Hello everybody,
I have count data and with these data, I would like to build a mixed
model by using the function glmer(). In a first time, I calculated the c-hat of
a simple model with glm() to verify overdispersion and I found a c-hat = 18. I
also verified overdispersion in the mixed model by checking the residuals of
random effects via the function glmmPQL and I found a c-hat = 15. Thus, the
poisson distribution does not seem suitable for my model. The problem is that
there is no negative binomial distribution in glmer(). Another advantage with
the package lme4 is the possibility to calculate easily conditional and
marginal R-square. The R-square is a way for me to do model selection. I cannot
use AIC because my models are not nested and come from different datasets. I
also
verify the package glmm.ADMB but I found that it is not much used.
What is the best solution to build a mixed model with overdispersion
under R ? Is it correct to use normal distribution with count data because I
have a high mean ( = 47) and a high variance ( = 1188) for my response variable
?
Thank you very much for your help
Have a good day
Marine
[[alternative HTML version deleted]]
