Is There a Problem with Polling? Not Exactly
Following Donald Trump’s surprise victory last November, the popular sentiment was that the polling industry had terribly botched things. What else could explain election eve projections giving Hillary Clinton a four-out-of-five chance, or better, of winning on Nov. 8?
Scientific polling provides a useful tool for measuring how people think and feel about a particular topic. Useful beyond gauging winners at the ballot box, polling also helps organizations of all kinds determine stakeholder preferences when developing outreach plans or communications strategies. But after what seemed like a monumental failure in traditional polling leading up to the 2016 U.S. presidential election, many were left to question how effective the process really is.
In reality, national surveys were highly accurate in predicting the popular vote. An aggregate of national surveys indicated that Clinton would win the popular vote by a 3 percent edge, according to an autopsy of the polling for the 2016 election performed by the American Association for Public Opinion Research. Clinton ended up winning on a national basis by 3 million votes, or by a 2.1 percent margin.
The shortcomings that threw off the results by undercounting support for Trump occurred almost entirely in polling in the state contests, where the election ultimately was decided, the study found. Several major factors worked against the accuracy of state polls, including the many voters who remained undecided until the last minute. Exit polls revealed that late-deciding voters in Florida, Michigan, Pennsylvania and Wisconsin broke for Trump by substantial margins, reported the Washington Post. Clinton lost the latter three states — and, in turn, the electoral vote race — by a total of 77,744 votes, or one-half of one percent.
Pollsters also failed to properly weight their samples by voter’s educational level, a problem because non-college educated voters backed Trump to a much greater degree than those with college degrees. The study also pointed to poll aggregates and the authors of prediction models to explain why so many on Nov. 9 felt they had been seriously misled. “They helped crystallize the erroneous belief that Clinton was a shoo-in for president,” it stated.
To avoid the flawed projections generated during last year’s presidential election, in the future, pollsters will need to do a better job on multiple fronts. They need to conduct better demographic modeling to ensure surveys accurately capture the views of a diverse electorate. Specifically, pollsters will need to adjust their estimates for voter turnout, and their propensity to respond to surveys in general or online ones in particular.
That’s the formula French polling firms followed in a conscious effort to escape the pitfalls suffered by the industry tracking the U.S. presidential race.
The key is to capture underrepresented voters, including young adults who often are not interested in answering pollsters, seniors who feel harassed by telemarketers, residents in rural areas and the less-well educated, Jean-Daniel Lévy, head of political polling at Harris Interactive, told Politico. In the first round of France’s presidential contest, Harris’ predictions were within one percentage point of the final readout for the four front-runners.
“Where we knew that we were getting less feedback from voters, particularly among the very young or the elderly, we tweaked our models accordingly,” said Lévy.
Based on the success demonstrated by French pollsters, there’s good reason to believe the U.S. polling industry can make the necessary technical adjustments to better capture the electorate’s leanings in a close race. But the larger task for the industry will be regaining the public’s confidence in polling as the surprise result of the 2016 presidential election almost certainly will remain vivid for years.
The midterms are only a year away and will provide the industry hundreds of opportunities to forecast races at the local and state level, and take the first steps to overcoming public skepticism. It will be critical to pay attention to how the lessons learned after the 2016 election were applied in this field. The result will shape how organizations use polling and how target audiences respond, which will have a broad and significant impact on the implementation and impact of outreach efforts.