Update: Talked to my friend earlier and between her, I, and one of her friends, we have three theories for this 2%:
1. The 2% (N~15 people) are strategic Republican voters who voted for Dean just because they thought a vote for him was the most subversive act. Or, if not strategic Republican voters, at least Republicans who went to the polls with their tongues in cheek.
2. The 2% are products of sheer errors in recording by exit poll interviewers or by the interviewees. This is my friend from Beauxbaton's theory, and if she reads this update she will likely be annoyed at the way I've stripped her idea of all nuance, especially when I go into detail about my own:
3. The 2% are people who had just made up their mind to vote for Dean and misunderstood the question as being about having had supported Dean for something longer for the day before the election. This was actually my theory. It's vaguely buttressed by these various analyses I have done of peculiarities in how respondents answer questions with "ever" in them and how they answer questions that seem like pose logical entailments with previous questions. With regard to the latter, I've long thought that all the failures of people to follow logical entailments when explicitly asked was simply a matter of people not being so logical as we might otherwise imagine them to be. More recently, however, I'm coming to suspect that part of what may happen is that if you give a person a question that might otherwise read like its answer is obviously logically implied by the preceding question, this may lead them to try some alternate reading of the question.
I mean, imagine a real conversation along these lines:
"Who did you vote for?"You can see where some people (say, 15 or so) might think that you are meaning 'support' in some sense beyond just voting for him.
"So, have you ever supported Dean's presidential bid?"