ______________________________________________
>> R-help at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
>> and provide commented, minimal, self-contained, reproducible code.
>> 
> 
> -- 
> Alain Guillet
> Statistician and Computer Scientist
> 
> SMCS - IMMAQ - Université catholique de Louvain
> Bureau c.316
> Voie du Roman Pays, 20
> B-1348 Louvain-la-Neuve
> Belgium
> 
> tel: +32 10 47 30 50
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

-- 
Peter Dalgaard
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com