[R-sig-ME] code for multiple membership models?

Jarrod Hadfield j.hadfield at ed.ac.uk
Sun Aug 14 11:59:34 CEST 2011

Dear Andrei,

If I understand correctly I think you can just add the level-1 and  
level-2 covariates into the fixed effects and they will be taken care  
of appropriately. With 150K observations MCMCglmm may be too slow but  
that depends on the complexity of the model. The presence of some  
level 1 units  across all levels of the level-2 units suggests that  
the mixed model equations may be quite dense and slow to solve, but  
you would have to try it out.  The implementation of multimembership  
models in MCMCglmm is not yet ideal because the design matrix for the  
multimembership part is initially dense and then coerced into sparse.  
This means that setting up the model prior to MCMC can be costly in  
time/memory. I have plans to change this.



On 5 Aug 2011, at 03:06, Andrei Boutyline wrote:

> Dear all (and especially Doug, Jarrod and Malcolm),
> About a month ago, there was a really useful discussion between  
> Doug, Jarrod
> and Malcolm about fitting multiple membership models (
> https://stat.ethz.ch/pipermail/r-sig-mixed-models/ 
> 2011q2/006320.html).  I
> was hoping that some of you could kindly offer some follow-up advice.
> I am trying to implement something very similar to what was  
> recommended in
> that thread, but with the addition of covariates on both levels.  It's
> pretty clear to me how to add covariates on level 1, but I do not  
> quite
> understand how to add covariates on level 2.  Could you please  
> explain how
> to do this?
> Also, I have roughly 120 level-2 units, and 150,000 level-1 ones.   
> Some of
> those level-1 units are members of almost all 120 level-2 units.   
> Should I
> be worrying that the resulting model will be too large to estimate in
> reasonable time (e.g., a day or two), or can these solutions handle  
> it?
> Thank you very much,
> Andrei
> -- 
> Andrei Boutyline
> University of California, Berkeley
> PhD Student, Sociology
> www.ocf.berkeley.edu/~andrei <http://www.ocf.berkeley.edu/%7Eandrei>
> 	[[alternative HTML version deleted]]
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

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