John’s recent musings about what all those intriguing-looking patterns in Wendy Tam Cho and Jim Gimpel’s county-by-county mapping of the relationship between economic hardship and the Obama vote might mean have led me to some musings of my own. The maps are cool, but like John I don’t know what they’re telling me. I think my sense of uncertainty goes deeper than his, though. Here’s the way I think about it.
Counties don’t vote. Individuals vote.
Individuals live in houses, apartments, or other such arrangements, sometimes by themselves, sometimes with others, as families or whatever.
Some of these people form groups of one sort or another – social, economic, cultural, religious, political, and so on.
The houses, apartments, and so on in which these people reside are located in neighborhoods.
These neighborhoods are located in cities, suburbs, or rural areas that may or may not be coextensive with counties.
The counties are located in states.
The states are located in regions.
Of course, it’s really more complicated than this. Your parents (or their ancestors) may have come from a place other than the one where you grew up, which could be different than the place where you went to school, before you got a job in a third place, moved the family to a fourth place, worked in a fifth place, and retired in a sixth place – or something along those lines.
This little recitation of the obvious isn’t intended to tell Wendy and Jim anything they don’t already know. They’re solid scholars who are better versed than I’ll ever be on topics like multi-level modeling and ecological inference. Rather, it’s simply to raise a question:
Why counties?
Why, if we want to understand the pocketbook basis of the 2008 vote, should we, as an early step, engage in mapping on a county-by-county basis? I expect that part of the answer is that, for now at least, they consider the county to be the lowest areal level for which data are available. Well, okay, but I’d offer an available alternative:
The individual voter.
Wouldn’t it make more sense to begin with the individual voter, model his/her vote with all the individual-level factors that are supposed to influence vote decisions, and then begin to move outwards toward incorporating family, group, neighborhood, area, state, and regional effects? If we’re committed to parsimony in explanation, shouldn’t we begin with the actual decision-making unit, see how far that assumption will carry us, and then progressively bring other analytic “layers” into play?
Note that I’m not arguing for a model of the vote that excludes group, areal, and other “contextual” factors — far from it. But unless and until we have a better sense than we currently do of the individual, family, group, neighborhood, and area dynamics of the vote, I don’t know what we can really learn from a county-level mapping.
I hope and trust that Wendy and Jim will approach my uncertainty as a teaching moment, and that they’ll clarify for me and perhaps for other “Monkey Cage” readers what they see as the payoff from their county-level focus.