[R-sig-ME] Fitting crossed random effects in lme (not lmer)

[Douglas Bates]

Sorry for the late reply.  As I indicated in another message, I was at
useR!2010 then on vacation.

On Tue, Jul 27, 2010 at 7:43 PM, Christoph Scherber
<Christoph.Scherber at agr.uni-goettingen.de> wrote:
> Dear all,
>
> Let´s assume I have an lme model with the following random effects structure:
>
> random=~1|block/plot
>
> Now, further treatments are applied at the next smallest level, but they
> are on the same hierarchy. In lmer notation, this could be parameterized
> as
>
> ...+(1|block/plot/treatment.type1)+(1|block/plot/treatment.type2)
>

I'm not entirely sure that would be successful.  The expansion of the
/ in the grouping factor in lme4 doesn't do a full symbolic analysis
of the terms.  Instead it is more of a macro expansion.  I think that
would probably end up with multiple random-effects terms of the form
1|block and 1|block:plot and a model for which unique parameter
estimates are not defined.

> How could I specify the same structure in lme?

It would not be easy to do that.  In fact, I'm not even sure it is
possible.  I think my advice would be to stay with simpler model
specifications.  You would need a huge amount of data to be able to
estimate parameters in such a model successfully.

> I know that I would need to use one of the pdClasses(), but I don´t
> understand how to set this up correctly. I tried of course the Pinheiro &
> Bates book, but maybe someone can give me a hint. That would be excellent!
>
> All the best and many thanks for any help,
> Christoph
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>