In the fabric of human events, one thing leads to another. Every mistake is in a sense the product of all the mistakes that have gone before it, from which fact it derives a sort of a cosmic forgiveness; and at the same time every mistake is in a sense the determinant of all the mistakes of the future, from which it derives a sort of a cosmic unforgiveableness. Our action in the field of foreign policy is cumulative; it merges with a swelling stream of other human happenings and we cannot trace its effects with any exactness once it has entered the fluid substance of history, George Kennan wrote in his classic American Diplomacy (1951). Maybe. Maybe not. “In a sense” sharply qualifies Kennan’s claims. The main problem with such a broad claim as Kennan’s is that it disregards the many different ways in which explanation and causation can be conceptualised.
What caused the First Iraq War? The most common answer to this question will surely be Iraq. Iraq invaded, occupied and annexed Kuwait in the summer of 1990. Technically, the subsequent military action by the US-led coalition turned what may have come to be known as the Kuwait War into the Iraq (or First Gulf) War. This technicality aside, however, is it correct to say that Iraq caused the Iraq War? In one sense, it is highly likely that that US and allies would not have attacked Iraq, had Iraq not annexed Kuwait. In another sense, the US and allies did initiate Operation Desert Storm, thereby transforming the Kuwait War into the Iraq War. The standard argument implies perhaps not so much that the US had no choice but to respond militarily. Rather, if the Iraqi invasion of Kuwait had not happened, the US would not have attacked Iraq and it could not have liberated Kuwait.
Psychologically, the so-called fundamental attribution error may be at work here. Others, and especially our adversaries, do unpleasant things out of an inner disposition (Kennan 1947). By contrast, we rationalise much of our own behaviour by way of a situational logic, especially if we end up doing morally reproachable things, for circumstances give us little or no choice. Psychologically, the fundamental attribution may also underpin the view that Iraq caused (or was responsible) for the Iraq War. After all, the United States was semi-forced into military action by prior Iraqi actions.
Epistemically, who or what caused the Iraq is far less straightforward. One way of identifying the case of the war is to analyse the chain of events that led to the military confrontation. Again, if Iraq had not invaded Kuwait, the US would not have gone to war with Iraq. This counterfactual is difficult to contest. But if one accept this, would it not be equally correct to say that – assuming this was in fact the case – Iraq invaded Kuwait because of any (combination of) the following: low global oil prices; Iraq’s need for revenue to rebuild the economy following the Iran-Iraq War; Kuwaiti refusal to curtail oil output to help increase oil prices. Few people would say that Kuwait’s refusal to cut oil production caused the Iraq War, or at least not in the same sense that Iraq caused it by invading Kuwait and forcing the US to enter the fray. Yet logically, the only difference is that Iraq’s decision to invade Kuwait is temporally closer to the Iraq war in this hypothesised chain of events than Kuwaiti policies or low global oil prices. But following this logic, the US decision to evict Iraq from Kuwait was even closer. After all, had Kuwait been more accommodating towards Iraq, Baghdad might have faced less of an incentive to invade Kuwait in the first place, not least given the significant risks such an action would (and did) entail. It might be argued that Iraq was less constrained and had more options to respond to Kuwaiti policies than the US vis-à-vis Iraq following the invasion of Kuwait. Maybe. Or maybe it is the sense that Iraq violated a norm by invading another country that leads one to focus on Iraqi actions rather than the rule-restoring actions of the US-led coalition. People (and analysts) often focus on agency and tie agency to motives. But why should one not attribute the cause of an event to the incentive that combined with a given motive led to the action that is regarded as the cause of an event? Let’s complicate things further.
Question: What caused WWII? Answer: France’s and Poland’s failure to develop nuclear weapons, for the existence of a credible deterrent, assuming Germany would not have possessed nuclear weapons itself, would surely have dissuaded Germany from attacking Poland and France. Why does this not feel like a good, intuitive explanation? After all, the counterfactual is difficult to dispute. It is highly likely that had France and Poland possessed nuclear weapons, Germany would not have attacked them. Again, part of the answer lies in our inclination to relate effects to agency and identify agency as the cause. It was Germany that invaded Poland (ask Basil Fawlty). Actions are often the focal point of the analysis and causal or explanatory power is attributed to them. We often reason from motives to action and attribute less causal power to circumstances and we are generally less inclined to attribute causal power to omissions and absences. An important change of a background condition may catch our attention, but less so a relatively constant constellation of background conditions, and even less so the absence of a specific background condition. Sherlock Holmes was aware of this (“dog that didn’t bark”). In this instance the counterfactual is not particularly plausible. After all, all other things equal, Germany, given its technical-industrial expertise, would have been far more likely to possess nuclear weapons in the late 1930s than Poland.
We may be slightly more inclined to agree that Germany would not have invaded Poland, had Moscow given Warsaw an iron-clad security guarantee, instead of signing the Molotov-Ribbentrop Pact. While most would agree that the pact “contributed” to Germany’s decision, they would be reluctant to accept that it was the USSR’s failure to provide such a guarantee that caused the German invasion of Poland. But logically it did. In terms of the counterfactual approach to causation, the argument is sound. At the very least, a Soviet security guarantees would have made a German attack on Poland much less likely. And, intuitively, do we not at least occasionally attribute causal power to the absence of a condition? Do we not say that it was the “absence” of petrol (cause) that led the car to come a screeching halt (effect)? Again, a more formal evaluation of what “caused” WWII goes against our propensity that attributes causal powers to agency and agents and motives, less to background conditions and even less so on absent background conditions. Similarly, we tend to regard a change as an effect and look for a cause. The petrol was there. It was only when it disappeared that it had an effect on a car in motion. We typically do not look for a cause when there has not been any change, even if on further reflection we might have had good reasons to expect a change given changes in background conditions that went unnoticed. Paul Schroeder (2004) raises this issue by pointing out that it may be less relevant to ask why WWI broke out in the summer of 1914 than to inquire why previous international crises had not led to great power war in Europe.
The logic underpinning the causal reasoning in these instances goes like this: If agent A with motivation/ reason B faces incentive C, then action D will bring about E. Change C (e.g. nuclear weapons) and neither D nor E will occur. (It may of course occur for different reasons and under different conditions.) Equally, modify A, B or D sufficiently, then E may not happen (again, at least in this instance). Once the causal chain/ constellation has materialised, we are more inclined to assign causal power to D and, by extension, to A or B (or both). This is more intuitive because C is a background condition and will be less noticed, the more constant it is. But it is fairly incontrovertible that a change to any one of the factors A-D may would have prevented E. Once more, we tend to attribute others’ behaviour to dispositional rather than situational factors. Logically, however, there seems to be little difference between attributing the cause to the agent or the background condition. It is the constellation A-C or A-D that leads to E, not any single factor on its own. The likely reason we find this counter-intuitive is that we tend to think of action D as un-pre-determined and of background conditions as brute facts. Here is another way of looking at this.
An INUS cause is an “insufficient but necessary part of a condition which is itself unnecessary but sufficient” (Mackie). This is a way to conceptualise more complex causal conditions and constellations. (Once the ontological issue is settled, the epistemological issue rears its head. More on this issue below.) Let’s take WWI as an example. A number of important background conditions were in place. For example, without any claims to historical accuracy, Russia’s rise, German concerns about encirclement and desire for a preventive war, Austro-Hungary’s decline, a rigid alliance system, the Schlieffen plan created a constellation where, it turns out, the assassination of the archduke was an INUS cause. It is not too much of a stretch to think that several of the conditions just mentioned can also be construed as INUS causes. Moving on.
What caused WWI and what causes wars are very different questions. The first question asks what caused a specific event (token), the second question asks what causes a class of events (type). In the first case, one seeks to an explanation of how and generally why a specific war broke out (please note semantics!), while in the second case one wants to know what leads to armed conflict in general. Roughly, this is what tends to distinguish IR scholars (at least of a certain ilk) and historians.
International Relations scholars typically strive for generality and middle-range (sometimes grand) theories to explain international politics. Historians tend to focus more on the particular and seek to understand how and why a certain outcome did come about in a specific instance (aka single case). IR scholars strive for generalisation, while historians are quite content to come to a thorough understanding of a single case. IR scholars seek to explain patterns of events, historians to understand singular events in their detailed complexity. IR scholars incline towards a variable-oriented and historians towards a case-oriented approach. IR scholars incline towards abstraction and theory, historians typically construct coherent, detailed narratives. IR scholars seek to discover the causes of phenomena, historians lean more towards understanding the reasons motivating actions leading to specific outcomes. Easier-to-account-for-and-quantify material conditions often feature more prominently in explanations by IR scholars, while less tangible and less quantifiable, more actor-centred reasons, beliefs and desires feature more prominently in historical scholarship. At risk of over-simplification, IR scholarship has epistemic aspirations akin to the natural sciences, while historiography is more attuned to interpretation, meaning and hermeneutics, and hence the “humanities”. Both IR scholars and historians offer accounts of what and typically why something happens and/ or happened.
IR scholars often talk about causes, while historians more modestly refer to the origins of an event (Joll & Martell 2006, Mulligan 2010, Taylor 1961). A pre-occupation with the causes of war, and particular World War I, gave birth to International Relations as a distinct academic discipline (Blainey 1988, Levy & Thompson 2009, Suganami 1996). Kenneth Waltz says it is the anarchical nature of the international state system that causes wars (1959, 1979). Thucydides, a historian, says war comes about because human beings are driven by fear, profit and honour. Historians, possibly purposefully, often talk about the origins rather than the causes of war. This makes good sense. “Origin” suggests that only a combination of factor can lead one to understand why an event occurred. “Cause” generally implies something stronger and an outcome that is less inevitable than something that is attributed to origins. Unfortunately, rarely do IR scholar or historians specify what causality is (ontology) and how one can recognise causes when one sees them (epistemology). Nor are they always very transparent about what makes for a good explanation.
To the extent that historiography seeks to explain rather than simply describe or understand (make sense, interpret), historical explanations need to rely on inferences about the causes of specifics, singular events (Mahoney et al 2009). Historiography typically does not aim for generalisation or generalisability. Historical research seeks to understand the ‘causes-of-effects’, that is, it explains and/ or seeks to understand how and why something has happened, the outcome is already known (Goertz & Mahoney 2012). Often historical reasoning proceeds in terms of necessary and sufficient causes, explicitly or implicitly, compared to ‘effects-of-causes’ approach that is probabilistic and is often geared as much to prediction as it is to an understanding of the past. Aside from quantitative history, both micro and macro history typically rely on the causes-of-effects approach.
Positing necessary and sufficient causes is not difficult. Making the case that a cause was, in a particular case, necessary or sufficient (or both) is. First, absent the necessary cause Cn, an outcome E cannot occur. Many necessary causes are trivial (e.g. gravity). However, the more frequently the factor Cn is present only when E is, the less trivial Cn becomes and the closer it is to the threshold of being sufficient cause. Second, a sufficient cause Cs is trivial if it never occurs, but it becomes less trivial and approaches the threshold of also being a necessary cause to the extent that it is the exclusive that produces E. The third logical possibility is a cause that is both necessary and sufficient. Nice if one can find them. Fourth, the so-called INUS cause is epistemically more interesting. An INUS cause refers to a factor that is an insufficient but necessary part of an unnecessary but sufficient condition. An INUS cause becomes more relevant as it becomes a sufficient cause (this can only be established quantitatively). Last but not least, there is a SUIN cause, that is, a sufficient but unnecessary part of condition that is insufficient but necessary. That’s all very well.
The problem however is how one can know or be reasonably certain that a factor is a necessary or sufficient condition, or even just a cause in a particular instance. Mill’s method has significant, well-known shortcomings as a tool to establish necessary and sufficient conditions. Equifinality, multi-causality and causal heterogeneity sharply limit the usefulness of Mill’s method. An effect E can have more than one cause. Multi-causality (number of causes for each instance of an effect) and causal heterogeneity (dissimilar causes across different instances of an effect) tend to make the method impractical-to-inadequate. It is easy to why for historians and social scientists dealing with a single case or unique event, the method is irrelevant.
Social-science-oriented historians have proposed a method of sequence elaboration (Mahoney et al. 2009) and process tracing (Bennett & Checkel 2015). Historical research “explains by tracing the sequence of events that brought them about” (Clayton Robert 1996). This is related to the notion of a narrative that itself is based on the notion of coherence, progression and empirical evidence. Establishing causal chains of sequences relies on a combination of plausibility and evidence and can, of course, easily subjected to criticism (counterfactuals, interpretation). Moreover, if each step in a causal chain is though to be probabilistic rather than deterministic, the link between a distant cause and an effect may be weak. But this is of course a problem inherent in all probabilistic explanation. Cleopatra’s nose may have had a significant impact on world history (by way of how attractive it made her in the eyes of various Roman powerbrokers). But it is difficult to demonstrate this convincingly, not least because the probabilistic nature of the various parts of the historical-causal chain. This is why Pascal actually used this example to demonstrate the role of change in history rather than deterministic causation. Last but not least, even a good causal explanation at every important juncture of a causal chain will often require prior hypotheses (theory) and the extent to which it in provides a good explanation may be debatable due to different interpretations of the evidence (evidence). But the method of sequence elaboration should be credited for forcing analysts and historians to be more forthcoming about what are often implicit assumptions.
What caused WWII? Germany’s drive towards world domination? A plausible cause, but was that what actually drove German expansionism before and during WWII (Simms 2020)? And even if the drive to world domination did in fact form part of the beliefs of important members of the German political leadership, can it be shown to have been a necessary or sufficient cause or an INUS condition of the re-militarisation of the Rhineland, the takeover of Austria and the Sudetenland, the destruction of Czechoslovakia and Poland, the invasion of the USSR. Even if it can be shown with the help of documentary evidence that such an aspiration was held by several influential policymakers, can it really be said to have been the cause of German expansionism? The imputed cause/ motivation needs to have existed and it needs to have been causally operative at critical decision points and junctures. Again, process tracing creates greater transparency, but practical challenges remain. At least, it spells the logic of the proposed argument and is explicit about the evidence required to support it.
What actually is an explanation? The Cambridge Dictionary defines explanation as “the details or reasons that someone gives to make something clear or easy to understand”. In other words, an explanation gives rise to understanding. Makes intuitive sense, but things are substantially more complicated than that. A mistaken explanation can leave one with a feeling of understanding, even though technically one does understand. Understanding something after an explanation is given leaves one in a psychological state of understanding. But this does not mean that one finds oneself on firm epistemological ground, too. Reducing the unfamiliar to something familiar evokes the experience of understanding. In order for an explanation to be epistemologically sound and therefore to understand in an epistemic sense, the explanatory account must also be true. This distinguishes a common-sense notion of explanation from its epistemological cousin.
Philosophers of science generally distinguish between five models of scientific explanations: (1) deductive-nomological model, (2) statistical-relevance model, (3) causal model, (4) unification, (5) pragmatics. The D-M and S-L models explain B by demonstrating that it is the logical consequences of a covering law (or laws) and some initial condition. Explaining an event is to say it had to happen or was likely to happen, not why it happened. One comes to “understand” an event when one can predict it based on empirical or statistical laws and initial conditions. Explaining (and understanding) are about nomic expectability. The rejection of causal accounts of explanation of course goes back to Hume’s arguments that causes are metaphysical entities for they can never be observed directly. At best we observe regularity. Due to their arguably lesser relevance for social sciences, we’ll ignore (4) and (5). So let’s move on to causal explanations.
And what actually is a cause? A cause is something that brings about something else. A cause offers an answer to a why question rather than merely to a what question. This leaves much unanswered. What does legitimately count as a cause and an effect (e.g. entities, structures, processes)? Are causes material entities (often quantifiable or codable) or can they also be immaterial ones (meaning, perception, reason, desire)? Do causal claims need to be generalisable? Causal realism posits that causal relations are out there in the world and one thing objectively causes another rather than causal relations being nothing beyond laws or regularities (causal reductivism or skepticism). This allows for the possibility that unique causes or causal conditions can be the causes of effects. A historian may claim that the cause of the Punic and Silesian Wars was, respectively, Hannibal’s and Frederick the Great’s decision to go to war. Here the cause does not imply any regularity whatsoever. Yet, causal realism claims, the decisions brought about the wars.
Causal models of explanation posit causes, causes being defined as something that brings about something else (and thereby explains the thing in question). Causal realism is not the only approach to causation. Different types of causal models exist, including (1) physical connection (or process tracing in the social sciences), (2) regularity view of causation, (3) counterfactual conception of causation (4) statistical causation and (5) manipulation. It is easy to see why IR scholarship, as defined above, inclines towards (2) and (4), while historians incline towards (1) and maybe (3). Put succinctly, the mechanistic approach relies on underlying (observable or unobservable) causal relations. The regularity model relies on observable, empirical regularities. The counterfactual model relies on unobservable, counterfactual conditions. The probabilistic model relies on changes in (empirically derived) conditional probabilities. Last but not least, the manipulability approach relies on (observable) invariance under (effected) intervention. All of them suffer from significant problems.
First, the physical connection, mechanical or process-tracing model can be used in a variety of ways. It is relatively uninteresting to claim that WWII or the Punic Wars started because Hitler or Hannibal gave the orders and the orders were then disseminated throughout the military bureaucracy. This seems to give an answer to a how question and maybe a why question as far as why the armies started moving is concerned. But process tracing also has the potential to provide an answer to a why question. For instance, the assassination of the archduke may have triggered a chain of events that led to the German attack on Belgium. Process tracing relies more on a logic informing a chain of events and decisions rather than a physical or social mechanism (e.g. bureaucracy) as such. This approach is very much related to the method of sequence elaboration. Given the difficulties alluded to above, little wonder that both historians and social scientists continue to disagree as what caused – or at least was primarily responsible for – WWI (Fischer 1961, Copeland 2000, Clark 2012, McMeekin 2012).
Second, the regularity view of causation simply states if A then B without implying necessity or any mechanism or process. Regularity theorists in the Humean tradition simply view this as an empirical fact. They tend to be causal sceptics (epistemologically) as well as causal reductionists (ontologically), meaning they believe that we simply cannot establish a necessary relation between A and B and are therefore forced to accept the reduction of causality to “co-variation”. This conception of causality typically posits precedence and contiguity as a necessary condition for A to be the cause of B – causes being defined in terms of regularity, not necessity. In the social sciences, this allows analysts to disregard causal mechanisms. All there is and all there can ever be established is a regularity relation. It only ever answers (or can answer) a ‘what’, not a ‘why’ question – a ‘why’ answer being an answer that relies too heavily on metaphysics.
Thirdly, the counterfactual conception of causation posits that if A had not happened, then B would not have happened, then A is the cause of B. “We think of a cause as something that makes a difference, and the difference it makes must be a difference from what would have happened without it” (Lewis). Randomised control studies, indirectly, allows one to establish a real-world counterfactual (manipulability concept of causation). In the case of historically relatively unique events, this is not an option (Ned Lebow). As we have seen, a lot depends on the plausibility of the counterfactual. For example, the archduke’s driver is said to have taken a wrong allowing the assassin to kill the archduke. This sounds plausible. As we have shown, most people would be reluctant to say the wrong turn was the cause of WWI. But on a counterfactual conception of causality, it was, for had the driver not taken this turn and had the archduke therefore not been killed during his visit to the Balkans, WWI would presumably not have broken out when it did.
Fourth, statistical causation posits that A makes B more (or less) likely. With the social world less deterministic than parts of the natural word, propensities and likelihoods. A rise in interest rates will lead to less on investment (than would otherwise have occurred). A state’s quest for hegemony will encounter a counter-balancing state or alliance of state. But sometimes states prefer to bandwagon rather than balance. Place in a broader context/ reference category (“reference class problem”). A (statistical) model may propose and find evidence for a correlation between This was Friedman’s argument saying that theory A is superior to theory B if, on average, it generates more accurate predictions. So mini-skirts and recessions – causal mechanism (physical or logic/ process tracing in social sciences). Country A will respond to country B imposing tariffs by retaliation – under a specific set of conditions (control). Difficulties arise when this approach is used to predict a specific event (“reference class problem”).
Fifth, the manipulation approach to causation is related to the counterfactual conception of causality. The central idea here is to change a condition and what effect such a change has. Large-N randomised control studies are based on this logic by creating the closest thing possible to real-world counterfactual. This is closely related to counterfactual causality because if changing one condition does an effect, it can be said to be the cause of the effect (all other things equal, of course). The problem is that randomised control is not an option when it comes to the past and/ or unique historical events. Too many things are simply different in order for the intellectual manipulability approach to work in explaining world wars, even if one were to include all those cases where world wars did not break out. As a tool to inform policy, it is also more than questionable, as it runs into the problem of reflexivity (Jaeger 2020).
A theory needs to be logically consistent, interesting and (partially) supported by empirical evidence (Walt 2001). Theories often describe, explain and/ or predict (Singer 1961). A theory lays out the reasons, causes, mechanics or processes and/ or relations among entities, events, phenomena, and a theory has something to say about what and (depending on how it is defined) why something has happened, happens or will happen. Theories allow for the generation of observable predictions that must be falsifiable, even if (pace Popper) the aim of (social) science is not falsification as such. A plausible hypothesis derived from a logically consistent theory may allow for correct predictions, yet mis-identify the actual reasons, causes, mechanism that bring about the predicted outcome. This is precisely why, where possible, analysing a question by using different approaches to explanation and causality is desirable. A nice example is to do with the understanding and explaining of French peasant marriage patterns (Nusser 1986). Quantitative approaches can establish the relationship between harvests and marriage patterns. A good harvest is a good predictor of an increased number of marriages. This would provide an explanation along the covering law model (S-L) and the model of statistical causation. Analysing individual cases on the basis of a causal approach to explanation as well as counterfactual conception of causation and/ or a process conception, perhaps relying on an understanding of how French peasants rationalise their own behaviour under various circumstances, would strengthen our confidence that the explanation allows to come to an epistemically well-founded understanding of the French rural marriage patterns. It would also leverage the Weberian Erklären vs Verstehen dichotomy to do so.
All of this is to suggest that analysts should be more transparent about how their research conceptualises epistemologically crucial terms like explanation and causality. Different disciplines and different schools within a discipline often rely on different conceptions of explanation and causality. Admittedly, often conceptions and related methodologies are determined by the nature of the issue under consideration. But in many cases, a more self-conscious reflection about the epistemic foundation of research would be desirable. Even where this is impractical or unnecessary, research should be maximally transparent, explicit and upfront about what conception of explanation and, if applicable, causation is used – and why.