Why were polls so wrong in the election? How analysts explain their failures

Why were polls so wrong in the election? How analysts explain their failures

In a few hours, all expectations for the 2016 presidential election collapsed. Projections that gave Clinton between a 72% and 99% chance of victory evaporated as Donald Trump surged to victory in Democratic strongholds.

An analyst working for Nate Silver's FiveThirtyEight, Sam Wang of the Princeton Election Consortium, Nate Cohn of the New York Times and other analysts offered autopsies as to why they failed spectacularly at their jobs: Predicting the election. 

FiveThirtyEight

Among the most trusted election analyses can be found at FiveThirtyEight, a website launched by Nate Silver that has predicted national elections since 2004. Silver was among the most bullish on Trump's chances, giving him a nearly 30% chance of victory heading into Tuesday. 

FiveThirtyEight took a fairly calm approach to the election projection failure. "The miss wasn't unprecedented or even, these days, all that unusual," Carl Bialik and Harry Enten wrote. "Polls have missed recent elections in the U.S. and abroad by margins at least as big. Every poll, and every prediction based on it, is probabilistic in nature: There's always a chance the leader loses."

As they note, the projections are only as good as the polls that go into them. And while detail drawn from aggregated surveys weighted by quality and historical accuracy has been right before, the underlying polls were wrong.

Princeton Election Consortium

The Princeton Election Consortium drew substantial attention in 2016. The approach of its leader, Princeton University professor Sam Wang, differed substantially from FiveThirtyEight. As Wang explained to Wired shortly before the election, his model gave Clinton a greater than 99% chance of victory because it factored in less uncertainty. He expressed great confidence in his model, telling Wired this was "the most stable election in a long time" on Nov. 7. 

His message had changed early in the morning on November 9. Because Wang cut out factors FiveThirtyEight incorporated to focus narrowly on state-level polls alone, the scale of his miss was magnified. "We all estimated the Clinton win at being probable, but I was most extreme," Wang wrote. "It goes to show that even if the estimation problem is reduced to one parameter, it's still essential to do a good job with that one parameter. Polls failed, and I amplified that failure."

The New York Times

In June, the Upshot at the New York Times gave Trump a 42% chance of victory, his highest odds of the campaign. By election day, the Upshot pegged Trump's chances at 15%. 

Months ago, the Upshot's Nate Cohn warned that data showed there were more white voters in the electorate than previously believed. Yet Cohn, the New York Times and essentially everyone else followed state and national polls to their months-in-the-making conclusion on Election Day: Clinton would win, probably easily. 

But Cohn's months-old siren about white voters proved true on Tuesday to a degree no one expected. "Donald J. Trump won the presidency by riding an enormous wave of support among white working-class voters," Cohn wrote. "It was always a possibility, but it had always looked unlikely. Hillary Clinton led in nearly every national poll — and in other surveys in the states worth the requisite 270 electoral votes. The traditional view of recent American elections gave even more reason to think Mrs. Clinton was safe. ... The truth was that Democrats were far more dependent on white working-class voters than many believed."