[R-sig-ME] Bug in mcmcsamp

Jarrod Hadfield j.hadfield at ed.ac.uk
Sun Aug 12 16:31:12 CEST 2007


With version 0.99875-6 I get the posterior variances for A and B both  
coming out with a mode around 1.5 (See Hadfield2.pdf)



-------------- next part --------------
A non-text attachment was scrubbed...
Name: Hadfield2.pdf
Type: application/pdf
Size: 337185 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20070812/c24ca3f5/attachment.pdf>
-------------- next part --------------

On 12 Aug 2007, at 15:06, Douglas Bates wrote:

> On 8/12/07, Jarrod Hadfield <j.hadfield at ed.ac.uk> wrote:
>> Hi,
>> I'm using lme4 0.99875-6 and I think there may be a bug in mcmcsamp.
>> When I fit 2 random effects, the posterior variances always seem to
>> be approximately the same irrespective of the true variances (or the
>> REML estimates of the true variances).  For example:
>>     A<-gl(100,10)
>>     B<-gl(100,1,1000)
>>     y<-rnorm(100,0,sqrt(1))[A]+rnorm(100,0,sqrt(3))[B]+rnorm
>> (1000,0,sqrt(3))
>>     model1<-lmer(y~(1|A)+(1|B))
>>     model1MCMC<-mcmcsamp(model1,n=5000,trans=FALSE)
>>     plot(model1MCMC)
> Thanks for reporting the problem, Jarrod.  I tried your script with
> the development version of lme4 and got the expected results, which I
> enclose Could you try the mcmcsamp.R script in lme4 0.99875-6 and tell
> me if the results are similar?
> <Hadfield.pdf>
> <mcmcsamp.R>
> <mcmcsamp.Rout>

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