The Geometer’s Angle

What John O’Loughlin talks about in his recent presidential address in Political Geography strikes me as substantially similar to many of the things I’ve read from him on the field. Indeed, it reminds me of the same track I got from him as a PhD applicant seeking to work in quantitative political geography. I’ll never forget; right at the time of (what I thought and still feel is) great ground-breaking work in political science focusing directly on the new understandings possible from aggregate electoral data [1,2,3] he suggested that electoral analysis needed to go beyond this.

Maybe sure, yes, this is true. But, right then (as is the case for right now) aggregate election data is cleaner, more useful, more spatial, and more interesting than it ever was before. The “analytical cartography” of previous geographical studies was actively being replaced with statistical analysis of high-quality, high-detail data with national coverage, and linked directly to individual or nearly-individual behavioral data. If there’s somewhere else where empirical electoral geography should be going, I was not quite sure how this was not it. Nor am I still sure there is a there there in alternative explorations of space & context in electoral geography:

The point of context is the person-in-space, how the systematic factors a person experiences in their surroundings affect the individuals political statements, beliefs, preferences, or attitudes.

And, with this data, we need to resolve ourselves to find the truth, not the expedient.

So, if there was some unobservable gestalt, some “thick” region which affected individual behavior, I wanted evidence, not exceptionalism. And, my colleagues & committee at Arizona State showed me is exactly how this type of evidence is constructed. If at the end of the day, your model has no room for space, what power does unmeasured or unmeasurable context have? The work I like to do, read, and that I find useful for others is work with didactic relevance (Ron’s term), or transferable insight. While it may not be directly replicable in a computational sense, it is designed to be replicable in a loose sense. Do what I did and the same systems should be apparent in your data.

I do not doubt that you can come up with some very convincing explanations of your quirky model, but at the end of the day, if you’ve missed a critical aspects of the process specification, what you’re interpreting is less than meaningful. It’s neither wrong nor right — it’s misspecified, and the truth can be anything depending on the structure of that misspecification. Spatial underspecification, intentionally leaving a model vague so that spatial effects can pick up the rest, is misspecification, too. And, it’s kind of ethically dishonest as a scholar, since finding that “regions matter,” but that there’s no clear explanation as to “why” they matter, “how” they matter, or how to move one from mattering to not mattering… that’s not really a transferable insight.

Isard, in his 1956 address on the question of whether regions really are things-in-themselves to be found by geographers

Having your cake and the other table’s, too

Thus, again, we are sucked back into the cognitive trauma of the Agnew-King exchange from 1996. O’Loughlin returns to this, pointing out places where this supposedly-weak view of contextual effects coming from political scientists contrasts with Agnew’s perspective of strong context within a nearly-numinous historical-geographical approach. There, he reiterates Agnew’s argument that “places and the meaning of regions drives our discipline.” Indeed, this would suggest (as Isard (1956) also does, but with whom O’Loughlin or Agnew would likely disagree on many other facets) that there is some “true” but unobservable region that impacts human behavior, bounds social systems, and is a context as a thing in itself. Is there truly a “region?” Or, is this just “regional,” i.e. underspecified, but leaning on surrounding areas to relieve this underspecification?

Dodging this question of whether or not there are (or are not) “true” regions, O’Loughlin goes immediately to the work by some of my colleagues at Bristol nearly a decade ago on drawing useful socioeconomic contexts; they find that much of the impact of one type of region in one electoral problem is mainly a statistical artifact of aggregation! While there are interesting context-based parts of the work there (the output being the local contextual regions), the point was that one set of regions long thought to be useful in modeling spatial heterogeneity in vote structure in the UK tend to arise not from the fact that these regions are meaningful in and of themselves. Given another better-specified regionalization based firmly in social processes, this alternative regionalization was not useful.

This irony is not even hinted at by O’Loughlin in the address. Is it not important that they find that the “meaning of regions [that drive] our discipline” is that, well, they might not?

Whose geography is geometry?

This brings me to the most confusing and frustrating aspect of this address; having read O’Loughlin’s previous discussion about divisions between political scientists and political geographers, it’s baffling that he highlights this work, but then uses outdated or confusing divisions of “weak” contextual explanations offered by political scientists that amount to “political geometry.” How is this any weaker than that offered by exploratory regionalization or naive distance-varying coefficient work?

How are multilevel methods or theories of distance decay, which have been used in some form in spatial social science for nearly forty years at this point, any “stronger” than the work cited by Gimpel, Tam Cho, etc, who actually theorize how the region works? that a region has a different effect in a multilevel model isn’t sufficient; Gelman’s work on “Red State, Blue State” (effectively ignored by many of the old-guard political geographers) does this, but then also provides a reason why multilevel is not the whole story!

If you do actually believe that context is best captured by some numinous function of distance to event or group (and find evidence of this under-specified geography), that’s fine. But, what is that distance or that group membership actually capturing? Is your model underspecified?

Surely, individuals are not themselves breaking out their rulers to measure the distances to all other relevant sites; the right construct may be measures of social distance, news diet, experiential familiarity with areas that suffer from violence, travel history, etc. As quantitative geographers Linke & O’Louglin (2015) understand that it’s often sufficient to instrument this using an explicitly geometric variable.

But, as a political geographer, O’Loughlin is somehow now in the position to claim that stronger theories rooted in social science about what makes place are too weak to publish? When political scientists (i.e. non-geographers) attempt to define a precise mechanism by which proximity may affect change (e.g. interpersonal interactions, friend group dynamics, social interaction in public transit), this is a “weak” geography?

This leaves me to wonder, is it a “weak” geography because it’s not treated as a thing-in-itself, left underspecified, and then simply attributed agency in a weak model? Or is it a weak geography because they’re not participating in the nearly 30-year-old canonical texts of political geography? The Agnew readers and the Taylor & Gudgin book, the Pred & the continual trauma of Berry’s “moribund backwater” remark?

… and, who wields the protractor?

Sometimes, the better angle is orthogonal to your own.

The thing I always heard in grad school is that sometimes, in some cases, a spatial model isn’t going to be useful. Sometimes, some mechanisms are so strong or so useful in prediction that it doesn’t really matter that the process has a spatial embedding; compared with the specific, testable, repeatable mechanisms that we can demonstrate have an effect, a numinous conception of the “region” or “contextual” effect has no space in which to operate. There is no large unstructured unknown that we can instrument using geography. But, since this “weaker” work, where geography is given agency through a specific social or behavioral process, is done by a non-geographer, it’s somehow weaker than work that does the same thing through imposing a theory of context-as-nearness or context-as-socioeconomic surroundings. Context as social environment is no weaker than distance decay, although it’s more specific and clearly articulated.

Why does this more-clear articulation, with a specific process and mechanism of action, constitute a “weaker” conception of space? Is it because it’s not the numinous region that does the lifting? To me, this really doesn’t look good… over time, it might appear that geographers are using the pseudo-scientific concept of the “region” to ostracize external work as insufficiently thick/reductive (“the region is more than a social network or measurable social context”) but then publishing their own work which is even more reductive in hopes of chasing an underspecification (“the region is a distance neighborhood/egohood around survey sites”).

The shape of the discipline

So, who gets to say this work on the social dimensions of microgegoraphies is “weak” political geometry? The same author suggesting a “weak” model using distance decay and multilevel modeling (entirely geometric concepts) is somehow a more noteworthy investigation than these alternative richly-parameterized conceptions of place as a social-spatial ensemble. It’s useful to know the properties of these contexts, assigned specific actionable processes by which they impact their members.

I’ll be honest; it’s not quite as useful to know that a $k$-nearest neighbor model heuristic improves accuracy, or simply that variable weights shift as $k$ shifts. If you don’t have a clear, supported mechanism for why that occurs, then how is that finding not geometry itself? I’m not buying this distinction between Geographers and Geometers that O’Loughlin admits to using in order to gatekeep Political Geography. Why stop these other “political geometers” but publish your own geometries elsewhere?

This is a bummer, since he’s (at multiple times in many different publications) lamented the lack of “good” political science work published in PG. And, I agree, there has been. But, that’s not because it’s “weak work,” it’s because there’s clearly turf here being used to reject “weak” work by the wrong people, who then must expected to consume our “weak” work. At the end of the day (and the data), this is what these “weak” perspectives of regional agency provide. They provide repeatable analyses with specific claims and distinct measures of uncertainty, focused on specific mechanisms, without reference to a “numinous” underspecified regional geography.