For decades, straight white voters have over-stated their willingness to vote for black candidates and other “Others.” How the marriage equality debate and some statisticians may have vanquished the Bradley Effect.
Image from Flickr via strangedejim
By Casey Michel
Tom Bradley was the first to experience it. Despite a healthy, late lead before the 1982 California gubernatorial election, Bradley, a black man, lost the race by a lone percentage point to the white attorney general, George Deukmejian. Pundits, angling off of post-election research, attributed the unexpected loss to white hesitancy in electing a black candidate—despite what they’d told pollsters. They took to dubbing the drop the “Bradley Effect.” The conceit is simple: Despite professions of equality, tolerance, and meritocracy, the white, heterosexual American Voter remains hesitant to see any difference in his or her elected officials. (The Bradley Effect can be seen even in both of President Obama’s elections: yes, he won, but he also received a smaller slice of the electorate’s votes than pre-election polls had predicted.)
The 2009 votes against gay marriage stymied even the famously accurate Nate Silver.
But in recent years, the Bradley Effect has been seen most widespread, and most notably, not in elections between candidates of different races, but in those focusing on gay and lesbian issues. Five years ago in California, polls showed a healthy opposition to a constitutional ban on gay marriage in six polls taken between May and October. But when the final results were tallied, however, the ban passed by a 52-48 percent margin. Maine, a year later, experienced the same statistical switch when a bill legalizing gay marriage was signed into law by the governor only to be struck down in a “people’s veto” referendum. The votes stymied even the famously accurate Nate Silver, who had pegged the Maine marriage equality camp’s success at 70 percent. Examining the results after the 2009 election he wrote, “I think we have to seriously consider whether there is some sort of a Bradley Effect in the polling on gay rights issues.” A 2010 paper out of NYU buoyed Silver’s claim, examining 32 different marriage equality ballot measures and calculating that the Bradley Effect would warrant a seven-point slide on Election Day.
Fast forward to November 2012, with marriage equality questions on the ballot in Maine, Minnesota, Washington, and Maryland. And in each state, in each campaign, came the knowledge that supporters have to combat not simply confident opposition, but that last-minute drop that had plagued every campaign prior. “We didn’t particular have name for this drop, but we were keenly aware that our side was probably going to lose several points on Election Day,” said Josh Levin, campaign director of Maryland’s 2012 marriage equality campaign. “We sort of assumed that that was not only a possibility, but that most of the undecideds would probably end up supporting our opponents.”
And yet, when the final results were in, not only did all four states that had marriage equality on their ballots see the electorate side with gay marriage (three states legalized and one overturned a ban), but they also saw the seeming disappearance of the Bradley Effect. In Maine, for instance, a poll from Public Polling Policy released November 2 showed support for legalizing gay marriage at 52 percent, with opposition at 45 percent. The state’s final tally showed nearly identical numbers, with 52.7 percent in support and 47.3 percent opposing. Polling in Minnesota, Maryland, and Washington showed the same stability between pre-election and exit polls. The Bradley Effect, remarkably, was not found in a single race.
Old markers showed that you needed a 55 percent margin before the election in order to win—if that isn’t true anymore, what is the new normal?
But then, perhaps it’s not remarkable at all. Continued polling shows swelling majority support for marriage equality. As same sex-marriage slides into normalcy—as elderly opposition shrinks, and religious opposition continues to fade—it’s only understandable that the Bradley Effect would dwindle with opposition.
Yet these are just numbers, and pointing to pure demographic shift is a bit facile. After all, opposition to gay marriage remained the status quo coming into 2012: Prior to this latest electoral season, there was only one electoral success pro-equality camps could point to— Arizona, in 2006—to bolster any confidence. Even that victory, however, proved to be transient, negated two years later in another election.
As Richard Carlbom, director of Minnesota’s campaign related, there was no catchall way of telling whether the Bradley Effect would hollow support in the campaign’s final days. And each campaign existed with its own populaces, requiring individualized tactics. “You have several campaigns where you saw the Bradley Effect, but it’s hard to compare apples to apples,” Carlbom observed, noting the national uptick in marriage equality support over the past three years. “You can’t compare Minnesota 2012 to Maine 2009—and, frankly, you can’t compare Maine 2009 to Maine 2012.”
Marriage equality advocates in each state provided their own tactics and targeted demographics, but each could point to lessons from other states’ previous failures. Minnesota reached out to its significant religious populations. Maryland targeted its minority, namely African-American, communities. Washington and Maine, with liberal-leaning majorities already, cited interpersonal relationships as reason enough for the passage of marriage equality. All four states found traction, and as the election approached, all four pointed to polls remaining steadily in their favor. And then the final numbers came through, and they realized that the numbers—for the first time in the history of gay marriage polling—had stayed true.
“If we had seen [pre-election] numbers varying wildly, I think we would have been lot more worried,” Levin said. “The big thing to figure out now is if Bradley Effect is dead. Old markers showed that you needed a 55 percent margin before the election in order to win—if that isn’t true anymore, what is the new normal?”
Indeed, with the successes seen in the four states last fall, it does seem as if a new normal has settled in. Advocates of gay rights, rather than being part of a fringe movement, are increasingly mainstream, and those pushing for marriage equality can speak with increasing confidence. And if polls and ballots are any indication, such support only continues to snowball. While opposition to same-sex marriage remains strong among the Republican leadership in the House, this week several prominent Republicans submitted a brief to the Supreme Court voicing their support for marriage equality.
Defining “normal” is a slippery undertaking, of course, but some pollsters believe they have developed a formula for quantifying it. Statistical researchers with the University of Washington, only three years after they misjudged a statewide civil union measure by nearly a dozen points, crafted a new “social desirability model,” aiming to not simply project assumptions about the Bradley Effect but to put a state-specific number on the potential drop, allowing them to predict respondents’ actual votes, even when these differ from their poll responses.
We also asked whether or not they ever felt pressure to be a bit more politically correct than they’d be with their friends.
“What we noticed [with polling on civil unions in 2009] was a huge gap with what our poll estimated and what actually ended up happening,” Betsy Cooper, a Ph.D. candidate specializing in public opinion research at UW, said. “As a researcher, you start to ask questions about what went wrong.” So during 2012, Cooper and her colleagues attempted to gauge how their previous work had gone awry. Looking at specific demographics traditionally opposed to gay marriage—especially registered Republicans and the highly religious—Cooper added a few qualifying questions within UW’s polls. Searching for a “social desirability bias,” Cooper and her colleagues hoped to discern whose support would leach come Election Day.
“We were able to get two additional questions approved for our surveys,” Cooper relayed. “The first was if any topics made [respondents] uncomfortable. If they said that they were uncomfortable with people who were gay, we moved them over to the ‘No’ column. We also asked whether or not they ever felt pressure to be a bit more politically correct than they’d be with their friends. If they said they felt that pressure, we’d also move them to the opposition.”
Cooper added that she and her team also moved, based purely on demographic assumptions, a handful of those suggesting they’d support marriage equality into the opposition column – including those who were highly conservative, “who had a 96 percent probability of voting ‘No,’” Cooper said.
Without the qualifying questions, Cooper found that, as of mid-October, a preponderance of Washingtonians backed marriage equality: 57.3 percent in support, 36.2 percent opposed. However, with the additional questions and subsequent shuffling, support shrunk to a 52.9-46.6 percent split. The search for that “social desirability bias” had painted a far less rosy picture for gay marriage supporters, but one far more in line with reality, as Washingtonians ended up voting in marriage equality with 53.3 percent in support and 46.7 opposed. With Cooper’s team’s method, Washington managed to help eliminate the Bradley Effect not only through effective campaigning, but with some of the most refined polling in any of the four states facing marriage equality questions.
“There continues to be social desirability bias,” said Zack Silk, campaign manager at Washington United for Marriage. “It’s just that pollsters now know that it is there and they can account for it…. It’s not over, but the Bradley Effect is greatly diminished.”
Thirty years after V. Lance Tarrance, Jr., the campaign manager for Bradley’s 1982 opponent, called the Bradley Effect “a pernicious canard,” the statistical phenomenon may have finally faded—at least from the issue of same-sex marriage. Cooper noted that success in analyzing a social desirability bias doesn’t have to be limited solely to campaigns for same-sex marriage. “We actually think this method can be extrapolated to other social issues, like abortion,” she said. The revised approach to polling will be useful wherever voters are uncomfortable giving voice to the way they’ll vote.
The next votes on gay marriage—in Oregon and Hawaii, presumably—remain 20 months away. Pollsters may never know whether the Bradley Effect has become irrelevant when it comes to gay marriage; the Supreme Court is currently hearing two marriage equality cases this session and could make same-sex marriage not just the new normal but also the new legal.
Casey Michel is a current editorial fellow with Houston Press, and a former polling fellow with Talking Points Memo. He’s also a former Peace Corps Volunteer in Kazakhstan, where he drank far more vodka than he would have liked. He can be found on Twitter at @cjcmichel.