I'm trying to decide which is more valid: a Friedman test or multiple Mann-Whitney-Wilcoxon tests with a Bonferroni correction. The latter produces significant results whereas the former does not. Which one is the correct one to use for non-normal data looking for the effects of one variable in the face of another? Is there another test I should be using?
More about my data: They do not approximate a normal distribution even with transformation. They are repeated measures (ie 3 treatments sampled at 2 sites many times). I want to figure out how to look for treatment effect removing any site effects. Basically, a nonparametric way of doing a 2-way ANOVA (treatment*site).
Thanks for any help you can provide!
Advertise on Talk Stats