Reporting credible/confidence intervals


#1

Hi, I’m just hoping to clarify what the 95% intervals in Stan’s summary output refer to. These are the intervals obtained after calling summary() on a stanfit object. Are the intervals defined by 2.5% and 97.5% percentiles quantiles from the distribution? Or obtained using mean +/- 1.96 * sd (I guess this is not the case)? Or are they High Density Intervals?

And maybe a relevant question, what is the suggested way of reporting credible intervals following a Bayesian modeling analyses; reporting 95% quantiles or reporting 95% HDIs?

Thanks!


#2

yes

Suggestions differ. In general, different intervals tell you different things. Central credible intervals are certainly the easiest to compute.