[R] Linear mixed models with custom link functions in R

[Ben Bolker]

Daniel Barton <daniel.barton <at> umontana.edu> writes:

> 
> Hello All,
>     I am looking for an R library/function that allows the specification 
> of a custom link function in a linear mixed model.

  You might want to try on the r-sig-mixed-models at r-project.org list.

  I would probably call this a *G*LMM because of the link function,
even if you are assuming normally distributed errors.

  I *think* (although haven't tried it) that if you use glmmPQL
from the MASS package, its 'family' argument is identical to
that in glm() [and in fact gets passed through to the glm() function],
so you should be able to adapt what you have already done
to glmmPQL.  The only down side (if it works) is that glmmPQL
(as the name suggests) uses penalized quasi-likelihood, which
may under some circumstances give biased estimates of the
random effects (see Breslow 2003) -- but it may not be important
in your application.

  If worse comes to worst, you may end up coding your own in
WinBUGS or AD Model Builder ...