# Kruskal-Wallis post hoc comparisons different results in different software used

I'm having some trouble analyzing my data because when I use StatSoft Statistica and run a Kruskal-Wallis, I get one set of values for multiple comparisons, but when the same analysis is run in XLSTAT (software my Master student uses), the p-values are different. In XLSTAT there are many options to choose for multiple pairwise comparisons but, in Statistica there are non, I just get the default output.

I'm wondering what I am missing here and why I get such different results?

These are my data:

G1  G2  G3  G4
26  11  12  20
36  9   7   35
22  28  43  8
14  21  30  14
23  11  9   14
13  17  55  22
10  9   9   27
16  6   6   41
25  10  59  18
5   10  15  14
11  7   10  9
8   7       44


In XLSTAT, these are the p-values:

p-values:

    G1      G2      G3      G4
G1  1       0,3672  1,0000  0,8063
G2  0,3672  1       0,7374  0,0946
G3  1,0000  0,7374  1       0,9268
G4  0,8063  0,0946  0,9268  1


In Statistica, these are the p-values:

    G1       G2         G3   G4
G1           0,879827   1    1
G2  0,879827            1    0,153257
G3  1        1               1
G4  1        0,153257   1


Anyone has any idea what the issue could be?

Thanks

• in Statistica there are non. How then did you get the pairwise p-values you are showing? – ttnphns Mar 11 '15 at 21:10
• What happens when you uncheck the "Multiple pairwise comparisons" option? – whuber Mar 11 '15 at 21:16
• Note that there have been proposed and developed many multiple comparisons post-hoc algorithms; some are found in many statistical software, some are specific to a particular software. – ttnphns Mar 11 '15 at 21:16
• @whuber ♦ I get a p value of 0.1110, the same as in Statistica. The pairwise comparison p values are different with using default options (and not present when I uncheck mpc) – praznin Mar 11 '15 at 21:19
• @whuber I have semi-basic knowledge of stats so I'm not quite sure to understand where the problem stems from: hence the question to learn something new. Basically, I want to know which software to use and which type of multiple pairwise comparisons, as I can have a difference between groups using some types, and no difference using other types. – praznin Mar 11 '15 at 21:26