ESS warning with censored predictive checks

Interfaces RStan
1

I often run models with censored data, and then wish to do posterior predictive checks. Some predicted data points may end up being above the right censoring limit with certainty. As a result, I get a warning about insufficient ESS from RStan; I then have to verify manually that this corresponds to the above-censoring PP check generated quantity, and that other parameters/quantities have sufficient ESS.

I could always generate the uncensored values and then censor in post-processing, but is there a way to do it under the hood in Stan without triggering the warning? Some way to tell RStan, in other words, that certain generated quantities should be ignored from the PoV of ESS calculation/warnings?

2

Can you provide more details, e.g. the model code with shows also the generated quantities? Simplified examples which still reproduces the example is fine. I like to understand why you get low ESS in case there is something wrong in the current computation.

However, I don’t need to understand the reason why ESS is low to recommend using posterior package and subset_draws to select which variables you want to diagnose, and summarise_draws to produce the convergence diagnostics for those. You will still get the warnings when RStan finishes, but with posterior package it’s easy to check that those ESSs you care are good enough.