[R-sig-ME] glmmPQL: random effects

[Ben Bolker]

  This is a good question - surprised I haven't seen it before.

  The general answer to your question is that people don't generally
worry about REML vs ML for generalized mixed models:

http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html#reml-for-glmms

  The proximal answer is that glmmPQL is a hybrid between lme and glm.
When you specify a method= argument, glmmPQL tries to pass it to the glm
function, which is expecting a function name. (i.e., "don't do this, it
doesn't work")


On 18-01-25 09:09 AM, Cueva, Jorge wrote:
> Hello everyone,
> 
> I am working with glmmPQL because have data count (richness and number of individuals), in both cases have mean >5 and overdispersion. The literature says is necessary to distinct between ML (for random effects) and REML (fixed effects), but I got one error even with the nlme package active:
> 
> Error in ML(x = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  :
>   could not find function "ML"
> 
> I am using:
> 
> glmmPQL(Spp~1+Cattle+Equine+Mth.Prec,random = list(~1|Formation,~1|Cluster),data = VariabRLplot, family = "quasipoisson", method="ML")
> 
> If I not use  --method="ML"--   the model runs without warnings
> 
> The questions are:
> 
>   1.  Can I distinct between "ML" and "REML" using glmmPQL? Or I must use some function like lme or lmer and later pass to glmmPQL
>   2.  By other hand, with the random effects, "Cluster" is nested in "Formation", the syntax should be right, but I am not 100% sure.
> 
> Thanks so much
> 
> Jorge Cueva Ortiz
> Ing. Forestal
> ECU: 0993085161
> GER: 0049 1631327886
> 
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>