[R-sig-ME] Error using MuMIn dredge with glmer.nb

[Ben Bolker]

Fiona Mae Caryl <fcaryl at ...> writes:

> 
> Dear R Users,
> 
> I receive an error message when using the dredge function in MuMIn
> on a glmer.nb glmerMod object.
> 
> "Error in dredge(cm.nb) :
>   call to 'global.model' contains '...' 
> arguments and cannot be updated: glmer(formula = CM ~ BUSH + FRAG +
> I(TREE^2) + (DWELL * TREE) + (1 | XSCAPE), data = ..2, 
> family = negative.binomial(theta = th))"
 
> Before I provide an example I thought it best to ask: is it possible
>  to dredge a glmer.nb glmerMod object?

  Well, at the moment (as you can see) no.  In principle we should
be able to make this possible, but glmer.nb has not been very thoroughly
tested, and involves another level of nesting of operations within
functions, so it doesn't particularly surprise me that this fails.
https://github.com/lme4/lme4/issues/176 highlights a closely related
problem.

> I have successfully dredged other glmerMod objects as well as
>  negative binomial mixed model objects from glmmadmb.
 
> Any pointers would be much appreciated as the fitted vs residual
> plots from glmer.nb are much less skewed than those from glmmadmb.

  Well, that probably has something to do with the way the residuals
are calculated (which also hasn't been carefully tested, neither in
glmmADMB nor in glmer.nb)!  It probably makes more sense to
double-check the way the residuals are calculated in both cases and
see why they don't agree.  Do your fits themselves agree (i.e., the
fixed-effect and variance-covariance estimates match closely) if you
try the same model in glmer.nb and glmmADMB?


> Best wishes,
> Fiona
> ---------------------------------------------
> Dr Fiona Mae Caryl
> Research Fellow
> Australian Research Centre for Urban Ecology
> School of Botany

 [snip]