Silver and the Cloudy Future

Head of election prognosticators fivethirtyeight.com talks at SXSW about the limits of prediction

By Richard Whittaker, 11:59PM, Mon. Mar. 16, 2009

Nate Silver: MVP of election forecasting
Nate Silver: MVP of election forecasting

Nate Silver might be the ideal dinner guest: A political expert with an encyclopedic knowledge of baseball careers.

Silver (read Wells Dunbar's interview with him here) delivered the SXSW Interactive keynote address on Sunday. He's revered amongst election wonks because of his number-crunching website fivethirtyeight.com. For people scared off by concepts like "regression analysis", "inferential process" and "rolling trendlines," he's the guy that predicted last November's election in March. For sports fans, he's the guy that created the PECOTA predictive algorithm (the bane of fantasy baseball leagues everywhere.) However, conference attendees now probably know him as the guy that admitted he wished he'd studied more programming, so he he could do more site maintenance himself.

Politics and baseball, as he pointed out, both have long seasons. Baseball fans, however, don't tend to write a player off because of one foul ball. Policy wonks, he said, tend to thank him when he reminds them that "one poll coming out in June or July means almost nothing."

It's a mind-set thing: policy wonks think issue-by-issue, while the general population doesn't tend to track elections at the same micro-level. So what are they looking for? "Leadership and personality, but I don't think people are thinking 'well, this candidate looks good in a suit,'" he added.

The biggest takeaway from his work may be that the old monolithic voting groups, like gender and race, have been pretty much debunked. However, poll trackers shouldn't be trying to build new stereotypical voters, and should instead look at the nuances of what he called "an infinite data set." After all, the biggest lessons aren't in looking at the 75% of evangelicals that went for McCain, but the 25% that voted Obama.

However, he warned, stats crunching for predictions has its limits. Where it could get ugly, as host and Business Week columnist Stephen Baker noted, is that firms are using predictive technology to find excuses to sack people who may not be great at their job five years from now. "That's a whole new can of worms," said Silver, adding, "I'm a little suspicious that a poorly-executed object-driven prediction is often worse than a subjective one."

write a letter