Non parametric N-way analysis of variance

Hi everyone! I had to perform a multifactorial test analogous to N-way ANOVA ("anovan" function) but suitable for non-normal distributed data. I know that there is Kruskal-Wallis test, but it seems that only the one-way option is available in Matlab. Do you know if there is an alternative for the N-way in Matlab?

Answer 1

[Jeff Miller]

No, there isn't one. This isn't actually a Matlab limitation, but rather a statistical one, because there simply isn't a clear nonparametric analog of ANOVA. One reason is that interactions in ANOVA reflect differences in differences, but these aren't at all meaningful with nonparametric data.

One thing people often do is to convert all of the original scores to ranks and then do ANOVA on the ranks. That can work pretty well if the problem with the original data is that there are a few really extreme scores that would unduly influence their condition means.

Probably more commonly, people just ignore the normality requirement and rely on the fact that ANOVA is pretty robust to violations of this assumption.

[Francesca Danielli]

Thanks for the answer, really appreciated!