[R-sig-ME] parameter expanded inverse Wishart prior for MCMCglmm multivariate model

[April Martinig]

Hello,

I am using MCMCglmm to run a multivariate model (4 response variables) with two random effect terms. I am particularly interested in the among- and within-individual correlations.

I posted my original question on Stack Overflow (https://stackoverflow.com/questions/67941391/how-to-properly-code-a-scaled-inverse-wishart-prior-for-a-mcmcglmm-model?noredirect=1#comment120089208_67941391) and received a very helpful answer. 

To expand on that question, what is the correct format for a parameter expanded inverse Wishart prior?

The prior as I currently have it:

prior.miw<-list(R=list(V=diag(4), nu=1), 
                G=list(G1=list(V=diag(4), 
                nu=1,   
                alpha.mu=c(0,0,0,0), 
                alpha.V=diag(4)*1000),
                G2=list(V=diag(4), #need to repeat to deal with second random effect
                nu=1,   
                alpha.mu=c(0,0,0,0), 
                alpha.V=diag(4)*1000)))

Also, how do I know if 1000 is an appropriate scale for my alpha.V? I understand that it should be large, but I am not sure if this is also appropriate.

Take care,
April Martinig


	[[alternative HTML version deleted]]