When voting to close a question as a duplicate, I have always considered all the answers when noting that the new question has already been answered. I don't necessarily pick the accepted answer on which to base my decision to VTC.
Is this acceptable? Or should we look only to the accepted answers when making the decision that a new question is a duplicate? ETA: And if the correct answer is missing a citation, is it appropriate to edit in the citation to improve the answer?
I disagree. Two accepted answers (for two different question) could be nearly identical and yet their respective questions could be dramatically different.
I do agree that identical accepted answers are likely to come from identical questions but that's not the benchmark. VTC for duplicate should be used where the questions are largely the same.
e.g. "Does Harry speak Parselmouth?" is not the same as "How does Voldemort tell Nagini what to do?" even though large elements of the answers may be identical.
If in doubt, there's always the chatroom to hash it out before dropping the hammer.
I think that the only thing that matters is the extent and the quality of an answer, not whether it's accepted or even its votes.
The fact of the closing as a dupe merely means that the information already exists and shouldn't be duplicated.
So, if the question is fully and exhaustively addressed by another question's answer, VTC as a duplicate, no matter whether the duplicate info is in an accepted answer or zero-score one.
Conversely, if the existing answer kind of touches on the info, but it's buried deep inside an unrelated answer; or, even more importantly, if the info isn't really complete (meaning, a new answer to a question you're VTCing is likely to be much more complete and better), then do not VTC - again, without regard to other answers' accept status or votes.
I definitely agree with your approach, but if you're VTC-ing based on an answer other than the accepted, I'd consider providing a direct link to the specifc answer in a comment. That way the asker gets confirmation of where the answer actually is (and therefore gets what they came looking for - an answer) and it seems less likely to provoke subsequent dispute.
