Post Hoc analysis for non-parametric analysis in Minitab: HELP

#1
Greetings

I wish to be able to use post hoc analysis on my non-parametric analysis on my data. I am currently using Krustal Wallis (as I have 5 sets of data that I wish to compare). As this will only give me a p-value and a z-value which informs me if the data is significantly different or not and the general ball park that the data sets are inn respectively. However, I wish to know which of my datasets are significantly different or not but theres no options to do so in minitab. Have I overlooked something?

PS: Basically what I want something like the ANOVA - Fisher pairwise comparisons.

Please advice.

Marius
 

Miner

TS Contributor
#2
Minitab's nonparametric tests do not currently provide post hoc tests. Your only apparent options are to see whether you can use a 1-way ANOVA, or subset the data, perform the pairwise tests on the sub-setted data and adjust your p-values manually.
 

hlsmith

Not a robit
#3
I typically use pairwise Wilcoxon Rank Sum tests after a significant omnibus test such as the Kruskal Wallis. Not a Minitab user, so no help there.
 

Miner

TS Contributor
#4
I typically use pairwise Wilcoxon Rank Sum tests after a significant omnibus test such as the Kruskal Wallis. Not a Minitab user, so no help there.
If they subset the data, to allow pairwise comparisons, Minitab has a 2-sample option for Mann-Whitney. Unfortunately, their Wilcoxon is a 1-sample test, so unless you have paired data and perform the test on the differences, you are out of luck.
 

rogojel

TS Contributor
#5
hi,:
maybe you can use the aligned rank transform . I do not know for sure, but I suppose the transorm could be implemented manually in Minitab and the resulting transformed data analysed with normal ANOVA.

regards
 
#7
Thanks everyone for replies, really appreciate it! Yeah it seems to be quite a big limitation with minitab so I just got a SPSS trial instead and used the built inn post hoc analysis "Pairwise comparisons" in the KW test in the "new" independent sample option whenver the p-value < 0.05. That makes sense right? It provides this pairwise comparison where each combination of the various groups has an assosiated p-value which quickly showes me which group are statistically differnet.
 
#8
Ah cheers gianmarco, thats pretty much spot on. I googled for ages and didnt find anything! I did turn to SPSS for this but I will definitly keep this in mind for future reference.