ESS computed on rank-normalised draws, which is what makes it survive
heavy tails and infinite-variance targets where the raw ESS is not
even defined.
Usage
morie_effective_sample_size_bulk(chains)
Arguments
- chains
matrix of draws, one row per chain.
Value
list with ess_bulk, efficiency,
sufficient (ESS >= 100 per chain).
Details
It governs CENTRAL summaries only – the posterior mean and median.
A good bulk ESS says nothing about the tails; that is
morie_effective_sample_size_tail.
References
Vehtari, A. et al. (2021). Rank-normalization, folding,
and localization: an improved R-hat for assessing convergence of
MCMC. Bayesian Analysis, 16(2), 667-718.
Examples
set.seed(1)
morie_effective_sample_size_bulk(matrix(rt(2000, 2), nrow = 4))$sufficient
#> [1] TRUE
