Whether it’s baseball stats or political polling results, Nate Silver has made a living sifting through scores of data and making predictions, most of which seem to come true (Silver gained considerable notoriety after correctly picking 49 out of 50 states in the 2008 presidential election). While writing his data-driven, election-focused FiveThirtyEight blog for The New York Times, Silver found the time to examine the phenomenon of modern-day prediction in his new book The Signal and the Noise, which is out Thursday (Sept. 27). He sat down with Adweek to explain why presidential elections are incredibly difficult to forecast and why you really don’t want him in your fantasy league.
Adweek: A central theme of the book is cutting through the noise to get closer to the truth. I'm curious then, how do you get your news?
I don't think you should limit what you read. I have 500 people I follow on Twitter. They're more data-driven but made up of both liberals and conservatives. If you're keeping yourself in the bubble and only looking at your own data or only watching the TV that fits your agenda then it gets boring. I think reading broadly, but skeptically. That's key. You have to try to avoid the temptation to jump on the bandwagon with political news, especially on Twitter where things go viral every couple of hours. In a campaign you don't have game-changing events every single day. Most of these small viral things burn out. So it’s kind of about waiting those things out but not blinding yourself to the news.
You interviewed hundreds of people for this book over a long period; have your feelings on the way that humans predict changed substantially?
The book was going to start out being more Freakonomics-y, then it got kind of darker as I was going through and looking at economic forecasting and the financial crisis. But there are these silver linings in the book as you go through it, and I think it makes things interesting. You end up seeing people who are working on models—not get-rich-quick schemes—who think slowly and take time to study, and they end up making huge progress in how we predict. In a way, the book has a "lowercase c" conservative moral to it, which says, Be humble, work hard and you could get your rewards. It's often the case, though, that people are latching on to flashy narratives, and think they can get to solutions easily. If we rely only on those narratives, it can get scary.
How do your findings in this book change how you work every day?
Well, it’s important to be mindful of how big your data set is. For example, you have 33 U.S. Senate races every other year. In a four-year span, you have 66 Senate races and only one presidential race. For the Senate, you can really mine the data and test variables and see impacts. But for the presidential race, there are so many variables that influence voting, and we only have a handful of cases to test them on. You have to find a good structure for this data and provide some backbone to the model because you can't tell too much based on the cases. Every four years in the presidential election, some new precedent is broken. It’s tricky, and you have to be humble about how much uncertainty there is in the long run. The presidential race requires a lot of humility. It’s a great exercise, but it'll never be as complete a data picture.
You were pretty spot-on in your 2008 predictions. How much of that, if anything, was luck?
We had three states (Missouri, Indiana, North Carolina) that were 50-50, and we got two out of three right [Silver missed Indiana]. But it isn’t the coin flips that you want to judge people on. People tend to judge on marginal cases, and it is not enough to really evaluate a forecast. You need to have a zen-like attitude about it. People ask me if I am betting on the election outcomes, and I say that I don't need to bet on it. If I try to calculate how much money I have on the line in terms of future [career] projects, it is probably a pretty big number based on some skill but a lot of luck, too. I can say to people that you're supposed to get this wrong or ask them to look at our predictions throughout the entire election and not just on election night, but I also know that if the forecast we have up on election morning does well, it will really make my life a lot easier.
You got a start with and were recognized for PECOTA, your baseball performance prediction system. Are you unstoppable when it comes to fantasy sports and NCAA tournament bracket pools?
I'm pretty good, but I’ve been publishing NCAA predictions on the NYT website for a few years now, and after a while you lose your competitive edge. I don't play fantasy baseball anymore now because it's too much work, and I feel like I have to hold myself up to such a high standard. I'm pretty serious about my fantasy football, though. I have evaluations for all my players and do risk adjustments. My league does an auction every year, and I try to gauge how our league’s market is evolving in real time.
People in your league must be terrified of you.
Last year my team missed the playoffs, but I have a pretty good track record in the long run.
The election’s over in less than 50 days, and the book is done. What’s next for you?
After the election I hope to increase the share of nonpolitics stuff that I'm doing on the blog. I kind of realized after writing this book that I don't particularly like politics that much. I definitely like elections, as they're fun to forecast and to watch evolve, but I don't particularly like the day to day of politics or some of the people who end up getting involved.