Nate Silver on What the Polls Got Wrong, and Right, About This Election

Nate Silver.Photo: Kate LaRue/FiveThirtyEight

The morning after America learned that Donald Trump will improbably be America’s next president, Nate Silver, over delicious scrambled eggs with lox at Friedman’s on West 31st Street, the sort of coastal-elite place a Trump voter might find objectionable, acknowledged that, along with category-9/11 shock, he felt some vindication, too.

The past few weeks had been rough on American politics’ most famous moneyballer. In the run-up to the election, FiveThirtyEight’s Silver, whose enthusiasm for food once inspired him to statistically rank the country’s burritos, had found it hard to eat. And for much of the campaign, Silver, normally a sound sleeper, had found himself waking up in the middle of the night “out of anxiety about whether there was a new poll and just general nervousness.” The previous night, he’d finally had a nice meal, pigging out on pasta at Lupa with his partner, but even this morning he had awoken at an uncharacteristically early 6:30. “‘Cause you wake up and are like, ‘It’s over.’ But it’s kind of not over. Everything is forever changed. Everything is different than we thought it was.”

Leading up to November 8, Silver had staked out the heretical position that Hillary Clinton had only a 70 percent chance of winning the presidency, rather than, say, the 99 percent odds that the Princeton Election Consortium was giving her. Silver has long maintained that forecasters tend toward overconfidence (he views cases where he himself has gotten everything right — such as in the 2012 election, when he accurately predicted the outcome in 50 of 50 states — as “a weird fluke”). But for nervous voters who seemed to think he should be a clairvoyant oracle rather than a rigorous sifter of probabilities, Silver was Nostradumbass. He had duked it out on Twitter with a Huffington Post reporter who’d disparaged his work as “political punditry dressed up as sophisticated mathematical modeling,” prompting Silver to deem the reporter’s article “fucking idiotic.” As it turned out, Trump won largely for the same reasons that Silver’s model had given him better odds than others: The high number of undecided voters had ended up breaking in Trump’s favor, and Clinton had underperformed in the Midwest. Silver told me the postelection reactions to FiveThirtyEight’s model, ranging from very favorable (from poker players and investors) to “you suck” (from laypeople), had been running two-to-one.

Nonetheless, over the past 48 hours, there had been much handwringing about the entire project of electoral science. “I’ve believed in data for 30 years in politics and data died tonight,” GOP strategist Mike Murphy tweeted. Silver mentioned an article just that morning in the Times: “How Data Failed Us in Calling an Election.” Even Silver, despite having been more accurate than almost anyone else, had been dragged into the breast-beating. (“Nate Silver Blew It Bigly on the Election - Can His Brand Recover?” a befuddled the Wrap mused).

Silver sees the elegies for polling as premature, noting that traditional reporters, seeing data as a threat to their craft, are always looking for ways to belittle it, and that polls still offer a much more accurate picture of reality than the conventional wisdom of the commentariat. But he allowed that, clearly, current polling needs to be improved upon. It failed to predict this year’s underwhelming voter turnout and also the emergence of Trump Democrats, and arcane methodological questions need to be answered, like whether “response bias” was a factor (people depressively not responding to pollsters after, say, a debate in which their candidate fared poorly, or ignoring calls from the loathsome MSM), and how best to reach people, as online polling eventually eclipses phone polling.

Still, though he hadn’t yet unpacked which state polls were “righter and wronger,” he saw the larger problem as being the tendency for people to equate probability with certainty, to ignore the fact that a 30 percent chance that Trump would win meant he literally did have a 30 percent chance of winning. “This was not an epic polling error,” Silver said. “The swing percentage is not all that large. For Clinton to go from a 2-point lead to a 6-point lead only takes 1 in 50 people to change their mind. So that’s not that big. It just so happens that a 3-point lead isn’t that safe, and a 6-point lead is a lot safer.”

Silver’s model had said very clearly that there was a 30 percent chance Trump would win, and he had. “I suppose I would have hoped the 30 percent involved something a bit less consequential for the country,” Silver said. “John Kasich becoming president or something.”