The Microfoundations of Control

The events in Afghanistan are tragic, they are heartbreaking, they are devastating… I cannot supply words that do them justice. I am not an expert by any means in Afghan politics, society, or history, and thus I have very little to contribute to a broader understanding of what went wrong or what happens next. All I can say is that I know that the human tragedy will not end when the Taliban finally secures the airport.

I will confess, however, that the collapse of Afghan security in the face of the Taliban resonates with a long-standing intellectual interest of mine; specifically, the way that social scientists and policymakers conceptualize large social phenomena such as “control over a country.” And I have written a bit on counterinsurgency warfare and the politics of local control, so I have had cause to think about this issue before. I also regularly teach students in my Southeast Asian politics course about the micropolitics of counterinsurgency warfare, drawing on three important cases in Southeast Asia: the U.S. war in South Vietnam, the Darul Islam and Permesta/PRRI rebellions in Indonesia, and the Malayan Emergency. These aren’t just academically interesting—modern counterinsurgency training in the U.S. Armed Forces relies deeply on these cases too.*

The analytical problem, to state it again, is how to conceptualize large social phenomena. Social science is replete with concepts such as “the state” and “democracy” and “power” and “control” and “institutions” and “markets” and “class” and “ethnic group” and so on and so forth, so it is inescapably part of the job of the social scientist to work with such concepts. But these concepts are always wooly, escapably so; these are essentially contested concepts.

To make sense of the world, we add precision to these concepts as necessary. We can delimit their scope (“a democracy is a …”) or we can define their alternatives (“the difference between a social convention and a social institution is …”). Or we can disaggregate them, breaking large social concepts into their component parts, which can seem simpler and easier to conceptualize than large ones.

To give an example, let us think of what “woodpile” means. A woodpile is a collection of pieces of wood, of course. So to make sense of what a woodpile is, we might choose to define it with reference to the things that comprise it. Woodpiles are comprised of pieces of wood. And, naturally, they vary based on what kind of wood is in there. Locust woodpiles are made out of pieces of locust wood, ash woodpiles are made out of ash wood, mixed woodpiles are made out of multiple types of wood, everything you need to know about how to classify woodpiles and how they vary, it seems, can be described with reference to the characteristics of the pieces.

But of course, that is not really true. Look at these two collections of pieces of wood:

Both of these would qualify as woodpiles under the definition given above. You could describe them with reference to the different kinds of wood, but obviously you’d be missing something else about the two piles, which is how they are arranged. To make sense of how these two woodpiles differ, you need to understand how the two collections of individual pieces relate to one another. In this case, it is their arrangement in a stack versus an irregular pile. You could describe all the parts of each of the two piles of wood, and still not see why one of them is special, because the characteristics of the pile cannot be reduced to just their components.

In the social sciences and the philosophy of science, we can speak of conceptual accounts that a reductionist, and those that are holistic. We can think of social science concepts that are microfounded, where the whole consists of nothing but the parts, so that the characteristics of the whole are epiphenomenal on the characteristics of its components. And we can also think of concepts where the characteristics of the whole supervene on the characteristics of the parts, so that you cannot describe the essence of the whole with mere reference to its components. The best modern defense of why we need to work with social concepts that embrace the possibility of supervenience is by List and Spiekermann, and if you’ve read this blog for a while, you’ve heard me make this point a lot.**

I propose that modern counterinsurgency doctrine has a very good way to conceptualize political control, but only in reductionist terms. And that is a problem for policy in situations like Afghanistan since the fall of the Taliban to coalition forces in 2001.

To explain what I mean, consider modern counterinsurgency doctrine and the emphasis on “clear hold build.” I will not debate whether this strategy is the right one (I don’t know the answer, and the debate is older than I am), but I will note that modern counterinsurgency doctrine has long been praised—including by me—for focusing on local control over territory. “Control and govern Afghanistan” is hard. Afghanistan is complex. And it’s hard to know where to start after you’ve dumped all of your bombs out of the planes and beaten the enemy into retreat.

But “control this valley” and “hold this town” and “build this bridge” are, conceptually, more manageable. Of course the implementation is hard. But these are engineering-type challenges: get the materials here, work with the local notable there, establish security in this well-defined pocket of territory. The U.S. government thought of this a long time ago through the Strategic Hamlet Program in Vietnam, as did the British/Malayan forces with New Villages. You build collective security one place at a time.

My point is that this approach lends itself to a kind of reductionist vision of what it would mean to control Vietnam Afghanistan. Let’s say you can exhaustively divide Afghanistan into 100 parts. Then you could solve the problem of controlling each one. Each one is more manageable than solving all of Afghanistan. And you might hold the view that if you could solve the local control problem 100 times, then you would have controlled 100% of Afghanistan.

That is the conceptual leap, to conflate “control each individual part” with “control over the whole that the parts exhaustively comprise.” To continue the woodpile analogy, “Afghanistan” is more than the territories that comprise it; to even conceptualize Afghanistan, one must understand the arrangement of those territories relative to one another, and the organizing principle (or lack thereof) that holds them together.

There is plenty of fodder here for criticism of the social science research on Afghanistan (and on counterinsurgency research more broadly), but that is not my goal here because I’m not the one to level such criticisms and I have no idea how to evaluate the motivations, incentives, and commitments of those who worked on these questions. My point is one step more abstract. In the context of the rise of modern methods for causal inference in the social sciences, credible causal inferences about large social concepts are basically impossible. But a lot of research, including a lot of my own, has followed the strategy of disaggregating a large social whole (say, “governance“) into smaller parts that are easier to measure and for which modern methods are better suited.

We can make progress that focuses on local concerns. True progress! And local concerns are important! But it could be that the instinct to disaggregate—fueled by analytical tools that require disaggregation—produces a system of knowledge creation and policy research that cannot tell the difference between a woodpile and a pile of wood. No amount of attention to ash versus locust can tell you if the pile is stable or not.


* This tweet, by a former teacher and coauthor of mine, captured a lot of frustration among political scientists at the moment as well, in a way that resonates with what I write here.

** Illustrative examples include “democracy is not government by democrats, and authoritarianism is not government by authoritarians” and “microfoundations for political science.”