[R-sig-ME] MCMCglmm diagnostics

[Iker Vaquero Alba]

   Dear Jarrod and group members,
   Thank you very much for your clarifying reply. Regarding 2, that's actually what I had understood on the first place. Which means, if I'm right, that if you are running 1000-2000 iterations, thin should be 1 (store every single iteration), thin=10 (at least) for 10,000-20,000 total iterations, thin=100 (at least) for 100,000-200,000 iterations, and so on. About 3, I had read it somewhere and forgotten it for the time I asked the question. So, autocorr() and plot() are the basic diagnostic tools for analysing a single chain, then. 

   Thank you very much again,   Iker
 __________________________________________________________________

   Iker Vaquero-Alba
   Visiting Postdoctoral Research Associate
   Laboratory of Evolutionary Ecology of Adaptations 
   Joseph Banks Laboratories
   School of Life Sciences
   University of Lincoln   Brayford Campus, Lincoln
   LN6 7DL
   United Kingdom

   https://eric.exeter.ac.uk/repository/handle/10036/3381


      De: Jarrod Hadfield <j.hadfield at ed.ac.uk>
 Para: Iker Vaquero Alba <karraspito at yahoo.es> 
CC: "r-sig-mixed-models at r-project.org" <r-sig-mixed-models at r-project.org> 
 Enviado: Jueves 29 de octubre de 2015 22:24
 Asunto: Re: [R-sig-ME] MCMCglmm diagnostics
   
Hi,

1/ yes
2/ no - it means choose the number of iterations such that the number  
of *stored* iterations is about 1000-2000.  You can of course save  
more, but then the memory for storage goes up.
3/ These are for assessing convergence from multiple chains. you can  
apply them to mcmc.list objects: for example mcmc.list(m1$VCV, m2$VCV)

Cheers,

Jarrod



Quoting Iker Vaquero Alba <karraspito at yahoo.es> on Thu, 29 Oct 2015  
20:51:57 +0000 (UTC):



>
>    Hello everyone. Just 3 quick questions about MCMCglmm diagnostic tools:
>    1. When using autocorrelation(), the result I get includes  
> several lines marked as "Lag 1", "Lag 10", "Lag 50", "Lag 100" and  
> so on. In Patrick Lam's fantastic "Convergence Diagnostics" I read  
> this: "The lag k autocorrelation ρk is the correlation between every  
> draw and its kth lag. So, according to this, "Lag 1" is the  
> correlation between one sample and the sample inmediately posterior,  
> "Lag 10" the correlation between a sample and the sample 10  
> positions after, and so on. Is that right?   2. In the Course Notes,  
> it says "I usually aim to store 1,000-2,000 iterations and have the  
> autocorrelation between successive stored iterations less than 0.1."  
> Does this mean thin=1,000-2,000? Because in that case, we would be  
> storing every 1,000-2,000 iterations, right?   3. Apart from  
> autocorr() and trace and density plots, I have seen other diagnostic  
> analyses described for mcmc objects, such as Gelman and Rubin,  
> Geweke, Heidelberg-Lewis or Raftery-Lewis. However, when I try to  
> implement this in my MCMCglmm models, R shows me the message "no  
> applicable method applied to an object of class "MCMCglmm" or other  
> error messages." Are there any diagnostics tools that can be applied  
> to MCMCglmm objects other than the ones mentioned in the Course  
> Notes, autocorr() and plot()?
>    Thank you very much in advance.
>    Iker.
>  __________________________________________________________________
>
>    Iker Vaquero-Alba
>    Visiting Postdoctoral Research Associate
>    Laboratory of Evolutionary Ecology of Adaptations
>    Joseph Banks Laboratories
>    School of Life Sciences
>    University of Lincoln   Brayford Campus, Lincoln
>    LN6 7DL
>    United Kingdom
>
>    https://eric.exeter.ac.uk/repository/handle/10036/3381
>
>
>     [[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.




  
	[[alternative HTML version deleted]]