Political Pollster Nate Silver

Nate Silver is a statistics savant. The managing partner of the influential publication Baseball Prospectus, Silver created PECOTA, considered perhaps the most accurate system for predicting player performance. Since March, Silver (a Barack Obama supporter) has run the political prediction website fivethirtyeight.com, which drew widespread attention during the presidential primaries for its ability to issue on-target projections. He spoke to TIME about the similarities between baseball and politics, the trouble with traditional pundits and why Obama will win the election.

What's the mindset behind your approach?

A lot of it is frustration with the kind of way things are oversimplified. For example, there was a perception in the primaries that Obama [was weak] with working class voters. If you actually drilled down, it tended to do more with education level than class. It's shoddy analysis. If you want to be a pundit, be a pundit. But don't cherry-pick polls. If six polls come out in a week and one says Obama's ahead by 10, guess which one gets talked about. But that's really the one you should probably ignore.

Is politics more conducive than baseball to this type of analysis?

Baseball is a little bit easier. Everything that takes place on the field can be represented statistically pretty well. In politics, our model still relies on polls. There are a lot of little tricks to make the polls more accurate than they are inherently—by weighting them, and by seeing which pollsters have Republican or Democratic leanings, and combining them with some semblance of demographic data. But if polls underestimate youth turnout, we'll have the same problem. The data isn't as pure as the fact that Johnny Damon hit a double yesterday.

What are the factors you use to predict election outcomes?

What we're doing is looking at how accurate different pollsters have been in the past. We weight polls based on past accuracy. For good pollsters, like Quinnipiac University, we might give them twice as much [value] as a poor pollster. We also look at the same size; certainly if you get 1200 people in a state, then that's better than sampling 400. But we're also adjusting for current trends, so sometimes we guess where gains will take place. We don't necessarily wait for new data; if Obama's gained four points nationally—and it looks like the demographics are pretty even on where he's gained and where he hasn't—we'll assume he's gained ground there as well. In cases like this, it does become more predictive than just averaging the numbers.

Are there any generalizations you can make about the candidates' strengths and weaknesses on specific topics?

I think it's mostly about the economy. If 70% of voters think the economy is the most important issue and Obama has an edge there, then McCain's in a lot of trouble. Inevitably, and probably smartly, [the Obama camp] probably understood that this doesn't have to be a complicated election. Focus on the economy. That's where they're going to win or lose. This financial blowup is probably favorable for him.

Your Electoral College projections have Obama winning around 330 electoral votes.

After Lehman Brothers blew up, a few days later, his polls shot up. McCain had made that comment about the fundamentals of our economy being strong. Clearly the events of the past 15 days or so have taken this from an election where it was basically a toss-up to one where more people are now asking, "Can McCain win?

But your margin of victory still seems wider than most pundits have forecast. Do you always adhere to the projections your system spits out, no matter what?

I might occasionally say, I think the simulation is out to lunch. But now we have Obama as about a 5-1 favorite and intuitively I think that's about right. A five-point lead doesn't mean a whole lot in February, but we have less than a month until the election. My guess is the election will tighten a little bit, and the model assumes it'll tighten some.

If you could give each candidate one piece of advice, what would it be?

With Obama, I'd say keep doing what you're doing. During the debate he came across as a nice guy and people seemed to react well to that. So he may actually want to ease up a bit and realize he has an advantage. That's one reason we assume races tend to tighten—one candidate starts to play it safe, but it's like a prevent defense. You give up some yards, but no touchdowns.

And McCain?

Well, he's played so many different cards—the Palin card, the suspending-my-campaign card. The most obvious advice is to trot out Jeremiah Wright, go negative. Maybe McCain apologizes to the nation for saying the fundamentals of the economy are strong. Be a little contrite, unorthodox. They're in a tough position now. The big mistake McCain made in the [first presidential] debate is he allowed Obama to identify himself as the home team for the middle class. It's not that Obama was especially eloquent. He just said "middle class" three times and looked at the camera. But McCain didn't do those things.

What do your projections say about November's Congressional races?

In the Senate things are changing pretty rapidly. We've had the Democrats picking up 5-6 seats. I think you'll see a lot of seats changing hands in the House. You may see an anti-incumbent revolt in general.

Sabermetricians have gone on to work in MLB (Bill James is one example) and campaigns obviously have a need for pollsters. Where do you see yourself winding up down the line?

Right now I'm having fun talking to people and informing the discussion. As a matter of practice, I think working for a campaign is probably more likely. Baseball is a small industry. Politics is a touch more open minded.