Andrew Gelman, guest-blogging at fivethirtyeight.com, uses survey data from before the 2008 US presidential election to ask the question: which states would have been won for John McCain if only rich voter's votes counted? (Answer: most of them, except mainly for California and New England.) By contrast, if only the votes of poor voters counted, Obama would have had almost a clean sweep. You can see the R code used for the hierarchical random-effects model (using the glmer function from lme4) at his blog.
Update Mar 4: Updated maps from Andrew's post.
