[R] glmmPQL

Ben Bolker bolker at ufl.edu
Fri Oct 10 20:29:09 CEST 2008

Jean-Baptiste Ferdy <jean-baptiste.ferdy <at> univ-montp2.fr> writes:

> Dear all,
> I am experiencing problems with glmmmPQL. I am trying to analyze
> binomial data with some spatial autocorrelation. Here is my code and
> some of the outputs
> > m.1 <- glmmPQL(fixed=cbind(nb_pc_02,I(nb_expr_02-nb_pc_02))~(PSDC99
> +I(log(RFUCQ202))+PCCHOM99+sd_RFUCQ202+PCFEMCHOM)^2+langue,
> +              random= ~ 1 | bidon,
> +              correlation = corExp(form = ~ Longitude
> +Latitude),             
> +              family=quasibinomial,
> +              na.action="na.exclude",
> +              data=d.glmm)
> iteration 1
> Erreur dans model.frame.default(formula = ~Longitude + Latitude + bidon 
>   les longueurs des variables diffèrent (trouvé pour 'bidon')
> Sorry about the french here... It is just saying that variables have
> different lengths. Which is obviously untrue, as they all are columns of
> the same data frame.
> The search() shows that I have attached the data.frame, as it is told in
> some FAQs.

  Hmm, that would seem unnecessary as long as you have the data=
> If somebody has an idea on how to sort this issue out, I would
> appreciate !

  I don't know. My only guess is that it _might_ have something
to do with your na.action?  What happens if you try the model
on na.omit(d.glmm) ?   You can try traceback() to see exactly
where the crash happens (although that may not be directly useful).
More usefully, can you provide a reproducible example (e.g.
post the data somewhere, or cut it down to a subset that still
causes the problem)?
> Another question: is the option correlation active in the latest version
> of lmer ? If it is, it might be an option to consider instead of
> glmmPQL.

   No, and not on Doug Bates' high-priority list.

  You might want to follow up to r-sig-mixed-models at r-project.org ...

   Ben Bolker

More information about the R-help mailing list