Learn how easy it is to sync an existing GitHub or Google Code repo to a SourceForge project! See Demo
I am fitting a parametric survival model, assuming a constant hazard. Therefore, survival times are exponentially distributed.
I'm trying to understand how to specify a non-informative prior on the rate. The winbugs examples and the textbook by ibrahim use a gamma prior around (1E-3, 1E-3). my understanding is that this is a conjugate prior. I also understand that small rate values generate a flatter prior on expectation of the exponential distribution (in median survival times are more relevant).
I was just wondering if there are limits to the shape and scale parameters for dgamma in JAGS? Would it be wise for me to truncate the distribution at some sensible value for my model?
Personally I like to put priors on scales which are easy to interpret. So I might put a uniform prior on the expected survival time, and then transform that to a rate. You can then choose the limits of the uniform to exclude survival times which are implausible for your application (e.g. over 150-200 for humans). There's rarely any need to restrict yourself to conjugate distributions in BUGS/JAGS.