The Bradley Effect

One of the issues that has become a subject of speculation is the Bradley Effect. Does this effect really exist? How does it affect the race?

First, what is the Bradley effect?

The effect is named after Tom Bradley's run for Governor of California in 1982. On October 7th, Tom Bradley, an African American, led his GOP opponent by more than 12 points and looked set to become the first African American governor. Even on election day, some exit polls had him ahead, and some news agencies called the race in his favor. Then the results came in and he lost. The popular explanation for the error was that people were not willing to tell pollster the truth about their voter preferences because of racism. If that explanation is correct, then polls not withstanding, Obama would need to lead by 6% or more in the polls to win the election.

So, is the Bradley effect impacting the polls in this race? Probably not. Why? Well, consider the following:

• It's not clear that the Bradley effect ever existed. It may have been a cop out by agencies that got things wrong. This post explains why.
  
• If there is a Bradley effect, it didn't show up in the primaries. Generally, except New Hampshire, the polls were spot on in the primaries. Even in New Hampshire, it was more some assumption errors than the Bradley effect that explains the discrepancy.
• Much has changed demographically in the US. African Americans and minorities now make up much more of the electorate. If we assume that there is a Bradley effect, there may in fact be a reverse Bradley effect where minority turnout and votes swing so dramatically that it completely offsets the effect.

... but, that should not make you more comfortable about the polls, because:

• Polling samples people's opinions and then projects it on the population using a segmentation based on historical data. This method works well as long as the dynamics of the campaign are similar to past elections. However, if there are changes in voter turnout or some key assumptions, e.g. like having a woman or an African American on the ticket, history is not a good guide for these assumptions, and the polls become uncertain guides. Case in point, the 1992 UK elections.
• A lot of the polls at the moment factor in the opinions of the under 30 age group, which overwhelmingly supports Obama. This age group is notorious for not turning out. Their low turnout cost Kerry and Howard Dean their elections. Their low turnout impacted Ron Paul, who didn't get close the the votes he was polling in the pre-election polls. Will they turn out for Obama in the general elections as they did in the primaries?
  

http://www.fivethirtyeight.com/ has some more interesting thoughts on this. Overall, what all this tells us is that while polls are interesting, their predictive value in this election is uncertain. They may well be spot on, but they may be wildly wrong.