Very interesting paper, which turns some common assumptions on their head:
http://volokh.com/archives/archive_2006_11_26-2006_12_02.shtml#1164605626
EXCERPT:
[i]Later in the paper I present the results of full latent variable structural equation models. The latent variable traditional racism (Model 1: r = .27) predicts the latent variable income redistribution. (I also find that the preference against income redistribution is not just the result of income or education; rather, the data are consistent with racism continuing to play a small but significant role in explaining the support for income redistribution.)
The data are broadly inconsistent with the standard belief in the social psychology literature that anti-redistributionist views are positively associated with racism. The results are a problem for the academic assumption that opposing income redistribution indicates hostility toward other groups and a desire to dominate them. Indeed, many social psychologists believe that the link between opposing redistribution and social dominance is so strong and clear that opposing redistribution can be treated as a measure of social dominance orientation. [/i]
To me, this conclusion that those who favor income redistribution are more likley to hold racist beliefs actually makes some sense, given a couple of my general assumptions: 1) that people with the lowest education levels are most likely to hold racist beliefs and also most likely to favor income redistribution; 2) Even though a lot of people at the highest incomes (which correlates to high education) also favor income redistribution (to a degree anyway), those with some college or just college educations do not, and are numerous enough to dominate the data for “higher incomes.”
ADDENDUM: I should note that the author points out specifically that he controlled for income and education, and got an independent relationship between traditional racism and favoring income redistribution. So the racism isn’t a proxy for income or education, even though those might be separately correlated.