[R-sig-ME] LMER vs MLwiN

Federico Calboli f.calboli at imperial.ac.uk
Fri Feb 17 14:06:44 CET 2012


On 17 Feb 2012, at 12:52, W Robert Long wrote:

> Hello
> 
> I'm new to using mixed models in R. Thus far I've been using MLwiN. I am trying to duplicate the results in MLwiN of a logistic mixed effects model.
> 
> At the moment I have no covariates, and a data hierarchy of
> pupil within class within school within commune with random effects at each level above pupil.
> 
> There are 9000 observations in total;
> 
> 300 classes
> 100 schools
> 20 communes
> These have been set as factors with as.factor(). I'm not sure if this is correct as they were not categorical in MLwiN (they were just ints) but I was getting this error before I did that:
> "Error: length(f1) == length(f2) is not TRUE"
> 
> I have tried to fit a model with glmer like this:
> 
> glmer(LOSS~(1|COMMUNE/SCHOOL/CLASS/PUPIL),data=dt,family=binomial(link = "logit"))

you seem to specify COMMUNE and SCHOOL as random effects, with no fixed effects. If I were you I would:

1) code PUPIL from 1:n so that you never have two different pupils in two different classes coded with the same code. 
2) code CLASS from 1:n (see above)

then I'd try

glmer(LOSS~ COMMUNE + SCHOOL + (1|CLASS) + (1|PUPIL) ,data=dt,family=binomial(link = "logit")) #coding class and pupil as I said will automatically take care of the nesting

as see what happens.

If you are using R 32 bits you might want to use R 64 bits to have more RAM available.

BW

F



> 
> However this generates the error
> "Error: cannot allocate vector of size 9.5 Gb"
> 
> I have also tried glmmPQL in the MASS package:
> glmmPQL(LOSS~1,data=dt, random = ~1|COMMUNE/SCHOOL/CLASS/PUPIL,family=binomial(link = "logit"))
> 
> However this generates /completely/ wrong estimates so I can only assume that I am specifying the model incorrectly in R.
> 
> If anyone can advise, I would be very grateful
> Thanks
> RL
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

--
Federico C. F. Calboli
Neuroepidemiology and Ageing Research
Imperial College, St. Mary's Campus
Norfolk Place, London W2 1PG

Tel +44 (0)20 75941602   Fax +44 (0)20 75943193

f.calboli [.a.t] imperial.ac.uk
f.calboli [.a.t] gmail.com




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