Power and cause

Abstract

Conceiving power relations as a subset of causal relations can be used to expose the problems of a certain behaviouralist take on causality and develop an interpretivist approach to explanation. The first section of this article shows that a behaviouralist approach ultimately clashes with a relational understanding of power, since the latter requires endogenising values and understandings in an analysis in which several causal paths to the same outcome can exist (equifinality) with radically different implications for attributing power. Power relations can be non-linear, and power dispositional or latent, as well as not translating into influence. The second section draws the consequences of these contradictions by conceptualising causal/social mechanisms for and in an interpretivist framework. Such mechanisms can be part of a wider analysis of contingent processes that answer ‘how possible’ questions. Although interpretivist process-tracing provides explanations without strict regularity, such processes include mechanisms which are transferable to other cases, hence generalisable. Finally, the article establishes a specific discursive mechanism of crisis reduction in foreign policy identity discourses, as developed in the comparative study of the processes that make us understand the unexpected return of geopolitical thought in Europe in the 1990s.

A re-conceptualisation of the nature of causation has been on its way in International Relations (IR).¹ While criticising common notions of cause as used in the social sciences, part of this turn aims to recuperate a version of causation² that is compatible with interpretivist approaches. In this context, this article proposes to introduce an interpretivist take on causal mechanisms in explanations of the social world. It will do so by the detour of discussing the behaviouralist take on power and cause in political science and IR, since the problems encountered in making power relations a subset of causal relations are more generally representative of theoretical problems in our usual approach to causality. The discussion on power and cause provides the rationale for developing a different view on causality.

The argument proceeds in two steps. The first section will show how cause and power have been tied into a package and will explore the contradictions this produces in the context and analysis of the social world. I will argue that scholarly attempts to subsume the concept of power under the more general concept of cause produced a number of tensions. These tensions result from the attempt to combine (1) a behaviouralist account of cause, where the presence or variation of A’s behaviour necessarily brings about/raises the probability of the presence or variation of B’s behaviour under conditions known a priori, something I will call ‘efficient causality’,³ with (2) a relational account of power where power cannot be read out of resources, but is ultimately constituted through the social relations in which it takes place.⁴ There are two related reasons for this tension. First, such relational analysis requires understanding of the specific values and preferences of the actors in the relation. They cannot just be externally assumed since doing so may miss the very role of power in the relations. Second, a relational analysis is characterised by equifinality: the same effect, i.e. B’s changed behaviour can be the result of several different causal pathways from A to B. As usual in social processes with several paths to the outcome, we only have an ex post assessment of multiple processes, possibly cancelling each other out, not ex ante prediction of effect or its probability. It is not a single causal line where conditions and process make a certain outcome more or less likely; there are possible parallel causal paths of different kinds which may come into play without us being able to know ex ante. Hence, such a relational approach will inevitably undermine this idea of efficient causality.

In the second section, I use the previous critique of power as a causal concept as an inspiration to develop a different view on causation. More specifically, I develop an understanding of causal mechanisms and their role in explanations that takes into account the idea of dispositions and the issue of equifinality. Taking cue from the relational character of power as a proxy for the specificities of the social world, this conception of causal/social mechanisms is made explicitly congenial to an interpretivist approach which I then briefly illustrate with an analysis of the revival of geopolitical thought in Europe in the 1990s.

In short, the article aims to show three things: how one can learn from the problems of defining power in terms of cause to improve our understanding of causation in the social world, how one can understand causal mechanisms even if one does not subscribe to positivism and how one can use causation in an interpretivist analysis in IR.

Power as efficient causation in IR

This section explores the promises and contradictions of casting power in terms of a cause. It harnesses the fact that such causal conceptualisations of power are embedded in an explicitly relational understanding of power, too. Using such a relational understanding as a proxy for the social world, the section then shows how the envisaged efficient causality of the behaviouralists (and others) cannot be upheld. Taking the relational aspect of power seriously eventually leads to an analysis which is driven from the receiving side, the normative context and individual beliefs. Such an analysis is ‘internal’, i.e. based on understandings, and can assess influence only ex post. Between the relational and (efficient) causal character of power, one side has to give. I suggest it is causation that needs to be redefined.

The section first develops the attempts to couch power as a sub-concept of cause within a relational understanding of power. It then criticises this conception, showing that it leads to the problems of equifinality that undermine a behaviouralist theory of action. Consequently, power is not conceptualised as a subset of cause, but as a disposition (i.e. ability/capacity).

Power as cause

Power features prominently in realist IR analysis, but not only there. Coming in different guises, political realism has used power in various ways, for instance, as the principal means/resources of action, its accumulation (or even: maximisation) as the main aim of behaviour, and the balance of power as the motive for action and the outcome of interaction – nothing less. In a vision of an anarchical world with no government, power is thus almost all we need to know to understand agent behaviour and world order. But although theoretical critiques of realism have questioned one or all of these assumptions (Aron 1962; Claude 1962, 1989; Haas 1953) and their empirical record (Wohlforth et al. 2007), they have not necessarily abandoned the central role of power in their analysis.

Important redefinitions of power beyond its realist take have tried to keep its central explanatory role. David Baldwin (1985, 1989) revived a Dahlian analysis of power to rescue a relational concept of power for IR, and Joseph Nye’s (1990, 2011) concept of ‘soft’ power similarly keeps a central place for power, yet with a widened or diversified conceptualisation. Scholars in International Political Economy (IPE), realists and others, have also relied on the ‘community power’ debate in political science to propose more ‘structural’ concepts of power (for an overview and critique, see Guzzini 1993). Post-structuralists and constructivists had initially resisted this tendency to confuse the revision of the concept of power with a revised theory and analysis of power (Ashley 1989; Guzzini 1993, 2000) – as if widening and/or qualifying the concept was enough to settle its place in our theories and explanations. But some have now joined the fray (Barnett and Duvall 2005). Power seems to be an explanatory concept nobody in IR can do without.

The reasons for this central place are manifold, but two stand out. First, power is a central concept in many theories of action and all theories of domination. It is connected to what may explain behaviour and used to understand the origins and mechanisms of domination. When interested in ultimate movers, scholars turn to power. Second, in our political discourse, power is connected to the idea of responsibility (Connolly 1974): only those who are able to do something can be held responsible for it. Therefore, there is also a tendency to see power as an indicator for political agency: if there is (a) power, then change is not left to God or nature, but politics can make a difference. Finding the ultimate mover and cause appears as a necessary step to enact change and assign responsibility (accountability). Identifying power is hence at the core of politics and political morals.

This double importance explains why a fresh conceptualisation of power became such an important issue for the behaviouralist study of government in the 1950s and 1960s. After all, behaviouralism was the attempt not just to add ‘political behaviour’ to the study in political science but to redefine the very science of politics.⁵ Major themes were re-visited, their meaning revised. In such a light, behaviouralists could claim that ‘the systematic study of power is very recent’ in the sense of formulating ‘the concept rigorously enough for systematic study’ (Dahl 1957: 201). And there, in the attempt to make it less empirically amorphous, ‘power’ became likened (again) to the idea of cause.⁶

Indeed, there is an almost self-evident sense in which power and cause ‘must be’ related. In his early definition from 1938, Bertrand Russell defined power ‘as the production of intended effects’ (Russell 1938/1960: 18). Power moves and affects things. In Dahl’s famous definition, ‘A has power over B to the extent that he can get B to do something that B would not otherwise do’ (Dahl 1957: 202–203). It is easy to see how this definition could be seen as a special case of cause: A’s behaviour causes B’s behaviour (Simon 1957: 5). And this causal view would become the explicit modus for understanding power in this tradition (Dahl 1968). For Oppenheim (1981), saying that B was influenced by A was no mere description, but also a partial explanation of B’s conduct: actor B did something because of A’s power. Power relations are a subset of causal relations, but qualitatively alike. This view also privileged a view of ‘power to’, having a role in explanations of behaviour at the expense of ‘power over’ or domination, which it claimed to subsume (‘the primitive notion that seems to lie behind all these terms’ [Dahl 1957: 202]).

Dahl and others developed their causal conceptualisations of power to overcome a view of power in terms of mere resources and instruments. Equating power with influence was necessary to retain a central importance of the concept, which mere resources would hardly permit. For as long as there cannot be found really determining outcomes, resources are not necessarily tied to power. Meant to oppose ‘elitist’ theories of politics, the unequal distribution of (economic and other) resources in any polity was a starting point, but no conclusive proof of political power. Resources alone do not tell us ‘who governs’.

And so, when Dahl and other behaviouralists developed a causal concept, and when they tied power to influence, they embedded it in a ‘relational approach’. A relational view of power means that power exists in and through a relation, not as a mere possession of any agent prior to the relation in which it is exercised. We cannot know ‘power’ before knowing the context of the relation and the people meeting/sharing it. This ‘relational’ character should not be confused with ‘relative’. It means something more than that one’s power is only understandable when taking into account the power (or worse: resources) of others as the net amount on a balance-sheet.

In IR, such an approach is well exemplified by David Baldwin, surely the IR scholar who has been most inspired by Dahl’s conception and the community power debate and who systematically tried to improve the debate in IR. His approach is driven by the concern about non-falsifiability which he sees featuring prominently in many studies marred by what he calls the ‘paradox of unrealised power’ (Baldwin 1979: 163). In such an explanation, prominent in the debate about the US defeat in Vietnam, even powerful actors can lose against less powerful ones because they did not, for one reason or another, use all their power. When allegedly overriding power does not translate into influence, it is not because an actor lacked sufficient power, but because of ‘conversion failures’ (lacking ‘political will’, for instance). But, according to Baldwin, such a move ends up making the same outcome explicable by opposite causes (power or powerlessness). And thus, whatever the outcome, it can never undermine the initial power assessment. The value of resources is ultimately objectified and all misfits in terms of influence are explained away via incompetent agency: power resources never fail; only politicians do.

Instead, Baldwin decides to keep the necessary causal link between power base and outcome. His proposed solution has two parts. First, the very assessment of what can count as a power resource or power base (and not just an instrument or vehicle) must be defined and qualified by the particular type of power relation under analysis. As a result, the analysis of power has to carefully delimit the domain within which a policy instrument can be assumed to have effect, that is, to count as an actual power resource in the first place (for stressing this argument, see Baldwin 1985, 1989). If power is causal and central to the explanation, it cannot fail, but the conditions for its presence need to be carefully specified. Second, since the analysis is relational, it needs to understand the values and preferences of the actors in the power relation before we can identify what can count as a power resource or base in this particular relation, and not just in a general domain. Additionally, skill and motivation must be added before we can establish the power base. As a result, much work goes into the specification of social context and contest in which power relations occur. The lesson is: power did not fail; the analyst did not correctly qualify the ‘policy-contingency framework’, as Baldwin called the particular context of the relation, for attributing power in the first place.

This tendency to heavily qualify the understanding of the very power resources or power base is further exacerbated by the need to analyse power relations from the receiving side. Carl Friedrich’s (1963) ‘rule of anticipated reactions’ stays prominent in relational analyses of power. Friedrich showed that it would be too narrow to conceive of power only where A actively did something to influence B, since A can affect B simply by what B supposes A’s preferences to be. This is a bit of a conundrum for a behaviouralist approach, since A may be powerful without doing anything. And yet, this qualification has been accepted, as can be seen in Nagel’s definition, where he starts from A’s preferences: ‘A power relation, actual or potential, is an actual or potential causal relation between the preferences of an actor regarding an outcome and the outcome itself’ (Nagel 1975: 29). Here, ‘the intermediate variable P’s behaviour (influence attempt) is to be replaced by R’s anticipation of P’s future reaction’ (ibid.: 28).

Thus, a relational conception reads power relations through the eyes of the recipient, or, more precisely, looks at all involved actors as potential recipients of power relations. By concentrating on the recipient of power exercises, and imputed preferences as causes, relational approaches also stress the ‘latent power’ of actors (‘having power’) who do not necessarily need to act (‘exercising power’); all it takes is that the recipients adjust their behaviour to pre-empt real or even imagined (negative or positive) sanctions, whether or not the power wielder is even aware of this in every case. The causal link from sanctions to changed behaviour of the recipient is upheld, but the explanation is driven from the recipient of power to its wielder, not the other way round.

Consequently, so the relational analysis would point out, no analysis of power can simply be made without knowing the relative importance of norms and preferences in the mind of the power recipient at least, if not also of the alleged power holder. The capacity to sanction and the resources on which the sanctions are based are a part of the analysis, but for themselves insufficient to attribute power, since what counts as a sanction in the specific power relation is itself dependent on the specific values in the minds of the people involved. Power resources are no proxy for power, since such proxy would end up losing sight of the actual power relation.

The theoretical contradictions between a relational approach and efficient causation

Yet, as this subsection argues, thinking this relational component to its conclusion exposes the difficulties of using power as a subset of efficient cause, i.e. where the presence or variation of A’s behaviour/preferences necessarily brings about the presence or variation of B’s behaviour under conditions known a priori (whereby statements in terms of probability do not change anything in the basic necessitous logic, which is then only moderated by circumstances and scope conditions). As said, power relations are seen just as a subset of causal relations: not all explanation of social interactions is reducible to power relations (although Simon’s and Oppenheim’s definitions do sometimes veer very close to this). Yet, as I will argue, what defines that subset, such as, for instance, the understanding of clashing ‘wills’ and ‘preferences’, is exactly what ends up contradicting a causal understanding of power.⁷ If those conditions are not known a priori, then we return to the conundrum that had attracted Baldwin’s critique, where the analysis can be fixed ex post again. But the problem with that fixing is now more in the attempt to uphold a causal explanatory link for power in the first place. We may all be interested in knowing what made B change behaviour, but, as this article will develop now in more detail, trying to do this with a causal concept of power overburdens the concept, underplays understandings of power different from ‘influence’ and misconstrues the role of causation in the social world.

The main characteristic of a relational approach is that it locates power in a human relationship, where one needs to know not only the expectations, intentions and preferences, but also the beliefs and shared norms of the participant actors in that relationship. The initial inspiration for the relational approach comes from Weber, who defines power as ‘any chance within a social relation to impose one’s will also against the resistance of others, independently of what grounds this chance’ (Weber 1921–1922/1980: 28), but also, and in more relational terms, as the fact when the ‘expressed will (“order/command”) of the dominating actors intends to influence the action of the subordinates and actually affects them in a way such that the latter act[,] as if they had turned the content of the command, for its own sake, into a maxim of their action (“obedience”)’ (Weber ibid.: 544, my translations). Weber was subsequently read either as saving a rationalist (strategic/interactionist) and causal approach for the social sciences, where his ‘chance’ would be translated as ‘probability’⁸, or, by instead stressing his interpretivism, as exposing the limits of such an approach.

Bachrach and Baratz’s famous example (1970: 20–21) may provide a good illustration of what a relational conceptualisation of power entails. Imagine the following situation. A sentry levels his gun at an unarmed intruder, and orders the intruder to halt or else he will shoot. If the intruder stops, it seems the threat has worked, the sentry has exercised power by the gun. Not necessarily so, Bachrach and Baratz say. If the intruder is himself a soldier, he may obey because that is what a soldier does when receiving an order from a sentry. The gun as an alleged power resource was ineffectual here, since the value system of the intruder made him obey, not the gun.⁹ Inversely, if the intruder does not obey, and gets killed, we may again not be seeing a power relationship. Strictly speaking, the killing of the intruder is not power, since the intruder apparently valued entering the base more than his life; the killing only shows the ultimate powerlessness of force (violence) in front of a suicide attack.¹⁰ Pushing the example to its extreme, the intruder may have wanted to commit suicide and gets the sentry to do it on his behalf. In this case, the intruder, by being shot, would have exercised power over the sentry.

The suicide example makes an important point for conceptualising power. In a causal approach, power is understood as influence on B, made visible by B’s changed behaviour. But thinking the relational approach to its conclusion in the suicide example means that, even if A has influence on B’s behaviour by killing B, that very influence may still not constitute A having power in this relation. A quite general causal definition à la Simon and Oppenheim (see above) would simply not work. But even Dahl’s early relational Weberian clause about ‘something B would not otherwise do’ needs to be taken in a strong sense. In a weak sense, A would exercise power, since killing B qualifies for the clause – the suicide candidate may not have had the courage to do it on his/her own. And yet, Bachrach and Baratz’s specification of a relational analysis would attribute power to B, not A. Only if the clause is taken in its stronger Weberian sense, where A was forced to do something against his/her will, is it then indeed B who has power.

Such a stronger version of a relational approach implies, however, that looking at A’s and B’s behaviour is simply insufficient to establish the locus of power in the relation, even in such a striking case where A’s act intentionally causes B’s death. Although this is usually granted by (the early) Dahl, an important implication is not drawn, namely that the analysis of preferences needs to be endogenised. Behaviouralists simply exogenise the actors’ values or preferences in the analysis. In the usual behaviouralist manner, scholars can infer such values and establish actors’ interests by increasing size: although interests may differ from one actor to another, in larger groups such differences wash out. The ‘typical’ actor is then constituted through an exogenously given theory of action, usually those forms of rationalism consistent with behaviouralism. Consequently, one can go the usual way of establishing some co-variation of antecedent conditions and behaviour and then apply these as the inferred interests of actors. Another strategy, closely related, is the idea of ‘revealed preferences’. Here, intentions or interests are simply extrapolated from previous behaviour. In terms of power relations, that implies that previous resistance to the imposition of an external will or the attempt to impose one’s will against such resistance can be seen as sufficient indicator of the actor’s intentions.

But such solutions obviously collide with the requirements of a truly relational approach, as outlined by Bachrach and Baratz. Taking the relational approach seriously undermines these choices, since we cannot know ex ante whether or not both shortcuts are permissive. The idea behind Bachrach and Baratz’s thought experiment is that a situation which may look exactly alike from the outside can stand for the power of the sentinel (although based on different reasons as seen from B) or for the power of the approaching soldier. The same outcome can be understood through different causal paths which, as in this case, can have opposite implications for understanding the power relation. In today’s methodological parlance, this is a case of equifinality, sometimes also called ‘multiple causality’, which requires opening up the ‘black box’ of the interaction between A and B and endogenising the analysis of beliefs, preferences and specific understandings, so as to establish the multiple possible causal paths. And that makes these shortcuts not permissive (see the discussion on equifinality in George and Bennett 2005: 157–62). This argument is similar to the critique of a quite common type of mixed-method research designs. Those designs consist in following up a quantitative correlation by a qualitative case study, usually an outlier case. But, if we have a case sample where equifinality of different causal paths could apply and cannot be excluded a priori, then even all those cases that ‘fitted’ the general correlation would need to be double-checked for the actual causal path in the first place. Any individual case is potentially a spurious correlation in terms of the general explanation usually provided in terms of a Humean constant conjunction.

This leads to a further implication. When equifinality’s implications for the status of causation in a relational power analysis are not sufficiently taken into account, scholars will tend to add factors into the conceptualisation of power to increase causal variance. This is a quite logical consequence of the conceptual shift in relational approaches to the power recipient. A causal analysis, which starts from the outcome, almost necessarily looks for a necessary and sufficient, and not just recipe-like, causality (for this distinction see Riker 1964: 346–47). Rather than focusing on the power wielder’s capacity to get someone else to change behaviour – i.e. the manipulative and agency-oriented view prominent in the underlying relational definition of power – the power analysis strives for a fairly comprehensive view of all factors that influenced the recipient’s behaviour. It is therefore no coincidence that Baldwin ends up adding even social norms to his catalogue of factors which define the specific environment within which power analysis takes place, although his approach is otherwise careful to avoid power as an ‘internal’ relation where the ideas and beliefs generate power in the first place. Pushed to its extreme, such analysis will embed so much into the qualification of the power base that only purely contingent factors, ‘luck’ in Dowding’s (1991, 2003) approach, are left as causes outside of power.

If, however, the relational aspects of the power relation and equifinality are taken seriously, then we cannot conceptualise power relations in a way akin to efficient causation. Let me re-hash the steps. A behaviouralist analysis of power sees power relations as a subset of causal relations in which A’s (imputed) behaviour is the cause of B’s behaviour. Hence, it conceives of power in terms of influence and control (which Dahl and Baldwin use interchangeably). This type of analysis opened up by including the possibility that no action is required for power to be present, and that a relational concept requires our knowledge of the individual values and social norms to distinguish between spurious and causal relations, and also the direction of it (soldier-sentinel example). Yet, that opening does not reflect on the type of causality that would be consistent with such a view of power. Power analysis is about actions which may have non-linear effects (the causal arrows can even revert); it can be about actions which have effects (‘influences’), but where those effects are no indication of power; indeed, it is about relations whose power component can only be assessed ex post through the understandings given to them by the actors involved; and, last but not least, it is about causes that can stay latent.

This can give rise to two types of answers. One answer is to say that power is simply not a causal concept and hence has no business in explanations. When delimiting the domains that give specific meaning to and require particular concepts of power, Peter Morriss argues explicitly against an ‘explanatory’ domain. He writes that power statements ‘summarise observations; they do not explain them’ (Morriss 2002: 44, emphasis in the original). And although actions can be exercises of power and be part of explanations, they cannot stand for power itself, as the behaviouralist setup has it. He therefore reconceptualises power as a mere potential or ability, in short, as a dispositional concept.

But to some extent, this answer begs the question. After all, these abilities are based on basic actions as Morriss calls them. They are related to resources and to outcomes; hence, they still seem to be part of the same causal path as in Dahl. Moreover, although they do not explain, they allow for predictions: ‘[…] all dispositional statements are handy summaries of an object’s properties – useful for prediction but not for providing explanations’ (Morriss 2002: 44). Admittedly, Morriss’ prediction is not more than a disposition with an army of conditionals attached; and in the covering-law model of explanation, there is no qualitative difference between the two. But what if one relaxes that conception of explanation?

Hence, the second answer would probe the relationship between power and cause in a different manner. If power analysis is said not to be explanatory because dispositions are not causes in the efficient sense, maybe we have too stringent an expectation of what an explanation in the social sciences can be. Rather than using causal relations as an inspiration for power analysis, one could use the previous findings about the analysis of power as an invitation to mobilise different conceptualisations of causation.

Power and causation re-visited

The reduction of power relations to a subset of causal relations has some appeal for the social sciences at large, but in particular for scholars in IR accustomed to seeing in power a kind of ‘ultimate mover’ or primary factor which explains why things happened the way they did. Moreover, as suggested above, that logic also began working the other way round: whatever was found to be an important explanatory factor had to be cast into a ‘new’ power concept (soft, productive, etc.). This is connected to the central place power holds in the discourse of political practitioners, pundits and media observers alike. If knowing (or rather, judging) correctly the causes of events is the trademark, and if power, as seen above, is a concept no one can do without, then their conflation will go on unabated. The dual pernicious effect will stay: thinking power in terms of (efficient) cause and thinking all significant causes in terms of an ever-widening concept of power.

The previous section untied the link between power and cause. Conceptualising power as efficient cause, where A’s behaviour/preferences cause(s)/raise(s) the probability of B’s behaviour, clashes with the relational character that requires including the social and interpretive setting of the interaction into the analysis itself. One could leave it at that. But the present section will pursue another possible conclusion. If subsuming power under a concept of (efficient) cause may not work because that view on causation is too demanding for the social world, maybe it is not thinking power and cause together which is the main problem, but the particular take on causality. If, in the previous section, efficient causation turned out to be misleading for understanding power even in its own relational terms, we can still use a re-conceptualisation of power as inspiration to explore different takes on causation.

Hence, this section will explore whether using the pointers of the previous debate on power could help rethink causality. The role of power in the explanation now takes a back seat. Rather, the questions informing this section are: How could a dispositional and relational understanding of power re-frame the idea of cause? What are the implications of equifinality for causal analysis? What could a more interpretivist and social conception of causality look like?

In the first step, I will introduce the idea of dispositions and then develop a non-efficient usage of causal analysis based on the idea of causal mechanisms duly distinguished from ‘intervening variables’. Then, I will illustrate how such a form of causation can actually be used in interpretivist or constructivist analysis.

Causation beyond efficiency: dispositions and causal mechanisms

Peter Morriss has suggested a way to untie the power–cause knot by re-conceptualising power not in causal, but in dispositional terms. He defines power neither as a resource or vehicle, nor as an event or its exercise (as behaviouralists would do), but as a capacity/ability to effect an action (Morriss 2002: 15–19). In such a conceptualisation, the stress is on the possibility of effecting outputs, not the full analysis of what affected outcomes. Dispositions can stay latent when not mobilised. The ability to read can exist even if one is not actively doing it. Yet, if put before a text (in a language one knows), only those able to read can do it.

Therefore, there is a special connection between dispositions and effects. They are neither probabilities nor efficient causes, yet, they play an explanatory role which needs to be assessed during the analysis (i.e. endogenously). Such an approach invites an analysis of process(es) in which such dispositions come into play (or not). This is nothing new for discourse-analytical approaches, which understand the ‘effect’ of discourses as pre-‘disposing’ for certain understandings.¹¹ For instance, when Ted Hopf (2002, 2012) uses discourse analysis in a constructivist foreign policy approach, he shows how identity discourses in the Soviet Union and then Russia define the terms within which the self and other are understood; they ‘pre-dispose’ to certain understandings of the self and other. This understanding is connected to, but does not literally ‘cause’, foreign policy behaviour. In a related manner, Jutta Weldes (1999) analyses the ‘security imaginary’ of US foreign policy discourses, which is the repository of collective memory and (legitimate) ‘lessons of the past’, the cultural ‘raw material’ that is mobilised when actors come to grips with foreign affairs. These representations pre-dispose for understanding the situation in certain foreign policy terms (and hence, for those who wish to analyse behaviour, indirectly for action). Their actual effect depends on being mobilised or not. They have one sure effect, though. Although any foreign policy identity discourse is plural, it is not unlimited: only a certain number of roles and identities resonate with it. As she shows in her study, the security imaginary cannot explain why a certain self-understanding had to become predominant – and hence there is no necessary causal line from the discourse to a single understanding and action. But she can show which possible representations were ruled out, since they are simply not part of the security imaginary. The analysis of dispositions in processes is open at the end, but at the same time shows what is excluded at the start: without certain dispositions, certain paths or processes could not have happened. They made them possible.

In the following, I propose using ‘causal/social mechanisms’ as a way to frame a version of causation that addresses the features of open process, the mobilisation of dispositions and the possibility of several causal paths leading to the same outcome (equifinality). Such a proposal may sound odd at first, since the term ‘causal mechanism’ is used in the correlational literature simply as the link between the independent and dependent variable, that is, exactly in the understanding of causality for the social world which we left behind in the last section. But such a view on causal mechanisms is not necessary. Indeed, in a congenial approach outside the interpretivist tradition, Renate Mayntz (2002) proposes thinking of explanations in terms of a ‘causal reconstruction’ which is characterised by its multi-causality with contingent links, its non-linear process and mechanism character, historicity (taking sequence seriously) and complexity (which is not just stochastic). It is an explanation but without regularity, and yet not just a description since it includes generalisable aspects. She derives the need for such a type of explanation from the specificity of macro-social ontologies. In the following, I join this and similar attempts. More specifically, I suggest a way in which even interpretivism can include a form of explanation, if redefined, by developing social/causal mechanisms in an interpretivist process-tracing that answers the ‘how possible’ questions.¹²

As a starting point, I take Jon Elster’s reference definition. He defines social mechanisms as ‘frequently occurring and easily recognisable causal patterns that are triggered under generally unknown conditions or with indeterminate consequences’ (Elster 1998: 45, 2007: 36).¹³ Like other sociological rationalists, Elster starts his analysis by openly repudiating the covering-law model of explanation and correlational analysis. It is necessary to open the ‘black box’ of why and how a particular outcome was actually reached. In Elster’s prominent approach to the matter, mechanisms are the intermediate level of generalisation available between universal laws, which are unattainable in the social sciences, and descriptions, which are too unambitious. Not being general laws, mechanisms can explain (ex post) why something happened, but cannot be used for predictions, since we cannot know whether a mechanism will be activated or not and/or whether it will always have the same effects (Elster 1998: 45; see also Grosser 1972), nor, finally, which other, possibly also countervailing, mechanisms may come into effect.

To really go beyond the covering-law approach, it is important to distinguish causal mechanisms from ‘intervening variables’, despite such use in prominent literature. Alexander George and George Bennett’s (2005: 225–27) analysis of process-tracing is based on such mechanisms, which they unhelpfully conceive as a series of small-range covering-law explanations (but see the change of mind in Bennett and Checkel 2015: 6–7). This seems to reduce mechanisms – and hence the causal paths – to a sequence of intervening variables. All that process-tracing does is add some intervening conditions to move from an independent to a dependent variable, or from input to output with a mechanism in-between (see the discussion of George and Bennett 2005: Chapter 3). A conception of mechanisms as intervening variables slides back into correlational analysis, now sequentially applied to however many different steps the analyst finds in the process traced (as in Gerring 2007: 172). Correlations are ‘explained’ by more micro correlations (as in Mahoney 2001: 578). The only thing process-tracing or mechanisms change in the analysis is the number of causal links.

In fact, such a reduction of mechanisms to variables does away with their very specificity. First, an intervening variable ‘is added to increase the total variance explained in a multivariate analysis’, but that is different from providing the links in a process (Mayntz 2004: 245). It starts from an assumption of static and additive causes, moreover understood as constant conjunctions, not of sequence and the relational configuration of factors. Second, reducing mechanisms to variables implies that they are ‘not just observable but observed’, which not all mechanisms are (Johnson 2006: 248). Third, at least in some approaches, mechanisms are not attributes of the unit of analysis, as they would be as variables. Instead, ‘[m]echanisms describe the relationships or the actions among the units of analysis or in the cases of study’ (Falleti and Lynch 2009: 1147). Again, reducing mechanisms to variables denies the possibility of a combinatorial or configurational explanation. Hence, a reference to mechanisms à la Elster gets away with the covering-law model of explanation and a conception of causality as regularity, where a certain cause is invariably connected to a certain effect. As Hedström and Ylikovski (2010: 53) write, talking about mechanisms excludes approaches which define ‘causality in terms of regularities (such as Hume’s constant conjunction theory or many probabilistic theories of causation)’.

Once so conceived, it becomes possible to see causal mechanisms as dispositions which can be triggered in an open process that defies an overall regularity (for the following, see Guzzini 2012b: 264). The challenge then consists of thinking causation in terms of what seems to be contradictory ideas, like openness and indeterminacy. Such openness can neither be absolute, because then the word ‘causation’ would make little sense, nor be reducible to probability, because that would fall back into an understanding of causation in terms of regularity. What the idea of causal/social mechanisms in an interpretivist process offers is to de-link the idea of causation from the overall explanation. In other words, mechanisms are but smaller causal but non-efficient components in an overall non-linear explanation.

At the same time, taking the issue of dispositions and equifinality seriously means that causal mechanisms themselves are also not causal in a deterministic sense. There are four main ways of approaching (in)determinacy in mechanism studies. First, the conditions under which mechanisms come into place are left open. In other words, mechanisms are conceived as latent or emerging capacities that need to be triggered by some initial conditions. Those conditions can be varied, and their triggering effect on a mechanism may be contingent. Here, indeterminacy is not in the mechanism, but in its triggering conditions. A second way consists in saying that certain conditions do trigger mechanisms, but they can trigger more than one; hence, the actual mechanism that is triggered in a specific case cannot be predicted. Here, indeterminacy derives from the alternative mechanisms that could respond to a certain input: the environment is underdetermining the response. A third way sees indeterminacy in the mechanism itself, which is said not to have determinate effects, since all depends on its interaction with other mechanisms and/or the process in which it unfolds. A fourth way, suggested by Ned Lebow (2014: 92ff.) in his analysis of ‘inefficient causation’ is ‘the degree of independence of mechanisms from the processes they set in motion and are triggered by’.

Indeterminacy therefore does not imply arbitrariness, but a contextual specification of the process which can only intervene ex post to establish the actual causal path. With this conceptualisation at hand, a last step is needed, since no understanding of causation can be specified independently from the theories and meta-theories in which they are used. The next section will briefly show how such mechanisms can be used in a constructivist framework.

Social mechanisms and process-tracing in constructivism: a brief illustration

In the 1990s after the end of the Cold War, geopolitical thought had experienced a revival in Europe. This is puzzling. The peaceful and unexpected end of the Cold War had undermined all determinist and materialist theories, most notably realism (Allan and Goldmann 1992; Lebow and Risse-Kappen 1995). And yet, even before realists started their response (see e.g. Wohlforth 1994/1995, 1998), geopolitics, the most materialist and determinist theory of it all, became prominent in several European countries. How come?

The research started by exploring a series of factors that could have been conducive to such a rise of geopolitical thought (Guzzini 2003). A realist theorisation would point to the Yugoslav wars and the European return to its nationalist past. Yet, it would not be able to explain why, for understanding the nature of IR, the truly global phenomenon of the peaceful end of the Cold War which had kept a grip over world politics over many decades in IR would be self-evidently considered of minor importance as compared to localised civil wars in the Balkans. A materialist sociology of knowledge would point to the advantages such a set of ideas would convey to some countries (and the Russian revival of geopolitics, from Marx to Mackinder, is mentioned in this context). Intellectual history-inspired analysis often looks at ideational path-dependency and would expect countries with a more materialist history to be more amenable to such a revival. Other theories concentrate on the institutional setup and context within which expert communities develop and reproduce the authorised common sense and dominant opinion. Here, the role of the military in politics could influence the presence and salience of geopolitical thought. Geopolitics can be a useful tool in political struggles, in particular where nationalist themes are prominent. Finally, constructivism would look at the role of identity discourses and see how geopolitics may help to stabilise or re-draw certain self-understandings or role conceptions.

After a first round of country studies, in which four countries have seen a revival (Estonia, Italy, Russia, Turkey) while two did not (the Czech Republic, Germany), the research focused on the constructivist idea of analysing geopolitical representations. It was inspired by scholars in critical geopolitics and constructivist foreign policy analysis who had forwarded the thesis before: ‘It is not surprising that the fading of Cold War antagonisms since the middle of the 1980s has put pressure on these countries to adopt a new, meaningful vision of external relations. The process evokes problems of identity and hesitations in establishing new foreign policy lines’ (Dijkink 1996: 140). It provoked an ‘ontological anxiety’ or insecurity (Agnew 2003: 115). The focus is hence on the analysis of identity processes and discourses in a context of other factors (Guzzini 2012a). In interpretivist manner, factors such as ideational path-dependence, institutional factors (the political economy of the field of expertise) and political struggles are part and parcel of social processes accessible through the meaning shared and given by the agents.

Such an analysis opens the black box of the relation between 1989 and the rise of geopolitical thought to trace the effects of and on meaning systems. Such process-tracing identified a possible causal mechanism which was about the reduction of a foreign policy identity crisis, or as Lupovici calls it, of ‘ontological dissonance’ (Lupovici 2012).¹⁴ A foreign policy identity crisis appears within foreign policy discourses when there is either a tension or a contradiction in the self-understanding of a collective actor or a mismatch between that understanding and the dominant external role perception (as in our study triggered by the way the end of the Cold War was understood). Although its logic is similar to ‘cognitive dissonance’, it is not to be understood as a scaling-up of individual psychological processes; rather, it consists of discursive ascriptions of collective identity.

If triggered, such a mechanism of dissonance-reduction would then effect more concrete responses. There are more than one. They include (1) denial that there is any dissonance to start with; (2) negotiation, that is, attempts to dissuade the other from the faulty identity vision (i.e. there is a perceived dissonance, but this is based on a misunderstanding); and (3–4) the acceptance of a real dissonance, which then spurs either attempts to change the international culture in such a way as to enable one’s own identity to fit (imposition), or efforts to redefine the identity as a way of adapting to external expectations or projections (adaptation). Geopolitical thought is particularly well suited to provide some easy fixed coordinates when one’s role in the world is in flux, since it relies upon environmental determinism from both physical geography (mobilised often through strategic thinking) and human/cultural geography typical of discourses essentialising a nation.¹⁵ Hence, the mechanism is both fundamental for understanding the revival, and yet, as a disposition itself, only indirectly related.

This discursive mechanism has of course been analysed with all but its name before. The US foreign policy during the Cuban Missile Crisis has been analysed as a case of denial (Weldes 1999), as was the UK’s caving into the US identity power politics (Bially Mattern 2005). Post-war and post-Cold War Germany’s politics towards Europe has often been seen as a case of negotiation (see e.g. Risse 2007), the Bush Jr administration’s foreign policy as an example of attempted imposition (Buzan 2004) and Apartheid South Africa as a case of adaptation (Klotz 1995: Part IV).

As an interpretivist mechanism, the trigger is dependent on certain understandings and the pre-dispositions in collective memory and ‘lessons of history’ within security imaginaries (the discourses that form foreign policy traditions). Hence, it is not the event itself, but the way it is incorporated (‘understood’) in the existing discourses which provides the potential trigger. Moreover, if the dispositions of the mechanisms are realised, several paths can ensue – whose number is not indefinite, but can be theoretically delimited. Thinking causality in terms of dispositions and process avoids pure contingency and yet allows for a causal analysis which is not constant conjunction or efficient causality. Such causal mechanisms are part of, but not reducible to a theoretically informed explanation of a wider process that remains open.¹⁶

Hence, providing explanation without regularity (at least in its regular sense in the social sciences), the mechanisms allow for some limited transferability of knowledge beyond the cases. Generalisation is not dependent on regularity. At the same time, the mechanisms are part of contingent processes which play out the sequential and configurational interplay of factors (for a similar take, see Jackson 2011, 2016). Hence they can also be used in interpretivist analysis focusing on the social ontologies of discourses in explanations of the ‘how possible’. All this is part of and constitutes an interpretivist process-tracing.

Conclusion: constitutive theory and interpretivist process-tracing

The present article is built upon the idea that certain ways of conceiving power relations as a subset of causal relations can be used to expose the problems of a certain behaviouralist take on causality. Hence, it harnesses the difficulties that a causal conceptualisation of power encounters with the aim of rethinking causality and causation in IR. The first section showed that a behaviouralist approach ultimately clashes with a relational understanding of power, since the latter requires endogenising values and understandings in an analysis in which several causal paths to the same outcome can exist (equifinality) with radically different implications for attributing power. The second section drew the consequences of these contradictions by conceptualising causal mechanisms for and in an interpretivist framework. Although the very language of ‘triggers’ typical of mechanisms should make interpretivists wary, it is important to stress that they are conceived here in a manner which avoids both determinism and total historical openness (for a similar take of closed/open systems, see Patomäki 2016). They are themselves not determinate in any strong sense of regularity, have multiple and hence open outcomes, and are moreover only part of a wider interpretivist analysis. Such mechanisms can be part of a wider analysis of contingent processes that answer ‘how possible’ questions. Providing explanations without strict regularity, such processes include mechanisms which are transferable to other cases, hence generalisable, but on a far less abstract level. Finally, embedding such mechanisms in constructivist theorising, the article establishes a specific discursive mechanism of crisis reduction in foreign policy identity discourses, which was developed in the comparative study of the processes that make us understand the unexpected return of geopolitical thought in Europe in the 1990s (well before 9/11).

The major implication of this article consists in adding a further item to the pedigree of interpretivist scholarship, or rather, in making such an item more explicit. In other words, it may be important to stress that this view on mechanism-inspired explanations for interpretivism may shed some light on the way to understand constitutive explanations or constitutive theorising (Wendt 1999: 83–89). When Wendt uses constitution, it comes in three different ways which are philosophically connected, but still quite different. For one, constitution is an ontological statement about social kinds. Masters are constituted by their relationship to slaves. Second, constitution can be used as constitutive theory, i.e. the analytical lenses with which we see the world or, rather, find matches for the concepts we already have: ‘seeing’ is active searching, not passive registration. And finally, there exist constitutive explanations which answer the ‘what’ and ‘how possible’ questions.

My insistence on making causal mechanisms available for interpretivism is connected to this third understanding. By amending the underlying conception of causality, interpretivism ‘can do explanation’, but of a different kind. This means that one can have constitutive explanations that stay consistent with constructivist assumptions. The discursive mechanism of identity crisis reduction is not reducing ideas or discourses to an object that can be modelled as an external variable having an effect on another. Intersubjective understandings are the basis of discourse analysis and also of this mechanism. Consequently, rather than seeing mechanisms of all kinds as being part of causal theorising as opposed to constitutive theorising, as Wendt does, the present article suggests that there are some types of mechanisms which are part of an interpretivist form of explanation, close to historical and configurational analysis. Process-tracing can be interpretivist. Whereas Wendt rightly insists on the specificity of explanations when social kinds are involved, not reducible to Humean causal analysis, the present approach sees a similar specificity, but derived from the ontology of open systems, non-regular processes and yet with generalisable elements (Guzzini 2017). Social kinds are important for understanding the constitutive role of ‘what’ questions, while open processes and intersubjective mechanisms address more directly the ‘how possible’ questions. The two are usually combined. As Wendt observes with regard to causal explanations in general, answering constitutive ‘what’ questions is a prerequisite for explanation. Nevertheless, the two components refer to different parts of the explanation. Mechanisms need to be added to the purely constitutive ‘what’ analysis to allow an interpretivist process-tracing of the ‘how possible’ analysis.

Notes

• 1 See, for instance, Kurki (2008), Lebow (2014), Patomäki (1996) and Suganami (2008, 2013), some of whom started this well over 20 years ago.

• 2 In this text, ‘causation’ refers to an explanatory path from cause to effect; ‘causality’ refers to the general conceptualisation of cause and causation. At times, the text may refer to both at the same time, since the second is the condition for the first.

• 3 Hume connected the idea of efficient cause to his view of causation as constant conjunction. See Kurki (2008: 38). For a reassessment of Hume’s more general take on causality, see Suganami (2016).

• 4 ‘Relational’ refers here to a version of interactionism as used by the authors under analysis, where it is the social context of interaction which is constitutive of power. It falls short of a truly relational ontology which, in the domain of power analysis, would be represented by Pierre Bourdieu’s approach, for instance.

• 5 As a representative example, see Dahl’s APSA Presidential Address in 1961 on the Behavioural Approach, in which he approvingly cites David Truman, who posited that political behaviour is no subfield, but a general orientation ‘which aims at stating all the phenomena of government in terms of the observed and observable behaviour of men’. To treat it as a special field in political science would defeat its aim. See Dahl (1961: 767).

• 6 As Terence Ball (1975a) shows, the causal conception of power is not new, but goes back to Hobbes (and Locke and Hume, who take it over from Hobbes). He also shows how behaviouralists follow the same mechanistic conception of causality that is then applied to power.

• 7 Indeed, as Keohane noted about Baldwin’s approach, if ‘we defined each issue as existing within a unique “policy-contingency framework”, no generalisations would be possible’ (Keohane 1986: 187).

• 8 The term ‘chance’ is often rendered by ‘probability’, which, however, as Talcott Parsons notes in his translation, should be stripped of all mathematical or statistical connotation in which that probability could be numerically measured (Weber 1978: 59, editor’s/translator’s note 13) – a caution not followed by the behaviouralist tradition.

• 9 For an early pointer to how this reference to the role of rules invalidates a mechanistic or covering-law view of causality for power analysis as used by behaviouralists, see Ball (1975b: 206–209).

• 10 Although in a more strictly Weberian reading, it is fair to say that the sentry exercised power in imposing his will – not to allow anyone unauthorised to enter the base – against the resistance of the intruder. But that is not particularly relevant even for his approach. Famously, Weber (1921–1922/1980: 28–29) said that the concept of power (of the same kind that inspired Dahl) was amorphous and of little use as compared to his concept of Herrschaft, which thinks power from the reasons for a command to be accepted, i.e. via processes of legitimation.

• 11 I am grateful to one referee for this pointer.

• 12 For a similar aim using process-tracing and mechanisms, see now also Pouliot (2015).

• 13 The following three paragraphs are excerpted from Guzzini (2011: 332–33).

• 14 The analysis is obviously indebted to the literature on ‘ontological security’ in IR, which is rich and developing. For early statements, see Huysmans (1998), Kinnvall (2004), Mitzen (2006) and Steele (2007). But its understanding of the dissonance is not necessarily informed by Giddens’ idea of ontological security, and its analysis is more directly focused on identity discourses as in Jutta Weldes’ (1999) study of security imaginaries, which provided the main reference of the analysis. For a related take on a state’s identity crisis, see Berenskoetter and Giegerich (2010: 424–425) whose analysis focuses on the role of friendship for ontological security and the ‘dissonance’ that (mutual) estrangement can produce.

• 15 Yet, although geopolitical thought fulfils that role very fittingly, there is no necessity for national security discourses or foreign policy elites to resort to it. Assuming otherwise would just exchange the problems of a behaviouralist regularity with a functional fallacy. Indeed, in Sweden whose non-aligned and neutral identity was undermined with the end of the two blocs in Europe, no geopolitical revival ensued. For an analysis of how neutrality is linked to the state-building project in Sweden, see Malmborg (2001).

• 16 For the full development, which also includes an analysis of several parallel processes, and their different speed, rather than the view of a single line, see Guzzini (2012b). The study also develops a second mechanism, called the ‘vicious circle of essentialisation’, which would trigger the self-fulfilling effects when geopolitical thought becomes the generalised blueprint for behaviour: the militarisation of politics and the essentialisation of identities.