Last month, just before she was promoted to Trump’s campaign manager, Republican pollster Kellyanne Conway told the Today show that some of his supporters were unwilling to reveal their real preference when interviewed on the phone.
“I think that the important point to note there is that when you have online polls as opposed to telephone polls, Mr. Trump tends to do better, and that’s because the online polls approximate the ballot box, where you’re issuing your vote privately,” Conway said. “We think there’s a big hidden Trump vote in this country.”
The “shy Trumpers” idea has been around for months, and a study run just before the primaries suggested that the effect was real.
Still, polling experts are skeptical that it will influence the coming election, with its broader electorate and division along party lines. And even if the effect was real in the primaries, a BuzzFeed News analysis shows that Trump hasn’t been performing better in online polls — where any reticent supporters should be more willing to express their true feelings — in his match-up against Hillary Clinton.
The presidential horse race, in online vs. telephone polls.
Speculation about “shy Trumpers” mounted in late 2015, as Trump consistently performed better — by five or more percentage points — in online polls of the Republican primaries than telephone ones.
But that didn’t prove that some of Trump’s backers were embarrassed about revealing their support in a live interview. It could have been due to differences in the people being surveyed — a strong possibility given that polls conducted by telephone usually select participants by randomly choosing from a list of landline and cell phone numbers, whereas online panels consist of people who take the initiative to participate, often by clicking an online ad.
That’s why the polling firm Morning Consult in December ran an experiment with people who had all been recruited in the same way. Its researchers split a sample of almost 2,400 Republican and Republican-leaning registered voters, originally recruited for online polling, so that one-third were interviewed by a person over the phone, one-third spoke with an automated telephone system, and one-third were surveyed online.
“It’s hard for me to see a flaw in their research design,” Charles Franklin, a pollster and political scientist at Marquette University in Milwaukee, Wisconsin, told BuzzFeed News.
The experiment found the largest difference between the online and live-phone groups, with Trump winning 38% percent support online compared to 32% in the telephone interviews. What’s more, this effect was largest for better-educated voters: Among those with a college degree, Trump performed 10 percentage points better online than in the phone interviews. This suggested that this section of the electorate may be particularly unwilling to admit backing Trump — perhaps because they fear being thought of as racist.
The study seemed to echo a phenomenon called the “Bradley effect,” in which voters say they will support a black candidate or that they are undecided when interviewed for a poll, but then vote for a white opponent. It was the leading explanation when Tom Bradley, the black Democratic mayor of Los Angeles, suffered a shock defeat to the white Republican George Deukmejian in the 1982 race for governor of California.
The effect was invoked again in 1989, when Douglas Wilder became the first elected black state governor, winning in Virginia by only a tiny margin after enjoying a comfortable lead in pre-election polls. But when Barack Obama ran for president in 2008, it didn’t seem to influence his support.
In 2009 Daniel Hopkins, now at the University of Pennsylvania, crunched the numbers on 180 gubernatorial and Senate elections from 1989 to 2006, finding that by the mid-1990s, the Bradley effect had disappeared as a factor in American politics. And it’s unclear even if it was the true reason for Bradley’s defeat: He won on votes cast on election day, apparently losing only because Deukmejian’s campaign had successfully mobilized absentee voters.
Some survey researchers argue that concerns about “social desirability” effects — people’s unwillingness in polls to admit to opinions or behavior that may cast them in a bad light — have generally been overblown.
“If you look at the full set of studies, it’s amazing how limited the evidence actually is.”
“If you look at the full set of studies, it’s amazing how limited the evidence actually is,” Jon Krosnick, a specialist in polling methodology at Stanford University, told BuzzFeed News.
There are also more specific reasons to be skeptical about shy Trumpers carrying the 2016 presidential election. As FiveThirtyEight and The Upshot have pointed out, when the results from the primaries came in, there’s little evidence that online polls were more reliable — if anything, the telephone polls more accurately predicted the results.
What’s more, the contest between Trump and Clinton is vastly different from the Republican primaries. Not only is the electorate much bigger, but the race is between two candidates with historically high unfavorable ratings — which means embarrassment could cut both ways.
In July, the Pew Research Center found little difference between the willingness of Trump and Clinton supporters to be open about their preference: 8% of Clinton’s supporters and 7% of Trump’s said they would rather other people didn’t know.
Even the basic premise that Trump performs better in online polls than in telephone interviews seems to no longer be true. “If you fast forward to today, it appears Donald Trump is doing as well or slightly better in live interviews than online,” Kyle Dropp, Morning Consult’s co-founder and chief research officer, told BuzzFeed News.
Trump’s support has been slightly stronger in live phone polls than in those run online.
But so, too, has Clinton’s.
The big difference is that in online polls, more people say they are undecided.
This mostly reflects differences in the way questions are asked online and over the phone. Live interviewers are better able to press respondents into making a choice, whereas online, pollsters tend to allow undecided answers because they don’t want the genuinely unsure to leave the survey.
There’s also no reason to assume that undecided voters are likely to flock to Trump on election day. Indeed, Dropp said that in Morning Consult polls run so far in September, 50% of the undecideds said that they were leaning to Clinton, compared to 45% leaning to Trump.
The acid test of the shy Trumpers effect, however, would be a repeat of Morning Consult’s December experiment with a sample from the current electorate, faced with the choice between Trump and Clinton. The firm plans to run this study.
Until the results are in, however, the available polling data indicates that any shy Trumpers must be very shy indeed.