Statistician turns from sports to politics, with striking results

CHICAGO — Nate Silver's political rags-to-riches story plays out on a BlackBerry brimming with e-mails and a schedule conspiring to run him into the ground.

Last spring he was a sports statistics savant virtually unknown outside the world of obsessive baseball fans. Now he dreams about hiring a personal assistant. With the election a week away, the Chicagoan has become one of the most sought-after election analysts in the country, thanks to a computer program he built to blend poll results, demographics and political trends.

Silver boldly predicted a Barack Obama victory months ago. As of Sunday, he's 96.7 percent sure of that. He breaks down the numbers into a dizzying array of scenarios: On Sunday, he sees a 4.75 percent chance of John McCain winning the popular vote. There's a 2.13 percent chance of Obama losing the popular vote but winning the most electoral votes. There's a 0.25 percent chance of a tie in the Electoral College.

Silver earned his baseball credibility in 2005 by predicting the White Sox would win it all. He earned his political stripes during this year's primaries with several smart forecasts, such as picking the Super Tuesday results within 20 delegates out of 1,700.

Now Silver's political Web site, FiveThirtyEight.com, is a must-read for political wonks.

"I don't know who his publicist is, but he's doing a hell of a job," said Charles Franklin, a political scientist and co-founder of the influential site Pollster.com, which gets 1.8 million visitors a week. Silver's gets 700,000 a day.

"I mean that mostly out of jealousy and admiration," Franklin added.

Silver has been a partner at sports think tank Baseball Prospectus since 2004. He was a high school debate champ in East Lansing, Mich., and got an economics degree from the University of Chicago. Back when he was a consultant at international tax accounting firm KPMG, a side fixation with baseball statistics led him to create PECOTA, his system to wrestle reliable predictions from raw numbers.

Since he created the computer model that powers FiveThirtyEight.com late last winter — the name refers to the number of votes in the Electoral College — demands on his time have grown.

Fresh numbers drive everything. Silver feeds 30 state polls into his computer model every 24 hours. He weights pollsters by track record and averages state polls, checking them against national polls and similar demographic slices elsewhere in the country. A statistical model allocates undecided voters and simulates the election 10,000 times a day.

Andrew Gelman, a statistician and political scientist at Columbia University, used a similar approach with a colleague from Harvard University in a paper on Sept. 21. Both Gelman's and Silver's models agreed: They gave Obama a convincing lead in the popular vote and more than enough electoral votes to seal the election.

"I'm meticulous," Silver said. "I like getting into details, and when I get into something, I get into it. I don't dabble in a lot of things."

So far, he has also been accurate when it mattered.

On the morning of May 6, he defied conventional pollsters who saw a 5-point lead for Hillary Clinton in Indiana and rapid momentum for her in North Carolina. On May 7, election results bore Silver out by giving Indiana to Clinton by 1 point and North Carolina to Obama by 17. The world came knocking at Silver's door.

Silver wasn't always so confident in his prediction program. He was stunned by a leap for Obama when Lehman Brothers declared bankruptcy in September. Markets fell, and so did McCain's chance of victor in Silver's model.

"I got a little bit worried," Silver said. "This isn't the actual condition of the election. Ordinary logic goes out the window."

And so Silver left to finalize a deal that sends text messages of his predictions to readers' cell phones, wondering what would happen to his numbers if emotional voters just changed their minds.