Will the "Bradley effect" help McCain? According to Realclearpolitics.com's poll of polls, Obama is now up about 5 percent points over McCain. At a moment like this, Republicans might be hoping that the "Bradley effect" is in operation; that a sizeable number of so-called bigoted whites are saying they plan to vote for Barack when they really intend to vote for the white guy. Data from Realpolitics compared with the recent AP-Yahoo News poll indicate that there is no such effect at work.
The AP survey was conducted between August 27th and September 5th, and is unique in that telephone interviewers direct respondents to the Internet or to loaned equipment so that answers are private. People might be unwilling to admit to others that they plan to vote for a white candidate over a black one (I'm skeptical about that) but why lie to a computer?
According to the results, Obama was up 4 to 5 points during that period of time. I looked at the other nine national polls that were taken during that period (data provided by Realclearpolitics), and I calculated a simple average which turned out to be a 5.1 point advantage for Barack. Respondents are telling telephone interviewers the same thing that they are telling the computer, so there is no dishonesty. There is some evidence from other races that whites might change their minds last minute, and that late deciders might break for McCain, but there is little reason to expect that current polls are off by much.
One factor that might benefit Obama is that surveys sometimes under-sample cellphone users. This is a younger group, and consequently is more likely to vote for Barack.