RM-ANOVA vs. Nonparametric Tests
Hi all,
I have a design with a pretest and two posttests testing multiple questionnaires and experiments.
In this case, I was interested in the outcome of perceived stress scores (PSS). Please note that group sizes are unequal (35, 33), as well as there is some missing data along the three timepoints.
So I ran an RM ANOVA and got a significant measurement x group interaction (yeey!), but looking at the descriptives, Shapiro-Wilk tests for all three timepoints were p < .05. I was not sure whether I should ditch the ANOVA and run a Friedman's test, but I tried anyway. When I did that, I got an error message stating "Specific ANOVA design is not balanced".
How to continue? Any alternatives? :)
Many thanks for your answers!
Cheers,
Bence
Comments
Hello Bence,
Good question. Apparently the Friedman test requires the ANOVA to be balanced?! Have you checked out any alternatives yourself?
Cheers,
E.J.
Hi @szaszkob ,
Unfortunately the design requirements for the Friedman test are pretty strict (it is the same in other software, like R). It also does not allow multi-way designs, and can only assess main effects.
You could consider an ANOVA on the rank-transformed data, although this is a very rough method (see Wikipedia and this article) and I think would be more rough/error-prone than simply continuing with the RM ANOVA and the slight deviation from normality in your data.
Kind regards
Johnny