Among the sea of partisan pieces in the days leading up to the election just past, the most interesting non-partisan piece I saw was the following post from October 13th about the so-called "Bradley Effect" -- that is, the idea that polls overpredict how well a black candidate will do, the theory being that people claim to the pollster that they're open-minded when they're really not.
The short version of this article is that, according to the pollster of the person who beat Bradley back in '82, there IS no Bradley effect. According to him, it was simply the result of bad polling by Bradley's people; the other side knew that the race had tightened to more or less even (45% Bradley, 44% Deukmejian) by election day; Bradley got slightly more votes from those who went to the polls that day (and thus those who talked to exit pollsters) but Deukmejian got a significant edge through the absentee ballots, which won him the election.
Here's the link:
Personally, I've always been baffled by the whole idea that racists CARE what pollsters think about them. If anything, I'd expect any misrepresentation to swing the other way (i.e., for racists who intend to vote for minority candidates to not be willing to admit to it in public).