Seth Stephens-Davidowitz, job market candidate from Harvard, has an interesting paper on this question:
Abstract: Traditional surveys struggle to capture socially unacceptable attitudes such as racial animus. This paper uses Google searches including racially charged language as a proxy for a local area’s racial animus. I use the Google-search proxy, available for roughly 200 media markets in the United States, to reassess the impact of racial attitudes on voting for a black candidate in the United States. I compare an area’s racially charged search volume to its votes for Barack Obama, the 2008 black Democratic presidential candidate, controlling for its votes for John Kerry, the 2004 white Democratic presidential candidate. Other studies using a similar empirical specification and standard state-level survey measures of racial attitudes yield little evidence that racial animus had a major impact in recent U.S. elections. Using the Google-search proxy, I find significant and robust effects in the 2008 presidential election. The estimates imply that racial animus in the United States cost Obama three to five percentage points in the national popular vote in the 2008 election.
The question and method of this paper are excellent. I cannot in polite company reproduce the Google key word used to proxy for negative attitudes about Obama. What Google key word might you try if you were looking for districts were the race factor boosted his vote total? Laredo, Texas is the area with the least interest in the negative search word, but I am not sure that is the best proxy for “support because of race.” (See the author’s p.19 for a discussion of related topics.) How about searches for the title of his autobiography?
Page 29 ranks the states by their interest in “racially charged searches.” West Virginia is the worst, Utah is the best, and Pennsylvania and Michigan and New Jersey are the worst northern states, coming in at #3, #6 and #10, respectively. The graphs and charts at the end of the paper are all interesting, including p.36.
Addendum: You might think I got the pointer from @RovingBandit, but actually the paper led me to him rather than vice versa.