'Bradley Effect' gone, but race still matters.

Public opinion scholars and practitioners call it the "Bradley Effect." That's Bradley as in Tom, the former Los Angeles mayor. As an African-American Democratic candidate for California governor in 1982, polls conducted shortly before the election showed him with a comfortable lead well above the statistical margin of error, but when all the votes were counted, he lost.

Why the discrepancy? Many argued that there were enough white voters, ordinarily Democratic supporters, who were privately against voting for an African American. But in responding to an interviewer, gave the more socially accepted response that they supported Bradley.

This explanation gained credibility when, seven years later, another African-American candidate, Douglas Wilder, seemed to be breezing toward winning the Virginia governorship.

Early November polls by the Washington Post and the Richmond Times-Dispatch had Wilder ahead by 11 and 9 percentage points, respectively, but the election a few days later saw him winning by less than 0.5 percent.

A scholarly post-mortem of this race by Steven E. Finkel and colleagues in the Autumn 1991 Public Opinion Quarterly, provided empirical support for the social desirability hypothesis especially among white Democrats and white undecided voters. They were reluctant to tell a black interviewer that they were not for Wilder.

For the past 15 years, however, the Bradley Effect has apparently vanished. In 2006, black and white candidates opposed each other in six statewide races--two for the U.S. Senate (Maryland, Tennessee) and four for governor (Maryland, Massachusetts, Ohio, Pennsylvania). In each contest, there were between two and four polls conducted within two weeks of the vote.

According to an analysis by the Pew Center's Scott Keeter and Nilanthi Samaranayake, all but one of the surveys had the African-American candidate's ultimate vote well within the margin of error. And the one exception, a Mason-Dixon poll in the Tennessee race for U.S. Senate, underestimated Harold Ford's vote rather than exaggerating it. They concluded that "fewer people are making judgments about candidates based solely, or even mostly, on race itself, and that relatively few people are now unwilling to tell pollsters how they honestly feel about particular candidates."

But what about 2008? Do findings that apply to statewide offices also extend to the presidency?

Earlier presidential candidates like Jesse Jackson and Al Sharpton were not good tests since they projected a sharply drawn racial appeal.

Until the Rev. Jeremiah Wright situation forced Sen. Barack Obama (D-Ill.) to address race directly in his March 18 speech in Philadelphia, his approach was to transcend color.

Both before and after that address, what does the Obama case tell us about the relationship between pre-election poll preferences and actual vote outcomes?

Two University of Washington social scientists, psychologist Anthony Greenwald and political scientist Bethany Albertson, have been following this connection. Their cumulative results, posted May 8 on the Pew Research Center Web site (www.pewresearch.org), show "a more complicated pattern (than the Bradley Effect)."

For Democratic open primaries (in which independents and, in some cases, Republicans can participate) in states with large white majorities, they find some confirmation for the Bradley Effect in about half the cases.

Making this conclusion shaky is that there are many exceptions, including Missouri, and one substantial outlier--Wisconsin--where the polls significantly underestimated the Obama share.

Greenwald and Albertson also postulate a "reverse Bradley Effect" in states where African Americans constitute a substantial share of the Democratic primary electorate. In Alabama, Georgia and South Carolina--to take the most striking examples--pre-election polls had the Obama proportion considerably below the ultimate outcome.

As a result, they both showed that "race is still strongly operative as a factor in America's state elections" and that "its impact depends in substantial part on the racial mixture of the state in question."

Terry Jones is professor of political science at UM-St. Louis