Skip to content
Prev 5288 / 20628 Next

MCMCglmm: ixn between binary family-level factor and us() random effect

Hi Jarrod,

If speed is the only problem I can tolerate it - the wait is more than outweighed by the advantages of using MCMCglmm. The slowest model I'm running (the one fitting an unstructured 4x4 covariance matrix in each of two groups) gives >1000 independent samples in 3-4 hours, which is fine as I only have to do one full run, and all the other models can be run in parallel with this one. Presumably if time was really a problem, I could run the same model several times in parallel and merge the MCMC output, as long as each run had enough burnin.

I've now compared the correlations estimated in MCMCglmm from a similar model to the one below (4x4 family VCV in a single group, mother and father residuals fixed to zero) with correlations estimated using FISHER and they are similar enough to reassure me that they're measuring the same thing.

I hope that's all my problems solved - thank you for all your help with this.

Cheers,
Paul


-----Original Message-----
From: Jarrod Hadfield [mailto:j.hadfield at ed.ac.uk]
Sent: 07 February 2011 09:23
To: Paul Johnson
Cc: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] MCMCglmm: ixn between binary family-level factor and us() random effect

Hi Paul,

Unfortunately fixing residual variances to zero (or close to) is not
going to work in MCMCglmm because mixing will become intolerably slow.
As you say, you can fix the variance to zero in the G-structure but
this also fixes the covariances within the sub-matrix to zero, so as
yet there is no satisfactory solution in MCMCglmm. Sorry. I'm pretty
confident a SIR model could be constructed that would fit the model,
but a the moment the SIR implementation is so rudimentary it will not
run on all data (e.g. if there are missing data such  as that
associated with phantom parents).

Cheers,

Jarrod
On 7 Feb 2011, at 00:51, Paul Johnson wrote:

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



The University of Glasgow, charity number SC004401