[This is a belated follow-up to posts by Bill Harris & Mike Day, Aug 30th & 31st]
I think Mike Day's probably right in recommending following up links between J, R & BUGS. Note also the recently published (August 2007) paperback "Bayesian Computation with R" by Jim Albert - but beware of misprints etc.! MCMC is inherently iterative, and can be relatively slow in J. However, it can also be relatively slow in SPlus, R or any other system. Some MCMC techniques such as "adaptive direction search" and "multiple try" methods are more suitable for parallelisation. My user-unfriendly code for this will be upgraded from J3.05 Real Soon Now! [I'm the still small Bayesian voice that uses cubature methods in preference to MCMC.] Aside: spot the REAL MCMC guru and the recent TV star beside me on the orphan web page http://www.ewartshaw.co.uk/keble.html J.E.H.Shaw http://www.warwick.ac.uk/statsdept Ewart Shaw http://www.ewartshaw.co.uk Scary portraits http://metalmanatee.deviantart.com/gallery/ ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
