Of course the results remained the same as in the pre-election polls, because more Blacks compensated the loss of Whites.
This was the point I was trying to make. That there may have been unexpected voters compensating for other "lost" voters - hiding the presence of a possible Bradley effect.
A fairly good paper on it is here:
http://people.iq.harvard.edu/~dhopkins/wilder13.pdfI do disagree with Hopkins that there is an over-reporting effect of the front runner in Patrick's case. However, I do agree that it was declining prior to 2008.
I think there are some masking factors:
1. Disproportional turnout. You expect group X to represent 10% of the vote cast and it's 13%. Was Gallup so far off because its model expected a 8% Black turnout and it was actually 10%?
2. There are not a lot of good polls in some of the states. Did Obama overpoll with whites in AL or NY? We don't know because we don't have a large number of good polls from AL or NY just before the election. Did Obama overpoll with Hispanics in CA or TX? We don't know for the same reason.
I would be looking to see if Obama underpolls with Hispanics of Mexican ancestry more than I would to see if Obama overpolls in general.