First, I think was a Bradley Effect, i.e. people lying to pollster because of race, in the 1980's and 1990's, but it was dwindling over time. It arguably was there in some statewide races in the 2000's but it was not strong.
Second, the states where I though it would be pronounced, PA, OH, it was not there. The PA numbers were right on.
Third, there were states where it looked like it happened, IA, where I never would have expected it. There also seem to be cases where voters lied to the pollsters, but they said they were voting for McCain and voted for Obama. That seemed to be in AZ and NM, and may be tied to voters of Mexican ancestry. (A study from Harvard indicated that during the primaries, Obama underpolled with African-American voters. Basically, they didn't want to look like they were voting for Obama just because he was black.)
Fourth, in trying to see if there were more states where Obama underpolled, there is the problem that a several large states were not polled. I had found one poll within a month of the election in NY. Obama overpolled and overpolled outside of the MOE. The poll was not in a week of the election, and it was not a major polling firm. Was that because of the electorate shifting toward the end? Was it because the methodology of poll was bad? Was it because a significant number of people didn't want the pollster to think they were voting for McCain because Obama was black?
The only thing that we can say is that Obama overpolled on some normally good nation polls, like Gallup, but didn't on some others, Rasmussen and PPP. (And 2008 made PPP.)
Summarized: If a black candidate ever overpolls, it must be assumed that the reason is the Bradley Effect and not any other plausible explanation (even it simply being a bad poll), even if there is plenty of evidence toward a different explanation.
