However, I do have a reservation regarding the method employed for calculating the credible interval. The paper uses
the quantile() function for this purpose. While quantile-based intervals are similar to Highest Posterior Density (HPD)
intervals in cases where posterior distributions are unimodal and symmetric, they differ when distributions are not
symmetric. In Figure 8, the case of prior_family = "npmle" appears to fall under the non-symmetric category.
Consequently, the application of quantile(fit_npmle, probs = c(0.1, 0.9)) on page 24 might not yield an accurate
HPD interval. This aspect warrants further clarification or revision in the paper.
I removed this part from the mauscript (Sec. 5), but it is still in the baseball vignette baseball.Rmd in the R package.
I don't remember exactly what @stephens999 suggested (you probably remember better), but from what I recall he suggested not referring to these intervals as "HPD intervals", but instead say that the function returns the "shortest intervals".
Correspondingly, we should fix help(confint.ebnm).
@willwerscheid Regarding this reviewer comment:
I removed this part from the mauscript (Sec. 5), but it is still in the baseball vignette
baseball.Rmd
in the R package.I don't remember exactly what @stephens999 suggested (you probably remember better), but from what I recall he suggested not referring to these intervals as "HPD intervals", but instead say that the function returns the "shortest intervals".
Correspondingly, we should fix
help(confint.ebnm)
.Can you please make these changes in the restructure-vignettes branch?