GUY RAZ, HOST:
If you look at any of the election polls, likely voters are usually asked this question: If the election were held today, who would you vote for? Well, Justin Wolfers, who's a political economist at the University of Michigan, has spent years researching these polls. And he concluded that this is the wrong question to ask, especially if you want to predict who's going to win an election. The right question, he says, who do you think will win?
JUSTIN WOLFERS: It's basically about treating the people you're asking a question, the respondents, with respect. They know a lot of things about the election. For instance, I know who I'm going to vote for. I know who my better half is going to vote for. I was around the water cooler at the University of Michigan the other day, and I learned who all of my colleagues are going to vote for. I know who my neighbors are going to vote for. And so when you ask me who do I think will win, I'm going to report not only my own intentions but also what I'm sort of hearing around the neighborhood. So we have basically 20 times as many people who are effectively in the room as they're actually responding to the survey.
RAZ: OK. But, I mean, what kind of evidence is there to show that that question gives you a more accurate predictor of the results?
WOLFERS: So we went through basically every election where a pollster asked both: Who do you think will win, and they asked the usual polling question: Who do you intend to vote for, questions going all the way back to the 1930s.
WOLFERS: And the data show in the 345 state races we looked at in the United States, if you looked at polls of voter intentions, the standard polling approach, you would have gotten a win 69 percent of the time. If you turned to my question and asked people who do you think will win, you would have gotten the answer 81 percent of the time.
RAZ: Ha. Now, there is evidence to back your argument from this election, right, going back to the Republican primaries.
WOLFERS: Oh, my goodness. This one was a roller coaster. At one point, they had Donald Trump as heading the pack. You had Rick Santorum.
RAZ: Herman Cain was leading, yeah.
WOLFERS: Herman Cain. None of these guys were serious candidates. And the whole time, when we're asking people who do you think will win, you know who they kept saying?
WOLFERS: You bet.
RAZ: OK. Gallup decided to ask this question recently, and they found that more Americans believe that President Obama will be re-elected on Tuesday.
WOLFERS: Yup. So when Gallup asked this recently, 54 percent of Americans said they expect Obama to win. Thirty-four percent said they expect Romney to win. There's a few others who basically had no opinion on the question. But that's a pretty sizeable lead. And so I have a statistical machinery where I can chug that through and try and come up with a forecast for the election. What that says is first of all, Obama's strongly favored to win and this is really - in other words, (unintelligible) if you ask the American people, Obama's strongly favored to win.
And it looks like his winning margin could be bigger than most pollsters are currently saying. Lots of clever pundits out there have much larger models, and they have many more respondents to their polls. So there's a lot of information what other people are doing as well. But I think this is an important piece of information that says we may just surprise on a stronger Obama performance come Election Day.
RAZ: Justin Wolfers is a political economist at the University of Michigan. Justin, thanks so much for coming on again.
WOLFERS: A pleasure.
(SOUNDBITE OF MUSIC)
RAZ: And you're listening to ALL THINGS CONSIDERED from NPR News.