inst/bootstrap.md

In vz-risk/verisr: Tools for working with VERIS objects

NOTE ABOUT bootstrap option for confidence intervals

verisr uses the binom package for simple bootstrapped confidence intervals (as a faster alternative to MCMC generated ones.). However, the binom package has a bug in the binom.bayes function (the one we use). An example can be seen below

test2 <- dput(data.frame(x=c(0,6,151,4), n=c(1699, 1699, 1699, 1699)))
#> structure(list(x = c(0, 6, 151, 4), n = c(1699, 1699, 1699, 1699
#> )), class = "data.frame", row.names = c(NA, -4L))
binom::binom.bayes(test2$x, test2$n)
#> Warning in binom::binom.bayes(test2$x, test2$n): 1 confidence interval failed to converge (marked by '*').
#>   Try changing 'tol' to a different value.
#> Error in `$<-.data.frame`(`*tmp*`, "method", value = structure(c(1L, 1L, : replacement has 3 rows, data has 4

Looking through the binom.bayes function (binom version 1.1-1), it appears to be the use of ci$error rather than error in the line

res$method <- factor(sprintf("bayes%s", ifelse(ci$error, "*", "")))

is to blame.

ci$error has had the ends removed from it while error is the full length of the input vector.

Unfortunately, this fix will require the package author to update the package (or me to clone it). The package dates from January 2014, and so while I have asked for an update from the package author, I can't be sure it'll be received and acted upon.

vz-risk/verisr documentation built on Aug. 5, 2023, 4:34 a.m.