If a candidate underpolled i don't know how you can exactly say it was because of the "Bradley effect". Maybe there were other reasons.
Wasn't Kerry polling better in North Carolina in 2004 than what it finally turned out to be the results? And i'm sure a lot of candidates in the past years have been underpolled in some places and overpolled in others. How exactly do you know why that happened?
Maybe the Bradley effect happened somewhere, maybe not, but i still can't figure out how you can be so sure J.J.
First of all, it has to occur across polls; it can not be just one bad poll.
Second, in some of the examples, it has to be consistently outside of the MOE or at least consistently at the upper end of the MOE. CA is an example. McCain did slightly better than expected than several late polls, but in the MOE. There were others IA, AR, GA, WV, and even NY, where McCain's result was better than the MOE. (It did occur in UT as well, though it wasn't great polling)
Third, it seems to be more likely to occur in races where one candidate is black and the other is white, more so than in a white/white or black/black candidates race. It occurs across party and on the ultimate winner, but it doesn't occur in all cases.
Fourth, in 2006, the pattern was the undecided vote looking like it would break heavily to the white candidate. In other words, the undecideds tended either to make up their minds for the white candidate, or they said to the pollster, "I'm undecided," but secretly thought, "I'm voting for the white guy."
That pattern occurred in 3 out 4 black/white candidate races in 2006 and seems to have occurred in some state polls and certainly in good national polls (and a bad one, Zogby).
My question is, did the polls that got it closer, in the MOE, do something that corrected for this? We know that Rasmussen uses robocalling and Hotline and TIPP doesn't so that's it.