Science of History and Counterfactuals

A the Santa Fe Institute

I am back home after a very intense and productive trip to Santa Fe. But in the meantime the world has made yet another step towards World War III … or at least that’s what some commentators say. Others claim that the bombing of Iran’s nuclear facilities by the USA was a “nothingburger.” It’s too early to tell (which is why I generally avoid commenting on news in real time).

At the Science of History conference, which I attended, there was a lot of discussion about the role of contingency in history. We didn’t talk specifically about Donald Trump, but the President’s decisions in the days ahead will clearly have an enormous impact on whether we end up in World War III, or some “lite” version of it—without nuclear explosions (I really hope so), but perhaps with huge economic and political disruptions. Or that the whole thing will be defused.

This moment, then, seems like an example of the key role an individual can play in history. Perhaps. On the other hand, judging by the support for war from most Democrats, President Harris, were she elected, would probably be an even more enthusiastic supporter for inserting America into the Israel-Iran war. Trump campaigned against unending military adventures but now he has started another one himself. Clearly, he is under enormous pressure to go against his election-year promises, as well as against the wishes of a large chunk of his “base” (see, for example, an NYT article, Trump’s Base in Uproar Over His Openness to Joining Iran Fight). So, does this example support the Great Man Theory, according to which history is largely explained by the actions of highly influential and unique individuals? Or the opposite, the strength of larger social forces that constraint leaders? My guess is that both are important, but how to make it more than a banal statement is a big question

Returning to the Santa Fe Institute (SFI) conference, one observation that came to my mind was that there were very few complexity scientists in the room, which is surprising since the SFI was the pioneer in establishing this research direction as a bona fide scientific discipline. Most speakers were historians, philosophers, and historians of science; plus a couple of archaeologists, an evolutionary psychologist, a sociologist, and one complexity scientist (yours truly). I am not complaining; this was a terrific conference and I learned a lot. But my insistence that we need to use mathematical models (broadly understood, including computational approaches such as agent-based simulations) was clearly a minority view. I don’t say that models should replace other approaches in history, far from it. But they are a necessary part of the whole.

Take contingency and a related idea, counterfactuals, which were much discussed in the conference. The idea of counterfactual history had fallen out of fashion amongst historians towards the end of the twentieth century, but was recently brought back by the historian Walter Scheidel in a recent book, Escape from Rome (see my review of The Great Escape). It generated considerable discussion, for example, by Mark Koyama in his review essay on Walter’s book (of course, Mark is an economist, so much more open to models than your typical historian). Counterfactuals, indeed, have a tremendous potential in historical sciences, in which we cannot do controlled experiments. In my view, however, Scheidel’s use of this approach shows the limits of such non-mathematical, “verbal cliodynamics.”

One of the major ideas in Walter’s book (and the one he uses the counterfactual approach on) is that a major factor in the rise of Europe was its political fragmentation. This “Fractured-Land Hypothesis” seemingly explains why China, which was most of the time unified by a mega-empire, fell behind Europe. But why was Europe disunited? A most popular explanation relies on the geographic differences between the west and east ends of Eurasia.

In a recent 2023 article Jesús Fernández-Villaverde and colleagues (including Mark Koyama) developed a dynamic model that explored the effect of topographical and environmental features on state formation in Eurasia. These geographic drivers included terrain ruggedness, climate (assuming that both too cold and too hot temperatures were detrimental to imperial expansion), and agricultural productivity. Most importantly, the simulation was run within a landscape representing actual continental shapes (that is, peninsulas and narrow seas dividing them).

The model outcome was to see what proportion of China and Western Europe were unified by a single empire as time evolves. They found that in multiple simulations China rapidly and reliably unifies, while Europe always stays fragmented. Furthermore, systematic variation of mechanisms and parameters that are included in the model allowed the authors to understand what factors are needed to generate this result.

Their conclusions were as follows. First, it is not differences in ruggedness that explains fast Chinese unification and persistent European polycentrism, because China is, actually, more rugged than Europe. The main factor explaining China’s recurring political unification, predicted by the model, is a core region of high agricultural productivity located on the East China plain. Europe, in contrast, suffers from two obstacles to unification. The Mediterranean Europe, unlike China, is divided up by seas. Northern European Plain, on the other hand, suffers from colder climates that slow imperial expansion. As a result, the model by Fernández-Villaverde et al. predicts formation of large states to the east of “Europe,” in areas corresponding to Ukraine, southern Russia, and Kazakhstan.

Here's a fine example of how counterfactuals should be done. The model and overall conclusions of Fernández-Villaverde and colleagues are not free of problems, and I discuss them in my forthcoming book, The Great Holocene Transformation. But the important advantage of writing an explicit model (and publishing its code) is that we can know precisely how the results follow from the model assumptions; and where we disagree with the assumptions, we can write an alternative model and investigate how the results change. In other words, this is a way that knowledge is cumulated.