Son of Voting
Somebody who goes by the moniker of "poblano" has run some multiple-variable regression on the demographics of the Democratic electorate and how the results have turned out for Obama vs. Hillary. As expected, Obama does better in states with caucuses, states with a lot of African-Americans, and states with a lot of young voters. But while we hear a lot about how Obama is the "upscale" candidate while Clinton attracts more working-class voters, what he's found is that if you control for level of education (which correlates positively with Obama), then higher incomes are either a wash or even slightly correlated with Clinton votes. He's also found that a better correlation than just "Latinos vote for Clinton" is that recent immigrants tend to vote for Clinton.
He did this analysis yesterday, before the results of the weekend races were known. He correctly predicted that Obama would sweep the weekend, but he was pretty far off on the margin of victory. He overestimated Obama's performance in Louisiana (he predicted a 35.8-percentage margin of victory in the two-way vote; the actual two-way vote share was 61.3 - 38.7 for a 22.6-point margin) and Nebraska (predicted 76-24, actual 68-32), and underestimated his victory margins in Washington (predicted 53.6 - 46.4, actual 68-31) and Maine (predicted 51.8 - 48.2, actual 59-40).
So obviously it's a pretty crude tool... I'd like to see what his calculated confidence levels are.