Counterfactual History & The Minimal Rewrite Rule

Guest blogger Helen Zhao writes...

Recent years have witnessed a proliferation of counterfactual histories -- that is, historical narratives assuming the truth of carefully chosen falsehoods.

One example is Peter Bowler’s 2013 book Darwin Deleted. It considers the scientific, political, and cultural landscape of the nineteenth and twentieth centuries had evolutionary theory’s Big Man never existed. On the one hand, Bowler concludes that evolutionary theories would still have emerged. Most likely, social and political conditions would have given rise to twentieth century Nazism and racism, even in the absence of Darwinian theory. On the other hand, he argues that natural selection would have played a smaller, if not nonexistent, role in evolutionary theories of the nineteenth century. Moreover, in the absence of Darwin’s harsh vision of nature -- “as full of cruelty and suffering” -- science and theology would have made for easier bedfellows.[i]

Other examples abound outside of history proper. In fields like paleontology and evolutionary biology, they include Dougal Dixon’s 1988 book The New Dinosaurs and Dale Russell and Ron Seguin’s 1982 paper on dinosauroids. (See Derek Turner’s post ‘What-if Prehistory’ for an introduction to this literature.) This is unsurprising. After all, paleontologists and evolutionary biologists, like historians, are interested in understanding contingencies of the past. And counterfactual construction provides a useful means for exploring these contingencies -- for reflecting on the kinds of intelligent life that would have thrived today had key events such as the Cambrian Explosion and the K-T extinction never happened.

Nevertheless, reception for counterfactual history continues to be mixed. Historians like Richard Evans, Bob Richards, and Allan Megill dismiss the enterprise as “a waste of time”, “grounded on “evidence [which] can only itself be imagined”, and “deeply permeated by under-supported assumptions about the real nature of the world”.[ii] Evans goes so far as to characterize the growing scholarly interest in counterfactual history as a symptom of a “postmodern world” -- a turn away from history towards historical fiction.

Why do they object so strongly? One thought is that, for example, as Bob Richards argues in his review of Darwin Deleted, it seems unlikely that anyone—even historians with a bevy of historical evidence—could predict the consequences of a minor alteration to historical trajectories:

in a complex matrix of interacting events, the conceptual addition or elimination of a significant cause must have unpredictable ramifying consequences...If the cause added or subtracted is magnitudes more powerful than the ripple of a butterfly wing for example, the deletion of Darwin from the intellectual world of the late nineteenth century the echoing perturbations must shatter all expectations.

By contrast, historians Geoffrey Hawthorn, Philip Tetlock, Aaron Belkin, James Fearon, Geoffrey Parker, and Richard Lebow are more optimistic.[i] They concede that counterfactual history is “dauntingly difficult”; however, they argue that adherence to methodological rules governing the writing of counterfactual history can produce narratives which are nontrivial and well-substantiated by evidence.

This seems right. Counterfactual histories lean on empirical generalizations about the past, which are often difficult to substantiate. Moreover, because they lack the evidential support of ‘traces’ such as archived documents, artifacts, and testimonials, they are harder to justify. Nevertheless, there does not seem to be any in principle reason for dismissing counterfactual narratives. Evidence for counterfactual narratives simply has to come from the actual past -- from empirical generalizations, traces, and inferences from the actual historical record.

Still, even supposing that this is right, two crucial questions remain. One, how do we distinguish between the legitimate counterfactual narratives and the illegitimate? Two, how do commonly prescribed methodological rules get us to legitimate counterfactual narratives in the first place? My aim today is to consider these questions. I point to examples where following what historians have called the ‘minimal rewrite’ rule (MRR) is taken to get historians away from illegitimate narratives. I propose an explanation for historians’ use of MRR, namely, that MRR blocks the construction of counterfactual narratives about ‘non-modular’ causal systems: causal systems to which the methods of counterfactual history do not apply.

To begin, MRR is the requirement that historians (and paleontologists and evolutionary biologists) ought to take as the counterfactual starting points of their narratives only unactualized possibilities implying minimal changes to the states of affairs before them. In other words, it is the rule that historians etc. ought to, as much as possible, avoid rewriting history up to the start of their counterfactual narratives.

Consider, for example, the way in which Bob Richards has criticized Darwin Deleted. Not only does he raise general worries about the project of counterfactual history -- “[r]eal history with real evidence is hard enough” -- he also argues against acceptance of Bowler’s narrative on the grounds that the narrative ignores too much of what we know of the past before and around the time of Darwin. That is, Bowler’s history presupposes that some theory of natural selection resembling Darwin’s would not have arisen in 19th century England had Darwin himself never existed. But according to Richards, it is likely that a theory of natural selection very similar to Darwin’s would have developed through the ingenuity of some other 19th century naturalist like Alfred Russel Wallace. So Bowler’s story purchases its plausibility at the cost of unsubstantiated historical assumptions about the 19th century naturalists.

That is, one way to put Richards’ historical objection to Darwin Deleted is as follows. The starting point of Bowler’s counterfactual narrative is implausible given what we know about Darwin’s contemporaries. It depends on more-than-minimally changing what we know of the actual past. Consequently, Bowler’s story is a bad one. It violates MRR.

Richards’ objection constitutes a case where following MRR rules out a counterfactual narrative because the narrative conflicts with facts about background characters. However, there are other ways in which an historian too might follow MRR. One way is by rejecting a counterfactual narrative because it contradicts well-established empirical generalizations about the actual world. Indeed, it is just on these grounds that Jon Elster has criticized Robert Fogel’s Railroads and American Economic Growth.

According to Fogel, 19th century America’s 1890 gross national product (GNP) would have been only slightly lower had America never developed railways. Elster argues, however, that Fogel’s calculations are based on an implausible assumption: namely, that 19th century America would have lacked a railroad economy in the first place. On Elster’s view, Fogel’s assumption runs in the face of empirical generalizations about technological change, according to which functional equivalents of railways, and thereby a ‘railroad’ economy, would have developed in 19th century America, even in the absence of railways. So Elster employs a different strategy than Richards against Bowler. He undermines the credibility of Fogel’s counterfactual narrative by appealing to accepted generalizations about technological change. By contrast, the latter objects on grounds appealing to accepted historical theories about historical actors.

Considering such examples, it should be clear that, as it stands, MRR is ambiguous. Which states of affairs ought to minimally change remains undecided. Nevertheless, one thing is transparent, namely, good counterfactual narratives do not stray far from what we know of the actual past. Moreover, however we specify it, as far as consensus among historians goes, MRR is an accepted measure of a counterfactual narrative’s warrant for acceptance.

Nevertheless, beyond the consensus among historians, is there a reason to accept the rule? I think there is. Specifically, I’ll argue that the MRR ought to be followed because it redresses a major worry among skeptics: that ‘modularity’ fails in history. Moreover, I suggest that it redresses this worry by guiding historians’ selection of modular causal systems for counterfactual investigation.

To begin, skeptics of counterfactual history worry that counterfactual narratives perpetually lack evidence. Because counterfactual pasts by definition leave no traces, historians in constructing their accounts of the counterfactual past must adduce traces and empirical generalizations about these traces from the actual world. But this method is possibly problematic. After all, what if the counterfactual pasts that historians are interested in describing do not sufficiently resemble the actual past? What then licenses historians to use data and empirical generalizations from the actual world to draw conclusions about counterfactual ones?

The answer, I think, emerges when we reflect on the modularity -- or lack thereof -- of sets of causal generalizations, i.e. causal systems.

‘Modular’ causal systems are systems in which every cause has only one effect, but effects can be multiply caused.[ii] They are ‘epistemically convenient’ because they can be subject to ‘the method of concomitant variation’. That is to say, in modular systems, it is possible to identify causal relationships by varying the value of one variable in a causal system and simply observing how a putative effect variable varies in turn. This is because, in a modular system, one can intervene upon a variable independently: one can hold fixed the rest while varying one variable within the causal system. Thus, one can infer that C causes E when changing the value of variable C changes the value of variable E, since the possibility of confounds does not arise. By contrast, in systems where modularity does fail, the method of concomitant variation is less reliable. We have to worry that C is the cause of both variables D and E, and that the observed change in E is in fact the result of both C and D causally bearing on E. That is:

So the method of concomitant variation is applicable to modular causal systems because the threat of confounds has been dissolved. I can infer the causal relationship between briefly touching a hot stove and burning my hand because in the set of causal generalizations governing hands-touching-stoves, the only effect of touching a hot stove (I assume) is that my hand gets burnt.

One way to put the skeptics’ worry about counterfactual history, then, might be as follows. The past is not amenable to changing the value of one variable and seeing how the value of another variable varies. This is because historical variables are just too interconnected: too many variables are perturbed by changing the value of one. The past comprises only causal systems where modularity fails. So the method of concomitant variation, used in the construction of counterfactual narratives, does not apply. Counterfactual history employs the wrong method for its material.

It would appear that skeptics like Evans, Richards, and Megill assume that failure of modularity is a universal characteristic of historical subjects. Consequently, they are dismissive of counterfactual narratives in toto. But perhaps we should distinguish between causal systems of the past where modularity does fail, and causal systems where it doesn’t. In addition, perhaps failure of modularity comes in degrees. If so, then so long as those historical systems selected for counterfactual analysis are sufficiently modular, the method of concomitant variation might too apply to them in degrees.

In short, the central worry underwriting skepticism of counterfactual history is the belief that modularity fails ubiquitously in history. Failure of modularity rules out counterfactual history by ruling out application of the method of concomitant variation to the past. But one way for historians to avoid this worry is to pledge to restrict their counterfactual narratives to parts of the past in which the failure of modularity comes in smaller degrees.

My task now is to show that MRR constitutes such a pledge. Counterfactual narratives minimally rewriting the past describe more modular causal systems. This is because we can reasonably infer that historical outcomes, the changing of which would require rewriting more of the past, are outcomes produced by trajectories exhibiting more historicity. Historicity, moreover, suggests failure of modularity. So, it seems likely that adherence to MRR avoids outcomes which are the effects of non-modular causal systems. 

This requires a bit of unpacking. To start, by ‘historicity’, I mean what Adrian Currie has argued in chapter 8 of his forthcoming book to be a feature of historical trajectories.[i] On Currie’s account, historicity has both ‘negative’ and ‘positive’ symptoms. On the one hand, its negative symptom is the decrease in the range of plausible alternatives to the actual outcome. For example, some logically conceivable morphologies are nowhere to be found today, such as nine-eyed vertebrates. Many events leading up to the present-day vertebrate phenotype have ‘entrenched’ the phenotype, making most other phenotypes, such as the nine-eyed one, evolutionarily inaccessible from where vertebrate phenotypes are now. So, over time, there has been a winnowing out of plausible alternative vertebrate phenotypes as a consequence of events leading up to the current vertebrate phenotype.

On the other hand, historicity also makes actual outcomes more likely. Currie illustrates this point using sauropod dinosaurs: specifically, by asking how did they get so big?

According to Currie, “sauropods were gigantic because they were the right lineage, in the right place, at the right time.” Their phenotype resulted from a unique set of cumulative evolutionary changes, as well as a unique timing and ordering of these changes¾the phenotype’s evolutionary path. What seems to have been improbable¾sauropod gigantism¾was made probable by many events leading up to the phenotype’s realization. So, a positive symptom of historicity is the increase in likelihood that an outcome is realized as a result of events leading up to it.

With this account of historicity, we can now see why following the MRR helps historians, etc. to avoid historical outcomes embedded in non-modular causal systems. Outcomes brought about by trajectories exhibiting historicity are not capable of change without substantially rewriting the past. It was after all the past¾the events and processes leading up to those outcomes¾that made them probable and alternatives improbable. So historicity explains why changing some outcomes rewrites the past more than changing others.

Moreover, historicity suggests failure of modularity in the outcome’s causal system. Consider that, otherwise, actual outcomes could not have been made nearly inevitable by the events leading up to them: the positive symptom of historicity. In the case of the sauropod’s phenotype, gigantism was the cumulative causal consequence of various developmental and environmental changes, such as the development of a distinct body shape, egg-laying habits, and increased predatory pressure. One could not counterfactually remove the sauropod’s gigantism without perturbing the values of all these variables that came before. Historicity suggests causal interconnectedness, because outcomes made nearly inevitable by their pasts could not have been made nearly inevitable without a large number of causal connections between them and other parts of their causal system -- connections rendering the method of concomitant variation inapplicable.

To be sure, this is no proof that following MRR avoids non-modular causal systems. Nevertheless, it is one hypothesis which, I hope, has sufficient prima facie plausibility to be evaluated on a case-by-case basis. Questions to ask are: (1) Are there concrete examples of historical causal systems in which modularity does not fail? (I suggest yes.) (2) Does historicity imply a non-modular structure of causal chains? (I also suggest yes.) 

To sum up, I have proposed today that global skepticism of counterfactual history is underwritten by the worry that the modularity of causal systems fails in history. It seems plausible that some causal systems of the past are more modular than others. That is, failure of modularity is not a universal characteristic of history. Supposing we accept this claim (incumbent on the discovery of concrete historical examples), then one justification for following the methodological rule MRR is that it facilitates the selection of modular counterfactual alternatives. This is because one can infer that outcomes, the changing of which violates MRR, are produced by trajectories exhibiting historicity, and are in turn the effect of non-modular causal systems. Historicity suggests causal interconnectedness: failure of modularity. Following MRR in counterfactual history is therefore desirable because it helps historians and others to avoid non-modular causal systems, to which the methods of counterfactual history do not apply.


Helen Zhao is an incoming PhD student in philosophy at Columbia University. She recently completed an MPhil in the history and philosophy of science at the University of Cambridge, during which she wrote her dissertation on function-talk and the ENCODE controversy. Her undergraduate degrees are in biophysics and philosophy. In the future, she plans to research topics in the philosophy of social science, philosophy of history, and philosophy of biology.

[i] Currie, Adrian. Rock, Bone & Ruin: an Optimist’s Guide to the Historical Sciences. Forthcoming.

[i] Hawthorn, Geoffrey. Plausible worlds: Possibility and understanding in history and the social sciences. Cambridge University Press, 1993; Lebow, Richard Ned. Forbidden fruit: counterfactuals and international relations. Princeton University Press, 2010; Tetlock, Philip Eyrikson, and Geoffrey Parker. Unmaking the West:" what-if" scenarios that rewrite world history. University of Michigan Press, 2006; Belkin, Aaron, and Philip Tetlock, eds. Counterfactual thought experiments in world politics: Logical, methodological, and psychological perspectives. Princeton University Press, 1996.

[ii] Cartwright, Nancy. "Modularity: It Can-and Generally Does-Fail." Stochastic causality (2001): 65-84.

[i]Love, Alan C., Robert J. Richards, and Peter J. Bowler. "What-if history of science." Metascience 24.1 (2015).

[ii]Evans, R. "What If’ is a waste of time." The Guardian, March 13 (2014); Love et al (2015); Evans, Richard J. Altered Pasts : Counterfactuals in History / Richard J. Evans. 2013. Print. Menahem Stern Jerusalem Lectures.