This election will test America's willingness to vote for an African American president. And there are at least two relevant questions: will Barack Obama suffer from a defection of Democrats and Independents who might otherwise have voted Democratic, but won't support a black candidate? And, secondly, will a Bradley Effect occur, or is it already occurring, in which the polls understate John McCain's support against Barack Obama?

A extensive new poll commissioned by the Associated Press and Yahoo and conducted by Knowledge Networks shows that Barack Obama could suffer in the final result from white hostility to his race. Knowledge Networks claims that by getting respondents to fill out answers on a computer screen (it provides computer access to people who don't have it), it can get more honest answers to questions about race than it would have gotten from in-person or phone interviews.

The poll, conducted  from August 27 to September 5, shows Obama in the lead. That shouldn't be surprising given that most of the polling took place before the climax of the Republican convention. But the poll still shows a resistance to Obama among white Democrats that appears to be based on race. Using questions from the psychologists' "racial resentments" survey, the poll finds that among white Democrats, a significant proportion of those who exhibit racial resentment say they are not voting for Obama. 

Overall, 71 percent of white Democrats back Obama.  Only fifty-nine percent of white Democrats who backed Hillary Clinton back Obama. And here are the racial resentment numbers: Forty-two percent of white Democrats agree with the following statement: "Italians, Irish, Jews and other minorities overcame prejudice and worked their way up; blacks should do the same without special favors." Of these, only  61 percent back Obama.

Twenty-eight percent of white Democrats agree that "it's really a matter of some people just not trying hard enough; if blacks would only try harder, they could just be as well off as whites." Of these, 56 percent back Obama.  Forty percent of white Democrats disagree that "generations of slavery have created conditions that make it difficult for blacks to work their way out of the lower class." Of these,  61 percent support Obama. It's hard to do the math for these figures, but according to the AP summary, they suggest that Obama may be currently losing as much as six percentage points in the polls as a result of prejudiced white voters who might otherwise back a Democrat.

Nate Silver is right to questions about these results. As he notes, they don't screen out unregistered or unlikely voters.  And the racial tilt is probably less than six percent. And what are described as "the full poll results" don't include figures for "white Democrats."  They are included in the subsequent AP stories about the results. So it is hard to make your own calculations about white Democrats--or independents--and racial resentments.

But it is also hard to know what to make politically of these figures. One would have to know where these prejudiced voters were concentrated. One would also have to know the extent to which the resentment uncovered is going to govern these voters' final choice. What's interesting from the polls is that more than half of the people who do score negatively on these tests are backing Obama. The question is how strongly the remaining 40 percent or so feel about not backing an African American. Will an Obama campaign that draws a sharp line between him and McCain on economic issues sway them? As our current editorial suggests, the Obama campaign would be wise to press these differences.

This controversy about racial resentment is often conflated with the question about whether a Bradley Effect is taking place. Racial resentment could affect results and show up in polls. The question is whether whites or other voters who harbor these resentments tell pollsters that they are "undecided," but end up voting for McCain; or whether in conducting polls, pollsters under sample these voters because they are more reluctant to talk to pollsters. When I was at the Republican convention, I did hear from one person who claimed to be close to the McCain campaign that they are counting on the Bradley Effect to pull them past the finish line in November. 

As former TNR intern Daniel Hopkins has shown, the Bradley Effect has not appeared unequivocally in elections in over a decade. But there is some controversy about whether it showed up in the Democratic primaries this year and will therefore appear in the general election.

One Republican website has made a case that McCain will benefit from the Bradley Effect in November. Sean Oxendine at The Nextright argues that outside of the deep South, McCain could benefit from a two percentage point Bradley Effect.  He arrives at this figure by looking at the discrepancy during the Democratic primaries between estimates of Obama's strength and the final results in primary states outside the deep South. Nate Silver has disputed Oxendine's findings. Silver argues that Oxendine did not have grounds to exclude the South from his calculations. In addition, he says that by using a different pre-election polling average from the one Oxendine rather than gets a statistically insignificant result.

I am not going to get into the controversy about polling averages, and I don't share Silver's insistence that Southern primaries be included. As my colleague Noam Scheiber has argued, these primaries may have exhibited a reverse Bradley Effect, due to African American voters who backed Obama saying they were undecided. Still, my overall position is closer to Silver. If you look at Oxendine's results, they are enlarged significantly by ten percentage-plus polling disparities in New Hampshire and California--two states in which there were other explanations for the polling disparities.  In New Hampshire, much of the difference was the result of a last minute pro-Clinton surge that didn't show up in polls that stopped polling two days before the election. In California, polls may have undercounted and miscounted Hispanic support for Clinton. So I think the actual spread is closer to one percent, which is not that statistically significant.

Still, there are states that raise questions about a Bradley Effect. As Hopkins argues, much of the disappearance of the Bradley Effect can be attributed to the reduction of popular concern in the late ‘90s about racially-loaded issues like welfare and crime.  In the primaries, racial issues only began to surface significantly after the revelations about Obama's connection to Pastor Jeremiah Wright. Democratic voters were talking about Wright in Pennsylvania, West Virginia, and Kentucky, all of which exhibit a three percent or higher underestimation of Clinton's support against Obama. But on the other hand, polls underestimated Obama's support in Indiana by 3.6 percent. 

So I am not sure what to conclude, except I have a lingering suspicion that a Bradley Effect could show up in some of these swing states where the votes of white ethnic Democrats are going to make the difference. I think if you were an Obama supporter, you would want him to be at 50 percent or more in the polls on the eve of the election.  If he is ahead 48 to 47 percent with a lot of undecideds, I would worry. And in that respect, I would see some connection between the AP-Yahoo findings on racial resentment and the possibility of a Bradley Effect. 

But a final warning: None of this is to argue that race will determine the final result. Even if there is a Democratic defection over race, and a pronounced Bradley Effect, it could be the case that the Democrats implicit edge in the election is sufficiently large that Obama will win. He just won't win by as much.

--John B. Judis