 |
 |
Dear Akos and list, The bug should only occur when the elements of the mean vector of a multivariate normal likelihood are given univariate normal priors. The full multivariate setup should be OK, but all MVN results for 0.6 should be treated with caution until we get this fixed. We are very sorry for this and hope it has not caused too much inconvenience. As a small recompense we can announce that the next release of WinBUGS will have the facility to run parallel chains with automatic Gelman-Rubin statistics with some very pretty pictures of converging chains. David Spiegelhalter > It is written in the Webside of BUGS, in the section News & Latest Developments > for BUGS that there is a bug in multivariate normal sampler for VERSION 0.6. I > used this version for models with multivariate normal sampler/Wishart prior, > and compared some results with Mln, but did not find any serious discrepancies > in parameter estimates. Could you explain what is the nature of this bug. > Thank you for any information. > > Akos Pap > Hungary > %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%