Could the polls be wrong? Yup. But they probably aren’t.
Remember, the state of play, at least until Sandy shut down a lot of polling, is that the national vote is basically tied (with Mitt Romney possibly having a very narrow edge) while Barack Obama appears to have a 2-3 percentage point electoral college advantage. What that means is that in a tied race nationally, Romney would have to win basically every close swing state to win. If the polls are correct and nothing changes between now and Tuesday, the chances for an Obama victory are good.
How likely is it that the polls are wrong?
They could be wrong, no question about it. For example: many polls still find better enthusiasm among Republicans than Democrats. This could yield a large turnout advantage for Romney, one which most polls are not picking up, but which Gallup’s likely voters screen has been projecting. If Gallup’s 5 percentage point difference between registered and likely voters is correct, then polls with looser likely voter screens might be harsher to Romney than the actual electorate. That’s one possibility. Another is the possibility that undecided voters will break sharply for Romney. The evidence suggests that “how the undecidedss ‘break’ may not be consequential in this election. But could it be consequential? Sure.
One more possibility: That old chestnut, the Bradley Effect — that is, people who don’t want to admit to pollsters that they don’t want to vote for an African American candidate. No, that didn’t show up in 2008. Could it show up this time? It might.
But the polls could also be wrong in ways that favor Democrats.
If Gallup is wrong, and it’s overstating the difference between registered and likely voters, then Obama is actually doing better than polling averages that include Gallup are showing. Perhaps the undecideds will actually break for Obama. Perhaps the underpolling of people with cell phones means that the polls are overly tilting towards Romney. And there’s some fairly good evidence that polling significantly undersamples Spanish-speaking Latinos.
The bottom line is that any of these effects are possible, but none of them are clearly real; if they were, the pollsters would have adjusted for them! Nor is there any particular reason to believe that the case for the pro-Romney adjustments is any stronger than the case for the pro-Obama ones. And even if one or more is correct, they could cancel each other out. All of which only means that the polls are probably right.
The margins are close enough that any systematic bias could matter. But in the end, the usual result is that the final polls are accurate.