Or, rather, defends state-by-state polling quite convincingly:
Of the 77 states with at least three late polls, the winner was called correctly in 74 cases. (I exclude Missouri in 2000, where the polling average showed an exact tie.) There has been little tendency for the state polling averages to overrate either Democrats or Republicans, or either incumbents or challengers. The state polls also performed fairly well in two years, 1996 and 2000, when the national polls were somewhat off the mark.
The chances of a miss are higher, of course, when the polls show a closer race. Even among the 33 cases where the final polling margin in a state was within five percentage points, however, the polling average identified the winner correctly in 30 cases.
Mr. Obama is at about 50 percent of the vote in the polling average in Pennsylvania, Wisconsin, Nevada and Michigan; at close to 49 percent in Ohio; and at about 48 percent in Iowa, New Hampshire, Virginia and Colorado.
There are not really any recent precedents in which a candidate has led by something like 49 percent to 46 percent in the final polling average, as Mr. Obama does now in Ohio, and has wound up losing the state. That does not mean such misses cannot or will not occur: there have only been a few elections when we have had as much state polling data as we do now, which is why the model allows for the possibility of a 1980-type error based on how the national polls performed that year.
But the reasonably high level of confidence that the model expresses in Mr. Obama’s chances of winning Ohio and other states reflects the historical reality that the polling average normally does pretty well.
There is only one way Mitt Romney wins on Tuesday: If the state polls are as biased towards Democrats this year as they were towards Republicans in 2010. Most swans are white, but black swans do exist.