warnings ranef and design of a binomial model
---------------------------------------- From: Douglas Bates <bates at stat.wisc.edu> Sent: Wed Sep 10 15:43:38 CEST 2008 To: Robert ESPESSER <Robert.Espesser at lpl-aix.fr> Subject: Re: [R-sig-ME] warnings ranef and design of a binomial model
I think we would need at least the output from traceback() to see where the error is occurring. Better yet would be to have access to the data so we could try out the model fits. Sometimes these errors show up deep inside another computation. Does
ranef(mod.glmer)
produce a result? I'm trying to decide if the problem is in the postVar part or in the ranef part.
The warning messages are output by dotplot(ranef() ). ranef(xxx ,postVar=T) by itself does not output any warning messages. The models: modG.glmer <- glmer( vd ~ dlms*expe +(dlms|subject) ,family=binomial, ....) modI.glmer <- glmer( vd ~ dlms*expe +(1|subject) ,family=binomial, ....) modS.glmer <- glmer( vd ~ dlms*expe +(0+dlms|subject) ,family=binomial, ....) have their dotplot(ranef()) OK. The problem is only with the model: modIS.glmer <- glmer( vd ~ dlms*expe +(1|subject) +(0+dlms|subject) ,family=binomial, ....) ranef(modIS.glmer)-> rr.glmer
dotplot( rr.glmer)
$subject Warning messages: 1: In min(x) : aucun argument trouv? pour min ; Inf est renvoy? 2: In max(x) : aucun argument pour max ; -Inf est renvoy? traceback() returns nothing :
traceback()
2: bar(2) 1: foo(2) (because I have tested it with the example in ?traceback) In addition, how can I extract the postVar component in a ranef object ? Dr Bates, in case you need the data , do I have attach them in a private mail to you ? here is the complete sessionInfo(): R version 2.7.2 (2008-08-25) i386-pc-mingw32 locale: LC_COLLATE=French_France.1252;LC_CTYPE=French_France.1252;LC_MONETARY=French_France.1252;LC_NUMERIC=C;LC_TIME=French_France.1252 attached base packages: [1] stats graphics grDevices datasets utils methods base other attached packages: [1] lme4_0.999375-26 Matrix_0.999375-13 lattice_0.17-13 loaded via a namespace (and not attached): [1] grid_2.7.2 nlme_3.1-89
On Wed, Sep 10, 2008 at 5:29 AM, Robert ESPESSER <Robert.Espesser at lpl-aix.fr> wrote:
Dear R users,
I have a question about model design.
One group of 20 subjects run the A experiment, with 17 binary responses by subject.
An other group of 20 subjects run the 2 experiments B and C, with 20 binary responses by subject,
for both of the 2 experiment.
I want to study the effect of the factor "expe"
I tried this covariance model:
mod.glmer <- glmer( vd ~ dlms*expe +(1|subject) +(0+dlms|subject) ,
family=binomial, ....)
where:
dlms is a numeric factor (-2,-1,0, 1,2)
expe is a 3 levels factor , coding the experiment (A or B or C)
vd is the success/fail matrix.
The model with a random slope is clearly better than an intercept-only model.
the results look plausible, ranef(mod.glmer, postVar=T) was OK for the
Intercept, but failed
to compute the postvar of the random slope , with these warnings:
1: In min(x): no argument found for min; Inf is returned
It's the same for max.
I'm suspecting that my model is perhaps badly designed , concerning the
unbalanced repartition of the subjects across the experiments. Is it correct ?
Thanks for any comments or suggestions.
####
sessionInfo() is:
R version 2.7.2
lme4_0.999375-26
Matrix_0.999375-13
Robert Espesser
Laboratoire Parole et Langage UMR 6057, CNRS
29 Av. Robert Schuman 13621 AIX (FRANCE)
Tel: +33 (0)4 42 95 36 26 Fax: +33 (0)4 42 95 37 88
http://www.lpl-aix.fr
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models