In a wide-ranging, two-hour conversation with Seth Mnookin, co-director of MIT’s Graduate Program in Science Writing, Silver said there was “no doubt” that other analysts would close ground on his highly accurate projections, adding that it’s “not that hard” to figure out his forecasting formula. He also described the political media as “stuck in the Stone Age” when he reached prominence in 2008, and still “terrible at what they do” today, and suggested that he may, at some point, take a deep data dive into urban studies.
Silver, speaking to an overflow audience of some 350 people in MIT’s Bartos Theater, said he was pleased to be speaking at a place he called “ground zero for nerds.” He recounted his episodic career, which began as a financial consultant, included a phase making money in online poker, then began to take off in the world of baseball analytics, where he made a name for himself with the publication and website Baseball Prospectus.
“In the long run, you’re so much more productive doing something that captures your attention,” Silver said, describing his escape from consulting work.
In 2008, during the intense, drawn-out Democratic primary between Barack Obama and Hillary Clinton, Silver started gaining recognition for his impressively accurate vote forecasts — the first solidly quantitative projections of their kind, aggregating polls together — which helped his fledgling website, FiveThirtyEight.com, get off the ground.
“Political coverage was still stuck in the Stone Age by comparison, not very data-driven,” Silver said. He forecast 49 states correctly in the 2008 general election, and two years later franchised his site to the New York Times, where it now resides.
A target in 2012
The increased visibility deriving from Silver’s new venue, and the more closely contested 2012 race between Obama and Mitt Romney, made him a target of some reporters and right-leaning pundits in the weeks before the election, which Silver discussed at length. Quantitative forecasting, he asserted, represents a threat to a world of political reporting in which fact-free punditry has long been a core product.
“The industry has a history of getting things wrong,” Silver said, citing a string of examples and adding that “mainstream pundits … are terrible at what they do.”
Silver was relatively modest about his performance in 2012, noting that while he forecast all 50 states correctly, he was off by about two percentage points on the popular vote.
“We had a good [election] night, and I’m getting too much credit now,” he said.
Silver, whose contract with the Times is up in mid-2013, accounted for almost a quarter of the paper’s online traffic at times last fall. He indicated satisfaction with the Times, but declined to say for certain that he would renew his deal.
“Anything can happen in a negotiation,” Silver said, but repeatedly said he was pleased with his editorial freedom as well as the Times’ commitment to innovative online graphics.
The sauce remains a secret
Silver answered a long string of questions from the audience. In response to one query from a longtime member of MIT’s student newspaper, The Tech, Silver said he would not reveal his proprietary forecasting model, but added, “It’s not that hard to reverse-engineer it.”
Silver’s method largely consists of aggregating public-opinion polls, while evaluating the quality of those surveys and weighting them accordingly. An earlier generation of analytics pioneers often revealed their methods, including baseball analyst Bill James — a “hero” to Silver — who published his statistical formulas in his annual “Baseball Abstract” books. But while Silver said he was “sympathetic” to the idea of disclosing work, he felt he had no choice but to keep his models under wraps, for business reasons.
Still, Silver added, “There’s no doubt you are going to have people catching up,” noting that the Princeton Election Consortium had similarly effective forecasts this year.
Looking to the future, Silver speculated that he might branch out from politics and sports, and try his hand at new work on cities. In urban studies, he claimed, “You have data that is somewhat rich but maybe underused.”
A key to making an impact in analytics, Silver added, was not merely finding a subject with solid data, but specifically one in which that information was not being studied fully — like politics prior to 2008. Mnookin wryly observed that for all of Silver’s complaints about stodgy political reporters, “You have a vested interest in bad pundits keeping their jobs.”
Thursday’s event was part of the MIT Communications Forum, and sponsored by MIT’s Comparative Media Studies program.