It's probably just polling error. The finding that the white % of he electorate went up leads one to the conclusion that the white Obama vote also went up (or vice versa) for the polling numbers to add up.
It's probably not even an error. Subsample margin of error should explain it.
As an AP high school statistics teacher I agree. Margin of error is at stake. Polls showed Romney getting 6% of the Black vote nationwide, up from 4% in 2008. But in Detroit Romney's share of the total vote was just 2.1%, down from 2.6%. Local issues and get out the vote efforts affect things too.