Created attachment 337356 [details]
Example of wrong Wilcoxon rank test

The Wilcoxon rank test (for paired samples) delivers wrong values. The S+ sum is correct, but the S- sum (which is in most cases the lowest, thus the used) is often wrong due to the fact that the total number of cases N is not correct (should be the number of data minus the number of eliminated ties).

Checking same data with gnumeric and R the difference is huge.

In the attachement a .gnumeric with some test data. Correct solutions are: 

S+=87.5
S-=17.5
N=14 (and not 12 as stated in gnumeric). Look at the attachment

Output of R for
wilcox.test(soliani$DataA,soliani$DataB,paired=TRUE,corrected=FALSE):

V = 17.5, p-value = 0.03012

Regards

The example is taken from cap. 8 of the on-line manual of Prof. Soliani pag. 26 and f.

http://www.dsa.unipr.it/soliani/capu8.pdf