Now that Hillary has won New Hampshire, there has to be a sinister reason. According to Matthews, Eugene Robinson and others, it is the Bradley effect. The idea that whites lie to pollsters and then go and do something different behind the curtains. There is a problem with this analysis: the facts.
Obama was averaging 38 points going into the election, and more or less go that amount. The difference in the election is the surge Hillary had. If people had lied to pollsters, His numbers would have dropped. Matthew Yglesias explains it well:
Commenter Brian makes an observation "No one is talking about how the polls actually nailed Obama's number. Obama didn't lose this election. He stayed steady and Hillary surged ahead." That seems to be true. Here's a chart comparing the actual results to the most recent Pollster.com current standard estimate polling average.
Just as Brian says, the difference between the Obama poll level and the Obama vote total level seems to just be your basic statistical variance. The pollsters underestimated Clinton's level of support. People who were undecided as of the last round of polling seem to have gone overwhelmingly in her direction.
We just out worked Obama's team on the ground in the face of an avalanche of negative media. This surge wouldn't have been possible in Iowa because it was a caucus where people needed to show up at a certain time and stay for hours. Hillary's based vote were workers, who probably had to WORK, and older women who weren't going to go out for hours on a cold night for ANYBODY.
This election season may be known for how much people who know are completely wrong.