Just curious if there’s any reason it was decided to have rstan::summary() show quantiles rather than highest posterior density intervals?

My guess is that it’s 1) easy 2) purely descriptive and 3) HPDI does not always yield one set of intervals (e.g., when bimodality is present), which complicates output. Using HDInterval on stan samples is relatively simple.

Quantiles vs HPDI from `rstan::summary()`

Interfaces RStan

avehtari May 29, 2019, 9:41pm 6

. 4) HPDI is not invariant to transformations

3 Likes

Testing an "omnibus" effect of a variable across the output categories

Topic		Replies	Views
Divergences and Quantifying HPD intervals Modeling	7	958	October 13, 2020
Rstantools extension Interfaces	1	339	May 14, 2018
Proposal: including a "canary" variable to illustrate poor exploration of the posterior General techniques	11	699	June 15, 2020
Multi-modality of posteriors Modeling fitting-issues	3	721	December 10, 2018
Comparing the posterior with the prior distribution RStan rstan , techniques	1	877	July 22, 2020

Quantiles vs HPDI from `rstan::summary()`

Related Topics