[R-sig-ME] observations in multiple groups

Jarrod Hadfield j.hadfield at ed.ac.uk
Thu Jun 9 16:18:07 CEST 2011


I think Jeroen is after a multimembership model. Levels of firstgroup  
can correspond to levels of  secondgroup and they should have the same  
associated random effect.  Doug has suggested before that it may be  
possible with some work:


Alternatively ASReml, MLwin, MCMCglmm and I'm sure others can fit  
these types of model.


On 9 Jun 2011, at 03:19, Ben Bolker wrote:

> Hash: SHA1
> On 11-06-08 11:39 PM, Jeroen Ooms wrote:
>> The standard assumption in multilevel modelling seems to be that  
>> every
>> observation is in 1 group. However, I have an application in which  
>> every
>> observation is in 2 (exchangable) groups. More precisely, I have k  
>> groups
>> and k^2 observations, and there is one observations for every  
>> combination of
>> groups. The groups are unordered, so I can't do e.g. 1 + (1|  
>> firstgroup) +
>> (1|secondgroup). Is there a way to model this with lme4?
>> So the schoolbook example would be a hierarchical model with  
>> children within
>> classes, however every child is in two classes at the same time.
>  Isn't this exactly a crossed random effect?  i.e., child k is in
> classes i and j, so their expected value (in the absence of any
> covariates etc.) is (grand mean) + eps_i + eps_j +eps_(error).  I'm
> sorry if I'm missing something, but it seems like (1|firstgroup) +
> (1|secondgroup) is precisely what you would want in this  
> situation ... ?
> Why does it matter if the groups are ordered?
>  Ben Bolker
> Version: GnuPG v1.4.10 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
> iEYEARECAAYFAk3wLckACgkQc5UpGjwzenP82gCfXlUt9CV+7d71aZzj3gUWV/vS
> a0UAnAqoKK3n44/vzv0z4LVfuP9dB4hJ
> =WDgT
> _______________________________________________
> 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