Nice catch! Thank you for your response.
I made what I think is the correct update to the Stan model:
real a[P]; // page level random effects
vector[P] alpha; // page level random effects
which are assigned in these components:
// page level
for (i in 1:P){
alpha[i] <- a[page[i]];
}
a ~ normal(a_mu, sigma_a);
If I’m understanding the syntax correctly, alpha and a are each P parameters and the loop should estimate all P of them. Let me know if that’s wrong.
However, my model is still failing to initialize. Curiously, the error message now indicates that “Random variable[1] is nan, but must not be nan!”, whereas in the last message, it was Random variable[5] - so it’s possible that fixed one issue.
Is there a way to map the Random variable[index] to the associated parameter in the model? That might help tip me off to where the issue is arising…