[R] Type III SS in generalized linear mixed models

[Alfonso M Sanchez-Lafuente]

Good morning everybody,

I would like to get some advice on the following issue:

I am analyzing a dataset obtained from a study on plant visitation 
likelihood related to experimental herbivory treatments and composition 
and abundance of the pollinator assamblage.

The experimental set up was as follows:

In a number of plants (N=73), I replicated a herbivory treatment with 3 
levels and a control. I wanted to know if the experimental treatment is 
responsible for differences in flower visitation likelihood, in 
pollinator attraction and in the interaction bettween these variables in 
3 different populations. All treatments were replicated within plants.

Summarizing:

Treatment effect: 3 levels + 1 control
Pollinator effect: 4 levels (i.e., 4 pollinator groups)
Site effect: 3 levels (i.e, 3 populations)

The response variable is binomial: Flower visited vs. Flower not visited 
(data obtained from 852 individual censuses).

The model selected was:

glmmPQL(Visited ~ Treatm*Pollinator*Site, random=~1|PlantID, 
family=binomial)

I want to test the effect of each factor independently of the other 
factors first. I guess this is Type III SS.
In other words, I do not want to test the effect of Pollinator once I 
have accounted for the effect of Treatment, because the relative 
abundance of different pollinators in a given site and season is not 
dependent on the treatment applied. However, I can only obtain Tipe I SS 
with this model, and the Anova function in package car does not accept 
this type of model with random grouping factors... Any suggestions to 
analyze this type of models to get Type III SS in R ?

-- 

----------------------------------------------
Alfonso M. Sanchez-Lafuente
Departamento de Biologia Vegetal y Ecologia
Facultad de Biologia
Universidad de Sevilla
Avd. Reina Mercedes 9
E-41012, Sevilla, Spain
email: alfonso at slafuente.es / slafuente at us.es