[R-sig-ME] Choosing nlme or lme4?

[Ben Bolker]

Filipe Carvalho <filipescpcarvalho at ...> writes:

> I'm analysing the selectivity of resting site use by forest
> carnivores through mixed modelling techniques and I wonder which
> will be the best r package to deal with several aspects
> simultaneously:

> binomial variable response;
> possible spatial and/or temporal correlation;

> I have tried nlme (lme function) and lme4 (lmer function) packages,
> however I realize that the results were different concerning
> regression coefficients estimates and p-values!
 
> In nlme package, despite I can add easily a variance structure
> and/or temporal and spatial correlations structures, the choice of
> family = binomial not allowed. On the other hand, with lme4 I can
> choose the binomial family, but no structures can be added!
 
> Am I wrong with these statements? Zuur et al. 2009 always used lme4
> (Mass or glmmML) with binomial data but never nlme!

 Unfortunately, you are correct.

 See

http://glmm.wikidot.com/faq#spatiotemporal

for some of your options (MASS::glmmPQL, AD Model Builder,
INLA, ...).  glmmPQL is one of the suggestions of Dormann et al
(Ecography 2007) I wish this were easier.