[R-sig-ME] Unclear output from MCMCglmm with categorical predictors

roim@or m@ili@g off post@t@u@@c@il roim@or m@ili@g off post@t@u@@c@il
Wed Nov 7 14:51:40 CET 2018

Dear list members,

I have a model with categorical response and categorical + continuous  

My model has two categorical predictors: "diet" (3 levels) and  
"habitat" (6 levels):

THRE1 <- MCMCglmm(Activity ~ -1 + Habitat + Diet + log(Mass) + Max.Temp,
                      prior = list(R = list(V = 1, fix = 1)),
                      ginverse = list(Binomial=INphylo$Ainv),
                      family = "threshold",
                      data = Tdata)

If I understand correctly, in this configuration the algorithm  
shouldn't return estimated values for the effect of each level of a  
categorical predictor, instead, it returns a contrast between that  
level and another level which was arbitrarily chosen as the base  
level. Each species (data point) has a value for each of these traits,  
so I would expect them to be estimated independently, meaning that one  
level of each predictor should be the 'baseline' and absorbed into the  
global intercept. In that case I expect 2 contrasts to be returned for  
diet categories and 5 contrasts for habitat.

However, I get 2 estimates (presumably contrasts) for diet categories,  
and 6 for habitat categories, i.e., no habitat category was designated  
as baseline, which makes me question whether the estimates are  
contrasts or actual effect sizes.

My questions:
- Is the algorithm pooling all the predictor categories as if they  
were a single trait with 8 levels?
- If the habitat effect estimates are contrasts - what are they compared to?
- If they are effect sizes - what did I do to not get the contrasts as  
I expected?

Any help would be much appreciated!
Roi Maor
PhD candidate
School of Zoology, Tel Aviv University
Centre for Biodiversity and Environment Research, UCL

More information about the R-sig-mixed-models mailing list