Milani

The exact mean parametrisation of Huang (2017) is more helpful here than the mode parameterisation. The mean form is now used in the glmmTMB wrapper for TMB and works well...

While based on the mode parameterisation, perhaps the fast rejection sampler proposed here for the COM-Poisson might be helpful: Benson A, Friel N (2017) Bayesian inference, model selection and likelihood...

If I understand your query correctly - then this discrepancy is probably because the approach using tidybayes (which is correct) is to `transform-then-summarise` using the posterior draws. But if you...