[R-sig-ME] issue with glmer() with the newest updated lme4 version
Ben Bolker
bbolker at gmail.com
Fri Feb 14 23:32:40 CET 2014
The best place for general or even possibly-general lme4 questions is
r-sig-mixed-models at r-project.org : I'm forwarding this answer there now
... You can go to
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models to subscribe.
On 14-02-14 01:42 PM, Colleen Arnison wrote:
> Hi Ben,
>
> I have fit a glmer () model to binomial data with 12 fixed (various
> landscape variables) and 2 random effects (elk ID and Area).
>
> I need to apply (ie predict) this model to new data (a subset of the
> original data I used to make it) and having some issues.
>
> I found on a few help forums to download the newest version of lme4 from
> your GigHub website (https://github.com/lme4/lme4/) and that made the
> predict () method for merMod objects work (GREAT!!!), however, when I
> then try to run the full script to create the original glmer model an
> error occurs (Warning message: In mer_finalize(ans) : false convergence
> (8)) which never occurred when I use the original CRAN version of lme4.
This seems quite strange -- that error message comes from OLD
(pre-1.0) versions of lme4. Somehow you're loading different versions
of lme4. Have you started from a clean R session? What are the results
of sessionInfo()? What are the results of
inst.pkg <- installed.packages()
grep("lme",rownames(inst.pkg),value=TRUE)
?
>
> Is there any other way around this?? Another way to predict a glmer
> model with new data, or make the function glmer work in the updated lme4
> on GigHub??
>
> Thanks so much for your help,
> Colleen
>
> PS sorry I am emailing you directly, I am new to R and discussion groups
> and did not know where I could post this question.
>
>
> model<- glmer(USE~Cover+Elevation+slope+
> prisk_s+BrnAge+FireSmt+NDVI+BurnCat+SummerUse+
> Type*DistLinear+DWater+
> (1|Area) + (1|ANIMAL_ID), data=data, family=binomial)
>
> testData <- data[sample(nrow(data))[1:10000], ]
>
> predict(model, newdata=testData)
More information about the R-sig-mixed-models
mailing list