I am running a Hurdle model that is integreated in a Hidden Markov Model. Indicated by Rhat, all model parameters converge quite good. However, the effective sample size of the individual-specific intercepts (alphai, alphaj) and the state-specific intercepts (mu, nu) decrease drastically if I add da…

Try a non-centered parameterization on these: alphai ~ normal(0, sigma_alphai); alphaj ~ normal(0, sigma_alphaj); I think you can do this by just doing: real<multiplier=sigma_alphai> alphaj[H]; // individual-specific intercept in Bernoulli part real<multiplier=sigma_alphaj> alphai[H]; // individu…

Thank you, Ben. I didn’t get any divergences. Just to be sure: Using NCP with the approach you are suggesting, there is no need to change the priors on alphaj, alphai, sigma_alphaj, and sigma_alphaj, they remain the same? Because when I applied the NCP in the past, I remember I had to use std_norma…

That’s good to know, I will give it a run like this and report back if it helped to increase ESS. Thanks for your support.

Hierarchical zero-inflated model: effective sample size decreases with number of individuals

Modeling

bbbales2 February 5, 2021, 2:12pm 4

Yeah, it’s a different way of doing the non-centered parameterization (Updated non-centered parametrization with offset and multiplier). Hope I got the syntax right on it :D

Topic		Replies	Views
Non-centered parameterization slows down the model Modeling hierarchical-model	6	519	February 23, 2021
Hierarchical model with some very small parameter values Modeling fitting-issues	7	799	June 7, 2021
Low effective sample size after running Bayesian cognitive model in Stan Modeling rstan , fitting-issues	8	960	August 18, 2021
Posterior of my non-centered parameterization (of intercepts hierarchical model) is less navigable than my centered parameterization Modeling rstan , fitting-issues , specification , biology	5	561	February 24, 2024
Non-Centered Hierarchical Model with Multiple Nestings Modeling specification	3	942	November 14, 2018

Hierarchical zero-inflated model: effective sample size decreases with number of individuals

Related topics