This project helps answer a query that Lee Siegelman raised on Monkey Cage a few weeks ago. What's the best level to analyze the vote? Should we be looking at individual data? County returns? State returns? There's no obvious right answer here. Yes, individuals, not counties or states, are the ones that cast votes. But people are not islands. They often think as members of communities and evaluate political events in terms of their impact on their geographic area. There's some reason to think that counties are the relevant area to examine. People identify with their counties (much more than they do with their legislative districts, which shift frequently) as counties have relatively fixed borders, are host to many political contests, have set tax structures and public services that are often different from those of neighboring counties, etc. And yet some people may identify more with their county than others do. A resident of Compton probably has a very different conception of "L.A. County" than a resident of Woodland Hills does, even though they're in the same county.
Another nice take on American political geography can be found in the TPM Book Club's discussion of Andrew Gelman's book Red State, Blue State, Rich State, Poor State. The whole book is basically an examination of an ecological inference problem: wealthy states vote Democratic, while wealthy people vote Republican. Another way of stating the problem is, as one commenter notes, "mistaking the map for the territory."