[R-sig-eco] Specifying contrasts for an interaction in mvabund/manyglm

[Kyle Tyler]

Hi Diogo,

>From my joining of factors it appears that way, but this is just to identify which season and site combo the sample comes from for the post-hoc test to run on for the interaction. Season is otherwise orthogonal with Site.

I will look at gllvm again, but I am not sure it can perform hypothesis testing, or if it uses likelihood ratios so that I can then also determine taxa contributions to model deviance; my other reasons for using mvabund.

Cheers,
Kyle

Get Outlook for Android<https://aka.ms/ghei36>

________________________________
From: Diogo B. Provete <dbprovete using gmail.com>
Sent: Wednesday, September 23, 2020 1:38:38 AM
To: r-sig-ecology <r-sig-ecology using r-project.org>; Kyle Tyler <kyle.tyler using cdu.edu.au>
Subject: Specifying contrasts for an interaction in mvabund/manyglm

Hi Kyle,
from the bottom of your message it seems that what you have is a nested design, with Season nested within Site. If you specify your model with just the additive effect and the interaction it doesn't take that into account.
I'm not sure if mvabund allows that kind of planned comparisons you want to conduct. I'd try the newer package gllvm that is much more flexible and can handle random factors.

All the best
Diogo

	[[alternative HTML version deleted]]