[R] Bayesian Information Criterion

[Martyn Plummer]

On Mon, 2006-10-23 at 20:59 +0200, Alexander Geisler wrote:
> Hello!
> 
> I calculate a Bayesian logistic regression with the function MCMClogit 
> from the package MCMCpack.
> 
> After the calculation I use the package CODA to test the model in 
> respect to convergence.
> 
> Is there a compatible package to calculate the BIC (bayesian information 
> criterion)? So that it is possible to write
> BIC(mcmc object) with an mcmc object from the function MCMClogit.

No. An mcmc object is not a fitted model object.  It is just a
collection of samples from the posterior distribution and does not
contain the data required to calculate the BIC.

You may want to have a look at the BMA package for logistic regression
with model averaging using BIC. 

> I found the package nlme, but the function BIC in this package does not 
> work with the mcmc object from the function MCMClogit.
> 
> Regards
> Alex
> 

-----------------------------------------------------------------------
This message and its attachments are strictly confidential. ...{{dropped}}