[R-sig-ME] false convergence (8) in glmer

Douglas Bates bates at stat.wisc.edu
Thu Sep 25 15:02:04 CEST 2008


On Thu, Sep 25, 2008 at 3:36 AM, Stijn Ruiter <s.ruiter at maw.ru.nl> wrote:
> Hi,
> The suggested (though horrible) hack to provide a greater number of
> iterations (see digest-list: /Thu Jul 10 18:49:15 CEST 2008/)
>
> .Call("mer_optimize", <failed_lmer_fit>, FALSE, PACKAGE = "lme4")
>
> does not seem to work anymore. In previous versions it did.

I imagine this is a property of the model and data set and not a
property of the call.

> Any suggestions on how to continue iterating, because I have an non-converged model after glmer ends its process.
> It results in the following warning message:
> In mer_finalize(ans) : false convergence (8)

That's a warning that the parameter estimates are suspect but it's not
an error. I believe that a fitted model would have been returned.

I suggest you turn on the verbose option in the optimizer and examine
the progress of the iterations.  In the original call to glmer you add
verbose = TRUE.  If re-calling "mer_optimize" as above then change the
FALSE to TRUE.




More information about the R-sig-mixed-models mailing list