Jay Cost counts up the racial breakdown and finds 10 million missing White Voters. The question remains why did they not vote?
One of my intuitions was that the Democratic non-white vote would not rise very much this year because of the big jump in 2008, in particular in non-competitive states like Illinois, Mississippi, and California. Looking at the hard numbers, that turned out basically to be correct (although the Latino vote looks to have increased modestly).
What I did not anticipate was a steep drop in the white vote. My back of the envelope calculation suggests that the white vote was off by almost 10 million votes relative 2008. [This is the primary reason why Mitt Romney will end up winning fewer votes than John McCain, but have a larger share of the total electorate.]
So, the polls that showed a big Obama edge, often due to a loose likely voter screen, were right for an ironic reason. Turnout was down, suggesting a tighter screen would have been better, but because turnout was down so substantially among whites, the actual electorate looked a lot like more like the broader population than it has in years past (even in 2008). Thus, a loose screen produced the better reflection of the voting public.
A tip of the cap to those who figured it would go the other way. Job well done!