Donald Trump has consistently led in polls of Republican primary voters since August, when he first took the lead. But present in those numbers is an interesting wrinkle: Trump consistently performed better in online surveys than in polls conducted live on the phone.
This difference confused many pollsters: Those surveys were all conducted separately, so it was possible to chalk up the differences to external factors like how questions were posed from survey to survey.
A new experiment from Morning Consult sought to put the question to rest. Pollsters interviewed 2,397 registered Republican voters and Republican-leaning independents about their favorite candidates in the primary. One-third of the respondents took the survey online. Another third answered the same questions posed by a live interviewer on the phone, while the final third heard the same questions in an automated phone call.
Overall, 36 percent of voters picked Trump as their favorite candidate after the last Republican debate. But his levels of support differed markedly among the modes of questioning.
Of the respondents answering questions online, 38 percent picked Trump for president, while only 32 percent of respondents named him when speaking to live pollsters. (Thirty-six percent chose him in automated telephone surveys.)
That pattern is unique to Trump. Ted Cruz did about 2 points better in live telephone surveys, as did Ben Carson. Jeb Bush saw no difference.
The gulf grew even starker among voters with college degrees: College graduates favored Trump in online surveys over live telephone by about 10 percentage points.
What’s going on here?
The most likely explanation for this chasm in levels of support is a concept in social psychology known as "social desirability bias."
"Social desirability bias Is this tendency for survey respondents to provide answers that lead interviewers to hold a more favorable view of them," said Kyle Dropp, a co-founder at Morning Consult and the study’s main author.
This tendency to lie to pollsters is present all over the place. People, for example, severely underreport their involvement in taboo activities such as using illegal drugs or masturbating. And the phenomenon works in the opposite direction: People are more likely to tell callers they have donated to charity, for example, than they actually are to donate.
In politics, the idea of social desirability bias first gained notoriety in the early '80s, when longtime Los Angeles Mayor Tom Bradley, one of the most prominent black politicians in the country, ran as a Democrat in the California governor’s race. Though polls projected his victory by a significant margin, he lost narrowly to the Republican candidate, a white man.
Several years later, Douglas Wilder successfully ran for governor in Virginia, becoming the first black governor to hold office since Reconstruction.
Wilder won his 1989 race by a half-point margin, but opinion polling in the months leading up to the election consistently showed him leading his white opponent by as many as 9 percentage points.
After the fact, political scientists discovered that in both races many white voters told pollsters they planned to vote for the nonwhite candidate but ended up voting against him. This became known as the "Bradley effect" (or sometimes "Wilder effect"): Voters voiced false support for the nonwhite candidate to avoid opening themselves up to criticisms that they were racist.
Political scientists believe the Wilder/Bradley effect has since dissipated, especially with the election of Barack Obama in 2008 and 2012.
"The Wilder effect declined to insignificance swiftly at about the time that welfare reform silenced one critical, racialized issue, and as crime’s national salience was declining," wrote Daniel Hopkins, a Harvard professor who found the effect disappeared at some point in the mid-1990s. If race once more becomes a salient political issue, the authors warned, the Bradley/Wilder effect could reappear.
Even in this Morning Consult poll, former neurosurgeon Ben Carson, the only black candidate in the Republican primary, does slightly better in live polls than in automated surveys. Dropp said he suspected social desirability bias might be at play, but the 2 percentage point difference is within the poll’s margin of error.
In the case of Trump, though, social desirability bias appears alive and well. It seems even Trump’s supporters understand that favoring him is not entirely socially acceptable. But that doesn’t diminish their backing — that Trump is loathed by political elites is part of his appeal.
What the survey ultimately indicates is that Trump’s support might, if anything, be slightly underrepresented in the polls we see make headlines.
Commentators set Trump’s hard ceiling of support at around his highest poll numbers, but if political correctness is holding back additional voters from publicly voicing their support, the ceiling could realistically be higher. When voters cast their ballots, political correctness won’t be on their minds.