[R-sig-ME] REML estimates not given

[Douglas Bates]

On Wed, Sep 17, 2008 at 7:11 AM, Renwick, A. R. <a.renwick at abdn.ac.uk> wrote:
> I am running an lmer model using a poisson distribution and log link.  However, when I ask REML=TRUE I do not get a REML estimate only the ML estimate.

> For example:
> mix3<-lmer(RoundedOverlap~Sex+Margin+sess+Sex:Margin+(1|Site), family=poisson, data=both,REML=TRUE)
>  mix3 at deviance
>        ML       REML
> 159.158662         NA

> mix4<-lmer(RoundedOverlap~Sex+Margin+sess+Sex:Margin+(1|Site), family=poisson, data=both,REML=FALSE)
>  mix4 at deviance
>        ML       REML
> 159.158662         NA

> I am using both the latest R (R version 2.7.2 ) and lme4 package

> I was wondering if anybody knows why REML does not seem to be being used and if this is an issue with poisson distributed models in general.

The REML criterion is not defined for generalized linear mixed models.
 Well, at least I don't know of a definition that makes sense for
GLMMs.