I've long been obsessed with surveys that ask not so much who someone is going to vote for than those that ask who they think is going to win an election. Yes, the two are related. We tend to believe our preferred candidate will win -- something called wishful thinking in the social science literature -- and it plays an important role in my surprised loser research of late.
There is even strong evidence that asking people who is going to win is more accurate than measuring who they're going to vote for. But -- and this is important -- it matters when you ask the question. Take 1992, for example. CBS News/New York Times polls asked respondents who they expected to win between George H.W. Bush, Bill Clinton, and Ross Perot (a populist third-party candidate). Unfortunately they only asked it March through July. See below.
See the trend? The percent predicting a Bush win go down the closer it gets to the election, and of course Bill Clinton's went up. Unfortunately they didn't ask this question in the fall. We see a difference, somewhat, in 2008. In July 2008, 53 percent predicted an Obama victory. In August it's down to 48 percent, and by September it's down to 45 percent. In October it's back up to 65 percent. In part these differences reflect a shift from "registered voters" to "likely voters." In part they reflect the ups and downs of the campaign. And finally they reflect people finally making up their mind.
Of course in some elections it's easy. In 1972, back in July, 77 percent predicted a Nixon win. by October it was up to 85 percent.