The variables used post-dicta 95% of the result. Specifically:
1. . . . Obama does do better in caucus states. . . .
2. . . . Obama benefits significantly from states with high African-American populations. . . .
3. . . . Obama benefits when young voters make up a relatively high fraction of the Democratic turnout. . . . I looked at was the percentage of John Kerry voters who fell into a given age range as determined from CNN exit polls in the 2004 election.
4. . . . Obama does better in states with a large number of college graduates. It is educational attainment, rather than income level, that appears to be the driving force behind Obama's "upscale" support. . . .
5. Fundraising. . . . I looked at the total amount of campaign contributions that Obama and Clinton have received in a given state, divided by the number of votes that John Kerry got in that state in 2004 (e.g. a broad Democratic base). . . .
6. Percentage of Southern Baptists. It's clear that Obama has some issues with Southern whites. . . . I came across data on the percentage of Southern Baptists in each state; this ranges from virtually zero in many places, up to 33.8% in Mississippi. And this variable proved to have a lot of explanatory power . . .
7. . . . Even controlling for all these other variables, Hillary does better in blue states, and Obama does better in red states. . . .
8. Percentage of Democratic voters who self-identify as Liberal. . . . Obama's very best states are states that overall are "red", but where the Democrats that are in the state are very progressive.
9. Percentage of naturalized citizens, e.g. immigrants. Surprisingly, I did not find that Obama performed worse in states with large Latino populations. . . . Obama performed slightly worse in states with a higher percentage of foreign-born, but now naturalized citizens.
Many other plausible variables turn out to be statistically insignificant.