It was Tom Bradley's 1982 race for governor of California, in which he lost to George Deukmejian in spite of leading in the public polls, that gave the Bradley Effect its name. But now Lance Tarrance, the pollster for Bradley in that race, has an article up at RCP suggesting that the Bradley Effect was merely a case of bad polling -- and that his campaign's internals had shown a dead heat:
The hype surrounding the Bradley Effect has evolved to where some political pundits believe in 2008 that Obama must win in the national pre-election polls by 6-9 points before he can be assured a victory. That’s absurd. There won’t be a 6-9 point Bradley Effect –- there can’t be, since few national polls show a large enough amount of undecided voters and it's in the undecided column where racism supposedly hides.Tarrance's article is a fascinating read into the way that polls are spun and campaign narratives are spread. It is well worth your time to read the entire piece.
The other reason I reject the Bradley Effect in 2008 is because there was not a Bradley Effect in the 1982 California Governor’s race, either. Even though Tom Bradley had been slightly ahead in the polls in 1982, due to sampling error, it was statistically too close to call.
With that said, the evidence is pretty strong that the Bradley Effect in fact used to exist in the 1980s and probably through some point in the 1990s. In this Pew Research article you will find several examples of it, spanning the window from Harold Washington in 1983 to Carol Moseley Braun in 1992.
The evidence is perhaps equally strong, however, that the Bradley Effect does not exist any longer. As can be seen in the Hopkins paper for Harvard University that I have referenced many times, at some point during the mid 1990s the Bradley Effect seems to be disappeared.
(A brief aside: This is not to suggest that there was no relationship between race an errors in polling during the Democratic primaries. There is clear evidence that Barack Obama overperformed his polls in states with a large number of African-American voters, a.k.a a Reverse Bradley Effect. There is not any statistically compelling evidence however that Obama routinely underperformed his polls in states with a large number of white voters).
If the Bradley Effect has disappeared or at least dissipated, it is worth thinking about why. I can think of several plausible answers.
1. As Hopkins suggests, racial hot-button issues like crime, welfare and affirmative action are largely off the table today.
2. It may be generational. Expressions of racism are strongly correlated with age, and is much more common among pre-Boomer adults. However, a smaller and smaller fraction of the electorate each year came of age in the segregation era. The Pew study that I linked to above reports that 92 percent of Amerians are now comfortable voting for an African-American for President. In 1982, when Bradley's race occurred, that number was more like 75 percent. (Although the Bradley Effect isn't about racism per se -- it is about people misleading pollsters because of social desirability bias -- racism is nevertheless one of its prerequisites).
3. Racism also has a strong inverse correlation with education, and the country is much more educated than it used to be. In 1980, 55 percent of the electorate had attended at least some college. By 2004, that number had increased to 74 percent. Most colleges are racially diverse, at least to a degree, and so the experience of interacting with African-American students as friends and classmates may be a significant deterrent to racism.
4. There may be some relationship to the revival of the religious right in the 1990s. For members of the religious right, there are now ample and automatic reasons to vote against any liberal candidate, a.k.a. their positions on issues like abortion. In addition, the religious right has made voting along cultural grounds (as opposed to policy grounds) more socially acceptable in general. So long as the voter believes he or she can articulate a "valid" reason for voting against an African-American candidate, there is little reason to deceive a pollster about one's intention.
5. Relatedly, there may also now be less overlap between those sorts of voters who are more likely to harbor racist sentiments and those who are more likely to vote for a Democrat. One test of this hypothesis would be to see whether black Republican candidates still suffer from a Bradley Effect, even if black Democrats largely do not.
6. Polling techniques may have improved. For instance, "pushing" leaners toward one or another candidate with an appropriate follow-up question may be a good way to tease out the preferences of voters who are shy to reveal that they won't support a black candidate.
7. People's attitudes toward polls may have changed. Our society has become more and more impersonal, and so when a pollster calls, the respondent may no longer regard the interviewer as a "neighbor" to whom he or she must seem socially desirable. This would be taken to the logical extreme by IVR polling technologies (a.k.a. "robopolls") in which there is no interaction with a human at all.
8. African-American candidates may have gotten smarter about how they market themselves to white voters.