[R-sig-ME] Fwd: R - specify estimated residual variance

Douglas Bates bates at stat.wisc.edu
Fri Apr 4 16:19:40 CEST 2008

---------- Forwarded message ----------
From: Douglas Bates <bates at stat.wisc.edu>
Date: Fri, Apr 4, 2008 at 7:38 AM
Subject: Re: R - specify estimated residual variance
To: Kelly Wauters <Kelly.wauters at kuleuven-kortrijk.be>

On Fri, Apr 4, 2008 at 7:11 AM, Kelly Wauters
 <Kelly.wauters at kuleuven-kortrijk.be> wrote:
 > Dear Prof. dr. Bates,

 >  I want to use lmer to fit a glmm with a dichotomous dependent variable
 >  (family binomial - link logit). The scores on the lowest level follow a
 >  binomial distribution. This indicates that the variance of the lowest
 >  level is defined by the estimated residual variance multiplied by
 >  pi(1-pi). This means that the estimated residual variance has to be
 >  equal to 1. In SAS you can do this by means of the code

 >  proc glimmix data= dichotoom noclprint noitprint asycov ;
 >  class school id item;
 >  PARMS 0.34 1/HOLD=2;
 >  model scores=/ dist=binomial solution;
 >  random intercept/ subject=id;
 >  random _residual_;
 >  run;

 >  How can I translate the code into R code?

 > > dichloso$school<-as.factor(dichloso$school)
 > > dichloso$id<-as.factor(dichloso$id)
 > > dichloso$item<-as.factor(dichloso$item)

 Transforming those variables to factors is a good practice but not
 strictly necessary to fit the model shown below.

 > > model <- lmer(scores~1+(1|id), dichotoom,

 >  I don't know how I can translate the code PARMS 0.34 1/HOLD=2; into R

>  code. Can you help me on this?

 I'm afraid I can't help you on this because I don't know what the SAS
 code does.  I don't use SAS myself.

 It is possible that there is no need to specify this parameter in R as
 having a constant value.  My guess, but this is just a guess, is that
 the distinction between holding that parameter constant at 1 and
 allowing it to vary is equivalent to using the binomial family or the
 quasibinomial family in lmer.

 May I send a copy of this reply to the
 R-SIG-Mixed-Models at r-project.org mailing list? ("SIG" == "Special
 Interest Group")? (I ask your permission to send the copy because I am
 quoting your original question.)   Some who subscribe to that mailing
 list may have experience with SAS PROC GLIMMIX and be able to describe
 the equivalent code for lmer.

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