Hacker News new | ask | show | jobs
by tischler 3074 days ago
As a reminder:

true positive rate (TPR, a.k.a. hit rate, recall, specificity)

TPR = TP / P = TP / (TP + FN)

specificity (SPC) or true negative rate

SPC = TN / N = TN / (TN + FP)

1 comments

You labelled both TPR and TNR as "specificity". True positive rate should be sensitivity.
Oops, thanks. I've always had struggle remembering these terms because they are so confusing. "Specific" means that something only relates to one particular element X in a set of more than one thing. Given that meaning, one would expect specificity to mean "how good is the test at correctly reporting the thing it is designed to report X rather than confusing a different thing (an element from the complement of X) for X", but that is sensitivity. Instead, specificity means that with respect to the complement: "how good is the test at correctly reporting all the other elements in the set ¬X as negative rather than confusing an element from ¬X for X".

"Sensitive" has a clearer intuition: For example when you want to detect a weak signal, then if your antenna is more sensitive, you can better pick up the signal, thus you are better at telling whether the signal is really there (you have a better true positive rate).

I cannot think of a better name for TNR, though, other than simply sticking with TNR and TPR.

Your sensitivity intuition is a fantastic way to remember it. I had to look it up myself, but hopefully this will finally make it stick :)