I came across this question and answer today: Recall and AUC of a binary classifier
And I think the accepted answer is wrong. So I wrote what I believe is a better answer, and downvoted the accepted answer explaining why I think it is wrong.
side note: if my reasoning went astray please correct me
side note 2: I actually think in terms of content it will be easy to fix the accepted answer and I also left a comment that I'll gladly upvote a revised version. I don't think it will be a matter where the author of the accepted answer and I cannot reach convergence - I think that he just didn't think of a particular situation which makes the "can never happen" answer incorrect.
It is not like trench wars about the correct prior or appropriateness of Bayesian vs. frequentist approaches
The concept of stackoverflow relies on the fact that other users can up- or downvote answers. Ultimately this should lead to good (including correct) answers floating on top. This concept also assumes that in case some issue discovered, all sides can react: the author of the answer can edit and the OP can even change the accepted answer. This relies on people staying active on CV/SX, but what if they aren't active any more?
Question and accepted answer are 1 1/2 years old. The OP's profile says that the OP never came back to the site after the day that question was asked and "answered".
(The author of the accepted answer was online here more recently (2 weeks ago) so he/she may edit the answer in reaction to my comment.)
I'll flag the question for moderator attention, but I think it would be good to write down here how to proceed
- in case the OP doesn't show up again and react on the new situation.
- in future if similar issues arise (the OP could have deleted his/her account meanwhile)
To be even more clear: I think it is a perfectly valid policy to say that we don't do anyting in these situations.
I just think that it would be good to have the policy discussed and spelled out here (whatever it is).