[R-sig-ME] Constraining error variance to 0 in an lmer() model.

[Douglas Bates]

In https://www.jstatsoft.org/index.php/jss/article/view/v067i01 the lme4
authors describe in some detail the numerical methods used in lmer.  You
will see that one of the transformations used to profile the log-likelihood
is working with the relative covariance factor.  This is the Cholesky
factor of the ratio of the covariance matrix of the random effects to the
variance of the per-observation noise.  It is assumed that that variance is
greater than zero.

The model that you wish to fit is not within the scope of models lme4 is
designed to fit.

On Sun, Mar 25, 2018 at 3:40 AM Rolf Turner <r.turner at auckland.ac.nz> wrote:

>
> On 25/03/18 12:40, Rolf Turner wrote:
>
> <SNIP>
>
> >
> > The model that I wanted to fit was (adapting Maarten Jung's notation
> > slightly to suit my own tastes):
> >
> >      (*) lmer(test ~ 0+person + (0+person | occasion),data=Dat)
>
> <SNIP>
>
> Berwin Turlach has just pointed out to me that I got the preceding
> expression arse-backwards, as is so often my propensity.  It should
> be:
>
>      (*) lmer(test ~ 0+occasion + (0+occasion | person),data=Dat)
>
> The fixed effect is *occasion* and the random effect is *person*.
>
> Likewise
>
> >      lmer(test ~ 0+person + (0+person | occasion),data=Dat,
> >                  control=lmerControl(check.nobs.vs.nRE = "ignore"))
>
> should be
>
>        lmer(test ~ 0+occasion + (0+occasion | person),data=Dat,
>                    control=lmerControl(check.nobs.vs.nRE = "ignore"))
>
> Psigh!  Sorry for the confusion, everybody.
>
> <SNIP>
>
> cheers,
>
> Rolf Turner
>
> P.S.  Berwin also asked me to provide the data I used.  I simply took
> the data provided by Ben Pelzer, in a posting in the relevant thread,
> just before the posting in which Maarten Jung explained how to get the
> fit to run, by setting the appropriate control variable.
>
> For the record, Ben Pelzer provided the data through a URL:
>
> > Dat <-
> > read.table(url("
> https://surfdrive.surf.nl/files/index.php/s/XfE3mtbFCTUejIz/download"),
> > header=TRUE)
>
> (Actually Ben Pelzer wrote "mydata" where I have written "Dat" above.  I
> *refuse* to use this "mydata" construction since it is *so*
> Micro$oft! :-) )
>
> R. T.
>
> --
> Technical Editor ANZJS
> Department of Statistics
> University of Auckland
> Phone: +64-9-373-7599 ext. 88276 <+64%209-373%207599>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]