Just a quick comment. Joint mode and funnel can even have the opposite behavior: as I showed in my simulation, in models quite like “Gelman’s funnel”, increasing group variance eventually eliminates funnel as tau is kept away from 0 but also leads to bimodality, in contrast decreasing group variance will avoid bimodality but gives rise to the funnel/metastability/slow mixing of HMC chains.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Neal's funnel on a simplex | 1 | 217 | May 17, 2024 | |
| Non-centered parametrisation for hierarchical log_mix (mixed) distribution | 4 | 948 | February 21, 2022 | |
| Zaps to avoid the funnel | 9 | 1625 | October 3, 2018 | |
| Problem with convergence when using different scale parameters for groups | 5 | 42 | October 21, 2024 | |
| Apparently equivalent noncentered parameterization, different outcomes | 2 | 52 | August 8, 2024 |