HPE is a fast-growing field. That may not necessarily be a great thing. Just as more data does not mean more information, more research does not mean more knowledge. Ultimately, the success of a field depends not on the number of uniquely interesting publications with representative keywords but on whether the array of results it produces can be aggregated to arrive at more general knowledge about the world.
A recent post by Sean Gailmard sharply characterizes one important issue: HPE asks narrow empirical questions, investigates those questions in narrow empirical settings and does not invest enough in theory to determine how widely the empirical results can be generalized. So we have the problem of knowledge-extrapolation across empirical domains. I think that any piece of empirical research suffers from this problem, but one can see how it can be exacerbated in HPE due to its overt emphasis on case-specific, highly contextualized research.
My goal here is to articulate a complementary concern of knowledge-aggregation in HPE. Knowledge is aggregated when we combine a set of findings, empirical or theoretical, to learn something that would not follow from any of them individually. My, hopefully unjustified, concern is that HPE, as it is currently practiced, yields surprisingly little general, theoretically appealing knowledge about its key object of affection: history. I will argue that by focusing on causality, which has its rightful place in the field, HPE is unwittingly forcing a model of history where the dominant role is played by accidental natural events–a model that most HPE practitioners would likely find objectionable.
Causes and roots
Like every other empirical field, HPE has gone full force to endorse, stylistically and methodologically, the “credibility revolution” with its emphasis on cleverness and the “leveraging” of “plausibly exogenous variation” and such. In many ways, this has served HPE well since it opened up a new front in how history can contribute to other social sciences: as a field that can be harvested for natural experiments to test general theoretical claims about politics or economics.
Dower et al., for example, use data from 19th century Russia to test a prominent theoretical proposition about the link between collective action and representation. While they provide a detailed discussion of the historical context, it is clear that Russian history is a nuisance for the main point they are making — any other context with similar data opportunities would serve their goal equally well. Since they test claims derived from a general theory that is not tied to any specific historical context, the contribution of these kinds of studies towards general knowledge is clear (though see Scott Gehlbach’s post expressing a reservation on this matter). It is a separate question, however, whether econometrics with archival data is what makes HPE “historical.”
Many, if not most, HPE studies are primarily focused not on testing general causal theories, but on tracing historical roots or origins of things. Implicit in such studies is a view that an explanation of a phenomenon amounts to a reconstruction of its concrete historical pathway. To understand Y, we need to identify its roots in the past and trace them to the present. Equivalently, one identifies a certain X in the past and argues that some Y in present is a legacy of X from the past. Here, history is interesting intrinsically (not as a sandbox of research designs), and the goal is not necessarily to identify the structural causes of Y, but to characterize (qualitatively or quantitatively) the sequence of processes through which Y came about.
Making general claims about historical roots beyond trivialities like “history matters” or “things persist” is complicated, because concrete roots are bound to be context-specific. There is not much we can learn in general about the origins of democracy, for example, from an observation that the European democracy has its roots in Medieval parliaments,[1]because the counterfactual to the entire path of European history is impossible to enumerate. It is not surprising that the classical writers in this field adopt a defeatist (and anti-scientific) view that “the social world’s order does not reside in general laws.”[2] What is surprising is that the quantitative and causal revolution in HPE might have strengthened this instincint towards particularism.
Instruments as roots?
The most impactful HPE studies usually do both: identify the effects of historical causes and also trace the historical roots. Usually, an attempt is made to (a) use a historical instrument Z to estimate the causal effect of X on Y and (b) to trace the historical roots of Y in X or Z. While it may seem like we have the best of the two worlds, this actually creates a tension: successful identification of causes makes it difficult to make general and theoretically illuminating statements about the roots. A plausible instrument must be exogenous to all political and economic factors that could also drive the outcome of interest. Hence, if the instruments are treated as historical roots, it must follow that the historical roots are confined to factors that are exogenous to politics and economics.
The most famous HPE paper, “The Colonial Origins of Comparative Development” by Acemoglu, Johnson, and Robinson (AJR), is a good illustration of this tension that is present across HPE literature. On the one hand, AJR tell a general causal story about the impact of political institutions on economic prosperity. That story could be told independently of any specific historical context, without any references to Europe, colonialism, and such. There are good abstract, theoretical reasons to expect that institutions which limit the ability of the government to expropriate would motivate investment and growth. The historical context in which AJR study the causal relationship between institutions and development just happens to be a sandbox in which they found data and empirical design. They use mortality of colonial settlers that was determined, as they argue, by natural factors independent of political and economic processes as an instrument for institutions. The instrument seems plausible precisely because it has little to do with the human-made world.
But on the other hand, AJR also tell — very forcefully — a political development story about how growth-promoting institutions spread from Europe to other parts of the world through colonial conquest. Those places where colonizers could easily settle by virtue of not facing deadly diseases became prosperous, because this is where they were able to establish institutions that later promoted development. It is telling that the title of the paper is not “The Institutional Causes of Comparative Development,” but “The Colonial Origins of Comparative Development.” The main substantive emphasis is not on the causal effects of institutions, but on concrete historical-political development: the roots of economic growth lie in colonialism, even though colonialism as such does not play any role whatsoever in the theorized link between institutions and growth.
The design requires colonialism to be incidental, but the conclusion makes it essential. Had AJR discovered another exogenous source of variation in institutions that had nothing to do with colonialism, they would have had to tell a very different political development story but an identical causal story. A different instrument would have given them a different historical root of prosperity while the cause of prosperity would have remained the same. The instrument is incidental, because it does not follow from any broader theory of institutions and development, but is an artifact of a very concrete and likely non-replicable historical condition.
The implied model of history
So why does it matter? It matters because if we zoom out and ask what general statements about historical development we can make from the body of empirical and, especially, causally oriented HPE work, we basically are bound to admit that the driving forces of history are random, often natural events that are orthogonal to political, economic, social, or any other processes that we are actually interested in. This must be the implied theory of history that HPE has adopted by virtue of focusing so heavily on causality and so on factors that are as exogenous as possible to the human world: earthquakes, pestilence, droughts, rainfall, droughts and floods, weather shocks, climatic shocks and so on.
This “germs and steel” version of history that is emerging within HPE has an important role to play, but at least we should have a debate whether we want to adopt geographic determinism as our underlying framework. Karl Marx was right that people make history, but not as they please. Surely, some of the constraints that people face are driven by natural forces, but I doubt that we want to adopt a view that these are the only constraints that matter because we just happen to be able to identify them causally.
To see this point from another angle, consider how we think about the role of instruments in non-historical studies. We would not say, for example, that the historical roots of income inequalities lie in quarters of birth or the Vietnam draft lottery even though these are used as instruments in regression equations that explain earning differentials. But this is exactly what we seem to be doing in HPE.
Can theory save history?
To enable a more coherent understanding of history from HPE, we would probably need to decouple the research on causes from the research on roots; or at least to avoid the practice where something that is merely incidental to our theoretical thinking about the problem becomes treated as a central historical root. Whether something may or may not qualify as a historical root should not be defined by its level of exogeneity, but by its theoretical reasoning that ideally builds from first principles.
Much of the theorizing in HPE is not abstract enough to enable generalization. Theories are often formulated not in terms of variables, but in terms of concrete facts. References to history are often used as a “theoretical” justification for a hypothesis (we test “X” because two historians said “X”), as if this were not circular. We should go in a different direction and build theories, inductively or deductively, that explain concrete historical facts in terms of more general principles. All I am really saying here is that we should follow Przeworski and Teune’s age-tested dictum that theories should not contain proper names.[3]
This brings a question about the role of theorists. The verdict by Sean Gailmard that empirical HPE does not theorize enough is well-deserved. Theorists have given many instructive and welcome lessons to empiricists on how to interpret their research. But what seems to be completely lacking is theorists taking a body of empirical work with diverse findings and figuring out a theoretical framework that can potentially synthesize it. Empirical HPE looks like a field on which everyone throws their own brick. Sometimes several bricks land by chance next to each other, and we have a “growing literature” around some topic. Theorists could be of great help in building a useful structure out of those bricks.
There are some great examples from non-HPE on how this can be done. In a recent working paper, Tara Slough takes a set of studies on accountability relations between voters and politicians, which has yielded highly variable results, and shows through a formal model how this heterogeneity could be explained by differences in bureaucratic capacity. This gives a coherent framework to generalize from a large body of empirical work that otherwise appears messy and disconnected. There are many areas in HPE that could benefit from this kind of theoretical synthesis.
[1] B. Downing, The military revolution and political change: Origins of democracy and autocracy in early modern Europe,1993.
[2] Ch. Tilly, Contention and Democracy in Europe, 1650-2000, 2004, p. 9.
[3] Adam Przeworski and Henry Teune, The Logic of Comparative Social Inquiry, 1970.
