According to polls, former Gov. Mitt Romney is supposed to run away with the New Hampshire primary with more than 40 percent of the vote. Wait, his numbers are down in a tracking poll. Except that, oops—they have crept up a bit, even after Romney made an unfortunate reference to enjoying firing people (he was talking about people selling him a service, but the remark stung).
And if Romney does not finish close to where pollsters put him in the last week of the campaign, the critics will dive in again: what's wrong with polling matrixes, or what's wrong with pollsters? The questions were louder in 2008, when Hillary Clinton scored what was seen as a surprise victory over Barack Obama, who had just won a somewhat-surprise victory in Iowa. In fact, to those who knew and know New Hampshire, the results weren't all that much of a shock. Clinton had strong ties and longtime loyal supporters in the state, going back to when her husband, former President Clinton, broke the historical "rule" of New Hampshire's predictive power and won the presidency despite not having won the Granite State primary.
There's nothing wrong with polls or pollsters, except that we place far too much importance on their findings. Polls are by definition incomplete assessments of the electorate, both because they reach a small percentage of the electorate and because they are taken ahead of Election Day. What was remarkable about a Suffolk University tracking poll on Monday was not that Romney was at "only" 33 percent (he's back up on Tuesday), but that a full twelve percent of those polled said they were undecided. That's a big chunk of the electorate to still be making up their collective minds so close to the election, but it's a reason why the polls aren't always predictive of final results.
Things can change quickly, especially in a crowded field. That's one reason Clinton ended up besting Obama in New Hampshire. The same is true of Republican Sen. Scott Brown's victory in the special election in Massachusetts in 2010: it's not that the press and pollsters "missed" Brown's ascension. It's that it happened quickly, and right before the election.
Polls can be useful indicators of the mood of the country or of a state—but only for that moment, given the unique circumstances of that particular day. There's only one poll that matters, and that's the one that happens on Election Day. That is, after all, why we vote.