Bootstrapped Confidence Intervals for (Negative) Likelihood
Computes appropriate confidence intervals for the likelihood ratio tests commonly used in medicine/epidemiology, using the method of Marill et al. (2015) <10.1177>. It is particularly useful when the sensitivity or specificity in the sample is 100%. Note that this does not perform the test on nested models--for that, see 'epicalc::lrtest'.10.1177>
bootLR v1.0 (Release date: 2015-07-10)
- Many minor fixes
- Replaced a grid search with a direct application of the binomial distribution, yielding marked speed improvements
- Reduced the number of bootstrap replicates to 10,000 (from 50k), but added an outer loop to average 50 bootstrap estimations, yielding speed reductions but more stable estimates.
- This is the initial full release, and coincides with the release of the corresponding methodological paper in SMMR: http://dx.doi.org/10.1177/0962280215592907