1) What variable(s) and how did you check for normality? Why is it not normal? Maybe there is no need to switch to non-parametric tests.

2) There are post hoc options available in SPSS for Friedman's test.

Hello Babette,

Yes, stay away from the Kruskal-Wallis in this instance.

The default go-to for following up a Friedman test with pair-wise tests would be the Wilcoxon signed rank sum test. Dunn's method is typically an option when one of your treatment conditions is the base/control condition to which the others are to be compared. If that's not the aim of your "focused tests," you'd be better served by implementing a follow-up procedure such as the Benjamini-Hochberg (https://www.statisticshowto.com/benjamini-hochberg-procedure/) which only requires that you rank all the specific tests you run based on the ordinary p-value (which Wilcoxon test will yield, in most statistics software), then compare them against a sliding significance scale.

As Rainer Duesing 's response indicates, newer versions of spss offer pairwise follow-ups to the Friedman anyhow.

Good luck with your work.

Hello, Dear @David Morse

As far as I know, the number of samples between groups for Hocberg's test should be quite different from one.

Rainer Duesing

Maybe it's usefull that I give more background information first. I was curious whether there is a difference in effect between different type of touches on itch.

I wanted to know the difference between affective vs. non affective touch and the difference between impersonal vs. interpersonal touch in different time points. The timepoints were measured to get more information in reducing pattern over time. So I wanted to define three factors in the RM ANOVA: 1. touch (affective vs. non affective), 2. personal (impersonal vs. interpersonal), 3. time (2min, 4min, 6min, 8min, 10min).

1) I analysed the raw scores of the itch sensations in different conditions for normality (so the dependent variable itch). The Shapiro Wilk test indicated that a few itch scores (5) were not normally distributed, so I tried to transform the data with a log transformation and square root transformation, since the data was positivily skewed. Unfortunately this did not work. That is why I wanted to test the data with a Friedman ANOVA. Only I am not really able to include different time points now.

Now I read in other forums that I was supposed to analyse the residuals instead of the raw scores for normality with a RM ANOVA. Is that true? I cannot find a lot of information about it in articles, so I am not sure if I did it wrong.

Kind Regards,