Hi,
I have two non-nested models that I am trying to compare by calculating
the Bayes Factor according to the Carlin and Chib method. I am,
however, having difficulty in ensuring that each chain visits both
models frequently (or at all!). In my model, a chain seems to instantly
pick a model and stick with it throughout the run, no matter how
lopsided I make the prior distribution on the selection parameter
against this. I can give a selection-parameter-value a prior
probability of 1E-8 or even 1E-16 and it will still be chosen on every
single iteration (even with no burn-in).
The only way I have managed to get both models to be visited by each
chain (and to converge) is to apply the cut function to the value of the
selection parameter in the model. I presume, however, that this goes
against the intention of Carlin and Chib's method because it enforces
that both models will be selected exactly accordingly to the prior on
the selection parameter, thus breaking the calculation of the Bayes
Factor. Is there anything else that can be done to make sure both
models are selected by each chain?
I have tried the model with two approaches: one is based on the pines
example, using pseudo-priors determined by previous converging runs, and
the other one is based on the approach in a post to the list by Martin
Upsdell
(https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind01&L=BUGS&D=0&I=-3&d=No+Match%3BMatch%3BMatches&P=132918).
Both models have sensible initial values (not far from the means of
those pseudo-priors). Both are able to converge when run by themselves
(i.e. if I set the prior probabilities on the model selection to 0 and
1), if one a little slower than the other. I can post the various
models if that would help.
Thanks in advance,
k
-------------------------------------------------------------------
This list is for discussion of modelling issues and the BUGS software.
For help with crashes and error messages, first mail [log in to unmask]
To mail the BUGS list, mail to [log in to unmask]
Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html
Please do not mail attachments to the list.
To leave the BUGS list, send LEAVE BUGS to [log in to unmask]
If this fails, mail [log in to unmask], NOT the whole list
|