UA study on Trump, Hillary Clinton support shows loyalty not as strong in private

Some political polls incorrectly predicted the outcome of this year’s presidential election, and a University of Arkansas economics professor believes it could be partly because they asked the wrong questions.

Andy Brownback, assistant professor of economics at the Sam M. Walton College of Business, recently conducted a national survey on feelings toward President-elect Donald Trump and his opponent, Hillary Clinton, that he says indicates “polling methodologies exaggerate differences in voter preferences.”

Brownback said the study, which quizzed 1,000 voting-age Americans on Nov. 1 through the Amazon’s Mechanical Turk website, showed individuals are resolute in pledging their political allegiance to a given candidate when asked directly, but soften their views and are less dedicated to the chosen candidate when able to mask their answers.

Within the study, Clinton supporters were more likely to privately agree with Trump than vice versa.

“If we interpret the more privately elicited preferences as indicative of true feelings and likely voting, this would directly represent an overstatement of support for Clinton from explicit polling,” Brownback said.

One solution to the discrepancy between polls and election results, he said, is to ask likely voters questions that take into account possible flexibility in the voters’ opinions, questions that gauge agreement with the candidate rather than explicitly asking who will likely get their vote. Brownback also believes some Clinton voters, and possibly voters in general, might not admit to agreeing with the opposition because the individual’s identity might be tied to his or her choice of political alliance.

STATE POLLS VS. ELECTION RESULTS
Jay Barth, Ima Graves Peace Distinguished Professor of Politics at Hendrix College in Conway, believes political environment could play a part in the polls being off in certain areas.

Two statewide polls fared much better than some swing-state polls and national data in predicting the Election Day outcome, where 61% of Arkansans voted for Trump, and 34% voted for Clinton. Barth helped conduct a Talk Business & Politics/Hendrix poll that showed Trump with 55% support and Clinton with 34% support among likely voters in Arkansas. The poll looked at 831 individuals between Sept. 15 and 17.

A few weeks later, the 18th annual Arkansas Poll, designed and analyzed by UA political science professor Janine Parry, had Trump leading Clinton among likely Arkansas voters 51% to 31%. The poll, conducted Oct. 18-25, looked at 800 randomly selected individuals, interviewed by phone.

Barth believes one reason Arkansas polls were more accurate could be that the overall view of Trump was more positive than it was in some of the other areas in question.

“In an environment where you perceive your candidate to be unpopular, you might be hesitant to be overt in your support,” he said. “There can be social pressures to hide one’s real beliefs.”

DOING BETTER
What does all this mean for the future of the polling industry? Barth said changes might be in order.

“Much-discussed polling errors in those key states in this election cycle have people discussing how to do polls better,” he said.

Barth agreed with Brownback’s assessment that a possible solution is to “figure out strategies to measure implicit, rather than overt support.” At the same time, the industry must find the best way to gather real opinions in a multimedia world, he said.

“I will always be concerned with how to truly and accurately gauge opinion. That’s the job of a pollster.”

Parry agrees with Brownback and Barth’s assessments of the polling industry and the recent election but warns against going too far in making changes in the polling process.

“We hear about when polls are incorrect because it’s so unusual,” Parry said. “I hope people won’t overcorrect, because this has been an unusual election.”

Barth agreed on that point.

“Trump’s statements and actions violated so many norms, which may have made some respondents less willing to articulate support for him,” he said.

If the election was unique, however, that doesn’t mean it’s not educational to pollsters.

“We learned that there are some subjects and apparently some candidates that some voters might be reluctant to reveal their true opinions on,” Parry said.

POLL CORRECTNESS
There are other theories as to why the polls proved wrong. For example, some say polls look at likely voters who have cast ballots in recent years, and many of Trump’s backers might not fall within that category.

Parry said the Arkansas Poll sidesteps this issue because its sample is a true representation of Arkansas adults, and the respondents are then asked to self-identify whether they plan to vote in the upcoming election. All in all, Parry said with the election being close, there is no way to find one key factor that went wrong in the polls that were off.

“When everything is at the margins, no one reason is going to suffice as an explanation,” Parry said.

If anything comes of out of the situation, Parry hopes its more transparency with polling. Parry said she wasn’t sure how many polls used her approach of a simple sample, rather than polling only likely voters who voted in recent elections.

“The fact that I can’t tell what method they’re using is telling,” she said, adding that polling companies guard their voter screening process as their “secret sauce,” leaving the details behind how they got their data murky.

BROWNBACK’S METHOD
For his study, Brownback used a method he believes might reveal voters’ true feelings. Of the 1,000 people surveyed, all were asked who they planned to vote for. Then, half were prompted to evaluate their feelings toward Clinton, and the other half were asked about Trump.

For each candidate’s group (500 looking at Clinton and 500 looking at Trump), half were asked directly whether they often agree with the candidate and were also instructed to say how many statements, out of four presented, they supported.

The other half, however, were presented with the same four statements, plus one, the last of which was “I often find myself agreeing with (the candidate in question)” and were simply asked how many statements with which they agreed, without having to specify which ones.

“We do this to understand how the observability of an answer affects the statements made,” Brownback said in a press release. “This is a mechanism to elicit preferences that people may be hesitant to reveal explicitly.”

The four statements given to the respondents were “red herrings,” a control group designed to shed light on whether they agreed with the final statement, based on the number of affirmative statements they reported, Brownback said.

For example, the team assumed if a person agreed with the statement, “I prefer presidential candidates who oppose the National Rifle Association,” he or she were likely to agree with the statement, “I think the threat of global warming is exaggerated.”

The other given statements were: “I agree with George H.W. Bush’s foreign policy,” and “I think small businesses are important for the economy.” Brownback said the team assumed most respondents would agree with the value of small business.

If Brownback’s project is not a blueprint for how to improve polls, he believes his data at least show Americans’ political allegiances might not be as deeply ingrained and inflexible as they seem, and, therefore, the country might not be as divided as individuals purport.

“If supporters and detractors of a given candidate are asked directly, their views of that candidate differed by about 85 percentage points. If their responses were elicited indirectly, however, their views differed only by about 57 percentage points,” he said.

That is still a significant difference, but Brownback said it closes the gap a bit.

“Hopefully, that’s comforting to some people,” he said.