I posted this as a reply to Markos in the latest polling thread, but after more thought, it should probably be repeated here. I believe Research 2000 is over-weighting Latinos as a part of the electorate, and as a result upping Obama's numbers by around 4%.
First, let's start with the Research 2000 race numbers, which have been stable across the entirety of polling, from what I can see. This suggests this is actually a polling weight by race - not just how they happen to come in.
White: 73%
Black: 13%
Latino: 13%
Other: 1%
According to the most recent census demographics, the population of the U.S. was:
White: 80% (Probably includes half of Latinos)
Black: 12.8%
Asian: 4.4%
Native American: 1%
Native Hawaiian/Pacific Islander: 0.2%
Multiracial: 1.6%
Hispanic/Latino: 15% (can be of any ethnicity)
However, many Latinos are undocumented. Even among those who are citizens, a disproportionate number are children, meaning they cannot vote yet. In 2004, CNN and MSNBC had the following racial breakdowns in their exits:
White: 77%
Black: 11%
Latino: 8%
Asian: 2%
Other: 2%
The tracking poll's numbers thus currently suggest the following:
White vote down by 5.2%
Black vote up by 8.3%
Latino vote up by 62.5%!
Other vote down by 75%
Seems off to me. If I recalibrate racial weighting according to 2004 exits, but retain the racial preferences show in the latest tracking poll, I come up with:
McCain: 46.2%
Obama: 45.1%
Now, I expect more African Americans will vote this year, upping the numbers to a dead heat. But I cannot think that Latino voting will be up 62.5% compared to 2004.
The only thing which helps Obama not slide dramatically here is that "other" for some odd reason, is extremely Pro-Obama. Presumably this is where Asians are lumped, who historically have tended to vote for Democrats, but not overwhelmingly so.