History as mind furniture

Historians don't build theories, they give us tools to build theories with

A debate has erupted in the blogosphere — Myles na Gopaleen used to write articles with every cliché in italics. A debate has taken place in the blogosphere about whether historians need to be more rigorous. Here’s Matt Yglesias and Noah Smith, and here’s Brett Devereaux who writes the excellent acoup.blog on mostly military history. His article spends too long being rude about Noah Smith, and it misunderstands what social scientists do, but it makes some interesting points. Here’s Andrew Gelman too. I was also provoked to think more by an exchange I had with Kaleberg in the comments on Noah Smith’s piece.

About how history the discipline actually works today, these guys underestimate the scale of the problem (I’ll stop now). Historians don’t just have vaguely-framed hypotheses. As I’ve argued here and here, they are seriously vulnerable to academic fraud and sharp practice, and unlike some social science fields, they have not yet collectively recognized the problem or done anything about it.

But the more interesting question in this debate is how history ought to be.

Non-competing magisteria clauses

There are two extremes to avoid here. The first is what theologians used to call “non-overlapping magisteria”. History has its own truths, social scientists have theirs, each produces them in their own way, and you can’t compare them. This argument makes no sense, because social science and history have the same subject matter: society. Here’s a chunk from my current reading, David Edgerton’s The Rise and Fall of the British Nation:

It is too easily assumed that the United Kingdom was at its most industrial in the distant past, when it was the workshop of the world in, say, the 1850s. The economic statistics tell a different story. It was in the 1950s and 1960s that the United Kingdom was most focused on industry, and on manufacturing industry in particular.

This is an economic question! But it’s also clearly a historical question. There’s no way to divide the disciplines up, and indeed the book cites several economic papers.

“Non-overlapping magisteria” is tempting because it lets everyone get on. In effect, it’s like a duopoly dividing the market. Each field writes its own theories of phenomenon X, and each agrees to pretend that the other field doesn’t exist. In practice, this happens a lot! But that’s a failure, not something to endorse. If academics in different fields say contradictory things about some phenomenon, they can’t both be right.

Scientology

The other extreme to be avoided is to think that history should just be economics (or political science). Historians should all be running instrumental variables regressions on long-run variables, all else is error and delusion. Brett Devereaux calls this scientism. I don’t find it easy to explain why scientism is wrong, but for example, take the murder of Julius Caesar on the steps of the Roman Forum in 44 BC.

BRUTUS: Great Caesar,—

CAESAR: Doth not Brutus bootless kneel?

CASCA: Speak, hands for me!

CASCA first, then the other Conspirators and BRUTUS stab CAESAR

CAESAR: Et tu, Brute! Then fall, Caesar.

This is a single event. It has huge consequences for world history, because it’s a step towards the end of the Roman Republic and the eventual emergence of the Principate under Augustus. In addition, our sources for this event are incredibly exiguous: some Roman historians writing a century later; a few references in Cicero’s letters; a contemporary Greek chronicler. Understanding what happened is more like solving a murder case than doing science with statistics. Well, it’s literally solving a murder case.

A scientism supporter might have two getouts here. The first is to claim that single events like this don’t actually matter. In the grand scheme of things, history is determined by deep structural forces; focusing on Caesar or Augustus is the “great man theory of history”; if Augustus hadn’t become emperor, some other member of the Roman elite would have.

I can’t literally prove that this is bunk, but there are very good reasons not to believe it. In particular, there is nothing “scientific” about it. The most scientific social theory we have, economics, provides no reason to think that only structural forces matter, and some good reasons against it. The reasons against it are that people need to coordinate their actions, and they regularly do so by following leaders. “Great men” matter, not just by chance, but because humans organize their entire societies in huge command structures in order to make them matter! That is why Vladimir Putin can send Russia to war in Ukraine, while many Russians and even many Russian elites would prefer otherwise.

Even in macroeconomics, the most “structural” economic discipline, many believe that political institutions are vital to determining a country’s growth rate; and political institutions are often shaped by powerful individuals. So, structuralism does not save scientism.

At your service, my queen

The second getout is that you can understand Caesar’s death with statistics. You could construct a database of killings of rulers and spot features that are reliably associated with them (younger, ambitious rivals; warnings from a mysterious soothsayer). Political scientists do just this with modern coups and ruler deaths, they indeed find some interesting results, and there are active efforts to extend this kind of work back in history. The economic historians running IV regressions on medieval pogroms or the spread of printing are doing something similar.

There’s nothing wrong with such work, and it may generate useful findings, but this argument misses the heart of the matter. History is the queen of the sciences, statistical work can only ever be its handmaid.

Think again of the murder case analogy. Imagine a real court case where the prosecution relied only on statistical evidence. Why’s the defendant guilty? Well, that sort of people often are. Obviously, that won’t work: you need actual proof that Colonel Mustard was in the library with the candlestick, not that X per cent of book-lined bludgeonings are committed by army officers. History is the same. It is full of one-offs. The death of Caesar is a one-off event. On a macro scale the Reformation is a one-off. The Industrial Revolution and the take-off into modern economic growth is essentially a one-off. They offer little or nothing for statistics to get hold of, yet they still matter.

But now we are left with the problem: if historical knowledge isn’t statistical knowledge, then what is it, and what’s it for?

Why history?

Statistical empirics have a story behind them: if you take a large enough sample, you can predict facts about the population. Sometimes this story fits the real world fairly closely, like when a pollster samples voters to predict an election. Sometimes it’s… a stretch: are the twentieth century’s conflicts a random sample from the same population as the twenty-first’s? But at least we know what we are trying to do, how it works under ideal conditions, and what happens when we deviate from those conditions.

Maybe history just describes what did happen, and makes no claims to predict the future. Historians sometimes take this modest line, but they usually can’t stick to it. It’s too tempting to say “this episode offers lessons for today”. And there is a good reason behind this. If history tells us nothing about today or tomorrow, then what is it for?

Most of us probably have a vague sense that it matters to know how the Republic fell or the Reformation happened. It just seems somehow worthwhile to know about such huge events in our own past. That needs justification, though: what makes it worthwhile? (I sometimes get the urge to do arbitrary sums in my head. For sure, doing so produces knowledge, but I would be silly to spend time on it, and I certainly wouldn’t deserve public funding.)

To be more concrete, I once read a letter from an officer serving in Iraq, who had a PhD in political science. Polisci has a large subfield studying war and conflict, but he reported that in understanding his surroundings, he found history more useful. Why might that be?

Here I get uncertain. I can only wave vaguely at what might be a promising way to think about this.

Ordinary knowledge of the past

For an analogy, consider everyday knowledge. I come down for breakfast. I know there’s bacon in the fridge because I remember buying it. I know it’ll taste good with eggs, because I have experience in the fried foodstuff department. I first learnt this, though, because somebody told me. These are all useful facts as I prepare to face the day. None of them are scientifically based. I have run no double-blind trials comparing B&E to muesli (what kind of monster…?) I’ve never taken repeated measurements of the mean tastiness of different breakfasts. And yet somehow I have come by useful knowledge.

In fact, ordinary unscientific knowledge covers a much bigger domain than scientific knowledge, and is more important. Modernity is based on science, but human civilization is based on “cumulative culture”, which is just knowledge transmitted over generations, and knowledge in general is the basis for human life itself.

What we learn from experience is vital, but it’s also limited. I’ve never broken my arm, eaten locusts or grown a radish. Luckily, others have and if need be, I can ask them about it. My country has never been invaded. Here, information is harder to come by. Few other Britons have experienced invasion first hand.

Here is a deceptively banal idea: history simply expands our ordinary knowledge to include other times and places. It gives us knowledge of large-scale situations like periods of civil war or the change from republic to empire; or of small-scale situations that are very different from our own, like an Occitan village at the turn of the 13th century. History rescues us from time-parochialism, by the simple means of telling us stuff.

The non-banal part of that idea is as follows. Understand how we get ordinary knowledge from ordinary experience, and you’ll understand how we learn from history.

That suggests some of the strengths and limits of history the discipline.

Ordinary knowledge leverages the capabilities of the human organism. We come into this world ready for it to have three dimensions, with inbuilt expectations about how objects interact (“folk physics” in the lingo). On top of this we rapidly build a reliable database of how things are, figuring out how to move around, picking up language as a means to learn more, and expanding the limits of our universe from the nursery to our society. We’re good at learning from raw experience, and from experience transmitted by others, even those who aren’t perfectly reliable. We take past experiences and fit them to the present to predict the future.

If history is just true stories about the past, you should expect us to be able to use it in that way. Perhaps that’s what the Iraq officer did.

Mental furniture

Put it another way. An engineer’s estimates are precise and specific: the load-bearing capacity of a certain material, say, expressed as a quantity with error bands around it. They feed into a precise process, like building a bridge that can take a certain tonnage of lorry. Macroeconomic forecasts are equally precise: estimates of future GDP growth, say, with 95% confidence intervals. They also feed into a tailored process. The Central Bank has a regular committee meeting, reads the forecasts and picks a number for the interest rate. There might be a bit less precision here than for the engineer; the economy is complex and not that well-understood, so there’s more legitimate scope for expert judgment.

History stands at the other extreme. It doesn’t feed neatly into a policy process. It’s less like a machine on a conveyor belt, and more like general mental furniture. We read history to learn a lot of stuff about the past. We don’t know which bits will be important, although historical authors will try to suggest how their period has lessons for today. But those lessons aren’t neat or predictable. Just as I pattern-match this morning against my experience and think “mmm, bacon!” so the politician or journalist pattern-matches the present day against historical experience to have a sense of what might come next. Both are mysterious, little-understood processes that can’t be summed up in equations, but both are real.

This theory might explain two things about history books. They are typically narratives, and they have lots of descriptive detail.

The narrative aspect matches how we actually experience events. Stuff happens, now what will happen next? Well, this happened, and that is how it turned out. Supplying history in this form might make it easier to extract the epistemic juice. Perhaps this is why Annales-style history, which draws a static picture of a social scene, seems to find it harder to hold the reader’s attention — at least in my experience.

The descriptive detail is because the historian isn’t sure what matters. It’s like asking your friend about Montenegro. If he knows exactly what you care about — the odds on the Baccarat tables which feed your spiralling gambling habit — he can focus in on that. But if he doesn’t know, he’ll just give a broad description of his weekend there, throwing in all kinds of information, so you can decide for yourself if Montenegro’s the place for a holiday.

Histories contain the kitchen sink. David Edgerton’s book tells me about the meat liners which sailed between Britain and the River Plate; the Valiant, Victor and Vulcan nuclear jet bombers; and the 1950s peanut-based wool substitute, Ardil. These eye-catching details would be out of place in an economics paper which aimed to test a specific theory. But they make sense on the view that histories provide material — mind furniture — for people to construct their own ad hoc theories.

Ardil: not a great British success story

What mean?

If historians are about a different business than the social sciences, that does not mean that the two are incommensurable and can just ignore each other. In fact, they are rivals, not just for undergraduate numbers and grant budgets, but as ways of explaining the world. My story is that historians hope to give people materials to build ad hoc theories and make predictions. Social scientists go about building theories and making predictions themselves. This is asymmetric warfare, with historians as the mobile but undisciplined guerrillas and social scientists as the powerful and organized, but lumbering, army. At present, I don’t see any reason for either side to win. These are some points against social science:

• Social scientists create rigorous theories which explain a clearly-defined dependent variable with a few independent variables. They test them with stats, typically with linear regression. Fine, but it’s hard to build a linear regression that can predict whether a particular image is the number six. So you may struggle to predict whether Russia will win in Ukraine. Life is complex, theories are (and should be!) simple.

• Social scientific theories also typically aim at predicting conditional means, i.e.  averages.  What’s the average effect of being an autocracy on a country’s growth rate? That’s an important question, but not all countries are equal. China’s ruling elite are not an average autocracy, and the effect of Chinese autocracy on Chinese growth may matter more in the long run than any average.

• Because of how social life works, rare events are very informative. Most people don’t commit crimes most of the time. But to understand the incentives involved, you need to look at the rare, “off-equilibrium”, cases when someone does commit a crime. Statistics is harder when there are not many cases. The same logic applies to the rare event of going to war. You can build a database of wars, but to get a decent N you might have to go back to the 19th century when there were no tanks or planes; or to include tiny civil wars in the same dataset as World War Two. Any resulting generalizations may be useful, but I don’t think they should make military intellectuals give up reading Thucydides or Sun Tzu. If rare events matter, then narrating them might illuminate things.

Bringing a knife to a Methodenstreit

History also has its pathologies. Our inborn folk physics can be misleading: objects do not, in fact, remain at rest unless subject to an external force. Equally, we might be misled by the intuitive way we build explanations from historical furniture. Narrative can be informative, but can also direct our attention away from regularities that a statistician might find. It’s easy to tell a story about how each European state gained or lost territory in some war, but from this collection of stories, would you spot that over a millennium, states with inbred rulers regularly lost a bit more territory? As Charles Tilly pointed out, narratives also naturally come with heroes and villains, just like fairytales. Sure, important individuals matter, but so do those anonymous forces which are hard to weave into a story.

Historians also do make theoretical claims and causal claims, and sometimes in doing so they illustrate why economics got so nerdy and data-crunchy in the first place. Let’s go back to part of Edgerton’s argument.

… the fall in the United Kingdom’s share in world exports of manufactures was a notable feature of the period…. Furthermore, the share of exports in GDP fell… into the late 1960s…. From the early 1980s the United Kingdom imported more manufactures than it exported. In the manufacturing-focused picture of the economy this was a calamity…. A different analysis is needed. The main determinant of manufacturing export shares was what happened not in the United Kingdom, but in other countries. It was perfectly possible to be succeeding in promoting manufacturing, and indeed manufacturing exports, and to lose market share at an expanding world level.

An economist would think about the issue in terms of estimating a model, say

Eₜ = a + bₜ

where E is British manufacturing exports, t indexes years, and bₜ is a set of time dummy variables. Or maybe

Eₗₜ = a + bₜ + dₗ + g(dₗ × postwar)

where now we include other countries, l indexes countries, dₗ is a set of country dummy variables, and dₗ × postwar represents country l during the postwar period. These two models ask different questions, so they can get different answers. The first asks “were exports faster in certain periods” such as after WWII? The second is the (in)famous diff-in-diff estimator and asks “were British postwar exports unusually high or low?” These aren’t the only equations you could estimate, and there are lots of complexities about how you do it, but the point is that you know what question you are answering.

If you compare the historian’s approach it looks much messier. What’s the question? What counts as “successfully promoting exports”? Is it just “increasing manufacturing and exporting in the economy”? But if all countries are doing that, what’s the evidence that Britain is successful, and not that, say, world trade is starting back up after a post-WWII low? It’s hard to know what the theory is here. Words are too imprecise and verbal theories are too Protean.

Consilience

I read a lot of history and a lot of social science. I’ve even written a bit of both. (A book of history, you say? I’m glad you asked: check it out.) I find myself puzzling over the same question that the Iraq officer had: history seems helpful, but how exactly? I’ve tried to get beyond just “agreeing to disagree”, but also to avoid thinking of social science as the One True Way to knowledge, to which history must bend. I think understanding how history helps readers know things might be a more promising avenue than how history builds and tests theories.

It could be that there’s rigorous metatheoretical work to be done here. The environmental sciences have put a lot of thought recently into answering questions like “did global warming cause this extreme flood?” – that is, into explaining individual events, rather than whole classes of events, which is the more typical case in science. But perhaps it is more likely that historians can build a coherent defence of what they do, which talks with not past social scientists, without needing a whole formalized statistical epistemology.

The past has a future

Here’s a historically-informed prediction: in the long run history’s prospects should be good.

There are reasons good and bad why its undergraduate numbers have fallen. On the one hand, academic history is indeed often ideologically extreme and narrow, and that is a shame. If the ideas above are near the mark, then the historian’s job is to let the present learn from the past. So it is strange when historians seems determined to do things in reverse, teaching the past a lesson from the present. It’s also weird that while contemporary historians are more than ever against “Whig history”, many seem to show more than ever the enormous condescension of posterity.

On the other hand, materialist attitudes to education are strong today, and history degrees are not as easily marketable as accountancy or law. But it hasn’t always been that way, and it needn’t stay that way forever. (As history shows.) After a period in the dumps, philosophy has come back to fascinate a new generation of trolley-bothering altruists. There’s no reason that history won’t do the same. Its problems aren’t permanent. For anyone who hopes to work at the level of politics and social change where nothing is predictable — where history is made — it still offers a unique treasure trove: the collected experience of humankind.

[Update: Noah Smith, not Feldman. Apologies.]

---

If you liked this article, you might like my book Wyclif’s Dust: Western Cultures from the Printing Press to the Present. It has a bit of history and a bit of theory, it’s available from Amazon as a paperback/hardback/ebook, and you can read more about it here.

You can also sign up for this newsletter (it’s free):

And if you’re very important and influential — 👉 which you are 👈 — feel free to share this post with your many followers: