As I noted this morning the New Hampshire polls didn't actually get the level of support for Barack Obama wrong. Instead, they undercounted Hillary Clinton's supporters. How'd that happen? Jay Carney has friends of friends who have the answer:

He didn't find any evidence that white respondents who were telling pollsters they planned to vote for Obama did not. What he found, instead, is that a certain percentage of Democratic voters in the last days of polling presumed Biden (especially) and (to a lesser degree) Dodd hadn't dropped out. By and large, come election day, those Biden and Dodd supporters ended up casting ballots for Hillary. Also, of the 5 percent or so who were still undecideds in the last polls, almost all broke for Hillary. And a tiny percentage of Edwards supporters switched to Hillary.

When you think about it, this makes perfect sense. Obama got 38 percent of the vote in Iowa. Not only is Iowa only one small state, but 38 percent of the vote is way less than half. Nevertheless, based on that plurality he was about to march to the nomination. As a result, while Obama continued to hold his own in terms of his baseline level of support, all the uncommitted people -- supporters of minor candidates, undecideds, some soft Edwards people -- voted for Clinton to keep the race going. In Iowa, a similar dynamic probably helped Obama. People knew that a Clinton win might end the competition, so Obama can, so to speak, the benefit of the doubt. Unlike most political bloggers, most voters haven't been following this thing since the first quarter of 2007. A lot of them want to see how the competition plays out.