I try to take the polls with a grain of salt, and I have tried particularly hard this year because of the distinct possibility that Senator Obama’s candidacy could unsettle some traditional assumptions about voter turnout and polling bias. The race has seemed over for a couple of weeks now, but there was always the chance the polls might be very, very wrong. And I should say up front: There still is that chance.
But it is hard not to marvel at this little item from FiveThirtyEight.com: Early voters in five swing states seem to have gone for Senator Obama by margins that are from 17 to 39 points more favorable to him than the polling results for non-early voters:
... Poll % Voted Non-Early State Date Early Early Voters Likely Voters ==================================================== NM 10/13 10% Obama +23% Obama +6% OH 10/13 12% Obama +18% Obama +4% GA 10/12 18% Obama +6% McCain +11% IA 10/9 14% Obama +34% Obama +10% NC 10/6 5% Obama +34% McCain +5%
What’s surprising about these numbers is that, in the past, early voters as a group have tended been older, more Republican, and more male than later voters. In addition, if there were going to be any appreciable “Bradley effect” in the 2008 election, we should expect to see it here, shouldn’t we? If Obama is cleaning up despite such pro-McCain factors, it is bad news indeed for the McCain campaign.
Nate Silver of FiveThirtyEight.com adds all the usual and very appropriate caveats about these numbers, which are after all from just one pollster, SurveyUSA. But the major sites that aggregate and publish state-by-state polling results seem largely in agreement that the race is no longer close. RealClearPolitics is, at this moment, showing Obama with 238 electoral votes solidly on his side (including Pennsylvania) and another 75 leaning his way (including Florida and Virginia) by margins of between 5 and 8.7 percent. Only six states are toss-ups on the RCP map, and Obama is leading in all of them except Missouri and Indiana. Indiana! A state President Bush carried in 2004 by over twenty percentage points! Pollster.com says that Virginia is still a toss-up, but Ohio leans toward Obama and Florida is solidly in Obama’s tally — which Pollster.com says would be 358 if the election were held today. CNN and MSNBC are more favorable for McCain, with only 264 electoral votes currently going to Obama and 100-111 still in the toss-up category, including both Florida and Ohio.
FiveThirtyEight.com itself adopts a more statistically involved method than the sites above, paying attention not only to the published result of each poll but also the margin of error. By treating the “result” of any given state poll as a probability distribution within a certain range, FiveThirtyEight.com can simulate the election over and over again and then tally the results of all simulations. Using this method, FiveThirtyEight.com currently projects that the five most likely electoral vote totals for Obama are 375, 380, 383, 338, and 381. Four of those five would be comparable to Clinton’s tally against Dole in 1996, and some consider 375 as the threshold for “landslide” status. Of course, in some of the simulations, the stars align for McCain and all the polls turn out to have overstated support for Obama. According to FiveThirtyEight.com, the current probability that McCain will pull out such a “margin-of-error” victory has slipped to 4.2% — way down at one tail of the bell curve.
That, I guess, is what makes the early-voter numbers from SurveyUSA strike me as significant. Barring some pretty big “October surprise” that trends wildly in McCain’s favor, he can only win if actual votes for Obama lag significantly behind the vote totals that the polls would predict. Since most of these polls track “likely voters,” the assumptions about likeliness are critical. And if Obama supporters are voting early in numbers that far outstrip the turnout the models would predict, it looks more and more likely that the junior senator from Illinois will have a very big night.