I am getting started with Stan, and I have the following state-process model coded in Stan: data { int<lower=0> maxIter; int<lower=0> numGears; int C[numGears,maxIter,2]; real totalArea; real a[numGears]; } parameters { real<lower=0> N0; real<lower=0,upper=1> juvProp; real<lower=0> …

You are expecting sampling statements like nJ[i,j] ~ poisson(a[i]*J[j]/totalArea); to populate nJ[i,j] with a value sampled from a Poisson distribution. But that’s not what these statements achieve in Stan. Instead, in Stan, these statements exist to define something proportional to the joint log-…

Thanks that is a great point! If it were the case that the math didn’t work out as in this case, would an alternative solution be to include nJ and nA as dummy variables in the data section and let Stan deal with them? Or are data mutations not allowed?

Data mutations are allowed in transformed data, but these must be deterministic data transformations that run once prior to sampling. You can never use latent discrete parameters in your likelihood computation in Stan; these must always be marginalized out.

Thanks! I’m guessing that in that case one would need to do something like what you mentioned in the other post .

Random Poisson variable is a negative but must be nonnegative!

Modeling

Topic		Replies	Views
Poisson Distribution always return erros Modeling	1	521	January 20, 2020
Hierarchical model with a poisson distributed parameter Modeling cmdstanr	12	1481	December 7, 2022
Help with Poisson model Modeling loo	45	8254	March 3, 2018
First Pystan Poisson Model Modeling	7	1512	January 10, 2020
Poisson distributions and int variables Modeling specification	7	1864	February 6, 2020

Random Poisson variable is a negative but must be nonnegative!

Related topics