Possible Problems With Exit Polls:

Mystery Pollster has an interesting discussion of this, building around one incident -- I'm not an expert on the subject, but it struck me as quite informative. Thanks to Mickey Kaus (Slate) for the pointer.

Ilya Somin:
There are certainly methodological problems with exit polls, as with other types of polls. But it is worth remembering that the much-derided early 2004 exit polls had Kerry winning the popular vote by about 2-3 points, whereas in reality he lost by about 3 points. This difference is well within the stated margin of error exit polls (and most other surveys).

To expect exit polls to give highly accurate predictions on the outcome of a very close election is to expect them to do something that they simply cannot do. Polls can be very useful for studying voter intentions and attitudes in situations where a 2-3 point error is unimportant to the analysis. Their usefulness is extremely limited if small errors do make a crucial difference.
8.17.2006 7:29pm
James Lindgren (mail):
At NORC and the Univ of Chicago, they taught us that student interviewers get (as I may imperfectly recall) about three times the level of variation in responses as older, nonstudent interviewers.

Jim Lindgren
8.17.2006 10:18pm
NickM (mail) (www):
Exit polls break down when it comes to predicting the behavior of absentee voters - you can phone poll people who you know to have requested absentee ballots (if you have their phone numbers), but the margin of error for such a poll or portion of poll is rather high - especially as many absentee ballots are mailed to addresses other than the voter's registered address.

In at least one recent Congressional special election I am aware of, a majority of votes were cast by absentee ballot.

Republicans have traditionally had a higher percentage of their voters vote absentee.

Is anyone really surprised that exit polls skew Democrat in their results?

8.18.2006 12:01am
Richard Aubrey (mail):
It is an interesting speculation that the character of the pollster person might have skewed the results. While the plural of anecdote is not data, the individual datum points are, actually, anecdotes.

So we can reasonably speculate on the larger impact of this issue. Who are you going to get to do this stuff? Best might be a retired life insurance agent, freshly barbered, wearing knife-creased dockers and a starched button-down shirt, and with a friendly, non-intrusive way of getting in everybody's face. But, why would he agree to do this? He's retired and can think of better things to do. If he needs money, he can bug somebody for life insurance.

Instead, you're more likely to get some version of this kid; a rumpled, diffident college kid who doesn't really want to be there. Maybe he did, but now that he is doing the equivalent of street-corner evangelism, he's wringing with flop sweat and wishing his shift were over. He won't be intercepting people who are legging it determinedly for the parking lot.

How many times do we get a scenario like this, and what would the impact be on the results?
8.18.2006 8:23pm