Add on top of that this analysis by Nate Silver at FiveThirtyEight.com. He assumes that Obama wins 95% of black votes, and 30% of non-black votes (including Latinos and Asians), and projects how increased black registration and turnout narrows the result:
Now suppose that black and nonblack voters each turn out at the same rates as they did in 2004, but that we account for the increase in black registration. According to our math, John McCain's 7.0-point lead is now cut to 4.9 points...Is this realistic? Well, 40% of early votes have been cast by African-Americans. Surely this can't hold, but it is encouraging. Apparently Martin is out-polling Obama, as local Dems tend to do. So if Obama loses by 1 point, the Martin could win.
Even this, however, may be too conservative. For one thing, the registration window in Georgia is not yet over ... it concludes today. The statistics I cited above only reflected registrations through September 30. There is typically a surge of registrations in the final few days before the deadline...
So suppose that by tonight, black voters have increased to 30 percent of Georgia's registered voter pool. Plugging that 30 percent number in, McCain's advantage is a mere 1 point.
I have heard that pollsters update their voter registration files for calling potential voters every month, and so these later poll numbers would favor Obama once these new registrants started appearing on the pollsters' lists. Is this a partial explanation of why these polls have tightened recently? That doesn't make them less viable, of course.
I, for one, am a huge skeptic. I still expect Martin and Obama to lose by about 6 points. As far as I'm concerned, GA might be the reddest state around - even Alabama and Mississippi have Democratic state legislatures. Kansas, Wyoming, and Montana have Democratic governors. Until Martin actually wins, I don't really believe these numbers.
FWIW, Pollster.com's poll of polls still shows a very wide gap: