[R-sig-ME] fixed effects clarification/confirmation (ZIP model)

[Ned Dochtermann]

Hi all,

I constructed a mixed-effect model about which I'm primarily interested in
the fixed effects and I wanted to make sure that I was interpreting the
output appropriately. Based on the MCMCglmm course notes I specified the
following model:

maleF.zip<-MCMCglmm(attack.male~trait-1+at.level(trait,1):trt*at.level(trait
,1):fphase*at.level(trait,1):day.1+
at.level(trait,1):temp.bar+at.level(trait,1):temp.dev,
rcov = ~idh(trait):units,
random=~us(1+tl.diff.f):subject,family="zipoisson",
data=Compiled,prior=mf.prior,
nitt=1080000,thin=480,burnin=120000,verbose=FALSE)

trt and fphase each have two levels (A&B for both)

The posterior mode estimates for the fixed effects from a shorter run are:
posterior.mode(maleF.zip$Sol)
traitattack.male       	traitzi_attack.male 	at.level(trait, 1):trtA 
11.74390983             	-2.52399057            	-0.23913921 
at.level(trait, 1):fphaseB    	at.level(trait, 1):day.1
at.level(trait, 1):temp.bar 
-0.14206666                    	0.02169066                     	-0.43889554 
at.level(trait, 1):temp.dev    	at.level(trait, 1):trtB:fphaseB
at.level(trait, 1):trtB:day.1 
-0.05774492                   	-0.24961100                    	-0.03413810 
at.level(trait, 1):fphaseB:day.1		at.level(trait,
1):trtB:fphaseB:day.1 
-0.02103302					0.11741711

My interpretation of the labeling of these effects is that
traitzi_attack.male is the inflation latent factor. Of more interest to the
questions I'm attempting to address, traitattack.male is the group mean for
trtB with at.level(trait,1):trtA ( henceforth a.l():trtA ) being the
difference between the group mean for trtA versus trtB. Picking an effect
for example/clarification, I'm interpreting a.l():trtB:fphaseB as the
difference from the global mean (traitattack.male) for trtB at fphaseB. Is
this the correct interpretation? Also, since besides the inflation factor
the effects are modeled as Poisson processes I've been assuming that the
standard transformation of the fixed effects (averaged over random effects).

The layout is a bit different from other mixed model outputs (i.e. trtA for
the first contrast rather than trtB being carried throughout) so I wanted to
double check. I tried extracting the sparse fixed effects design matrix for
clarification but it wasn't what I was expecting and didn't end up helping
me.

Thanks for any help,
Ned

--
Ned Dochtermann
Department of Biology
University of Nevada, Reno

ned.dochtermann at gmail.com
http://wolfweb.unr.edu/homepage/mpeacock/Ned.Dochtermann/
http://www.researcherid.com/rid/A-7146-2010
--