How to test whether my data meet Permanova assumption

Hello everyone,

I was wondering if anyone could help me test Permanova assumptions.

From the original paper I realized that there is an assumption of homogeneity of multivariate dispersions. I would appreciate if first anyone could explain it to me in a plain language and help me test it.


Hi @ptalebic,

So, if I have two groups of samples, one from :chicken:s and one from :t_rex:, and I do a test on them, I hope that permanova is testing the hypothesis that d:chicken: vs :chicken: and d:t_rex: vs :t_rex: is smaller than d:chicken: vs :t_rex:. (Because then I can say that the difference in the microbiome between :chicken: and :t_rex: is larger than the differences in the microbiome within each group.)

…This works pretty well if variation/dispersion within each group is close to the same size, because then it becomes a pretty easy comparison. But, if d:chicken: vs :chicken: >>> d:t_rex: : vs :t_rex:, then when you test the distance, d:chicken: vs :t_rex: > d:t_rex: : vs :t_rex:, because the larger dispersion in the :chicken: group shifts that dispersion.

…Here’s the good news! You can apply the permdisp test (it’s an option under qiime diversity beta-group-significance) and that will test that assumption around dispersion within your groups. It also gives you a handy little boxplot that can let you look at the within and between group distances and get a sense for yourself.