So I have a truncated normal in my model. But how do I do a posterior predictive check with such a model? I know there are two ways to force a truncated distribution in the model block, either by parameter bounds or by explicit truncation wtih the
T[a,b] modifier. But the explicit truncation approach does not work for the
_rng functions. So the question is: how do I draw samples from truncated distributions using
_rng functions in the
generated quantities block?
Or is there a simple trick to transform draws from non-truncated distribution to a truncated one, similar to the desugaring of the
T[a,b] syntax for sampling statements shown in the docs? The best I have so far is rejection sampling which luckily is not slowing my model down a lot, but it feels hacky.
Thanks for any ideas.