Convergence and efficiency diagnostics for Markov Chains

These functions are improved versions of the traditional Rhat (for convergence) and Effective Sample Size (for efficiency).

Rhat(sims)
ess_bulk(sims)
ess_tail(sims)

Arguments

sims

A two-dimensional array whose rows are equal to the number of iterations of the Markov Chain(s) and whose columns are equal to the number of Markov Chains (preferably more than one). The cells are the realized draws for a particular parameter or function of parameters.

Value

The Rhat function produces R-hat convergence diagnostic, which compares the between- and within-chain estimates for model parameters and other univariate quantities of interest. If chains have not mixed well (ie, the between- and within-chain estimates don't agree), R-hat is larger than 1. We recommend running at least four chains by default and only using the sample if R-hat is less than 1.05. Stan reports R-hat which is the maximum of rank normalized split-R-hat and rank normalized folded-split-R-hat, which works for thick tailed distributions and is sensitive also to differences in scale.

The ess_bulk function produces an estimated Bulk Effective Sample Size (bulk-ESS) using rank normalized draws. Bulk-ESS is useful measure for sampling efficiency in the bulk of the distribution (related e.g. to efficiency of mean and median estimates), and is well defined even if the chains do not have finite mean or variance.

The ess_tail function produces an estimated Tail Effective Sample Size (tail-ESS) by computing the minimum of effective sample sizes for 5% and 95% quantiles. Tail-ESS is useful measure for sampling efficiency in the tails of the distribution (related e.g. to efficiency of variance and tail quantile estimates).

Both bulk-ESS and tail-ESS should be at least \(100\) (approximately) per Markov Chain in order to be reliable and indicate that estimates of respective posterior quantiles are reliable.

References

Aki Vehtari, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian Bürkner (2019). Rank-normalization, folding, and localization: An improved R-hat for assessing convergence of MCMC. arXiv preprint arXiv:1903.08008.

Author

Paul-Christian Burkner and Aki Vehtari

See also

monitor

Examples

# pretend these draws came from five actual Markov Chins
sims <- matrix(rnorm(500), nrow = 100, ncol = 5)
Rhat(sims)
#> [1] 1.000929
ess_bulk(sims)
#> [1] 550.2937
ess_tail(sims)
#> [1] 485.0769