Unbalanced PERMANOVA

good afternoon,
I’d like to ask how Qiime2 deals with unbalanced replicates in PERMANOVA. I’m asking as I have an unbalanced design but PERMANOVA seems to work anyway. Does Qiime2 replace the missing replicates with the corresponding group centroids?
many thanks

Hi @massi,

I’m not a statistician, but here is my understanding:

When computing a particular within-group sum-of-squares, the sum is divided by the number of samples in that group. You can see the implementation here. That doesn’t mean large group differences couldn’t impact the results still, but the formula itself expects the group sizes to be potentially different.

I did a bit more reading and it looks like homogeneous dispersion between the groups is the more important assumption of PERMANOVA. There’s a paper by Anderson & Walsh that tests PERMANOVA in different conditions, including unbalanced designs and heterogeneous within-group dispersions that will probably be of interest to you.

many thanks Evan!
very clear and useful answer

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.