Khazen

In 1982, California voters were supposed to elect former Los Angeles Mayor Tom Bradley as the state’s governor. 

Bradley went into the election with a sizable lead over George Deukmejian. Exit polls projected a Bradley win. But when the ballots were counted, Deukmejian came out the winner.

Thus the "Bradley effect" was born — named as such because many white voters, who told pollsters they were voting for an African-American (Bradley), ended up breaking for the white candidate (Deukmejian).

Could Donald Trump be the 2016 version of a reverse "Bradley effect?"

That’s the theory of a new study released earlier this week by Morning Consult, a DC-based data and technology company.

The study examined a mystery that has confounded polling analysts over the past few months: Why does Trump perform better in online-based surveys than polls that include live-telephone interviews?

The study posited that voters, when interviewed by pollsters via telephone, are reluctant to admit their support for a controversial candidate whose critics have painted him as racist. In self-administered online interviews, on the other hand, they will be more likely to admit their support. And that will more likely reflect their eventual decision in the privacy of a voting booth.

"Much work remains to better understand which types of polls are actually right in predicting Trump’s support levels, but a key implication of the study is that many national polls may be underestimating Trump’s support levels," the study concluded.

If true, the results could indicate that Trump’s support in polls, which has risen to new heights over the past few weeks, is actually understated. Trump has climbed as high as 41% in a recent Monmouth survey of national Republican primary voters. On average, he has a 17-point lead over his closest Republican contender, Sen. Ted Cruz (R-Texas).

Overall, the study found that Trump performs about six percentage points better in online surveys compared with those conducted by phone. In a sample of nearly 3,000 Republican voters, the Morning Consult study found that Trump garnered 38% support with online respondents, 36% with respondents who were robo-called, and 32% among live-interviewed voters.

People’s level of education may have something to do with whether they’re willing to openly back Trump in live interviews, the Morning Consult study suggests. Polls have shown that less-educated voters have constituted the bulk of Trump’s support. But college-educated voters are more inclined to back him in online surveys than by phone.

The research suggests this is due to the "social desirability bias" — wherein, in live-survey interviews, respondents provide an answer they believe will be viewed more favorably by others. It’s the same bias that is said to have fueled the so-called Bradley effect.

In the case of the "Trump effect," blue-collar voters aren’t embarrassed about their support — their support is consistent in both live-interview and online surveys. But there’s a clear difference among college-educated Republicans.

"Among adults with a bachelors degree or postgraduate degree, Trump performs about 10 percentage points better online than via live telephone," the study said.

Among pollsters and other political analysts, the theory makes some sense. Public Policy Polling, a Democratic polling firm, conducts polls via automated robo-calls. This method also reduces the chances of bias that result from live-interview polling, since all respondents hear the exact same questions posed exactly the same way through a technology called Interactive Voice Response. 

Tom Jensen, the director of Public Policy Polling, told Business Insider the firm would often find results that would more accurately reflect the public’s attitude on the issue of gay marriage than more traditional, live-interview polls. The speculated reason: Respondents are more likely to want to appear sympathetic to the issue when talking to a real, live person.

"I think the theory that people are more reluctant to tell a live interviewer they support Trump is certainly plausible," Jensen said. "For instance, on gay-marriage referendums over the years we always showed less support for gay marriage than live-interviewer polls and were proven to be right by the election results — because people thought the socially desirable answer was to say they supported gay marriage and were more comfortable telling us how they really felt."

Other experts, while intrigued by the idea, cast some skepticism on the final thesis — the primary reason being that many online-based surveys have proven unreliable.

Sam Wang, a polling expert and professor at Princeton University, told Business Insider that he was "attracted to the idea." But he sent over a list of caveats. 

First, as he pointed out, the live-interviewer surveys featured slightly different samples than the online polls. The live samples were comprised of 4% to 5% more men, 6% to 8% more from the 30-and-older crowd, and 3% more Mitt Romney voters. If any of those groups are less favorable to Trump, the end results could contain some bias.

Second, we don’t quite know how Trump’s results compare to other candidates in the field. Is this phenomenon unique to Trump, or is it seen with other contenders too, such as Cruz?

"Bottom line: I think the effect’s no more than three percentage points, it’s therefore not enough to make a difference in how we interpret surveys, and it might disappear as the primary campaign season goes on," Wang said.

University of Michigan political scientist Michael Traugott, meanwhile, told Business Insider that he wondered to what extent the "computer-literate" sample represented the overall Republican electorate.

"This is an interesting piece of analysis of a cleverly designed study," he said. "… One issue that is not discussed is the properties/quality of the original sample drawn from an internet panel and how well they represent the US population of Republicans or likely Republican voters. Even though the experiment involves random assignment to the three different modes for the candidate preference part of the study, I would wonder about how well these relatively computer literate respondents represent the overall population of interest."

Other polling analysts posit that support for Trump may, in fact, be overstated: He often performs better in polls that feature Republican-leaning independents, for instance. And many experts question whether Trump supporters are likely to turn out to vote in primaries and even more complicated caucuses next year. 

If one thing is for certain, it’s that Trump’s candidacy is rewriting the rules of politics and polling.

When Trump first rose to front-running status in the summer, a number of theories prevailed: He had a "ceiling" of support. He wasn’t looked upon favorably enough to win. His poll numbers were overstated because of his celebrity status and high name recognition.

Said Jensen, the Public Policy Polling director: "I don’t think there’s any way to know for sure until we actually start doing some voting!"