Perhaps there is something in the nature of primaries that make them more unpredictable than general elections? People deciding to vote or making up their minds at the last minute? During a general election, I suspect most people make their decision pretty early in the process while in a primary, especially with the nominations as wide open as they are in both parties this year, people are changing their minds frequently. From my discussions with friends, it seems that a lot of people are losing their enthusiasm (what little they had)for Clinton. Much of her strengths was predicated on the seeming inevitability of her nomination and now that it doesn't seem inevitable, the ship is spinging leaks. Bill's campaigning is dredging up memories, not so pleasant, of the last Clinton years. This may be causing people to move away from her and toward Obama at the last minute.

The problem is that Obama is messing with all of their assumptions on turnout. He creates massive but uneven changes in turnout from their baseline assumptions (and not only in his own supporters). Since they can't accurately predict those changes in turnout, they can't accurately sample.

Kaus has been all over this, and suggesting its the 'Bradley effect'. Named for Tom Bradley, former (black) mayor of LA, who lost a Cali gubernatorial race (to an Armenian) despite polls showing he was leading. The original hypothesis was that some whites were wary of telling a polster they wouldn't vote for an African-American. Interestingly, this can be generalize in all sorts of ways. In Iowa, Obama would have gotten votes (according to the hypothesis) because it was a public, show of hands vote (people afraid to vote against the black guy and be labelled 'racist'). In South Carolina, some blacks might have suffered the conventional Bradley effect, afraid to be seen as 'race men' and 'race women' , thus telling polsters they would vote for Hillary.

Of course this is testable in all sorts of ways.

Sebastian's right. The Democratic turnout in the early contests is so far up that it renders obsolete all the previous models used by the pollsters. They've never beaten the Republicans in turnout in New Hampshire before. It has been a generation since they beat the Republicans in turnout in South Carolina.

The turnout theory is interesting. Understandable for black voters in SC, but why NH, which as I understand it has few African-Americans? Can't be the Obama effect, if polling was showing him winning. Could it be an anti-Obama effect?

Some time back I decided that I was tired of people calling me and taking my time to give me multiple choice answers that "most closely" resemble my opinion without allowing me to actually express my opinion, and knowing that they were scoring this multiple choice thing on some formula that I was not privy to..... So, I decided that mendacity in the face of polling is a virtue. I wonder if others have come to the same conclusion, thus leaving all polls as questionable.

Can anyone explain to me exactly what the value of these polls are? They are so often dead wrong and people tend to formulate their opinions on them. How many times have you hreard someone say, I like Biden but he can't win so i'm backing Barack, or I like Ron Paul but he is polling low so I am voting for someone else? I really think the polls corrupt the process for the sole purpose of giving news analysts something to talk about.