Why even Nate Silver ‎didn't foresee the rise of Donald Trump

Share

They filled a Harvard lecture hall beyond capacity, spilling into an overflow room – overwhelmingly young and white, disproportionately though not quite as overwhelmingly male, as heavily bespectacled as one might expect. And as the nerd king took centre stage here among his people, the crown was still firmly atop his head.

But for Nate Silver, the inaugural Political Analytics Conference – a Harvard University event this month at which he was the biggest attraction, and one that probably wouldn't have existed without him – was not quite the triumph he might have hoped. Not when he had to sit there, alongside other data journalists he had helped inspire, and spend so much time defending his work rather than touting its brilliance.

In the previous two presidential election cycles, Mr. Silver was about the best story going outside the candidates themselves. A former baseball statistician who became a pioneer of applying "predictive analytics" to campaigns, he used a polling-aggregation model – one that significantly took into account past trends and demographic voting patterns to weigh current polls – to coolly and successfully predict results down to each state.

As he was more than willing to point out, it was a rebuke of traditional media's breathless overhyping of campaigns' ups and downs. Becoming a household name among followers of U.S. politics, he offered reassurance that however tumultuous a campaign might seem at any given moment, there was a predictable order to how things would play out.

Then the 2016 campaign happened.

Mr. Silver and the growing predictive-analytics industry around him – much of it anchored around his website FiveThirtyEight.com, with a growing editorial staff of more than 40 people and eight million unique visitors last month – proved flummoxed by the Republican nomination process. And as it missed and dismissed the Donald Trump phenomenon, what it is selling began to look like a false promise.

As it turns out, voter behaviour can't always be understood from numbers alone. Knowing how to read polls and interpret them based on past trends is of limited use when your country's electorate is going through a tectonic shift. Highly rational data journalists might have their own biases. If they're not careful, they can do the public a disservice – in this case, arguably, by contributing to the fateful underestimation of Mr. Trump.

In retrospect, it may have been too much to expect anyone to accurately predict the outcome of a nomination campaign that at one point had 17 candidates. As Mr. Silver (who declined an interview request) acknowledged during his conference panel, such races are vastly more challenging than general-election campaigns, because there are so many variables. In fact, he said, "no one I know has really tried to build a model of the primary process," along the lines of the sophisticated weighting model he uses once nominees are set.

The problem is that such shortcomings didn't stop him and others from making predictions anyway. Through last summer and fall, Mr. Silver staked his credibility to telling anyone who would listen that Mr. Trump had no serious chance of getting the Republican nod. Whatever modelling he had was enough for him to peg the populist billionaire's nomination prospects at 5 per cent. In August, drawing on the trajectories of flash-in-the-pan Republican candidates past (Herman Cain, Steve Forbes, Pat Buchanan), he penned an article headlined "Donald Trump's six stages of doom." In September, confident Mr. Trump would wilt under scrutiny, Mr. Silver told a New York audience to "calm down" about his high poll numbers.

Should Mr. Trump lose a floor fight at the Republicans' convention this summer, it won't offer much vindication. The predictions were that he would flame out long before now, not at a minimum define the race, and Mr. Silver's site was initially just as dismissive of the candidate who now stands the best chance of beating him. Ted Cruz was "the media equivalent of a juicy rib-eye that robbers use to distract a guard dog during a heist," scoffed Harry Enten, one of FiveThirtyEight's more prominent analysts, last spring. He would get lots of attention, but he was too extreme and unpopular with Republican elites to win.

Other than all the variables in a nomination race, one of the better explanations Mr. Silver offers for such flubs is the small sample size. Political analytics draw much inspiration from sports analytics, which are more evolved. (This month's conference was modelled after the more established Sloan Sports Analytics Conference.) A baseball season has 162 games, offering plenty of room to spot patterns and anomalies; a presidential campaign happens every four years.

But to listen to Mr. Silver (and to a lesser extent fellow travellers such as The New York Times' Nate Cohn) at Harvard was to get the sense that where adequate statistical evidence falls short, prognosticators who made their mark with data have been falling back on what they want to think about the electorate.

Mr. Silver may have tipped his hand on this front during the election year he came to prominence. "The conventional wisdom underestimates voters," he said in a 2008 New York magazine profile. That electors in the parties' primaries had picked two strong candidates, in Barack Obama and John McCain, proved that they were "pretty smart."

Plainly, Mr. Silver – a highly rational, unabashedly left-of-centre urbanite – thinks voting for Mr. Trump is the opposite of smart. On the panel, he highlighted that despite Mr. Trump having won a plurality of Republican votes so far, only a very small segment of the electorate cast ballots for him. He chalked that support up, in large part, to rank bigotry on the part of Mr. Trump's voters, which he suggested traditional media have been too squeamish about covering. He insisted that the most under-covered aspect of the campaign thus far, again by traditional media, has been Mr. Trump's unpopularity with the majority of voters, inflating perceptions of his electability in November.

It seemingly adds up to a basic premise that the media have been manipulated into affording Mr. Trump shallow and sensationalist coverage, which has duped a stupid segment of an otherwise smart electorate into voting for him.

It's a plausible theory, but it also fits suspiciously neatly into Mr. Silver's view of his country's discourse. It suggests little willingness to challenge his own assumptions about the electorate – about why the Republican side of it is so susceptible to populism (or demagoguery), even as a larger-than-expected number of Democrats have been open to a previously obscure 72-year-old socialist. And that's part of what frustrates even some fellow journalists broadly committed to better understanding the relationship between politics and data.

"It's amazing how few of the people who do these election predictions are interested in actually understanding politics," Sasha Issenberg said over the phone this week, likening it to friends of his who play fantasy sports but aren't really fans of the sports themselves.

The author of The Victory Lab, a popular 2012 book about how campaigns have become more scientific in their attempts to win votes, Mr. Issenberg argued that the data crowd should be going out of its way to identify and learn from anomalies such as the Trump phenomenon – not trying to explain them away, and in the process giving "statistical cover" to those who want to believe the unexpected isn't really happening.

Ryan Enos, one of the two Harvard professors who organized the analytics conference, agreed to a point with the need for a rethink. "We need to perhaps step back from saying things are predictable," he acknowledged, and consider whether the right tools are in place to identify and understand deviations. His event, he contended, enabled conversations about how to do just that – and indeed, over the course of the day, you could learn a good deal about the electorate and the way the data world is interacting with it.

There was Democratic operative Jeremy Bird, who explained why the ability to immediately analyze data from canvassers' contact with voters is reducing campaigns' reliance on polls and focus groups – indirectly pointing to the difficulty in keeping up, for those of us relying on publicly available data. There was Kristen Soltis Anderson, a young Republican pollster, pointing out media outlets spent so much money polling the first couple of primary states, there wasn't much left to keep tabs on the electorate over the course of a longer-than-expected campaign. There was a doctoral candidate presenting a research paper on just how little campaign contact there still is with the large number of unregistered U.S. voters in the U.S., and how that particularly disadvantages Democrats.

There was even, on the same panel as Mr. Silver, a colleague of his whose magazine-style stories stand out a bit at FiveThirtyEight.

Last month, Clare Malone wrote a lengthy piece about what motivates Trump supporters that was refreshingly short on pat answers, touching instead on a range of sentiments – from distrust in institutions to a sense of empowerment – that added up to more radicalization than she had previously believed was out there. She had travelled the country to research it, drawing lines between data and what she was seeing and hearing on the ground.

Perhaps assigning it was proof that Mr. Silver has recognized you can't just understand this campaign from inside a data lab. It's debatable how many of the conference's attendees felt that way, considering how much more they were buzzing about hearing from him than from Ms. Malone. But remaining atop his kingdom will be no easy thing, as the world around it changes in ways nobody can predict.