the poor were more likely to vote for Obama, but once household income
hits $30k the income/Obama gradient is basically flat, so I don't think
"the rich vote Republican" is very accurate. Controlling for race
diminishes the effect of being poor by roughly a third. After
controlling for race, controlling for age, education, sex, and religion
have practically no effect on the income gradient.
Controlling for race and other demographics or not, the highest
income group ( > $150k ) were slightly more likely to vote for Obama
than all groups between $40k and $150k, that is, the rich voted for
Obama at slightly greater rates than the middle class (I do not know
how to reconcile this result with the maps. I did not weight anything,
perhaps that explains the difference.)
This is very much in line with my critique of Andrew Gelman's work. The real news isn't that "the rich vote Republican." It's that the effect of income on voting is small, especially after adjusting for race.