Interpreting the results of Friedman test
Doerte wrote:
Hello, I have problems interpreting the results of a Friedman test. It seems to me that the p-value resulting from a Friedman test and with it the "significance" has to be interpreted in another way than the p-value resulting from e.g. ANOVA? Let me describe the problem with some detail: I'm testing a lot of different hypotheses in my observer study and only for some the premises for performing an ANOVA are fulfilled (tested with Shapiro Wilk and Bartlett). For the others I perform a Friedman test. To my surprise, the p-value of the Friedman test is < 0.05 for all my tested hypotheses. Thus, I tried to "compare" the results with the results of an ANOVA by performing both test methods (Friedman, ANOVA) to a given set of data. While ANOVA results in p = 0.34445 (--> no significant difference between the groups), the Friedman test results in p = 1.913e-06 (--> significant difference between the groups?). How can this be? Or am I doing something wrong? I have three measured values for each condition. For ANOVA I use them all, for the Friedman test I calculated the geometric mean of the three values, since this test does not work with replicated values. Is this a crude mistake?
Hi Doerte, There is a non-parametric repeated measures analog to ANOVA developed by Edgar Brunner available at: http://www.ams.med.uni-goettingen.de/de/sof/ld/makros.html Unfortunately, the test that you (and I) would like to have does not appear to have been translated to R code. I intend to contact Professor Brunner and try to complete this, or at least contribute to the effort, but have not had the time to do so. There are several other methods, notably that of Joe McKean and Tom Hettmansperger, but I don't have the URL for their code at hand. I'll try to forward this from work next week. Jim