Samir Paul '06 worked on the Obama campaign in Wisconsin this fall. After the election, he spoke with Drew Linzer '94, whose website, Votamatic.org, correctly predicted the result of the Presidential vote in all 50 states.
Along with Nate Silver of FiveThirtyEight and Sam Wong of the Princeton Election Consortium, Linzer and his website, Votamatic.org, are at the forefront of a growing movement of statisticians developing complex statistical models to forecast election outcomes. Though the math isn't new, conditions have ripened just right to create an opening for these methodologies. “The data just wasn't available in the past,” Linzer said. “It was only in 2008 that we saw an explosion of state-level polling released. There were about 1700 state-level polls and a million interviews in 2008, and that's just at the state level.” There were not as many this time around, Linzer added, but there was still an incredible amount of information available.
Linzer's model took the current and historical conditions and forecasted an expected Election-Day outcome at both the state and national levels. The result – a forecast neither based on gut-feeling assessments nor careening wildly with each individual poll in the field – was a highly accurate call that put many broadcast news talking heads to shame. And while 2008 and 2012 may have been the first election cycles in which this brand of stats-intensive methodology took a front seat, these are only the after-shocks of a revolution that has been happening in many other fields. “In many ways, politics is behind,” Linzer said. “We've seen it in business, in sports – with Moneyball – and, increasingly, it's happening inside the campaigns.” The Obama campaign's sophisticated microtargeting received much fanfare for its ability to focus the Democrats' ground game and deliver close wins in most battleground states. “The Republicans will be playing catch-up, and I don't have any doubt that they will. Hopefully reporting will catch up, too.”
Still, Linzer said this does not spell the end for traditional campaign reporting. “We all depend on that sort of information, and no data analysis can replace going to the campaigns and talking to people and figuring out what's happening on the ground,” he said. “But I hope our work will not replace but rather inform the commentary and the punditry.” TV news reporters will often say things that are demonstrably false, Linzer said, and he hopes that stats-heavy analysts get enough recognition this time around that this kind of information is treated more seriously in 2016.
After studying politics at Pomona University, Linzer worked on the congressional campaign of Ralph Neas, a Democrat challenging long-time Maryland Republican Rep. Connie Morella. During that campaign, he recognized that his interest lay less in political organizing and more in the strategy, tactics, and research behind campaigns. So he proceeded to work with several public opinion polling firms doing research consulting. “When I was at Pomona, I was doing math for fun, and nobody told me you could do statistics to study politics,” Linzer said. “Nobody said, 'Why don't you put those two things together?' I did that a little as a pollster, but not in a sophisticated way.” So when he arrived at UCLA to pursue his doctorate in political science, he was pleased with the amount of quantitative work he was doing.
Now, as an assistant professor at Emory University, Linzer is a political methodologist. “I'm in the political science department, but I'm really a statistician,” he said. His work focuses on the statistics behind public opinion and voting behavior, and a paper about his model is slated for publication in the Journal of the American Statistical society.
Though Linzer has done this work from academia, forecasting like his doesn't require a degree or a title. Taking classes in statistics and statistical programming helps, but “I'm not a public figure in any way. I bought a domain name and put up a website.” This work, he said, rewards creativity and mathematical ingenuity. “There is so much public information and free software that you can download that anyone who is entrepreneurial and has an interest in data analysis can do this. Anyone with that creative spirit can take relatively basic statistical techniques and do interesting things,” he said. “My advice: Try stuff. Have ideas, see if they work, put them together in a package, and share them with people.”