Gelman et al. (2007) argue that the marginal effect of income on voting Republican decreases in richer states---that "income matters more in Red America than in Blue America''. I model individual-level and state-level racial resentment for the 2008, 2012, and 2016 elections, and find that the Gelman et al. picture is misleading in several ways.

A model that considers only income is biased as a result of excluding racial resentment. Racial resentment better explains US voting patterns than income, and the magnitude of the effect of changes in racial resentment is much greater than the magnitude of the effect of changes in income. The marginal effect of income decreases as a function of both state-level racial resentment and state-level income, but, in two of the last three elections, not as a function of state-level income alone. Connecticut isn't more Democratic because it is richer, but because it is richer and lower in racial resentment. There is also evidence of dynamic changes: racial resentment as a determinant of voting intention has increased in predictive power from 2008 to 2016.