Reply #54: it's actually kind of a cheap response, because...

[OnTheOtherHand]

it hinges on the interpretation of "raw." If memory serves, one of the 1988 exit polls (CBS, I'm pretty sure) did have more Dukakis respondents than Bush respondents, but that apparently is because of a deliberate oversample in black precincts. The model projections were closer to the outcome, although the network did subsequently blow one state call. I think CBS blew Illinois, and another network blew Maryland. But overall, the 1988 exits were less discrepant than the 1992 exit poll.

Ruy Teixeira wrote a little "President Dukakis?" blurb soon after the election, and it made a good point, but it doesn't tell the whole story.

Many people are under the impression that they saw, or have since seen, "raw" results from the 2004 exit polls as posted on cnn.com. Those never were "raw" results. They are based on a composite model which incorporates demographic and probability-of-selection weights, past results, and pre-election expectations.