[R-sig-ME] MCMC prior issues

[Ben Bolker]

David Bradley <bradleyd at ...> writes:

> I am trying to use the MCMCglmm package to run a bivariate plasticity model
> to assess plasticity in the timing of egg laying and the relationship with
> reproductive success. The issue I am having is with the prior
> specification, which keep returning errors.
> 
> #The prior I am using is a flat prior:
> 
> prior.bi<-list(R = list(V = diag(2), nu = 0.002), G = list(G1 = list(V =
> diag(4), nu = 0.002)))

  [snip]
 
> *Error in MCMCglmm(cbind(savg.nf, sld) ~ trait - 1 +
> trait:st1 + trait:factor(min.age) +  :ill-conditioned G/R structure: use
> proper priors if you haven't or rescale data if you have*


> I thought that I was using a proper prior, and I have already scaled the
> data. What is the meaning of "ill-conditioned G/R structure"?
> 

  I'm not sure (a reproducible example would be nice), but have
you tried increasing nu to something larger and seeing if that
fixes the problem?  That may not necessarily make you happy --
you might prefer a flatter prior -- but that would indicate that
your structure is OK, but your priors are too flat ...