[R-sig-ME] glmmadmb function
bbolker at gmail.com
Mon Feb 2 17:15:56 CET 2015
-----BEGIN PGP SIGNED MESSAGE-----
In general, fitting models with link functions that allow
non-feasible results is difficult. In particular, the Gamma likelihood
is only finite for non-negative values of the x variable (only
positive values for some parameter ranges)). So any inverse-link
function with a range that includes negative values is going to be
problematic; this includes the identity link as well as the default
(canonical) inverse. It's interesting that it works for the default
in this case -- there are often problems, a log link is often much
Specifying starting values might help.
Depending on your reasons for fitting a Gamma model, I often
recommend just log-transforming the response and fitting a linear
mixed model instead (i.e. that generally gives results very close to a
Gamma with a log link).
Changing the directory doesn't make the model run any better, it
just dumps the output in a way that may make it easier for AD Model
Builder experts to figure out what's going on.
On 15-01-28 04:57 AM, Simona Picardi wrote:
> Dear members,
> I'm using the glmmADMB package version 0.8.0 on R version 3.0.2
> (2013-09-25) to run the following mixed effect models with gamma
> mod1 <- glmmadmb(dist~test+(1|id),data=dataHS1,family="gamma")
> mod2 <-
In the first case, using gamma with the default link, everything
> works fine.
> As I try to specify link="identity", I get the following error
> Error in glmmadmb(formula = dist ~ test + (1 | id), data =
> dataHS1, family = "gamma", : The function maximizer failed
> (couldn't find STD file) Troubleshooting steps include (1) run with
> 'save.dir' set and inspect output files; (2) change run parameters:
> see '?admbControl' Inoltre: Warning message: running command
> './glmmadmb -maxfn 500 -maxph 5 -noinit -shess' had status 21
> Incompatible bounds in dvar_vector& dvar_vector::operator = (const
> dvar_vector& t)
> I already tried to set the directory, but I keep getting the same
> error message.
> Any help would be hugely appreciated.
> Thanks in advance, Simona Picardi [[alternative HTML version
> R-sig-mixed-models at r-project.org mailing list
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
-----END PGP SIGNATURE-----
More information about the R-sig-mixed-models