Two quick questions:
Are the confidence intervals calculated for sensitivity, specificity, PPV, and NPV asymptotic or exact?
What method is used for estimation if no false positives occur?
If you are referring to the MCMC add-in, the equal-tailed credible intervals for 95% take the 2.5th and 97.5th percentile of the posterior samples as the lower and upper bounds, respectively. The HPD intervals are the narrowest intervals that contain 95% of the samples.
Thank you for reply.
Actually I was referring to the Medical Diagnostics add-in credited to Karen Copeland. I believe it has a 2012 date on it.
I wondered if you were referring to my add in. It uses score confidence intervals for Se/Sp, etc. (so the same CI you would get if you used the distribution platform for each measure individually - with the right counts) and for the likelihood ratio confidence intervals are based on the method in Simel, et. al. (J Clin Epidemiology 1991).
NOTE the version of the script that is on the file share was written for JMP10 and has buggy behavior in JMP11. I am working on an update that is getting very close to being ready for sharing.
Note to JMP developers - the need for such a script (other than summarizing everything at once in a nice way for a particular field) comes from needing confidence intervals after running a contingency table and then wanting confidence intervals on row or column percents.