[R-sig-ME] generalized mixed models with proportion data using lme4

[Elizabeth Crone]

Sacha:

The way you have implemented the model uses dots, not leaves, as
independent observations:

> y<-cbind(chewdots,leafdots)
>
> mod1<-glmer(y~pair + treatment + (1|treeid), binomial(link = logit), data =
> damage).

This accounts for your weirdly small P-values, and also possibly for
any problems you have with overdispersion.

If I were you, I would calculate the proportions from your dot data
(or trees as suggested this morning), then either use family =
gaussian(link = logit) or arcsin-sqrt transform and use
gausian/identity.

If you are really hard-core, you could input the dots as independent
observations, then include leaf as a random variable.

Good luck,
Elizabeth

-- 
****************************
Elizabeth E. Crone
Senior Ecologist, Harvard Forest
Harvard University
Petersham MA 01366
phone: (978)724-3302
FAX: (978)724-3595
email: ecrone at fas.harvard.edu