Super Model

Sign up for The Media Today, CJR’s daily newsletter.

It was a slow and staggered end for FiveThirtyEight, the site that made everyone into armchair experts on the art of data modeling. Less than two years ago, Nate Silver, the site’s founder and editor, made his exit, and some two-thirds of the staff were laid off. On Tuesday, Disney—its corporate owner, via ABC News—announced company-wide layoffs that closed FiveThirtyEight for good. That hardly kills America’s polling obsession, or statistics nerds’ pursuit of just-right renderings—Silver, for one, has a Substack, Silver Bulletin, on which he wrote, in response to FiveThirtyEight’s demise, “‘Data journalism’ has a bad name but a bright future.” It does mark the end of a certain frenzied political chapter marked by popular attention to the work of finding answers in numbers.

In March of 2008, Silver, a KPMG consultant turned baseball analyst, started a site under the pseudonym Poblano, which he named after the number of members in the Electoral College. Silver had dabbled in election-data journalism before, as a contributor to Daily Kos; striking out on his own, he began to attract attention from tens of thousands of readers, including prominent political figures. He made a pivotal call, during that year’s Democratic primaries, predicting, against overwhelming dissent, that Hillary Clinton would win Indiana by 2 percentage points and lose North Carolina to Barack Obama by 17. When the results arrived—Clinton won by a point in Indiana, and lost by 15 in North Carolina—the rush to find out how Poblano had made his forecast was immediate. 

That May, he unmasked himself to readers: “No, I’m not Chuck Todd.” By Election Day, Silver published on FiveThirtyEight the final results of his model’s prediction: Barack Obama 349, John McCain 189. Hours later, the outcome was known: Obama 365, McCain 173—uncannily close to his projection. Silver’s fame, and FiveThirtyEight’s influence, skyrocketed. By the following year, he was on Time magazine’s list of the world’s most influential people. A year after that, the New York Times licensed the site, which became FiveThirtyEight: Nate Silver’s Political Calculus. He successfully forecast the 2012 presidential election result—factoring in, among other variables, the fallibility of phone surveys that did not reach cell numbers. He also dabbled in other data: sports, economics, entertainment. FiveThirtyEight moved to ESPN in 2013, and ABC News after that. 

What made FiveThirtyEight innovative was its dexterity with modeling. Unlike news outlets that merely reported on the latest polls, Silver integrated a great number of polls into a model of his invention, to reveal an overall political picture. RealClearPolitics had aggregated polling data, and there were other forecast models, such as PollyVote, which had started during the 2004 election. FiveThirtyEight stood out, in the early days, for its startling accuracy. Instead of treating all of the data equally, Silver gave polls different weights, depending on their past performance, and took into account other political factors, such as demographics and prior voting patterns. This was an approach with which Silver had already gained credibility among baseball fans, with his creation of the Player Empirical Comparison and Optimization Test Algorithm, or PECOTA, a sabermetrics system for forecasting ballplayers’ performance. (“PECOTA is now recognized as the most accurate system for forecasting how athletes and teams will perform in the future, down to the number of singles,” a Newsweek profile hailed. “In 2007, Silver’s algorithm enraged at least half of Chicago when it said the White Sox—2005 champs—would post a 72–90 record. Turned out PECOTA was exactly right.”)

The core audience of FiveThirtyEight were fans of quantitative analysis—those who reveled in, for instance, a data-driven investigation into plagiarism in crossword puzzles, a comprehensive analysis of popular hairstyles at the World Cup, and graphs categorizing the five types of Nicolas Cage movies. But the aim of data journalism is to make numbers widely accessible. Before long, as readership swelled, FiveThirtyEight was joined by the Times, PredictWise, and Silver’s old haunt, Daily Kos; the Princeton Election Consortium, formed in 2004, received more attention. Coverage of polling analysis followed along. And so emerged a paradox of success: an explosion of visibility drove interest in tracking election models for the purpose of filling out political coverage, including in superficial cable news analysis of horse-race polls, and online election betting, much of it happening in unregulated markets. 

FiveThirtyEight did not create fascination with political data, but it popularized the idea that, by turning the right dials, a statistician could come away with insights about which way the wind was blowing. When the story was more complicated—say, on election night of 2016, when FiveThirtyEight’s final forecast showed Donald Trump with a 28.6 percent chance of winning and Hillary Clinton with a 71.4 percent chance—probabilities, expressed as percentages, felt distorted. For many, Trump’s victory came as a shock. It didn’t help that the polls on which FiveThirtyEight’s model largely relied were off the mark that year. But the psychology of modeling misinterpretation was laid bare, and Silver was blamed. (“Predictions can be off, to be sure,” Sam Wang, the director of the Princeton Election Consortium, wrote for CJR. “The key is to keep coverage simple, put facts in context, and not overreach.”)

For the 2018 midterms, FiveThirtyEight changed how its forecast was represented, leading with a histogram—which may have been difficult to comprehend for the non-data-inclined. (“That one was misinterpreted much less, but it also was interpreted much less, period,” as Dhrumil Mehta—a data journalist who worked at FiveThirtyEight, and is now a professor at Columbia—told me.) In 2020 came Fivey Fox, a cartoon character who appeared alongside charts to help readers understand visualizations (“Don’t count the underdog out! Upset wins are surprising but not impossible”).

FiveThirtyEight’s changes, over the years, showed its responsiveness to the evolving realities of politics and media. All the while, more players in data modelling have emerged, in testament to its legacy: at The Economist, Split Ticket, and of course, Silver Bulletin. “Collecting and maintaining a database of public polls is a lot of work, requiring diligence, meticulousness, and dealing with constant complaints about edge cases from readers and pollsters,” Silver wrote. “But it’s also a public service.”

Has America ever needed a media defender more than now? Help us by joining CJR today.