What is political space made of, and how can it be represented in statistical models of political processes or behavior?
That question might sound academic (if not psychedelic), but it’s one that doesn’t get the attention it deserves in quantitative analysis of political events.
In recent years, political science has gotten better about considering the effects of physical space on politics, especially in the study of violent conflict. Intellectual trends in the study of civil wars have combined with technical advances in geospatial analysis to encourage observers of conflict processes to be more explicit about ways that things like distance, terrain, climate, and weather might shape where and when violence might occur. (See here for one prominent example.)
The single word that probably best captures the current push on this front is disaggregation. Where the 2000s saw a boom in quantitative studies of civil wars using country-year data to look at the onset and termination of episodes of large-scale violence, a more recent boomlet has shifted the focus to the level of the district and even the locality, sometimes attempting to model the occurrence of the specific events—battles, killings, kidnappings—that comprise those large-scale conflict episodes. The units are getting smaller, but they’re still usually geographic.
As I think about where this research might take us, I wonder if we aren’t atoning too much for past sins. The complaint that studies using nation-states as units of observation are naïve to physical geography may itself be naïve to the profound importance of the state as a political space.
Consider coups d’etat, for example. By definition, these events virtually always happen in a country’s capital city, but that’s not because of differences between the geography of the capital and the rest of the country. They occur in the capital city because it is the locus of national political authority, the point in political space that must be occupied to lay claim to national power. A similar logic applies to the location of battles in civil conflicts. Certain areas will have strategic or symbolic value that is unrelated to their situation in physical space or the character or their terrain. It’s not either/or, but we shouldn’t stop thinking about the one because the other is easier to observe and measure.
The good news is that improvements and innovations in analytic techniques are making this easier to do. For starters, researchers are increasingly using multilevel (a.k.a. hierarchical) models to incorporate factors at multiple levels of analysis in a single estimation. While still computationally intensive, versions of these models with unit-specific slopes let us search for general patterns without making the strong assumption that every variable has the same effect in every region/country/city/person/whatever.
Techniques developed to measure connectivity and distance in other dimensions–such as economic distance as reflected in trade flows, or cultural distance as indicated by populations’ languages and religious practices–can also be applied to political relationships. Mike Ward and Peter Hoff, for example, have done interesting work extending concepts from gravity models of international trade to other aspects of international politics, like alliances and membership in intergovernmental organizations.
Also, when designing research, we should remember that we don’t have to include all physical space in every analysis. Sociologists studying the occurrence of protests and riots in the United States in the 1960s and beyond often restricted their analysis to major cities (see here, here, and here for prominent examples). Riots rarely happen in rural areas in wealthy countries nowadays, but social and economic processes occurring in those rural areas may contribute to the likelihood of riots in nearby cities. So, we might look for ways to design studies that take all of those elements into account, and geographic proximity might turn out to be less relevant than many other things.
There’s certainly no grand solution to this problem. More than anything, this is a plea to researchers, including myself, to think carefully about the spatial dimensions of their theoretical models when designing empirical studies to probe or test them. The fact that a lot of data is available at the state level doesn’t mean states are an appropriate unit of observation, but subnational units aren’t automatically better, either. In many cases, it will help to start by conceptualizing the relevant political space and then looking for data that represent important features of it.
* Title shamelessly stolen from Kate Miller-Heidke.