Polls have faced heightened scrutiny since the 2016 election, when they were widely panned as inaccurate in failing to predict Trump’s victory. The criticism was partly misplaced: National polls actually proved to be accurate in estimating a small advantage for Hillary Clinton, who won the popular vote by 2.1 percentage points. But polls in key battleground states consistently underestimated Trump’s support, providing an uncomfortable reminder that such surveys have historically been less accurate.
Early skepticism played out prominently on CNN on Tuesday night. “I’m struck at how wrong polling was in so many places,” said host Jake Tapper. “Obviously there were polls that showed Andrew Gillum winning the governor’s office in Florida.” Tapper concluded that “once again, it does seem like there are some Republican votes that people are not able to pick up. . . . They’re always polling wrong in that direction.”
Shortly afterward, Democratic strategist David Axelrod predicted that poll performance “is going to prompt another round of soul-searching about whether and how you can poll accurately, because a lot of these races that were blowouts tonight or apparently blowouts tonight polled as tough races.” He noted, however, that the Florida poll results, which Tapper had criticized, “actually fell within the margin of error.”
Axelrod offered a more positive assessment Wednesday morning, saying in an email to The Washington Post, “Within margins of error, I think polls were reasonably good.” CNN declined to comment Tuesday on Tapper’s characterizations of poll accuracy.
Two polling experts reached by The Post on Tuesday praised the Tuesday showing.
“Election polling had a pretty good night,” said Scott Keeter, a senior survey adviser at the Pew Research Center. “The big-picture story from the polls turned out to be accurate: Polling showed that the Democrats had a very good chance to take the House back and that Republicans had a very good chance to hold or increase their advantage in the Senate. Both were correct.”
Final pre-election surveys tracked by RealClearPolitics found Democrats led Republicans by seven percentage points when voters were asked which party’s generic House candidate they would support. The latest congressional vote totals put Democrats’ advantage at four points, though this is expected to grow as votes from California and Washington are fully tallied. The New York Times forecasts that once all ballots have been counted, Democrats will win the popular vote by seven points, identical to the average margin in public surveys.
Democratic pollster Andrew Baumann called the pre-election polls “quite accurate, particularly for a midterm that ended up being totally different than any previous midterm.” Baumann, who conducts surveys for the Global Strategy Group, pointed to Florida as an exception.
Republican pollster Kristen Soltis Anderson was less positive, saying that she initially gave polls a “C-plus” for their accuracy but that “a B-minus or B is perhaps more fair.” Anderson, a partner at the firm Echelon Insights, said that even though pre-election polls largely met her expectations, a handful of high-profile surprises occurred, with election results diverging from final polling averages, including the Republican wins in Florida and large Republican winning margins in Indiana and Missouri.
In Florida’s Senate race, Sen. Bill Nelson (D) held a lead of roughly three points across 17 polls conducted since mid-October. At one point Wednesday afternoon, he was 0.3 points behind his Republican opponent, Gov. Rick Scott, in the vote count. While most surveys’ results were within the range of sampling error — which is nearly twice the traditional margin of sampling error when comparing both candidates’ support — as a whole, they underestimated Scott’s standing.
Keeter, of Pew, noted that while errors occurred in some of the most prominent Senate and gubernatorial races, polls did not always underestimate Republicans. He pointed out that although polls were a bit too Democratic in Florida, polls in Texas had understated Democrat Beto O’Rourke’s chances in the Senate race there.
Final Texas polls before the election found the Republican incumbent, Sen. Ted Cruz, leading by between three and 10 percentage points. O’Rourke lost by less than three percentage points, according to Wednesday’s vote tallies.
The Post’s analysis of the battleground U.S. Senate and governor polls tracked by RealClearPolitics finds that on the basis of preliminary vote counts, state polls erred by an average of 4.1 percentage points in estimating the vote margin between Democratic and Republican candidates. That is lower than an average error of 5.1 points in 2016 state-level presidential polls, but is just about average for state-level polls in presidential elections since 2000. Polling errors in key races for the Senate and governorships were also somewhat lower than the average of 5.4 points for Senate and governor elections since 1998, according an analysis by the website FiveThirtyEight this spring.
As in 2016, some of the larger polling errors this year occurred in states where polling was scarce. In Ohio, surveys by Emerson College and Gravis Marketing were the only two polls within three weeks of the election, both conducted using less-expansive automated and online methods. They found Democrat Richard Cordray up by three to five percentage points, while on Tuesday Republican Mike DeWine won by a similar margin.
On average across competitive elections for the U.S. Senate and governorships, polls underestimated Republicans’ vote margins in these contests by two percentage points, the equivalent of a poll’s finding a Republican trailing 49 percent to 51 percent when the vote result is an even 50-50.
One bright spot this year appears to have been congressional district polling, which saw a boon this year despite the greater difficulty of conducting surveys at this narrow geographic level. Bauman pointed to the ambitious effort by the New York Times and Siena College to conduct more than 90 surveys in congressional battleground districts as a success, noting that they “ended up being quite accurate and really did a good job of educating the political public about the difficulties of doing polling well.”
Monmouth University’s increased effort to survey congressional districts also appears to have paid off, with final polls proving largely accurate or consistent with the directions in which races were trending.
Emily Guskin contributed to this report.