02-12-2012 07:21 PM
It appears that one must calculate a normalizing constant (in a DATA step for example) for the posterior density estimates given with the OUTPOST= option of PROC MCMC. In other words the LogPost variable is off by a constant. The LogPrior density variable seems to be fine. Am I missing something?
I'm using SAS 9.2 TS Level 2M2.
03-02-2012 09:22 AM
Yes, you are correct. The LOGPOST variable is not normalized w.r.t. the parameters. It is simply the sum of the log of the likelihood + prior evaluated at each iteration (drawn of parameters).