Silver has argued that it is highly unlikely that a candidate who is 3 points behind in a very highly polled race (e.g., Romney in Ohio) can win that state. But in 2010, Harry Reid won his re-election to the Senate by more than 5 points, though the polls (and there were many of them) all pointed to a 2-3 point defeat for the majority leader. Silver also missed on the Colorado Senate race in 2010 and was behind the curve in predicting the GOP wave in the House that year.
Surprises happen. Polls can be wrong, particularly when votes are cast over a period of a month or more in some states, and when the percentage of those who respond to polls has dropped to an all-time low (below 10% in human phone surveys; below 5% in automated surveys). Many national and state polls have samples with a large edge for Democrats compared to Republicans, similar to the voting results in 2008 when Obama won nationally by 7.3%.
The latest CNN poll, which shows a 49-49 national tie, contains 41% Democrats and 29% Republicans; the only way you can get a tie is with Romney holding a big edge among independents (59% to 37%). Count me as highly skeptical of both the party ID split in the CNN poll and the size of Romney lead among independents.
For Romney to prove the pollsters wrong, there are a few paths to victory. If he carries North Carolina (quite likely, though not certain) and Florida (where he is favored but could lose if there is an Obama national win by 2% or 3%), then he would be at 235 Electoral College votes and would need to find 35 more to get to 270 (and 34 to get to 269, which would almost certainly be enough if the Electoral College race were tied at 269 and the election were tossed to the U.S House). The path to victory has increasing degrees of difficulty by state, with Virginia and its 13 Electoral College votes the easiest (polls are about even on average) to get Romney to 248.
Obama is ahead by 2 points or less in New Hampshire (4) and Colorado (9). Winning both would get Romney to 261. Obama’s lead is about 3 points on average in Ohio (18), Nevada (6), and Iowa (6). Pennsylvania (20), Michigan (16) , Wisconsin (10), and Minnesota (10) are all showing Obama with a 4 to 5 point lead on average, though individual polls in each state have Romney either tied or 1 point ahead. Wisconsin might be the best shot for Romney of the four given the strength of the get-out-the-vote effort put together by Governor Scott Walker. In Pennsylvania, the Romney campaign and its super PAC backers are heavily outspending Obama in the last week.
At the moment, the Intrade betting line favors Obama by just over 2 to 1. Nate Silver’s model suggests the odds are more like 6 to 1 or 7 to 1 against Romney. I would take the Romney side with Silver’s odds if he offered them (which of course he does not), and I think the Intrade numbers better reflect the current state of the race. Obama is favored, but it should not be a total shock if Romney wins and carries Ohio to do it. I do not share the optimism of George Will, Michael Barone, or Dick Morris. These pundits see Romney winning comfortably with over 300 Electoral College votes. These forecasts suggest the current national and state polls are off by 3 to 5 points, if not more.
I think the oversampling problem exists, but the average error is not that high. The Obama team is counting on a national voting sample that is only 72% white. Even if Romney wins white voters by 20%, he will lose by more than 2% overall if Obama win by 60% among the other 28% of voters. A turnout with 75% white voters, and the same preference splits, produces a tie in the popular vote, which might not be enough for Romney given the Electoral College math. Romney may need to win by more than 20% among white voters, or hold Obama’s edge among minority voters below 60%, if the final split is 75-25 white/minority or below 75%.