Can McCain still win?
I haven't followed the polls in the US presidential race in any detail - there are dozens of polls every day and I could never hope to do half as good as job as Mark Blumenthal over at Pollster.com, with just a couple of days to go however, I thought it worth having a look to se whether it really is all over. With the polls consistently showing Obama ahead, can McCain still win? Let's have a look at the possibilities.
1) The late swing. The most famous surprise result of a Presidental election was Harry Truman who was able to pose the next day with newspapers that had gone to print assuming his inevitable defeat and hailing his opponant's victory. What happened then was that Gallup finished polling too early and, in the gap between the end of opinion polling and the actual election opinion swung to Truman. We will have polls up until the day itself so unlike in 1948 we would know. Could it happen?
It is obviously a logical possibility, however strong an Obama lead the polls show, it is only public opinion now, it could change by Tuesday. Since McCain's convention bounce receeded Obama's lead has been pretty solid and there is no obvious sign of it. The last slice of data in Zogby's rolling tracker showed a shift to McCain...but there is no obvious pattern in other national match-up polls.
Besides, even it if did happen, it might be too late for McCain. Projection sites like Electoral Vote and Five Thirty Eight have Obama with a solid lead in states collectively worth almost enough electoral votes to win, in the next tranche of states, those leaning towards Obama, the ones McCain would have to get hold of to prevent Obama winning, five have early voting. In Florida a third of the total number of votes cast in 2004 have already been cast, in North Carolina, Colorado and Nevada over half of the total number of votes cast in 2008 have already been cast. If there was some event that suddenly swung public opinion to McCain in these key states, unless it was a major swing it would be too late, people have already voted for Obama.
2) So if it is unlikely the polls will change in time, what if they are wrong? There are lots of ways they could be. First up is the Bradley effect. This is commonly named after Tom Bradley, but refers to the apparant way that polling in a contest between a black and a white candidate in the USA used to underestimate support for the white candidate, presumably a social desirability bias in the answers because people are worried they may been seen as racist if they declare they are voting against the black candidate. There is some debate about whether this happened at all in US polls in the first place, but there is plenty of evidence of polls in recent years which accurately predicted elections between a black and white candidate. If there was a Bradley effect, it seems to have gone. The only way it could suddenly come back, I suppose, is if there are stronger social pressures in a Presidential race, but I can't see it.
3) Shy McCain vote. Leaving aside the race question, could the undecideds split in favour of McCain. Indeed might there be a shy McCain vote in the way polls underestimated the Conservatives in this country in the 1990s? The answer is probably not - looking back to 1992 here, ICM found in the aftermath that the don't knows were disproportionately people who had voted Conservative in 1987. In the USA Mark Blumenthal and Charles Franklin have got hold of a huge chunk of data on undecided voters from Financial Dynamics to analyse. When the don't knows are pushed for which way they are leaning, they break for Obama. When Franklin analysed the demographics of those who refused to even say how they were leaning, looking at their race, party id and so on, his prediction is they are a pretty even break, in fact, if anything they'll lean slightly to Obama.
4) Faulty Turnout Model. In the UK pollsters tend to base turnout projections on how likely people say they are to vote. In the USA turnout projections are more complex and factor in whether people have voted in the past (and sometimes various other factors as well). Early voting and current polls suggest a higher level of voting than normal in US elections. Could this help McCain? Probably not, as if pollsters are underestimating turnout it is likely to be black and/or poor voters coming out for Obama who are being understated. It may not make any difference at all: Gallup are producing parallel numbers, ones based on their traditional US turnout model, and ones based on a turnout model we are used to in this country based only on how likely people say they are to vote. In the latest figures Obama has an identical 10 point lead on both measures.
5) Weighting. Tracking how pollsters in the US are dealing with weighting is a bigger task than I can do here. Pollsters use a wide variety of weights by party ID, or in many cases none at all. Whereas in the UK demographic weightings are much the same between pollsters, in the US there is some variety because some target models of likely or registered voters, rather than the demographic make up of the whole US. We do see pollsters with different proportions of African-Americans in their samples for example. The sheer variation in the way US pollsters do things is why I suspect this is not disguising a McCain lead: with such variation, it is unlikely everybody's method has wrong. Equally, having a black candidate is new and probably having increased turnout is new, so it is reasonable to ask whether polls can cope with these developments. There has not, however, been a sudden massive change in US demographics.
Those are the most likely reasons why McCain could still win, and I don't think any of them work. If anything, I'm expecting a bigger Obama victory than the polls indicate. Of course, when polls do get it wrong it is normally something that we haven't predicted and it's perfectly possible that something completely different will come out of the woodwork. If miraculously the polls are wrong and McCain wins on Tuesday there would be one hell of a post-mortem about why the polls got it wrong, but really, I'm not expecting it.
