ROC curve induced madness

Calling all fellow egg heads

I’m getting obsessed with ROC curves.

Why is the horizontal axis of specificity flipped to 1 – specificity (false positive rate) rather than just keeping it simple as in (sensitivity versus specificity)?

I’m guessing it’s something to do with history and convention but can’t find the answer. Please can someone put me out of my misery!!


Omega Contributor
I am also pseudo-obsessed with ROC curves. What program do you use?

I believe the ROC came out of the use of Bayes Theorem in the Pacific during WWII. My guess would be that you want your sensitivity to be as high as possible and your false positive rate as low as possible as well. And that just made better sense to them given their content (signal detention). I would also wonder if they even used the same terms SEN and SPEC? Let us know if you come across the original stats paper introducing it - that would be an interesting read.

Just for fun, I just flipped the values on the x-axis and it is obviously a mirror image of the ROC curve, however it just did not feel right :)