Are Polls Overstating Obama’s Support?

Are Polls Overstating Obama’s Support?
There is a long history of black Democratic candidates doing worse than pre-election polling would suggest. Two recent studies disagree on whether this problem is still with us.
By: Lee Drutman | September 30, 2008

California. 1982. Los Angeles Mayor Tom Bradley, a Democrat, is running to be the first black governor in his state's history. Days before the election, he is up by almost 10 points in the polls. But when the returns come in, Republican George Deukmejian is declared the winner.
Virginia. 1989. Lt. Gov. Douglas Wilder, a Democrat, is running to be the first black governor in his state's history. Days before the election, polls give him a nine-point lead. When the returns come in, Wilder is indeed the winner — but only by a few thousand votes.
Some call it the Bradley Effect; others call it the Wilder Effect — this idea that pre-election polling tends to overstate the support for black candidates, especially Democrats. This happens, it is alleged, because there are a significant number of voters who tell pollsters that they'll support their party's candidate. But then, because they are ultimately uncomfortable voting for a black candidate, they don't.
With polls suggesting America is on the verge of electing its first black president in just a few weeks, both campaigns have largely been quiet on the issue of race. Race is, of course, a factor: A widely cited poll by AP-Yahoo News in late September found Obama's race is costing him 6 percentage points in support before the general election.
But the question lingers: Should the Obama campaign be worried that polls are overstating his support? Is his narrow lead safe from private racist sentiments that white Democrats refuse to share with pollsters?
Two recent academic studies have come to different conclusions on this question by examining past senatorial, gubernatorial and mayoral races with black candidates.
Daniel J. Hopkins, a postdoctoral fellow at Harvard's department of government, is convinced that the Bradley/Wilder Effect is largely a thing of the past. His analysis suggests that, at least since 1996, pre-election polling on black candidates has been accurate. Racism may still exist, but it no longer skews polling.
But Christopher Stout and Reuben Kline, both Ph.D. candidates in political science at University of California, Irvine, think that the Bradley/Wilder Effect is still very much with us, and that, all else being equal, polling numbers will overstate support for black Democrats by about three percentage points.
Stout and Kline argue that in order to understand the Bradley/Wilder Effect, one should pay attention to how competitive the election is. They find that in close, high-turnout elections, the effect tends to be more pronounced.
The reason, they argue, is that in a close election, Democratic voters feel more social pressure to support the Democratic nominee, fearing that any stated deviation from their party would make them look racist, since what other justification could they have for voting Republican in such a close race?
"They don't want to look racist to the pollster, but then they end up not voting for the black candidate or just not voting at all," Kline said.
Controlling for closeness of election, turnout and a few other contextual factors, their model predicts that, on average, pre-election polls will overstate support for black candidates by 3.3 percentage points.
Though a simple plot of elections over time makes it look like the Bradley/Wilder Effect fades post-1996, Stout and Kline argue that this is actually an artifact of other trends. One, there haven't been that many close elections involving black candidates in recent years. And two, there are more black Republicans running, and Republican voters are more comfortable telling pollsters they will vote against a black candidate (hence, no Bradley effect).
"People say the Bradley Effect is going away, but they're looking at elections that are not competitive and more Republican black candidates," Stout said.
Hopkins, however, argues that the post-1996 disappearance of the effect is indeed real but easily missed because of a frequently overlooked aspect of polling bias — the tendency to for polls to overstate the frontrunner generally (his best guess is that there is a 2.5 percentage point bias in favor of a leading candidate). "I think the Wilder Effect can sometimes be confused with the frontrunner effect," Hopkins explained.
Taking the frontrunner effect into account, Hopkins finds that evidence for a Bradley/Wilder Effect of about 2.3 points but only until 1996. Then, he finds, it just drops off pretty suddenly.
"One thought is that this is a generational effect, and that there is a particular generation that harbors racial biases but an anti-racist norm, and the Wilder Effect is a factor of those two processes together," Hopkins said. "But if that were true, it should disappear gradually."
Rather, Hopkins thinks it has something to do with the way the political agenda changed around that time. "By 1997," he noted, "welfare was off the agenda, crime's prominence as an issue had declined; so two racial issues are less salient. And starting in the 1990s, Republican candidates have run fewer racialized campaigns."
Hopkins also points to the work of Zoltan Hajnal, whose recent book, Changing White Attitudes Toward Black Political Leadership, has shown that white voters who have concerns about black political leaders tend to lose those concerns once they actually experience black political leadership. So, Hopkins wonders: "Did Wilder himself help to eliminate the Wilder Effect by showing that there wasn't much to fear from an Afro-American executive?"
And what about the upcoming presidential election? What should we expect?
"It's a close election and a high-turnout race, and all this indicates there's going to be a decent Bradley Effect," Stout said.
Kline, meanwhile, noted that in swing states such as "Ohio, and probably Michigan, you have socially conservative Democrats, and that's probably a problem for Obama because it seems like those should be the voters most likely to falsify their preference (i.e., lie to pollsters). They should be voting Democratic, but they may be reticent to do so."
Indeed, a recent AP-Yahoo News poll found that 40 percent of white Americans and more than a third of white Democrats and independents maintain negative attitudes toward blacks. But this doesn't mean necessarily that Obama's support is overstated. In fact, it may be one reason why Obama is not doing better.
"The Wilder Effect is often misunderstood," Hopkins said. "People tend to think that if there is no Wilder Effect, then there's no racial bias, but the Wilder Effect is one form of racial bias, and race can in many ways influence voter choices without producing a Wilder Effect."

http://www.miller-mccune.com/article/732