According to most national polls, Hillary Clinton should have ended Election Day on Tuesday as the president-elect. The surveys showed Clinton leading Donald Trump by anywhere from two to five points – a bigger lead than President Barack Obama enjoyed before his reelection in 2012 – and consistently ahead in important swing states. The Huffington Post, New York Times and Fivethirtyeight.com data models were strongly signaling a Clinton win, too.
So what went wrong?
Pollsters were trying to answer that question Wednesday after what may have been the greatest polling failure since they missed Ronald Reagan’s easy election victory over incumbent Jimmy Carter in 1980.
Clinton may have eked out slightly more votes overall – she was expected to lead by about 1 percent when all the votes were tallied. That would put polls indicating a slight Clinton popular-vote win within their margins of error.
But the polls were wildly off in states like Michigan, Wisconsin and Pennsylvania – bricks in the “blue wall” that Clinton was counting on to reach 270 electoral votes and the presidency. They also missed in critical swing states such as Florida and North Carolina, both of which went to Trump.
The short answer may be that pollsters gave too much weight in their surveys to college-educated white voters and not enough to non-college-educated whites, said Scott Clement, polling manager for The Washington Post, which gave Clinton a 4-point advantage in its final poll. The former group didn’t shift to Clinton as much as late-breaking polls anticipated, while the latter came out for Trump “in a big way,” particularly in northern industrial states, Clement said.
To illustrate: The last Washington Post-ABC News Tracking Poll said Clinton held a nine-point lead among white college graduates just before voting began; the exit poll found Trump won this group by four points.
The reliability of presidential polls has fluctuated wildly over the past five elections, and there’s no clear pattern or evidence of systemic flaws. It’s not necessarily that “the polls are phony,” as Trump put it in an interview with Fox News on Tuesday. The National Council on Public Polls rated presidential polls in 2004 and 2008 as the most accurate in the history of scientific sampling.
On the other hand, Gallup – a pioneer in politic research – said Republican Mitt Romney had a one-point advantage over Obama in the 2012 election; Romney actually lost the popular vote by four points, a blown call that prompted Gallup to get out of pre-election polling.
However, as the number of polls has expanded over the years, there may be some degree of poll pollution in averages of all these surveys, said Lee Miringoff, director of Marist College’s polling operation, which conducts polls for NBC, the Wall Street Journal and the McClatchy newspapers. By combining polls – some good, some bad – into running averages, news organizations may be distorting where the race stands. “You’re tossing different modes of data collection into the stew,” he said. “And some of those methods are unproven.”
The final Marist-McClatchy poll had Clinton with a one-point popular-vote advantage on Tuesday; as of Wednesday afternoon, she had a 0.2-point edge in the overall vote.
Polls in 2016 may have been thrown off by relatively low turnout. Despite many reports about Democrats’ ability to get its voters to the polls, preliminary figures indicate that 55.6 percent of eligible voters cast a ballot, the lowest total since 2004, according to early estimates. Low-turnout elections tend to complicate polling because they throw off assumptions about who and how many people will show up to vote.
Another potential factor: Trump voters, on the margins, may have been more reluctant to tell pollsters they were voting for their man than Clinton supporters were to acknowledge their choice.
In a survey of voters taken this summer, the USC/Los Angeles Times Daybreak tracking poll found some evidence of this, notably among women who backed Trump. It also found that voters who favored a third-party candidate were even more reluctant than Trump voters to say so when asked by a pollster. The Daybreak poll – which is conducted online, without human-to-human contact – has been an outlier, consistently favoring Trump (its final tally had Trump leading, 47 to 44 percent, a result that was no more accurate than polls showing Clinton with a four-point lead).
But Miringhoff doesn’t buy the shy-voter theory: “People are generally proud of their candidate,” he said. “It’s a pollster cop-out to say, ‘I can’t help it if my model didn’t work because people wouldn’t tell me who they were voting for.’ I think Trump’s supporters were a pretty enthusiastic group.”
Saying “the polls clearly got it wrong this time,” the American Association for Public Opinion Research said it will conduct an autopsy on 2016 presidential polls. Its findings won’t be completed until next May.
Data guru Nate Silver, proprietor of the data-crunching Fivethirtyeight.com site, had pegged Clinton’s chances of winning at 71 percent Tuesday, based on a statistical model of polls. “This reflects a meaningful improvement for Clinton in the past 48 hours as the news cycle has taken a final half-twist in her favor,” he wrote.
He favored her to win between 302 and 323 electoral votes. In fact, she appears headed for just 239.
But Silver noted on Wednesday that he had repeatedly flagged the uncertainty behind such a forecast and had called attention to Trump’s relative strength throughout the campaign.
“In an extremely narrow sense, I’m not that surprised by the outcome, since polling – to a greater extent than the conventional wisdom acknowledged – had shown a fairly competitive race with critical weaknesses for Clinton in the Electoral College,” he wrote.
“But in a broader sense? [Trump’s election is] the most shocking political development of my lifetime.”
(c) 2016, The Washington Post · Paul Farhi