The Monkey Cage

Good question: Why counties? Haven’t we moved beyond such analyses with survey data? Surely none of those political scientists who pored over aggregate electoral data from the 20s through the 60s have anything to say to us today. After all, they were using the only thing they had. But we can use surveys and surveys are a sure sign we have progressed.

Indisputably, surveys are helpful, and political science has accumulated a great deal of knowledge from them. But they too have limitations. It is impossible to infer from individual respondents, even if you secure permission to place their residential addresses onto the map, how particular locales voted. Nor could we detect any kind of ecological or regional patterns that may be shaping the sociopolitical climate of a place from looking at survey responses. This is because in most surveys individuals are far too sparsely sampled to permit us to interpolate a surface that might capture prevailing political sentiment in a place or location.

So if we are interested in what is happening at particular places in distinct parts of the country, it is exceedingly rare, even as survey data collection procedures have gone on-line, to find a survey with a sample large enough to address such questions.

Aggregate level data permits an examination of local and regional patterns in data — even in very rural areas — that would be impossible to study using a survey. South Dakota has come up on this blog a number of times, given Lee’s origins (and mine). How many South Dakotans might appear in a typical survey — perhaps 8 or 10? That doesn’t give us much of a sense for South Dakota, much less Pennington County. But we can examine the county map and get a pretty good sense for how South Dakotans voted as a whole, as well as particular counties.

Of course counties are convenient data containers too, so this is why they are commonly used. Many data elements are reported at the county level, but not at any other level. In other applications, we have used precincts and zip codes, and in one paper our observations were New England towns. These units of analysis are not always available for the entire nation, although the availability of data is improving with every passing year.

Is there evidence that counties are contextually meaningful environments to citizens in a social, economic and political sense? In at least one paper, Bob Huckfeldt found that they were, in spite of the fact that they range vastly in population and area. Certainly as noted in the above post, counties aren’t the only ecological units that matter, and probably not the primary ones, either.

I probably don’t have to remind anyone of this, but regression estimates based on counties are subject to all of the ecological inference problems discussed in that extensive literature. And it is probably true that if our units of observation were precincts or zip codes, our estimates and the local variation we display in the Forum paper, would be different.

Once upon a time in political science, scholars were more interested in entire places, locations and regions. Sometimes they used data from locations to infer the behavior of individuals, but more often, they were asking questions that, in both their time and ours, only aggregate data could address.

Jim:

Your response is almost certainly more thoughtful than the comment that motivated it. And I accept virtually all of it. So thanks.

I’d argue, though, that the cure for lack of usable data on units in which we’re interested is getting data on units in which we’re interested. If we really want to know why South Dakotans voted the way they did (as you and I in particular want to know), then let’s do what we can to put together an appropriate survey or exit poll or whatever of South Dakotans, add to it all the contextual data we can possibly add, and have at it. For me, starting in the middle — at a level like that of counties — skips over too much that’s too important. Perhaps, though, this difference between us reflects the fact that my interests lie more in political psychology and yours, more in political geography or political sociology.

Lee:

From a statistical perspective, it’s all about individual-level voting, and you can think of demographic, geographic, or other factors as attributes of individuals. Thus, we can talk about how “the people in Montgomery County” vote, just as we can talk about how “African Americans between 18 and 24” vote.

I agree, though, that geographic factors seem to feel different than demographic factors, at least for a national election where everyone’s voting for the same candidates. In that case, I think what’s interesting is to see how the importance of different demographic factors varies geographically, and that’s what’s interesting about the Cho and Gimpel graphs. (But I’d nitpick on John about his phrase, “the effect of John Kerry’s vote share on Obama’s.” I’d rather call a correlation a correlation.)

A couple of comments on the thread thus far:

1. While Lee has a reasonable point in saying we should survey South Dakotans if we want to know about South Dakota, it becomes intractable to do so with a survey if we want to compare it to other locations. This intractability becomes even larger if we want to add in contextual variation, which might spur heterogeneity in individual voting.

2. Again while I think that the individual is a perfectly reasonable unit of analysis, one could almost always suggest lower levels of analysis (cognitions, genes, etc.). Where do we stop? For that matter, we could suggest higher levels of analysis are endowed with political meaning, but there is almost always a higher level of analysis…where do we stop? I’m not sure there is a clear methodological imperative for one unit analysis over all others. Some units are more appropriate to some questions than others, but I don’t think I believe some units are inherently inferior to others. Another way of saying this is that if we want to study IR by looking at countries (certainly a reasonable thing), why can’t we want to study American by looking at patterns among counties? We could, afterall, study individuals in IR too.

3. As someone who does some county-level research, at least one justification that’s come to mind is that politicians tend to think of electoral strategy in terms of geographical aggregates. For presidential campaigns, counties are one of those principal units they consider, along with media markets and states (see Stanley Greenberg’s last book for examples of this). Surely if the subjects of our investigation see merit in understanding these units, then there is merit in us studying them. Past that, however, political scientists should be cognizant of the modifiable areal unit problem and choose aggregate units with care.

According to a growing literature in social and environmental psychology, it would appear that psychology is endogenous to geography — or a very large part of it is. (The basic insight here may not be totally new, if we go back and read Berelson, Lazarsfeld and McPhee.)

So maybe psychologists will increasingly turn to studying places as a means for understanding the development of personality and other psychological states germane to political behavior?

Jim:

As the co-author of a book titled RACE AND PLACE, which focuses on the impact of place on racial attitudes, I certainly hope you’re right. I’d still take the individual, not the aggregated individuals in an areal unit, as the basic unit of analysis, though.

To follow up on Jim’s points above, there’s some evidence from Joseph Aistrup and others that voters identify with their counties. Counties generally have fixed borders going back a century or more, many elections are held at the county level, people may choose to live in (or to avoid) certain counties because of the taxes they collect or the services they provide, etc.

As noted in other comments, which level one analyzes should reflect the nature of the questions one seeks to answer. But there’s good reason to think that people’s voting behavior is affected by conditions in their county — certainly more so than by conditions in their state or federal legislative districts, which shift frequently. If there have been a lot of home foreclosures in your county recently, chances are that you know it, even if it hasn’t happened to you or your immediate neighbors, and it might affect how you think about politics.

In follow-up to Seth’s comments (and in a bit of self-promotion), Brady Baybeck and I published a piece in APR where we found that people had pretty good knowledge of their neighborhood economic, social, and political characteristics. There is undoubtedly some slippage when you move to higher levels of aggregation, but I think his point about understanding the world through the lens of what goes around you intuitively makes sense to me.