In Virginia and Colorado, two traditionally-Republican states that Obama flipped in 2008, Catalist models show likely Democrats and Republicans running roughly even in the early vote.

But in Virginia they show that unmarried women - a constituency that should favor the president - account for 30% of absentee ballot requests. According to Catalist models, 60% of those single women requesting absentee ballots can safely be identified as Democrats.

They point to more signs of hope for Democrats in North Carolina, where a registered Democrat can't necessarily be counted as an Obama voter. In North Carolina, 30% of early voters are African-American. Exit polls in 2008 ultimately pegged African-Americans at 23% of the North Carolina electorate.

Ultimately, thanks to a little thing called the secret ballot, nobody knows exactly which candidate these likely Democrats and likely Republicans are voting for.

The campaigns are reluctant to share their own micro-targeting scores with reporters, although a Republican National Committee official told CNN last week that their modeling showed likely Republicans leading likely Democrats by a five-point margin among unaffiliated voters in Ohio. That number was quickly disputed by Obama officials.

But micro-targeting is driving each campaign's assessments of who their voters are, and the Catalist data offers some hints as to what the shape of the early vote electorate actually looks like.

"This is based on modeling of who is a Democrat and who is a Republican that we've been doing for well over five years, and we think its pretty damn close," Ickes told CNN. "We think it's very accurate."