[R-sig-ME] Convergence diagnostics for MCMCglmm

Jarrod Hadfield j.hadfield at ed.ac.uk
Thu Nov 19 14:36:42 CET 2009

Hi Wayne,

You should pass the mcmc objects from >2 models as a list. For  
example, the fixed effects:

diag.gelman(mcmc.list(m1$Sol, m2$Sol))

Ideally the models should have over-dispersed starting values, which  
you can specify in the start argument. This will stop MCMCglmm finding  
"heuristically" good starting values.

If you have access to ASReml you could always fit the model using REML  
and check to see whether the estimates are the same/similar.



On 19 Nov 2009, at 13:20, Dawson Wayne wrote:

> Dear R users,
> I've managed to get to grips with using MCMCglmm for phylogenetic  
> meta-analysis, thanks to help on here from Jarrod Hadfield. However,  
> I haven't yet worked out how to use convergence diagnostics on  
> MCMCglmm output.
> I am currently running models for 50,000 iterations, with a long  
> burnin of 25000, and a thinning interval of 10. Whilst my trace  
> plots look good, I would like to use the Gelman-Rubin or Raftery- 
> Lewis diagnostics in the coda package to check that my burnin/no. of  
> iterations are adequate. I've read relevant sections in Ch. 7 of Ben  
> Bolkers ecological models book, and the coda/MCMCglmm package pdfs,  
> but I'm still not sure what mcmc output from MCMCglmm I am supposed  
> to pass to the coda diagnostic functions. Apologies for the simple  
> question, but hopefully there is a simple answer!
> Any suggestions appreciated as always,
> Thanks,
> Wayne
> -- 
> Dr. Wayne Dawson
> Institute of Plant Sciences
> University of Bern
> Altenbergrain 21
> 3013 Bern
> Switzerland
> +41 (0)31 631 49 25
> _______________________________________________
> 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