paul-buerkner/brms

add meaningful convergence statistics to summary.brmsfit() for class 'brmsfit_multiple'

Open

#582 aperta il 4 gen 2019

Vedi su GitHub
 (2 commenti) (0 reazioni) (0 assegnatari)R (220 fork)batch import
featuregood first issuepost-processing

Metriche repository

Star
 (1402 star)
Metriche merge PR
 (Merge medio 7g 21h) (1 PR mergiata in 30 g)

Descrizione

The output of summary.brmsfit() for objects of class brmsfit_multiple reports an Rhat statistic that aggregates across chains fit to different imputed datasets, often giving rise to a false positive indication of non-convergence. Although this behaviour and the way to check dataset-level convergence is well-documented in a package vignette (ht: @paul-buerkner), it'd be nice to have some indication of these diagnostics in the summary output (and a more nuanced warning when some aggregate Rhats > 1.1).

Perhaps the least intrusive would be to add a max_Rhat column to the summary output? It would signal non-convergence clearly and succinctly with low likelihood of breaking changes for other functions (e.g. tidiers in broom and it's relatives).

Guida contributor