Whew! With Obama back ahead, however narrowly, in the daily tracking polls, and hitting McCain hard for his incoherent utterances regarding the collapse on Wall Street, I am free to move on to new worries. For example, the following observation of Larry Sabato, who, having just tipped his cap to the "pro-Democratic electoral conditions," blithely continues:
On the other hand, with the first African-American presidential nominee, [the Obama campaign] must worry about racial leakage on Election Day. There isn't a person in the country who can truly measure that effect in advance.
The phenomenon referred to here is the Bradley effect, so-called on account of the California gubernatorial race of 1982, in which Tom Bradley, the black mayor of Los Angeles, lost to George Deukmejian by around 100,000 votes, or 1.2 percent.
![Tom%20bradley1[1] Tom%20bradley1[1]](https://erictheblue.typepad.com/.a/6a00d8344ecfa453ef010534d1541b970b-800wi)
The result was a considerable surprise, because Bradley held what appeared to be a safe lead in all major polls conducted over the closing days of the campaign, and exit polling on election day tended to confirm that he would win. It developed, however, that (in the artful words of a Wikipedian) "a smaller percentage of white voters actually voted for Bradley than polls had predicted, and that previously 'undecided' voters had voted for Deukmejian in statistically anomalous numbers." White voters tell interviewers that they intend to vote--or even did vote--for the African-American candidate, but, in the privacy of the voting booth, they do not.
The tendency has been observed in other contests. Andrew Hacker, writing in the current New York Review, reminds us that in 1989 David Dinkins, an African-American candidate for mayor of New York, won election by 2 percent of the vote after leading his opponent by 18 percent in polls. The same year, in Virginia, another African-American, Doug Wilder, was elected governor by about a half of one percent of all ballots cast after enjoying a 9-point lead in the latest polls. Both their opponents were white. Moreover, when affirmative action has been on a statewide ballot, the percentage of voters who actually vote to outlaw affirmative action programs has consistently been 6 to 10 points higher than the percentage who tell public opinion surveys that they will vote that way. Hacker suggests that the Obama campaign, when weighing strategies with an eye on polling results, should adopt as a rule, "ALWAYS SUBTRACT SEVEN PERCENT!"
The Bradley effect has reared its ugly head in this year's primary season--for example, in the New Hampshire primary, when Clinton defeated Obama after trailing him by a significant margin in the last batch of public polls released before the balloting. In general, the Bradley effect could account for the fact that Obama tended to win caucus states, where one's vote is public, while Clinton did better in primaries, where one's ballot is marked in private. The notion that the New Hampshire result can be attributed to the Bradley effect is rebutted, I think convincingly, In a New Republic piece by John Judis that is available online. Still, I worry. Sabato, who knows more than I do, thinks it is a factor that cannot be measured in advance. Hacker demurs--not because he disputes that in 2008 the Bradley effect is still in force but, it seems, because he thinks it can be estimated beforehand: seven percent.
So if, on November 1, Real Clear Politics is showing Obama with a 7-point advantage in Pennsylvania and Michigan, and slightly smaller leads in Virginia, Ohio, Colorado, and Wisconsin, I'm still going to be among the nervous handwringing bedwetters referred to in Sabato's article.