I found out that there is a handy option in the Kruskal-Wallis test, to do pairwise comparison between groups, in SPSS version 22. Anyone knows what statistics does it use?
In Kruskal-Wallis Test as one of the Nonparametric test (no assumption about distribution); Assumes three or more independent groups, Dependent variable is at least ordinal; Reports the median and interquartile range of each group. The significant result will be reported with Chi Square statistics.
However, Kruskal-Wallis is not for pairwise test, it is the parallel test for ANOVA in parametric statistics.
The non parametric paired test is the Wilcoxon Matched-Pairs Signed Rank Test
In Kruskal-Wallis Test as one of the Nonparametric test (no assumption about distribution); Assumes three or more independent groups, Dependent variable is at least ordinal; Reports the median and interquartile range of each group. The significant result will be reported with Chi Square statistics.
However, Kruskal-Wallis is not for pairwise test, it is the parallel test for ANOVA in parametric statistics.
The non parametric paired test is the Wilcoxon Matched-Pairs Signed Rank Test
Wilcoxon Rank Test, Its present in SPSS 16/ graph pad prism in non parametric analysis
The Wilcoxon Matched-Pairs Signed Rank Test present in SPSS 21, and mostly 22 under the following order:
Analyze,--nonparametric test---legacy dialogue---2 related samples
The previous answers are correct that Wilcoxon signed rank test is the way SPSS does paired nonparametric comparisons. If you are looking for post-hoc approach to multiple comparisons (similar to how we might use Bonferroni or Scheffe after ANOVA), you can still use Wilcoxon signed rank test if the Kruskal-Wallis test is statistically significant. Just keep in mind that SPSS will not automate this as it would with multiple comparisons tests after ANOVA. You'll need to select the appropriate alpha level and run multiple tests to build your comparisons.
Different from other versions of SPSS, ın the 22nd version SPSS can perform post-hoc tests in both Kruskal-Wallis and Friedman test.
To get the pairwise comparisons you can use the following path:
Analyze--> Nonparametric tests-->Related samples
In the output double click on the table, another window will be opened. In this window select the "pairwise comparisons" under the selection pop-up menu from "view"
So the original question remains.. does anyone knows what is the test used in KW post-hoc (SPSS 22version, pairwise) to correct for multiple comparison ?
Thanks in advance
In tech-doc there is no information about it, but You can check it with http://astatsa.com/KruskalWallisTest/ calcullator - it's for 100% Dunn test with Bonferroni correction.
Edith, i'm sorry, but that's not true. Wilcoxon is not recommended due to lack of number-of-groups correction. It's the same situation like using t-test to pairwise comparision in ANOVA. You can do it, but You'll get wrong answers. :)
Recomended is pairwise test with f.e. Bonferroni correction.
Yes, you are right is a wilcoxon test with bonferroni correction for p value adjustement
Edith, You can perform wilcoxon test with bonferroni correction, but it's not recommended (see other threads). BUT question here in thread was about SPSS module, and answer is: this module performs dunn-bonferroni test. EOT. ;)
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis