In the latest Gallup tracking poll, Mr. Obama leads Mr. McCain 50 percent to 43 percent among registered voters. Mr. McCain’s deficit in that survey has remained seven percentage points or more for most of the last two weeks.There were 18 elections between 1936 and 2004, and in just one of those -- the 1980 race that Harwood mentions -- did a trailing candidate come back from a deficit this large in mid-October to win the election. One divided into 18 is 5.6 percent, which almost exactly matches our 5.9 percent estimate for Mr. McCain.
Since Gallup began presidential polling in 1936, only one candidate has overcome a deficit that large, and this late, to win the White House: Ronald Reagan, who trailed President Jimmy Carter 47 percent to 39 percent in a survey completed on Oct. 26, 1980.
There has also been at least one other election in which a candidate made up at least 7 points worth of ground this late, albeit in a losing effort. That was 1968, when Hubert Humphrey had trailed by 15 in Gallup's poll in early October, and 8 points in late October, but wound up losing the election by less than a point. (If you want to see all these numbers for yourself, by the way, Gallup has them here).
Gerald Ford in 1976 also made up significant ground in his re-election bid with Jimmy Carter, but most of that came in August and September. By the first few days of October, Ford had already cut Carter's advantage by 2 points -- the margin he eventually lost by. Ford than made another mini-comeback after Carter's lead expanded again to 6 points, but it wasn't enough to save his re-election bid.
Harwood also mentions Al Gore's comeback in 2000, although that is harder to evaluate since the Gallup poll was exceptionally erratic that year (Gore trailed Bush by an average of 3 points over all polls that Gallup conducted that October). The Pew poll, which was far more stable, showed Gore with small leads in early- and mid- October, although Bush had pulled 2 points ahead by the end of the month.
If 1980 and 1968 do offer a couple of favorable precedents for McCain, they also come with some caveats. If 1980 is the template, it's not clear which candidate gets to play Ronald Reagan, who on the surface would seem to share more circumstances in common with Barack Obama. Although it's relatively uncommon for a candidate who is already ahead to further build his lead in late October (1936 and 1988 fit this definition, but only to a degree), there is nevertheless no guarantee that the next large momentum swing -- if there is one at all -- will favor McCain. And secondly, McCain could very easily come close without winning. The chances are significantly greater than 5.9 percent that McCain will come close enough to make Obama sweat, but like Ford or Humphrey, he might wind up a little short.
Ford, Humphrey, and Reagan, also, did not have to deal with early voting, whereas McCain is pushing back against the fact that Obama is banking votes every day with a substantial national lead. And McCain's deficit in the key battleground states exceeds that in the country as a whole, such that Obama, by our math, has the equivalent of a 1-2 point buffer zone in the Electoral College. If he were to come back, McCain's fate could very easily resemble that of Humphrey, who lost the popular vote by just seven-tenths of a point to Richard Nixon but was beaten handily among the 538 electors.
EDIT: As several commentators have pointed out, my math is a bit misleading since in not all elections did a candidate have a lead of 7 points to lose. Another way to reach the same number, however, is as follows. In two elections -- 1968 and 1980 -- did a candidate make up at least seven points' worth of ground versus where he stood in the Gallup poll with three weeks left to go until the election. That is 2 of 18, or 11.1 percent. However, if McCain were to gain 7 points on Obama, that does not mean that he'd win the election -- it means that he'd be roughly tied, and that we'd play electoral roulette. So perhaps McCain has an 11 percent chance of having a 50 percent chance of winning, which works out to the same 5.6 percent.
All of these approaches, of course, are fairly quick-and-dirty. The more rigorous way to do it is the way that our model does it, which uses just this sort of historical data to build what amounts to a margin of error on the current popular vote estimate. What I am trying to make clear, however, is that my numbers are not drawn out of thin air: it would be quite unusual (though it is hardly impossible) for a candidate to overcome a 7-point gap in mid-October and come back to win the election.