[R] Why warnings using lmer-model with family=binomial

[Douglas Bates]

On 4/11/07, Åsa Granberg <asa.granberg at emg.umu.se> wrote:
> Hi all!
>
> My question is why, and what I can do about that
> I sometimes, but not always, get warning-messages like
>
> nlminb returned message singular convergence (7)
>   in: LMEopt(x = mer, value = cv)
>
> or
>
> IRLS iterations for PQL did not converge
>
> when trying to fit a model looking like this:
> lmer<-(cbind(Diseased,Healthy)~Fungus+(1|Family)+(1|Fungus:Family),
> family="binomial") to four similar data sets?
>
> All four data sets consists of four columns;
> Fungus (1 or 2), Family (1-30), Diseased and Healthy.
> An example:
>
> Fungus  Family          Diseased        Healthy
> 1               1               3               7
> 2               1               1               5
> 1               2               2               16
> 2               2               2               14
>
> ...and so on up to totally 30 families.
>
> One of the data sets works perfectly fine and
> gives no warnings, but for the others i get
> different amount of warning messages. Between the
> data sets it is only the number of diseased anf
> healthy for each family and fungus that differ.
> Does anyone have any idea why this happens and
> how I can get around it (I'm mainly interested in the interaction term...)?

Do you have the latest released version of the lme4 package (version
0.9975-13) installed? You can check with

library(lme4)
sessionInfo()

My memory may be failing me but I thought that I changed the default
estimation method for generalized linear mixed models to minimization
of the Laplace approximation to the deviance before that release, in
which case no PQL iterations are being performed.

In any case it is recommended that you include the output from
sessionInfo() in messages dealing with specific behavior of R or R
packages.