convergence in GLMM (lme4 package)

Hi Ben,
Thank you very much for all your help, so far. Nonetheless, there is
still something not working. If you could have a look at the following
code, I would highly appreciate it. It's about how to count convergence
problems of function glmer within a simulation. I tried to adapt your
proposal to my models, but something's not working, probably to me lack
of knowledge of functions like tryCatch and withCallingHandlers.

I'm thinking about the topic and I don't know if it?s useful to obtain
the warnings/errors, because I should  look at each error and then I
should decide what I can do.

I only want to inform aboutpercentage of convergence of GLMM, that is
the converge is true or false.  I understand that a model is considered
as not convergent  either if the estimation process did not converge or
if the estimate or its standard error was not provided. For example, in
some cases the parameter can be estimated but the estimation process was
unable to provide a positive definite variance-covariance matrix of the
parameters (problems with Hessian), mainly due to the instability of the
model.

Thank you very much in advance for your suggestions.

Mart?



2012/10/23 Mart? Casals <mcasals at aspb.cat <mailto:mcasals at aspb.cat>>

    Thank you so much Ben!!

    Mart?


    2012/10/23 Ben Bolker <bbolker at gmail.com <mailto:bbolker at gmail.com>>

        Mart? Casals <mcasals at ...> writes:

        >
        > I?m working on the simulation results and I would like to know the
        > convergence of the GLMM model. For example, if I want to

        obtain the

        > converge in the glm case with R, I can extract:
        >
        > model1<-glm(................family=" ",data=)
        >
        > *model1$converged*
        >
        > [1] TRUE
        >
        > I don?t know if exist a similar function in the lme4 package

        (glmer ).

        >

          I think not.
          I believe the only way to do this (it may have been discussed
        before)
        is to use options(warning=2) to promote warnings to errors, then
        to use
        tryCatch() etc. to intercept the errors and store the
        information somewhere.

        http://comments.gmane.org/gmane.comp.lang.r.lme4.devel/8691
        https://stat.ethz.ch/pipermail/r-help/2012-February/302767.html

          I've added a bit on this to http://glmm.wikidot.com/faq, since
        it's
        not the first time it has come up ...

         Ben Bolker