Editor’s Note: Last month, I published a post about the race card factor in the presidential campaign and I referred to research by Anthony Greenwald, a psychology professor at the University of Washington, and Bethany Albertson, a fellow faculty member in the political science department, who believe the race factor matters. This new post throws more light on the issue including updated analyses by the researchers which carry ramifications for how pollsters administer their surveys and for how media should be reporting polls and their meanings in major campaigns.

In the final month of this presidential campaign, especially as Barack Obama seems to be gaining statistically significant momentum in the polls, viewers and readers will be hearing a great deal from the media interpreting the meaning of these polls. Some, in fact, will likely tell you — without benefit of any statistical or empirical evidence in hand — that the race card factor could translate into Obama’s margins being trimmed by as much as five percent.

We also know that many in the media are approaching election night exit polls with a great deal of apprehension, recalling that Obama’s exit polls numbers were better than the actual vote tallies. Meanwhile, a good many pollsters will go on record, defending the integrity of their procedures especially in the way they sample historically under-represented groups. They also will tell viewers and readers that the widely-cited Bradley Effect — observed in such high-profile gubernatorial races as California and Virginia during the 1980s and early 1990s — is no longer a concern. The phenomenon refers to 1982 when survey scientists noticed that Tom Bradley, an African-American mayor of Los Angeles, lost a close gubernatorial election in California despite pre-election polls showing him with a solid lead over his opponent.

Greenwald and Albertson’s research, however, suggests several significant conclusions:

1. In analyzing the polls and the results of the Democratic primaries, they found a more complex manifestation of the Bradley Effect. Of 32 states, Greenwald and Albertson found 13 states in which a reverse Bradley Effect was manifest where polls significantly underpredicted Obama’s ballot box performance. That is, the prediction error was 7 percent or more, well outside of the margin of error accepted for most polls. Meanwhile, three states showed evidence of a Bradley Effect.

2. Of the 13 states where a reverse Bradley Effect was seen, eight are in the South, have an African-American population of 15 percent or more, and went for Bush in 2004. They are Alabama, Florida, Georgia, Mississippi, North Carolina, South Carolina, Tennessee, and Virginia. In three — Alabama, Georgia, and South Carolina — Obama’s underpredicted performance was more than 15 percent off, and Georgia’s neared 20 percent.

3. Pollsters also underpredicted Obama’s performance in Delaware and Maryland, both of which have an African-American population of 20 percent or more, and went for Kerry in 2004. Again, the prediction error was significant: more than 12 percent for Delaware and 7 percent for Maryland.

4. Pollsters also missed the mark in states with small percentages of African-American population. In the case of South Dakota and Montana, which went for Bush, the prediction error was in the double digits. Wisconsin, which went for Kerry by a very thin margin and has just more than 5 percent in African-American population, also showed a substantial reverse Bradley Effect, even higher than Virginia’s double-digit prediction error. Greenwald admits that Wisconsin’s showing cannot be easily explained and doesn’t fit any particular hypothesis neatly. As an aside, Wisconsin is one of the few states that allows same-day voter registration.

5. The three states showing a Bradley Effect are California, New Hampshire, and Rhode Island, all of which went for Kerry in 2004 (a one-percent margin in New Hampshire) and have comparatively small percentages of African-American population. Only in tiny Rhode Island was the prediction error magnitude more than 10 percent.

6. A regression analysis line also shows a strong relationship (r=.77) between prediction inaccuracy and the percentage of African-American population. In other words, pollsters, contrary to their claims, are not doing a good job at accounting for African-American voters in their polling.

In his separate analysis, Nate Silver of the FiveThirtyEight electoral projection Web site suggested similarly last week:

“What is actually quite clear … is that the polls did a rather poor job of accounting for the black vote. Not only did essentially every ‘undecided’ African-American voter wind up voting for Obama, but some of those who told pollsters they were going to vote for Hillary also wound up voting for Obama. The reverse Bradley Effect, in other words, was fairly manifest.”

Greenwald, a social psychologist, suggests that the effects of social desirability cannot be ignored here. That is, respondents may be giving answers that are socially acceptable or politically correct and the impact works both ways. “When you call for a poll, a voter may give the response more desirable to the organization, especially if it gives them the opportunity to show that they support someone of a different racial or ethnic background,” Greenwald explains. “This was the case in states showing the Bradley Effect, especially in the northeast and in California.”

Meanwhile, social desirability also would be manifest in the opposite way, according to Greenwald, who adds that in southern states, respondents may shield their implicit judgments and respond differently to a pollster. Thus, a reverse Bradley Effect was observed and the evidence is striking when one looks at the eight southern states in the research above.

Certainly, prejudice is one of those areas where explicit statements and implicit beliefs may contradict. After all, few people consider it socially desirable to admit openly to attitudes or negative stereotypes against groups such as minorities. Greenwald is well known for his work on the Implicit Association Test. It involves participants categorizing words or images on a computer screen with a touch of the keyboard and then the test moves into a more complex phase as categories become combined. Meanwhile, researchers measure the time it takes to sort out the stimuli from the combined categories which, in turn, provides some information about how participants construct and frame their mental associations. While the test measures associative knowledge, Greenwald has said the test could be used to measure implicit prejudices or biases.

Skepticism about the Bradley Effect — or more importantly, its reverse manifestations — is not good for the integrity of the polling industry nor for the media which clearly are grasping at straws on this issue. There remains a fair bit of intransigence on this point. Skeptics allude to the Tennessee U.S. Senate race in 2006, when Republican Bob Corker defeated Harold Ford, an African-American Democrat by less than 50,000 votes, and most polls showed Corker with a lead of between 1 percent and 4 percent, typically within the margin of error. Only one poll (Mason-Dixon/MSNBC-McClatchy) in the five days before the election had Corker with a 12-point lead. Ford’s campaign staff, however, believed the final margin was going to be exceptionally close, based on their own internal polling.

However, in 1991, when Douglas Wilder ran for governor of Virginia, he led comfortably in most polls up to the election but won with a margin of just 0.1 percent. A similar effect occurred in an earlier statewide race as well for Wilder. In fact, some refer to the effect in his name, not Bradley’s.

When Wilder was asked about the large margin of victory for Deval Patrick when he was elected governor of Massachusetts in 2006, he believed that the time had come to re-evaluate the Bradley Effect. In a January 4, 2007 Boston Globe article:

“I warned him, you’ve got to watch those polls,” Wilder chuckled. “But I think people are becoming less resistant to saying, ‘I’m going to vote for the person whether it’s a woman, or gay, or whatever. There’s more openness — but we’ve still got to watch it.”

Of course, the distinction here is that Obama is the first African-American nominee for national office. And, while the challenge of earlier researchers has been limited primarily to a single fixed geographical frame, the research analysis of Greenwald and Albertson is well worth the attention — not just for political junkies, media, and pollsters but also for every citizen who likely is seeing this election as a major transformative moment in racial attitudes throughout this country. Perhaps in future elections, pollsters will finally adjust their techniques to account for a truly diversified, enlightened voter base.