[R-sig-ME] MCMCglmm rcov specifications

[Jarrod Hadfield]

HI,

Sorry - I was to quick. The residual variances will not be  
proportional to mev, they will be mev. Its not possible to fit the  
proportional model currently/easily, although I could imagine with  
some thought you could trick MCMCglmm into doing it via a SIR model. See

https://stat.ethz.ch/pipermail/r-sig-mixed-models/2010q4/004916.html

for a solution to a similar problem. Not very elegant I'm afraid!

Jarrod





Quoting Jarrod Hadfield <j.hadfield at ed.ac.uk> on Thu, 2 Feb 2012  
18:05:34 +0000:

> Hi,
>
> see the schools example in the course notes:
>
> rcov=~idh(units):units
>
> prior=list(R=list(V=diag(mev), fix=1))
>
> Cheers,
>
> Jarrod
>
> On 2 Feb 2012, at 17:53, Ryan King wrote:
>
>> Hi list,
>> If I want to specify heterogeneous variances proportional to a known
>> factor in MCMCglmm, it seems like mev is the correct option, but
>> looking at the code it appears to add person-level random effects with
>> variance fixed at the specified value:
>>
>> random = ~us(leg(MCMC_mev, -1, FALSE)):MCMC_meta
>> prior$G<-list(G1=list(V=as.matrix(1), nu=1, fix=1))
>>
>> I've used the same trick to specify a known co-variance function.
>> However, the updates for this specification seem to go slowly and
>> induce bad mixing in my binary outcomes problem. The unidentified
>> residual variance certainly isn't helping. Is there a trick to
>> directly specify a matrix R and avoid inducing the identification
>> headache and slow MME solving?
>>
>> Thanks,
>> Ryan King
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>
> _______________________________________________
> 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.