[R-sig-ME] Bug?: var estimates on exp scale for gaussian(link=log)

[Douglas Bates]

The development version of the lme4 package has some of the families
(binomial or quasibinomial with logit or probit link, poisson or
quasipoisson with log link) coded in C for efficiency.  The general
form will need to allow for evaluation of the family from R functions
and I haven't incorporated that yet.

On Jan 7, 2008 4:23 PM, Gregor Gorjanc <gregor.gorjanc at bfro.uni-lj.si> wrote:
> Gorjanc Gregor <Gregor.Gorjanc at ...> writes:
> >
> > Hi!
> >
> > I would like to fit log-normal fixed model with lmer. Fixed effects estimates
> > are returned on log scale, while variance components are returned on exp scale.
> > Why is this the case?
>
> I also tried with newest version from R-Forge (svn checkout), but I can not fit
> the log-normal mixed model at all. Using the same code as in previous mail I get
>
> > tmp <- lmer(y ~ ai + bi + (1 | ci),
> +             data=pod,
> +             family=gaussian(link="log"))
> Error in mer_finalize(ans, verbose) :
>   General form of glmer_linkinv not yet written
>
> Gregor
>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>