I've used the basic formula :
95% confidence interval = sensitivity +/− 1.96 (SE sensitivity)
Where SE sensitivity = square root [sensitivity – (1-sensitivity)]/n sensitivity)
95% confidence interval = specificity +/− 1.96 (SE specificity)Where SE specificity = square root [specificity – (1-specificity)]/n specificity)
But this is giving me values that are greater than 100%, and being probability 100% is upper bound. Any suggestions? This is for diagnostic tests
Confidence intervals for a binomial proportion are cursed. The textbook formula gives nonsense results because it doesn't recognise that proportions are bounded by (1,0), so you can also get negative proportions, which are also meaningless.
You should have a look at this paper.
http://stat.wharton.upenn.edu/~tcai/paper/Binomial-Annals.pdf
Stata and R are both capable of calculating Wilson or Jeffrey's confidence intervals, which generally perform better than the alernatives.
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases