Nonparametric ANOVA w/ Between-Subjects Variable: Advice Appreciated

[Hello Kitty]

I'm currently looking to use a nonparametric alternative to the parametric repeated-measures with a between-subjects variable. I'm aiming to investigate three sets of subscores of a test taken by three groups of participants. I was initially using the parametric repeated-measures ANOVA using the within-subject variable tested at 3 levels, together with the between-subject variable Group tested at 3 levels. However, since it violated some of the assumptions to be met, I started considering using a non-parametric alternative.

To this end, I ranked each set of the subscores separately, and performed Kruskal-Wallis tests using the Tests for Several Independnt Samples in SPSS. Is it the right way of going about this? Or should I rank scores for each participant and perform Friedman's ANOVA? If I go with the latter, since this is a non-parametric equivalent of a one-way repeated-measures ANOVA, how can I investigate group effects? Should I carry out multiple independent-samples tests afterwards?

I hope my questions make sense and are not too silly.
Any suggestion or advice would be deeply appreciated.

Thanks.

 

[ych22]

I am interested in experiment design and statistics although I am pretty new. Hope to learn something from some discussion and forgive me if I sound too dumb.

Am I right to say that the main criterion for choosing between the Kruskal Wallis test or the Friedman test is whether the number of experiment units within each group/treatment are the same. So you should use the Kruskal Wallis test if you have unequal number of experiment units within each group, otherwise you may choose to use the Friedman test.

I guess the way you have to go about with investigating group effects depend on the differences between the groups? Are the groups different in a measurable (speed, IQ, weight) or non-measurable sort of way(color, gender, race)?

If your nonparametric anova suggests that group/treatment has an effect on the results, I suppose you have to use a nonparametric regression model also, to investigate group effects.