I'm raising this issue because it seems to me that when we find a new question that has an obvious duplicate, it should not necessarily be the new question that is closed.
We already have a good precedent in that occasionally someone will do a self-answered Q&A (actually the only one I know of is Gilles, and Gilles does a lot of the best Q&As on the site -- but I hope at least a couple of other people have tried) addressing a theme that already has one or more potential duplicates, and then elects those older questions for closing. This is a great practice, since we do have oodles of idiosyncratic/localized questions that could really all have the same general answer, if only someone had posed the question properly to start with.
We can edit such questions to start with, of course, except that in many cases the editing required would be very aggressive and I worry about introducing that as a common practice. But maybe that is worthwhile anyway? There are frequently terribly written questions that do still involve a legitimate issue. Perhaps a good policy would be: only do this if you can also provide an appropriate answer. If this is done well, the OP will usually be appreciative, I think.
One of the problems here is that a self-answered Q&A is cut n' dried -- it already has an answer that clearly matches the question, and since the author presumably knew what s/he was talking about, it is general enough to obviously cover the more localized dupes.
Based on that criteria, however, I think it is possible to identify new questions that are much better than old dupes. Too localized is the opposite of too broad; it refers to questions of the sort that involve an excess of irrelevant particular details, e.g.:
On Thursday I was playing around in a subdirectory of
/usr
on my CentOS server and noticed that...[some obviously more general phenomenon]
These questions, particularly when their titles are excessively localized as well, are serious problems because we end up with too much in the way of things that aren't even easily made into dupes, even though one good question could be the one ring to bind them all.
So, my proposal that I present here for your evaluation is: When you find a duplicate to a new question, if the new question is clearly better and more generalized, with less irrelevant detail, and it already also has a decent answer, close the old question instead. If it does not already have a decent answer, try one of the following two options:
If the old question has enough of an answer that you can provide a general answer to the new question (I think this should usually be the case, if you've found a legitimate dupe), then do that and nominate the old question for closing. Remember, this is easy points for you!
If you can't do the former, the situation is obviously a little ambiguous. If the new question is less than 4 hours old, leave it open, and add a comment to the effect that this is the question we should really want as the canonical one. Doing so will also help to disseminate a good, if slightly complex-sounding-at-first, practice.