I am using sigmoidal non-linear growth models (logistic,weibull,Gompertz)
to model radiological dose versus time. I am running into convergence
problems (monitored with Brian Smith's BOA) as I get farther from the
asymptotic dose (that is, as I try to predict the asymptote early in an
event). I have transformed individual parameters to the real line in an
attempt to orthogonalize the joint posterior. A paper by Hills and Smith
(Parameterization Issues in Bayesian Inference) recommends iterative
orthogonalizing and centering and scaling transformation of the resulting
real parameters. I was wondering if any BUGS users have
examples/experience with this procedure. I would appreciate any help.
John Neal
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|