Skip to content
Prev 6308 / 20628 Next

code for multiple membership models?

Hi,

The chain is getting "stuck" at values of zero for the variance, hence  
the spike. This should happen less with proper priors, and parameter  
expansion in particular increases mixing under these conditions. Using:

prior=list(R=list(V=1, nu=0.002), G=list(G1=list(V=1, nu=1,  
alpha.mu=0, alpha.V=1000),G2=list(V=1, nu=1, alpha.mu=0, alpha.V=1000)))

as your prior should help.

Note that the variance in the data due to the second set of random  
effects is not equal to the variance component because the diagonal  
elements of ZZ' are on average ~0.07. Depending on model assumptions  
you might want to square root your weights.  The diagonal elements of  
ZZ' are then all one and the variance component can be interpreted  
directly.  I can't see how the weights enter into the lme4a code, but  
presumably they do?

Jarrod



Quoting Malcolm Fairbrother <m.fairbrother at bristol.ac.uk> on Tue, 28  
Jun 2011 17:04:03 +0100: