The Signal and the Noise: Nate Silver and the limits of prediction

The patterns in the stars: Nate Silver makes amazing predictions by crunching huge volumes of data. But even he says we’ll never be great at it.

Nate Silver has the origin story of a nerd superhero. Back in 2008, an anonymous writer using the pseudonym “Poblano” began posting a series of articles to the liberal blog Daily Kos. It was the middle of primary season, a time when the web is thick with political commentary, but Poblano’s writing stood out for its sophisticated treatment of data and the accuracy of its predictions.

In May, the polls and professional commentators were pointing to a five-point Clinton win in Michigan and a too-close-to-call toss-up in North Carolina. Poblano saw it differently. He had crunched the numbers, looked at the demographics and predicted a landslide victory for Obama in North Carolina and a slim two-point Clinton win in Michigan. When the votes came in, the results were clear: Obama had won big in North Carolina, Clinton had squeaked one out in Michigan and the guy blogging under the name of a chili pepper had bested the pros.

Later that month, “Poblano” made a dramatic reveal—the man behind the stats was none other than Nate Silver, the wunderkind known to baseball stats nerds as the inventor of PECOTA, a remarkably accurate system for predicting baseball players’ future performance. Visits to his blog FiveThirtyEight exploded. He got profiled in national magazines, hailed as a stats guru. Over just a few months, Silver went from an anonymous blogger to broadcasting election-night coverage with Dan Rather. Aside from Barack Obama and Tina Fey, it’s hard to think of anyone who benefited from the 2008 campaign more than Nate Silver.

His timing couldn’t have been better. With the airwaves full of ostensible experts yelling at one another in cynical displays of political theatre, people were eager for a wonky stats guy who promised to cut through the bullshit. The fact that Silver came from the world of baseball sabermetrics—where people like Moneyball star Billy Beane were ignoring conventional wisdom in favour of hard stats—only increased his credibility. A New York magazine profile called him a “number-crunching prodigy” and a “spreadsheet psychic.” More than just a political analysis, Silver seemed to offer an appealing vision of the world—the promise that, with enough rigour and the right statistical models, we could sift through the data and see the future with clear, dispassionate eyes.

If it’s tempting to see Silver as some sort of statistical clairvoyant, his new book, The Signal and the Noise, goes to great lengths to discourage that kind of thinking. The man known for his ability to see into the future has written a book that’s largely about the limits of our predictive powers. Time after time, Silver reveals our crummy forecasting skills. He finds that when political scientists say there’s zero chance of an event occurring, it actually happens 15% of the time. Earthquake predictions are similarly a mess. And most economists can’t “predict” a recession, even after it’s begun. “We have a prediction problem,” Silver writes in his book’s introduction. “We love to predict things, and we aren’t very good at it.”

Silver offers some perspective on his own seemingly miraculous success. “I had been lucky on a few levels,” he writes. First, making a sound prediction based on probability doesn’t mean you’ll always be right. Accurately forecasting that Obama will win a state nine times out of 10 still means there’s a one in 10 chance he’ll lose. Second, Silver admits he carefully chooses the arenas for his forecasts. Baseball and politics are two areas perfect for someone like Silver—data-rich systems in which the level of competition (CNN pundits and drive-time sportscasters) was appealingly low. In more complex and competitive systems—such as the stock market—consistently making a winning prediction is far less easy.

If we’ll never be good at predicting the future, Silver’s aim is to show us how to become less bad. In this era of Big Data, with more and more information available to us, we need to get better at teasing out the signal from the universe of noisy data that surrounds it. Doing that means becoming comfortable thinking probabilistically and dealing with uncertainty. It also means being suitably humble and acknowledging our own role in making predictions. “We can never make perfectly objective predictions,” writes Silver. “They will always be tainted by our subjective point of view.” The very worst, most dangerous projections tend to come when we’re supremely confident our model is objectively true.

As The Signal and the Noise zips from the world of competitive poker to climate change to baseball and beyond, it can look superficially like yet another Malcolm Gladwell–esque work of non-fiction. The formula has become familiar: take a catchy, easy-to-grasp concept—“the tipping point,” or “the 10,000-hour rule”—and then use a series of expertly plucked and beautifully told anecdotes to advance that thesis. These kinds of big-idea books are hugely popular, and it’s easy to see why. It’s tempting to believe that complex systems can be explained with a few grand theories. It’s natural, too. With such a mass of information, the only way for humans to make any sense of the world is to make some approximations and assumptions, to look for the patterns, and try to find the constellations in the mess of stars.

We get in trouble when we become too seduced by the elegant explanation. “The simpler statements seem more universal, more in testament to a greater truth or a grander theory,” Silver writes. But they don’t work. “They leave out all the messy bits that make life real and predictions more accurate.”

At 450 pages, The Signal and the Noise leaves in the all messy bits, sometimes to a fault. It’s full of caveats and counter-examples, of charts and tables detailing the awkward exceptions to Silver’s ideas. It’s messy at times, but with a purpose. Clean things up too much and you might get a beautiful theory. You might also get someone like Jonah Lehrer, the bestselling science writer who was caught literally making up Bob Dylan quotes to better fit his thesis. You’ll also get predictions that will almost certainly turn out to be wrong.

Nicholas Hune-Brown is a National Magazine Award–winning writer based in Toronto