R help - Feb 2010 - Poisson and neg. bin. regression with random effects

Hi there,

I have relative abundance data for 13 mammal species that I collected at various
sites that ranged in road density. I'm trying to determine the effect of
road density on animal abundance across body sizes. For most species, I have
data that was collected in one year but for a few species I have two years of
complete data, and would like to use both. Since I have count data, I'm
trying to use Poisson regression and incorporate random effects for body size
using lmer. I would like to run 4 models and use AIC to select the best model
amongst:  a) fixed effect for road density, 2)  randoms intercepts for body
size, 3)  random slopes for body size, and 4) random slopes and intercepts for
body size. Since I have two years of data for some species however, I would like
to treat each year separately so I believe I need to account for this by nesting
year (as a factor) within body size (also a factor). My first question is
whether or not the code below is correct for the models outlined above?

data= (mixed)

fBodySize <- factor(mixed$BodySize) 
fYear<-factor(mixed$Year)


groupedData(ABUNDANCE~RoadDensity | fBodySize/fYear, data = mixed) 

#random intercept
m1<-lmer(ABUNDANCE~RoadDensity +(1|fBodySize/fYear), family=poisson(log),
data = mixed)


# random slope

m2<-lmer(ABUNDANCE~RoadDensity +(-1 + RoadDensity |fBodySize/fYear),
family=poisson(log), data = mixed)


#random slope and intercept
m3<-lmer(ABUNDANCE~RoadDensity +(1+RoadDensity |fBodySize/fYear),
family=poisson(log), data = mixed)

And secondly, I do have slightly overdispersed data, but since I would like to
compare models, I believe using a quasi-Poisson model won't work since it
does not provide a likelihood function, therefore I cannot do a model comparison
using AIC?? However, a negative binomial may be appropriate but I'm having
trouble figuring out the code for the same models above but using the glmmADMB
library (which I believe can handle neg. bin family with random effects??). Can
help with the code for the same models above but using neg. bin regression?

Thanks very much,
Bob


 		 	   		  
_________________________________________________________________


	[[alternative HTML version deleted]]