Discussing Ohio on a right-wing blog, I ran into someone who said
this: "When the polls taken in the weeks up to the election, the exit polls, and the actual election results are all taken in to account, it is the exit polls that provide the anomaly."
I wondered if this was true, so I did some checking. I wasn't surprised to find out the statement was false, but I was surprised to find out how false. In particular, the last 23 polls taken before the election, averaged, come within 0.1% of predicting Bush's support, as indicated in the exit poll. Here's more detail.
The final result in Ohio was 51% for Bush.
Here you'll find 76 Ohio polls, covering roughly 9 months prior to the election. In 67 of these (88% of the polls), Bush scores lower than 51%. The last time he scored higher than 51% was in September.
If you restrict the analysis to the 23 polls taken 10/12 or later, there's not a single instance where Bush manages to reach 51%. The highest he manages is 50.1%. His average in this group of polls is 47.8%. Amazingly enough, Ohio exit polls showed Bush at 47.9% (see pdf here). In other words, conventional polls indicated a result for Bush virtually identical to what was indicated by the exit polls. But somehow his final tally was more than 3% higher. Hmm.
Maybe you'd rather look at spread. The final spread was 2.5%. In 17 of the final 23 polls leading up to the election, Bush fails to achieve a spread that large. If you average the last 9 polls immediately before the election, he fails to achieve a spread that large.
In making the case that the exit polls should be taken seriously, it could be helpful to point out this example of the exit poll being so precisely corroborated by earlier polls.