For those that want a more detailed technical explanation of the truth behind exit polls, I ask you to go read
this article from a person who teaches statistics. But I'll do my best to summarize it.
In normal polling, you call people randomly, and you call enough of them that you can be reasonably confident that the sample reflects the overall population.
That's not how exit polls work.
For one thing, when they visit the precincts, they don't select them by random. They pre-select the precincts. They also pre-decide how many people to ask at each precinct. But, what they don't know is how many people will vote at each precinct. They don't know this until the polls close. This means that exit polling isn't truly random. And you NEED random samples for them to be representative of a total population.
So, this basically means that - by design - the raw mid-day exit polls are badly weighted. Remember how we were all discussing pre-election polls that weighted Republicans too much? This is the same sort of thing, except that they don't even have a way of knowing where the weights are bad, until the polls closed. It might give too much weight to women (as it did), or too much weight to minorities, or to people who go to church three times a week. But these are not flaws in the exit poll data - this is actually expected behavior, because they are using a completely different statistical technique.
The whole point of exit polls are to get detailed demographic information of a subgroup of the voting population. Then, after the polls close and we get an idea of real numbers, THEN the exit polls can be weighted properly - normalized - to create accurate data.
In short: professionals expect the mid-day exit polls to be innacurate. That's how they are designed. Raw, mid-day exit poll data is not even intended to be a representative sample of the voting population.
Therefore, studies that rely on pre-closing exit poll data are flawed by definition. It's just not how the numbers are supposed to be used. It doesn't make mathematical sense to directly compare mid-day exit poll data to the final results of an election.
