Oh yes my bad, you are right. I also checked with the bivpois package in R, so i’ts correct now
I would start with a simpler model, just scalar priors for the 3 parameters of the bivariate poisson and a bunch of observations, to confirm the logp is implemented correctly for the core case.
If that works, it may be that your model has parametrization / data issues. Too broad priors, redundant parameters, etc…, which may take some thinking to figure out.
1 Like