Ah, theories abound.
1. There was a conspiracy, somehow, because pre-election polls are just so much more valid than actual vote counts.
2. The Bradley/Wilder effect -- voters were afraid to tell pollsters they didn't want to vote for a black person, so a certain percentage of them lied about their preferences. But wait -- the pre-election polls did NOT overstate Barack Obama's support. He averaged 36.7%, according to Mark Blumenthal's compilations.
If anything, they understated Hillary Clinton's support by nine points. Let's name this phenomenon the "Dorothy Zbornak" effect -- for some reason, older women voters refused to disclose their preferences to pollsters, or refused to admit that they favored Hillary Clinton.
Mickey Kaus postulates an application of the reverse Bradley effect -- that Iowa Democrats somehow felt social pressure to stand up in front of their peers and cast a vote for a viable black candidate.
3. The exit polls were wrong. -- No -- by the third wave, they were basically correct. In Iowa and New Hampshire. So something is working for Edison-Mitofsky Research.
4. Ballot placement helped Clinton. Probably true, but it does not account for the discrepancy.
5. Obama actually won New Hampshire because both he and Clinton were allocated the same number of delegates (true) and he has more New Hampshire superdelegates than Clinton (true). Cute. But by that score, Clinton has a national lead because she has more pledged superdelegates nationwide than Obama.
6. Feiler's Faster Thesis with the Skurnik appendage. That is -- voters process information rapidly -- and they process through the information even more quickly. And uninformed late deciders usually rapidly assimilate and process late-breaking news. Again, from Mickey Kaus. Combine this with some evidence that working women are unusually late deciders...