I gave the OP an answer to what I thought was his question. However, I didn't realize he wanted embedded submanifolds rather than immersed ones. Thus, my answer is invalid. I didn't know whether to delete it. Incidentally, I think the result I pointed to is rather interesting in itself, just irrelevant here.
I was curious what the protocol was when one has answered something badly, especially when there are comments below. For now, I just made an edit telling people that the answer did not address the question.
If others read the question, would they want to see your answer or not? If it would waste their time to read even an edited version which points out the flaws, tend to delete it.
It's particularly important to delete an unhelpful answer if it is the first answer to a question. Unanswered questions may get more attention. Also, you can only get the "enlightened" badge if your answer is first and well-received. That shouldn't be too important, but it's an extra reason to delete an answer, so that you don't block someone else from having the first answer.
If the comments others make on an answer are particularly insightful, then lean toward trying to salvage the answer.
@Jonas: Whoops, you're right. I think this was a slightly different issue than David Speyer's, though.
@Douglas: I agree and have deleted it.
Interestingly, the answer by José Figueroa-O'Farrill seems to be along similar lines (at least, the terminology "analytic subgroups" is the same). In any case, his is a much better answer than mine was, so I do not regret having deleted it.
@Akhiil: I was wondering why you deleted your answer. You linked to a paper I didn't know and I think it would have been useful to leave it. As it it turned out, I too misunderstood the question and failed to realise that the OP wanted to restrict to embedded submanifolds. I am about to edit my answer to point this out, in fact.
I put it back but as a community wiki answer and with an explanation that it was not technically relevant, but answered a related question and had a useful reference. Do you think this is better?