POLITICS

Pollsters And Forecasters Had A Rough Night

It’s time to figure out what went wrong.

09/11/2016 6:34 PM AEDT | Updated 10/11/2016 2:35 AM AEDT
NEW! HIGHLIGHT AND SHARE
Highlight text to share via Facebook and Twitter

When Donald Trump won the Republican primary nomination, he did so against the predictions of many pundits, but in line with months of polling data.

His win Tuesday, by contrast, represents a stunning upset, going against the vast majority of public polling and every major political forecast. Although predictions of Democratic nominee Hillary Clinton’s margins varied, few doubted she would win.

Trump’s argument that he held a potent well of white supporters hidden from the polls, long mocked by pundits, turned out to be correct.

Unlike in Britain’s Brexit referendum, where even “shy” voters were willing to admit their preferences in online surveys, Trump’s support seems to have passed entirely under the radar.

There were no polls that showed Trump ahead in some of the states that he won. Clinton’s evening started with decent-looking early returns, but things quickly went south, starting in Florida, which the HuffPost Pollster poll aggregate and forecast model thought would go toward Clinton. North Carolina soon followed the same path, although it was slower to be called.

Both states were close according to the polling averages. Clinton was up by 1.8 percent in Florida and 1.6 percent in North Carolina. As of early Wednesday morning, Trump won the states by 1.4 points in Florida and nearly 4 points in North Carolina.

Trump’s path to victory continued through Ohio, which he was widely projected to win. Wisconsin was the state that sealed it for him, though. That state was supposed to go Clinton’s way, with the HuffPost Pollster average showing her leading by 6.1 points. Not a single poll ever showed Trump winning in Wisconsin.

The wave continued down-ticket, where Republicans pulled out races in Indiana, North Carolina, Pennsylvania and Wisconsin. Results are still outstanding in New Hampshire. Democrats held Nevada and took Illinois, but that won’t be enough to take the majority. Democrats needed to flip five seats in order to win the majority.

Of course, all of these individual polling misses mean that HuffPost Pollster’s forecasts for both the Senate and the presidency were wrong. It’s that simple ― the forecasts relied on the polls. One of us (Natalie) said a few times before the election that if the polls go down, our model is going down with them. It’s not just ours, though. Most forecast models had Clinton up and Democrats with a good chance of winning the Senate. 

Some national pollsters will come out within their margin of error for the popular vote. As of early Wednesday morning, the New York Times Upshot projects that Clinton will have about a 1-point advantage in the popular vote despite losing the Electoral College. IBD/TIPP, which looked like an outlier, will be right on if that 1-point popular vote margin holds. So will McClatchy/Marist. A 1-point result would be within a few others’ margins of error. The LA Times/USC poll, which HuffPost Pollster didn’t use because of its question wording, predicted a Trump popular vote win by over 3 percentage points.

But that’s small comfort given the systemic poll failures at the state level.

Misses in as many as five well-polled states (depending on final outcomes in Michigan and Pennsylvania), and botched margins in a few others (Virginia, for example, was a lot closer than expected), should lead to considerable introspection about the state of the field. We hope organizations like the American Association for Public Opinion Research will help pollsters look into the questions that have arisen as the British Polling Council did when polls missed the parliamentary election in May 2015.

It’s too early to know what happened. The exit polls will shed some light on it as we move forward, but we don’t have final exit polls yet. Claims that there was a “silent majority” or “shy Trump” voters can’t be ignored. If those are indeed where the polls missed, it’s time to take a good, hard look at surveys’ extremely low response rates, as well as how we locate voters. And we’ll want to look at the effects of voter identification laws and voter registrations being purged as well. 

In the coming days and weeks, we’ll be working on analyzing the polls ― what worked, what didn’t, and where to go from here. Certainly not all polls failed, and certainly the high numbers of undecideds and self-proclaimed third-party voters mattered. But there are some major issues that we look forward to helping address.

More On This Topic

Advertisement
Advertisement