Priors for and estimating heritability from an ordinal and a Poisson animal model

(I thought I responded to this e-mail a while ago... but my response 
might have not leave my computer !)

Concerning the priors, two things :
- For the first model, it is not very clear, but I assume you use 
"deformed toes" as a binary variable. In that case, a better prior would 
be the chi2 distribution. You can have it by using the following prior :

prior2=list(R=list(V=1, fix=1), G=list(G1=list(V=1, nu=1000, alpha.mu=0,alpha.V=1)))

This prior is a bit skewed toward little heritabilities, but the one you 
used is actually skewed toward strong heritabilities.
- For the second model, what is "p.var" ? If it is the phenotypic 
variance calculated from the data, you should not use it. It is 
ill-advised to use the data under investigation to set up the prior. I 
think you try there to use your first prior or the following one :

prior1.1<-list(G=list(G1=list(V=1,n=0.002)),R=list(V=1,n=0.002))

I hope this is answering some of your questions. Don't hesitate to check 
different priors to see if your posterior distribution is influenced or 
not by the prior.

Also, normally, the heritability being an intra-class coefficient, it 
should be independant from the fact that the residual variance is fixed. 
You should check the MCMCglmm Course Notes from Jarrod Hadfield 
concerning this point.

Finally, to add the random factor, using the chi2 distribution, you 
should just use :

prior2=list(R=list(V=1, fix=1), G=list(G1=list(V=1, nu=1000, alpha.mu=0,alpha.V=1),
G2=list(V=1, nu=1000, alpha.mu=0,alpha.V=1)))

This would not alter much the shape of the prior on the heritability.

Cheers,
Pierre.

Le 06/09/2012 17:13, Helen Ward a ?crit :

Priors for and estimating heritability from an ordinal and a Poisson animal model

Thread (4 messages)