33 Comments
User's avatar
Matt Wansley's avatar

This post resonates with me as a law prof who teaches torts! In tort lawsuits, we ask juries to determine factual causation by asking if the defendant's negligence was a "but-for" cause of the plaintiff's injury--an explicit appeal to counterfactuals.

But I'd like to offer a qualified defense of the historian's aversion to counterfactuals: it might serve as a useful check on the human tendency toward overconfidence about our own knowledge.

Consider, as an analogy, the use of "historical cost" in accounting. When accountants put together a business's balance sheet, they value many of the business's assets by their historical cost--what the business paid for the asset when the business bought it. In many cases, the historical cost of an asset will be out of step with its market value. For example, a building that the business purchased 20 years ago might have appreciated in value dramatically as the neighborhood around it improved. But the accountant will use the decades-old historical cost rather than attempt to estimate the building's current market value.

Why does accounting use this fiction? The market value of many assets is hard to estimate, and businesspeople might--consciously or unconsciously--be overconfident when they value their business's assets. Historical cost accounting checks that impulse.

That's how I understand the historian's aversion to counterfactuals. We might be overconfident when we speculate about what might have happened. So historians have adopted a norm that tamps down on speculation, even though, as Dan explains, they are necessarily reasoning about counterfactuals when they make causal claims.

fox's avatar
Feb 12Edited

Edit: I should add I don’t think this is exactly unique to the humanities but those have been the debates where I’ve been annoyed by it recently.

Edit 2: I was posting on my phone and ended up screwing up my original comment so I just deleted it then reposted.

Great post that gets at one of my biggest gripes with some humanities types. Their refusal to be explicit about the counterfactuals implied by their causal claims, in particular as they pertain to the present state of affairs, lets them smuggle in entire worldviews without ever testing them against reality. Does capitalism cause poverty? Is colonialism responsible for all the woes of the global South? Both judgments pivot entirely on the realistic alternative, for which they substitute a sort of utopianism that justifies their political project.

Also, wrt the move away from laws in the sciences, I was reading a paper yesterday that referenced Cajal's laws of neuronal optimization and thought, what a fun anachronism. Even if we end up talking about similar types of optimization objectives in the modern abstraction-correspondence framework, the linguistic tic reminded me of the physics envy of the past, which has since been replaced by all new kinds of even more advanced physics envy.

Felice's avatar

>the physics envy of the past, which has since been replaced by all new kinds of even more advanced physics envy.

Come on, now; the physics envy also has to obey the march of history, and progressive-overload itself into ever-more-evolved forms.

Paul S's avatar

I did not know about the timing of the USSR declaration of war against Japan! That's an excellent thing I can add to my bag of tricks for winding up undergraduates when I tell them (genuinely) that I think Truman was justified in dropping the A bomb in 1945.

"well actually, Truman probably did the Japanese a favour because the alternative was Soviet invasion then domination"

Great timing as well, we're doing the political theory of terrorism next week, and the attacks on Nagasaki and Hiroshima inevitably come up in discussion.

Chaotic Neutral's avatar

Completely agree, if you care about causality then you must care about counterfactuals.

And it is a rare historian indeed, even on attempting a strict “how the facts unfolded” telling of history, that refrains from causal explanations. “The battle was fought for 3 days, initially one side was losing for X reasons, but in the end Y was superior and resulted in victory.”

You can’t say something even so simple without considering the counterfactuals of what would have happened had X and Y been different.

Yusri's avatar

I think that when humanistic historians express opposition to counterfactuals, they are specifically referring to the second type of counterfactuals mentioned at the end of your article, rather than rejecting counterfactuals as a whole. After all, they are not analytical philosophers. In your example about Truman, it might be more accurate to say that humanistic historians have an interpretive framework rather than an implicit model. and there is large difference between the two.

Regarding the distinction between general and particular explanations, this distinction lies in whether the explanation focuses on reconstructing locally meaningful reasons (particularist) or on applying broad, portable regularities about human behavior (generalist). The humanist typically believes the second approach is not good or typically misleading.

Kenny Easwaran's avatar

This is great! Coincidentally, this week in my seminar on knowing how and AI, we were reading Dreyfus (and some relevant bits of Heidegger and Hegel!) and his discussion of skill and expertise makes some similar points. I think his main points are really good - that when you think about the “rules” for playing good chess (trade pieces with equal point values, advance your pawns, control the center of the board) or for driving (follow at 1 car length per 10 mph of speed, travel at a few mph above the speed limit) it’s clear that these rules mainly explain what the novice is doing, but the expert has a kind of fluency that they don’t know how to express in rules. But then he goes on to say that what the expert does is learn to recognize tens of thousands of types of board configurations or driving situations, and apply practices that worked in those cases in the past.

But it seems to me that this assumes too much of a dichotomy - either it’s rules expressive in public language or it’s just particular cases. I think neural nets give us a much more useful model, where we get a facility for processing the many kinds of information through meaningful intentional systems that aren’t exactly conceptual, in ways that let us recognize many types of cases similar to ones we have seen before, but also hopefully generalize to new cases.

Rob Sica's avatar

Wonder if historians will take up this recently articulated challenge by Tetlock & co. to predict what new evidence will show:

"Predicting the Past: Testing Expert Historical Judgement"

https://academic.oup.com/ahr/article/130/4/1615/8404686

Barry Lam's avatar

Do you remember when Bush offered the Taliban and then I think Hussein some kind of ultimatum, something to the effect that they relinquish power and there would be no invasion? I remember thinking at the time that the invasions were so overdetermined at that point that there was no way it wouldn’t happen however those regimes responded. It shocks me to think that historians would not be at all interested if I asked them whether my suspicion was true, and whether there was evidence for it.

Daniel Greco's avatar

Totally. While the idea that proper history doesn't traffic in counterfactuals is sometimes endorsed by professional historians, I doubt anyone is really consistent in sticking to it. How could you be?

Barry Lam's avatar

Seems like historians practice reasonable history and endorse unreasonable philosophy of history. It’s just like Richard Kroll who I just posted about saying that the Romantics and 60s rockers made great art but had a very shitty philosophy of art. And scientists walking around venerating Popper….

Age of Infovores's avatar

> It contains a piece by Winston Churchill himself, speculating about how World War One might have played out differently—or as it turns out, not at all—if Robert E. Lee had won the battle of Gettysburg... But the problem with such exercises is not that they involve counterfactuals. Rather, it’s that the counterfactuals at issue involve such distant possibilities, in which so many background conditions are simultaneously altered

In other words, this is not an indictment of counterfactual history but an invitation for historians to think more precisely about causal identification. Perhaps they should read Angrist and Pischke?

Jordan Nuttall's avatar

You’re a great writer and a great mind Daniel.

Mary M.'s avatar

I think this is right on! I recently restarted the World War Two in Colour documentary (narrated by Robert Powell), and I was once again impressed by the constrained counterfactual reasoning it demonstrates. The method behind its narration involves taking a few proximate variables (Hitler’s stubbornness, winter in Stalingrad, Britain’s geography, limitations of the Luftwaffe) and considering their likely impact. To me, this seems like a responsible approach that gives a dynamic picture of the events—while avoiding the speculation that would seem unwarranted. In other words, I think this is history done well :)

Radek's avatar

All claims of casual relationships are claims about counterfactual histories

Tanya Polarbear's avatar

Apologies if this is a naïve question, but aren’t there non-counterfactual conceptions of causation? Could historians be thinking about history in those ways?

Daniel Greco's avatar

Not naive at all. It’s certainly the case that not everybody tries to analyze causation in terms of counterfactuals. But I do think pretty much everybody accepts that there will be tight connections between them. For example, the ellipsis in that Hume quote I offered covered up what’s typically thought of as the seeds of a pretty different, non-counterfactual account of causation: a “regularity” account, in the lingo. (He treats them as equivalent, and connects them with “in other words”, but most people think they’re interestingly different.)

Roughly, a regularity account analyzes causation in terms of regularities in nature, so that we can say A caused B when events of the A-sort are regularly followed by events of the B-sort. (How does this sort of view distinguish causation from correlation? Does it at all? Good questions!) But people who like regularity analyses of causation will typically go on to explain counterfactuals, somehow or another, in terms of regularities. So it’s still the case that causal claims commit you to counterfactual claims, but the explanation of how that works will be a bit different.

My suspicion is that something similar will hold for whatever non-counterfactual account of causation you’re working with. That is, people who think causation can’t be defined in terms of counterfactuals—maybe counterfactuals have to be defined in terms of causation!—will still typically think that claims about causation will imply or at least strongly support claims about counterfactuals. So going for some non-counterfactual account of causation won't give historians a way to remain agnostic on how history might have gone, if their accounts of how history did go include claims about causation.

Tanya Polarbear's avatar

A follow up if you don’t mind (and no problem if you do). I know that a prominent if not popular view is that all non-trivial counterfactuals are false (Alan Hajek has argued this pretty forcefully, among others). On that view, it is false that Truman would have dropped the second bomb even if the USSR hadn’t declared war maybe because the antecedent is hopelessly underspecific to pick out a unique possible world. That seems like a really compelling view to me. But I’m sure anyone who accepts it would still want to be able to tell causal stories, including about history. Perhaps without appealing to counterfactuals they could also happily talk about the probability Truman drops the second bomb given the USSR declares war?

This makes me wonder if the dilemma for the historian is a bit less metaphysical than it seems at first be. Perhaps it’s just that the kind of claims that are historically interesting to make imply certain counterfactuals.

Daniel Greco's avatar

Yeah my understanding of the Hajek view is that he thinks what counterfactuals strictly mean is: "If it had been the case the P, then it would have been the case that Q with 100% probability" and then those are all false because the probabilities are always less than 100%. It's a lot like the view that no surface you'll ever encounter is truly describable as flat, because "flat" means "having absolutely no bumps," and everything has some bumps (at least on the microscopic level). Or if you say that some event started at "four o clock", that's always false, because to say it started at four o clock is to say it started at 4:00.00000000...and that's basically never true (it will always have started a bit before, or a bit after).

I tend to these sorts of views are false, but even if I accepted them, I'd think that in general we're not trying to say stuff that's strictly true, but only approximately true. And if the truth is that some probability is reasonably high, saying something that strictly implies that it's 100% might be close enough to true, just like calling something "flat" when it has some small bumps might be similarly close enough to true for most conversational purposes.

Tanya Polarbear's avatar

Thanks! That’s a helpful way of looking at it. I think a standard response to the first move is that you get agglomeration failures, where you have sets of counterfactuals that are individually true (because you’ve set the threshold for that below 1) but their conjunction is false. If a bunch of counterfactuals were almost true but their conjunction almost certainly false that would bother me which makes me think like you that we’re not just trying to say things which are almost true.

Sorry, I didn’t mean to make you digress into a discussion of that issue. My point is rather that it seems like a sensible if radical position, but not one which would preclude making any historical claims. Perhaps they would have to be claims about how actual events changed the probability of what followed from them, I’m not sure, but I think that historians would have a lost less of interest to say, maybe because counterfactuals are necessary for narratives in a way that they aren’t necessary for science, say.

Tanya Polarbear's avatar

That all makes a lot of sense. Thanks for giving me such an informative and detailed answer, and for such an insightful post too

Felice's avatar

Helpful piece! I wondered about how one would go about counterfactuals in some discussion on slavery's impact on the South's economic development (thread in Muñoz Note iirc), so it's nice to see this discussed.

Just from a mundane quantitative (computational) POV, I think a lot depends on constraining the sphere of consideration to avoid getting into doing "speculative fiction". Which means, practically speaking, that -- bc alternate possibilities grow exponentially over time -- you can only really do this sort of analysis and make sense for recent history, if you care about effects on the present day. But then again, that's probably all we really want? We can better understand the current world order if we think through what did and didn't happen during the Cold War, but trying to figure out what branch of history we would have gone down if something or other didn't happen to Alexander the Great is just some exercise in fiction bc we probably couldn't even imagine such an alternate present day in any case.

Even something like, "What would the English language be like today if Shakespeare hadn't lived?"; sure it would have developed in one way or another, but it's very different from trying to figure out how East Asian languages would be today without some major initiatives and events in the 20th century.

ETA- I do think maxims like "Those who don't learn from history are doomed to repeat it" suggest some longstanding inclination towards finding "laws", just maybe not in any particularly systematic way.

Kenny Easwaran's avatar

I’m a bit confused by footnote 2 - are there some negations missing?

Daniel Greco's avatar

I don't think so? But I admit it's not super clear! Oakeshott is saying, wrongly in my view, that historians shouldn't draw the essential/incidental decision, and I'm just trying to say in the footnote why that amounts to saying they shouldn't think counterfactually.

Daniel Greco's avatar

Yes, there was a typo! I don't know why I missed it the first time. Missing negation, just like you said. ("had" should've been "hadn't"). Fixed now.