tag:blogger.com,1999:blog-3240271627873788873.post5846828664606063149..comments2024-03-26T06:46:11.752-04:00Comments on Doing Bayesian Data Analysis: Negative Binomial for Count DataJohn K. Kruschkehttp://www.blogger.com/profile/17323153789716653784noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-3240271627873788873.post-52965165480367616382012-04-07T13:33:54.592-04:002012-04-07T13:33:54.592-04:00Thanks very much for your suggestion, Mark. Altern...Thanks very much for your suggestion, Mark. Alternatively, a simple reparameterization in JAGS also does the trick. See the follow-up post now linked at the top of this post. Thanks again.John K. Kruschkehttps://www.blogger.com/profile/17323153789716653784noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-24423619964403512252012-04-07T07:12:55.838-04:002012-04-07T07:12:55.838-04:00I don't know a transformation that will allow ...I don't know a transformation that will allow you to use a Gibbs sampler. But this looks to me like a good case for a random-walk Metropolis sampler. Your scatter plots show that m and p are pretty much uncorrelated, so use them. First form the posterior kernel in terms of r and p, next substitute r -> m*(1-p)/p, then multiply the result by (1-p)/p to account for the Jacobian of the transformation, and finally make joint draws of m and p. (I think that's right.) Now you can compute anything you want with those draws.Mark Fishernoreply@blogger.com