[R-sig-ME] prediction with glmer

Phillip Alday phillip.alday at mpi.nl
Thu Feb 22 12:32:14 CET 2018


In terms of the random effects: it depends on the other arguments to
predict() (in particular re.form) as to whether they are marginalized
out or not (i.e. whether it's a pure fixed-effects prediction).


My understanding is:
If you only the same levels of your grouping variable occur in the new
data as occurred in the old data (i.e. the data used to fit the model),
then the conditional modes / BLUPs (BLUP is a bit of a misnomer in the
GLMM case, and conditional mode is a fairly Bayesian perspective in a
frequentist model) estimated for each level of the grouping variable
(study_no in your case) are used.
With new levels, predictions are made using the variance estimates and
not the particular offsets / conditional modes / BLUPs.

However, in your particular case, it's even easier: predict simply
returns the fitted values.

Phillip


On 22/02/18 11:43, Schlattmann, Peter wrote:
> Dear all,
> 
> Apologies for a potential second post, I am not sure my first post came through.
> 
> 
> 
> I am fitting a generalized linear mixed model with lme4 using glmer
> 
>  with binomial errors and logit link. I am using the “predict” function
> 
>  to obtain predicted values for the current model and data set.
> 
> 
> 
>  Here is some sample code
> 
> 
> 
>  m.age50<-glmer(ct_pos~cath_pos+(1+cath_pos|study_no),data=test, amily=binomial,na.action=na.omit)
> 
> 
> 
>  result.age50<-predict(m.age50)
> 
> 
> 
> My question is now: How exactly are the predictions calculated? I  could not find any details in the documentation.  Are these  just
> 
>  based on the fixed effects setting the random effects to zero? Or are  these empirical Bayes estimates?
> 
> 
> 
> Is there any documentation available?
> 
> 
> 
> 
> 
> Thank  you very much in advance.
> 
> 
> 
> Peter
> 
> 
> 
> Universitätsklinikum Jena - Bachstrasse 18 - D-07743 Jena
> Die gesetzlichen Pflichtangaben finden Sie unter http://www.uniklinikum-jena.de/Pflichtangaben.html
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>



More information about the R-sig-mixed-models mailing list