Skip to content

fixing covariances to 0 in MCMCglmm

5 messages · Pierre de Villemereuil, Jarrod Hadfield, Celine Teplitsky

Original
Celine Teplitsky
# 
Dear all,

I would like to run an animal model in MCMCglmm with 3 traits, 2 male 
traits and one female trait, but I do not know how to fix the 
covariances between sexes for Vpe and Vr to 0. From what I understood 
from an earlier post by Jarrod, it was not possible 2 years ago, but any 
chance it is now?
Many thanks in advance for help,

All the best

Celine
Jarrod Hadfield
# 
Hi Celine,

Sorry, this is still not available. Given that there is no information  
in the data to estimate the covariances you want to set to zero, I  
wonder whether there is anything wrong with just having a fully  
unstructured matrix and then ignoring the covariances? The information  
for those covariances will come solely from the prior, but I would  
expect that the posteriors for those (co)variances that can be  
estimated would remain valid. It may reduce mixing, but would be one  
solution.

Cheers,

Jarrod




Quoting Celine Teplitsky <teplitsky at mnhn.fr> on Wed, 30 Jan 2013  
19:09:57 +0100:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Pierre de Villemereuil
    

    

      
      #
    

  


  

      
Hi !

What about the following :
random = ~idh(sex):PE + us(sex):animal, rcov=idh(sex):units

I did not test it though, so I'm not sure it will work, but I see no 
reason why not ! ;)

Cheers,
Pierre.


Le 30/01/2013 19:09, Celine Teplitsky a ?crit :

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Jarrod Hadfield
    

    

      
      #
    

  


  

      
Hi Pierre,

I think Celine wants matrices of the form

  V1 C12 0
C12 V2  0
  0   0   V3

it could be done for the PE term. I think this should work:

random =  
~us(at.level(sex,"Male"):at.level(trait,1:2)):PE+us(at.level(sex,"Female"):at.level(trait,3)):PE

where traits 1 and 2 are measured on males, and trait 3 on females.

The R-structure however cannot yet have multiple terms.

Cheers,

Jarrod






Quoting "Pierre B. de Villemereuil" <bonamy at horus.ens.fr> on Thu, 31  
Jan 2013 11:07:34 +0100:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Celine Teplitsky
    

    

      
      #
    

  


  

      
Hi Jarrod and Pierre,

thanks a lot for the help!

Cheers

Celine