Individualism

The idea that the individual should be the ultimate unit of social analysis goes back at least to social-contract theorists such as Hobbes and Locke and featured in J. S. Mill's claim that the laws of society must be derived from the laws of individual psychology (Mill Reference Mill1974 [1843]; Udehn Reference Udehn2001, 43–49). Weber famously criticized the use of collective notions such as “Volk,” “class,” and “society” in social explanations and argued that we can achieve understanding (“Verstehen”) only by interpreting the intentional actions of individuals:Footnote ¹

This view, which Schumpeter (Reference Schumpeter1908) called “methodological individualism,”Footnote ² was echoed by Hayek and Popper, who held that “abstractions” such as “society,” “economic system,”, and “capitalism” should be avoided in social explanations (Hayek Reference Hayek1942, 286; Popper Reference Popper1944, Reference Popper1945a, Reference Popper1945b).Footnote ³ On this view, “explanations of large-scale social phenomena (say, inflation) in terms of other large-scale phenomena (say, full employment)” were at best “half-way” explanations but never “rock-bottom” ones (Watkins Reference Watkins1957, 106; cf. Heath Reference Heath and Zalta2011; Lukes Reference Lukes1968; Udehn Reference Udehn2001, chap. 7):

More recently, Elster (Reference Elster1985) defended “the doctrine that all social phenomena. . .are in principle explicable in ways that only involve individuals—their properties, their goals, their beliefs and their actions” (p. 5). He rejects Marxist functionalist tendencies to make claims such as “the capitalist class keeps unemployment high to suppress the working class,” which ascribe intentions to collectives without recognizing that these are not purposive, intentional actors, and that even if they were, this would require further explanation (Heath Reference Heath and Zalta2011). Individualistic explanations, Elster (Reference Elster1985, 5) argues, “reduce the time-span between explanans and explanandum” and thereby the risk of mistaking spurious correlations for explanations. Ontologically, “the elementary unit of social life is the individual human action” (Elster Reference Elster1989, 13). As can be seen, methodological individualism has been endorsed from a combination of ontological and explanatory perspectives (for recent reviews, see Udehn Reference Udehn2002 and Hodgson Reference Hodgson2007).

Holism

The term “holism” was coined by the controversial South African politician and part-time academic Jan Smuts in his Reference Smuts1926 book Holism and Evolution. The association with Smuts may have given the term a bad repute. He notoriously supported South Africa's racial segregation (but also co-authored the preamble of the UN Charter, corresponded with Einstein, and advocated humanitarian values abroad). Nowadays the term is often used by its critics as a label for implausible metaphysical claims. We nevertheless use “holism” to refer to the opposite of individualism, without any negative connotation intended.

Nonindividualistic approaches to the social sciences can be traced back at least to Durkheim (1982 [1901]), who introduced the notion of a “social fact”:

Social facts, for Durkheim, go beyond their “individual manifestations”:

Similarly, Durkheim suggests, a society can have properties that are irreducible to the properties of individuals. This kind of holism about properties, however, is consistent with the claim that “society is made up of nothing except individuals”; we return to this observation later. How exactly individual agency and social constraints relate to each other is often called the “structure-agency” question. Parsons (e.g., Reference Parsons1937) sought a middle ground between structural and individualistic explanations, interpreting social structures as systems of norms that can serve as reasons for individual actions, but still considering social structures causally relevant.

In recent research, we find at least three motivations for nonindividualistic explanations. First, there has been a “systemic” or “institutional turn” in several areas of political science and related fields. Properties of “systems” or “institutions” are seen by many not only as explananda (or dependent variables, which are to be explained) but also as explanantia (or independent variables, which play some explanatory role). Theorists of democracy, for example, have turned their attention to large-scale “deliberative democratic systems,” subsuming “the norms, practices, and institutions of democracy” (Mansbridge et al. Reference Mansbridge, Bohman, Chambers, Christiano, Fung, Parkinson, Thompson, Warren, Parkinson and Mansbridge2012) and spanning across a range of domains, from local to global (see also Dryzek Reference Dryzek2010). “Social-ecological systems” are the units of analysis in recent social-scientific studies of sustainability (e.g., Ostrom Reference Ostrom2009), and institutions are sometimes seen as the units of selection in social evolutionary processes, though it remains controversial whether such explanations can ultimately be reduced to an individual level.Footnote ⁴

A second motivation for holism comes from a re-awakened interest in social ontology (e.g., Bratman Reference Bratman1999; Gilbert Reference Gilbert2006; List and Pettit Reference List and Pettit2011; Pettit Reference Pettit and Schmitt2003; Tollefsen Reference Tollefsen2002; Tuomela Reference Tuomela2007, Reference Tuomela and Dieks2010). Although most scholars accept that phenomena such as collective intentions, social norms, conventions, institutions, and group agency are ultimately brought about by individual attitudes and actions, many doubt that these phenomena are fully explicable in terms of individual attitudes and actions alone. A statement such as “The Conservatives want to come across as a party that cares about the environment” need not be straightforwardly reducible to a set of statements about the attitudes of individual party members. The collective intention may exist even though not all, or a majority, of them have a matching individual intention. The party may be committed to certain views because of earlier commitments, irrespective of its members’ attitudes right now (List and Pettit Reference List and Pettit2011). Similarly, our acceptance of printed slips of paper as money, with all the conventions that go along with this, may be hard to explain without reference to joint intentions or social ontology (e.g., Searle Reference Searle1995, Reference Searle2010).

A third motivation for holism is the problem of complexity (e.g., Jervis Reference Jervis1997). Holists maintain that a reduction of social explanations to the individual level, while perhaps logically possible, is not possible in practice, if this requires that “our current theories could be extended or replaced with well-developed ones that allowed for reduction” (Kincaid Reference Kincaid1986, 494). Holists may grant that all social properties are ultimately determined by underlying individual properties. But this does not imply that a reduction is feasible, especially because the same social property can be instantiated by many different constellations of individual properties: the problem of “multiple realizability” (Fodor Reference Fodor1974; Putnam Reference Putnam, Capitan and Merrill1967). The fact that “Japan has a high savings rate” may be explanatorily significant for certain phenomena, yet it may be realized in so many different ways that a reduction is not feasible (Little Reference Little1991, 190–5). Think of all the different ways of holding money, the different social and legal practices related to it, its different possible distributions in the population, and so on. Many if not most social properties are multiply realizable. Similarly, complex social systems, such as the social networks in which political opinions form and spread, may have to be analyzed in terms of aggregate properties. Attempts to understand them by reference to individuals alone, without appeal to their systemic properties, may be hopeless (see, e.g., Epstein Reference Epstein2006).

A Thesis about Facts

Scientific theories usually refer to, and offer explanations of, certain domains of facts. As already anticipated, different theories employ different concepts and categories to describe the facts they are concerned with.Footnote ⁷ That a hydrogen atom contains only one proton and that human beings are capable of linguistic communication are examples of physical and psychological facts, respectively. The most basic physicalist thesis is as follows:

Supervenience physicalism is the mark of most scientifically grounded worldviews: though described in nonphysical terms, psychological facts are fully determined by more fundamental physical facts.Footnote ⁸ It is hard to imagine a world in which all physical particles, forces, and so on are configured in exactly the same way as they are in our world—including all the particles making up all living creatures and their environments—but where some psychological features are different. Supervenience physicalism becomes more or less demanding depending on how narrowly or broadly we demarcate the realm of facts that count as “physical.”Footnote ⁹ The social counterpart, as already foreshadowed, is as follows:

Examples of individual-level facts are that the individuals in the U.S. acted in certain ways on November 6, 2012, and that some went to the polling station. Examples of social facts are that Obama was re-elected President and that voting turnout decreases as the laws make voter registration harder, other things being equal. In analogy with supervenience physicalism, supervenience individualism can be more or less demanding, depending on what we take the individual-level facts to be. On a narrow definition, these may be confined to facts about the attitudes and behavior of each individual in isolation. On a broader definition they may also include facts about relationships between individuals as well as the state of their environment. Different purportedly individualistic theories can differ in how restrictive or permissive their descriptive concepts and categories are.

The appeal of supervenience individualism should be clear: unless we employ an unreasonably restrictive criterion of what counts as an individual-level fact, it is hard to imagine a hypothetical duplicate of our world in which all individual-level facts remain unchanged but which nonetheless differs in some social facts. Indeed, it is hard to see how there could be differences at the social level that could not be traced to some differences, however subtle, at the individual level. Accordingly, supervenience individualism seems to be an important “no mystery” constraint on social facts (see also List and Pettit Reference List and Pettit2011). The negation of supervenience individualism would be a very radical—and we think implausible—form of “social-fact holism”:

A Thesis about Particular Objects

A second physicalist thesis concerns particular objects. Any scientific theory has some ontological commitments: it takes some kinds of objects as “ontologically primitive.” Particle physics has electrons, neutrons, and protons in its ontology, for instance, while psychology has an ontology of mental states, beliefs, intentions, and so on. Furthermore, the objects in a theory's ontology need not be restricted to entities in the ordinary sense, but may also include events or processes. Mental events or processes are objects in the ontology of cognitive science, just as historical events or processes are objects in the ontology of history or political science. The physicalist thesis about particular objects is as follows:

The idea is that every entity, event, or process in the ontology of psychology—say, a particular intentional attitude, perception, or thought process—coincides with some physical entity, event, or process, notwithstanding its conventional description in psychological terms. In principle, one could accept token physicalism with respect to some kinds of objects, say events, while rejecting it with respect to others, say entities or processes. The popular research program of trying to find neural correlates for all sorts of mental phenomena reflects a token-physicalist view. Here, brain-imaging techniques are used to identify the brain state or neural process with which any particular mental state or process—for instance, a particular experience of pain or visual perception—coincides. The social-scientific counterpart of token physicalism is as follows:

For this to be true, every particular entity, event, or process that we conventionally describe as “social” must admit a redescription in individual-level terms. Thus any entity such as a state, political party, or university must be identical to a specific collection or configuration of individuals, and any social event or process such as the 2008 U.S. presidential election or the 2011 E.U. bailout of Greece must be identical to some event or process that can be fully described in individual-level terms. As with token physicalism, one could accept token individualism with respect to some kinds of objects, such as events, without accepting it with respect to others, such as entities or processes.

Whether token individualism with respect to any such objects is defensible depends on how narrowly or broadly these are construed. If we take a snapshot of a university or nation at a given point in time, then perhaps it is not absurd (though still demanding) to identify it with some collection or configuration of individuals at that time. Similarly, if we construe a particular bailout as a specific sequence of individual actions, then the token-individualist stance may seem theoretically viable.

But once we view a social entity as something that can persist over time, or a social event or process as something that could have occurred under different circumstances, any such identification becomes a stretch. An indefinite number of individual-level constellations might give rise to what counts as the same social entity, event, or process.Footnote ¹⁰ An entity such as Facebook clearly occurs in our social ontology, yet, it seems practically impossible to pick out any entity described at the individual level with which it exactly coincides. If this observation is correct, it may support a form of “token holism”:

A Thesis about Properties

A third and even more demanding physicalist thesis concerns properties. Scientific theories refer not only to particular objects, such as particular entities, events, or processes, but also to general properties of those objects. Property ascriptions play important roles in explanations, as discussed below. Examples of physical properties are “having an electrical charge” or “having a certain momentum,” while examples of psychological properties are “believing or desiring something,” “experiencing fear or hope,” “perceiving an object,” and so on. Physicalism about properties is the following thesis.

This is a demanding claim. General properties can usually be instantiated in myriad ways, and it may be impossible to spell out in physical terms what, for example, all the different brain states that may encode the belief that Kennedy was assassinated in 1963 have in common. The social-scientific counterpart of type physicalism is as follows:

A paradigmatic example of a social property may be “owning 20 dollars.” To see how demanding it is to claim that this property is identical to some individual-level property, consider all the different ways in which someone may instantiate it. One may have certain slips of paper or coins in one's pocket or somewhere else. One may have a bank account, which, in turn, can be instantiated in many different ways (in the past, it was registered in a physical book, while today it is stored in a computer system). More generally, one may stand in a particular contractual relationship with others, which can again take innumerably many different forms. Even if every token instance of the property of owning 20 dollars could be translated into a complex individual-level configuration, such a translation is bound to fail when it comes to the general property itself (for detailed arguments, see Fodor Reference Fodor1974 and Sawyer Reference Sawyer2002). This supports the following:

A Thesis about Causal Explanations

A final physicalist thesis whose social counterpart we wish to consider concerns causal explanations:

The idea is that genuine causal relationships, which “rock-bottom explanations” should identify, exist only at the physical level. Apparent higher-level causal relations, in this picture, are at most “epiphenomenal.” For example, when someone forms the intention to lift his or her arm and then acts on this intention, the cause of the resulting bodily movement is not the person's intention (a higher-level, mental state) but rather his or her brain state (a lower-level, physical state). The so-called “exclusion argument,” to which we return in the next section, asserts that, in a causally closed physical world, all genuine causal relationships are physical ones. The corresponding social-scientific thesis is as follows:

An implication of this thesis is that any higher-level social patterns or regularities that we observe can be interpreted only as correlations and not as genuine causal relationships unless we have identified the underlying individual-level mechanisms. When we see evidence for an inverse relationship between the rate of unemployment and inflation in an economy, for example, we are only entitled to interpret this relationship, the so-called “Phillips curve,” as a correlation. In the absence of microlevel foundations, we cannot legitimately claim that there is a causal connection between the two. This view is widely accepted, though seldom critically scrutinized, in political science and microeconomics. In the next section, we examine the case for causal-explanatory individualism in more detail. The negation of the thesis, which we come to defend, is as follows:

How Are the Different Theses Logically Related to Each Other?

The need to distinguish between the different variants of each of individualism and holism is underscored by the fact that they are far from equivalent to one another, and that although each individualist thesis has a holistic negation, some individualist theses are consistent with other holistic ones. The logical relationships are as follows:

In an online appendix,Footnote ¹² we sketch an analytic argument for all of these claims (in line with Stoljar Reference Stoljar and Zalta2009). We now focus on the most central claim for our reconciliation of individualism and holism (included in part (i) of the proposition and its corollary):

The Exclusion Argument

The argument is a version of what philosophers call the “exclusion argument” (e.g., Kim Reference Kim1998, Reference Kim2005). Consider a system that can be described at multiple levels:

1. (1) a lower level, e.g., the level of the brain in the mind-body case or the level of individuals in the social case; and

2. (2) a higher level, e.g., the level of the mind or the social level.

Let us further accept supervenience physicalism (in the mind-body case) or supervenience individualism (in the social case), i.e., all higher-level facts supervene on lower-level facts. Suppose we wish to explain a particular property of the system—call it E for “effect”—by identifying the property or combination of properties that causes it. We face the following methodological question:

In response to this question, two principles are frequently invoked:

From these two principles, it follows that the cause of E must be some lower-level property C, rather than some distinct higher-level property C* that supervenes on it.

Suppose, for example, we wish to explain why immigration queues at London Heathrow Airport have got longer—an effect—and we cite the government's tightening of immigration regulations—a higher-level property—as the cause. The causal-closure and exclusion principles would imply that this is the wrong causal attribution or just an epiphenomenon of the real cause. The real cause must be a complex set of individual actions by various officials leading to the immigration officers’ extensive checks of each arriving passenger, which in turn lead to increased waiting times. In this picture, the higher-level property serves only as a “shorthand” for the real, microlevel cause on which it supervenes. In sum:

Since we accept supervenience individualism as a “no mystery” constraint on social facts, we can deny the conclusion of this argument only by denying either the causal-closure principle or the exclusion principle or both. Whether these withstand scrutiny depends on what we mean by a “cause.” Drawing on recent work in philosophy (List and Menzies Reference List and Menzies2009; Menzies and List Reference Menzies, List, Mcdonald and Mcdonald2010), we now argue that if we define causation in the way that is most common in the social sciences, neither of the two principles is generally defensible.

Two Conceptions of Causation

Broadly speaking, there are at least two approaches to defining causation (e.g., Hall Reference Hall, Collins, Hall and Paul2004): the “production- or mechanism-based” approach and the “difference-making” one (sometimes called the “counterfactual” approach). Both can be seen as attempts to elucidate the traditional notion of “cause and effect,” but they do so in very different ways. The first focuses on the processes or mechanisms “producing” certain effects, the second on the regularities in which certain events stand in actual and counterfactual conditions.

The production- or mechanism-based approach is best illustrated by the traditional idea that causation paradigmatically involves physical objects or bodies impacting on one another, transmitting forces, and thereby pushing one another around. (Think of a billiard ball colliding with another.) Thus causation is understood as a process or mechanism that produces certain outcomes. Developing this idea further, Russell (Reference Russell1913) argued that modern science, such as Newton's physics, does not “carve up” physical processes into discrete events (such as discrete “causes” and “effects”) but describes continuous functional relations that capture “general causal laws”: mechanisms by which systems evolve over time, for instance when objects, such as the planets in the solar system, interact under various conditions.Footnote ¹⁴ Social-scientific approaches that emphasize the step-by-step individual-level mechanisms by which social phenomena are “produced,” such as the microfoundations program in macroeconomics, are best interpreted as following a production- or mechanism-based approach (Hedström and Ylikoski Reference Hedström and Ylikoski2010; Tilly Reference Tilly2001).

The difference-making approach, by contrast, defines causation not in terms of processes or mechanisms, but in terms of the regularities in which certain events or event-types stand. This approach is particularly useful in many special (i.e., nonphysical) sciences, especially when intentional decision-making or other higher-level phenomena are involved. In sciences ranging from medicine and ecology to political science and economics, we are often interested in how changes in some “independent” (or “causal”) variables (e.g., through interventions) systematically relate to changes in certain “dependent” (or “effect”) variables. On the difference-making approach, causal relationships are robust regularities between certain variables or properties (Brady Reference Brady, Collier, Brady and Box-Steffensmeier2008; Lewis Reference Lewis1973; Woodward Reference Woodward, Beebee, Hitchcock and Menzies2009; see also Fearon Reference Fearon1991). So, a property C (within a system of interest) is the cause of another property E if and only if C systematically makes a difference to E. In the illustrative case in which C and E are binary properties, this can schematically be spelt out in terms of two conditionals:

The way in which we obtain evidence for causation in statistics, for instance in regression analysis, is very much in line with this way of thinking about causation: if causation is defined as difference making, then robust correlations—controlling for sufficiently many other factors—are indeed evidence for causation.

A Response to the Exclusion Argument

The exclusion argument is most compelling when causation is understood in the production- or mechanism-based way (as intended in Kim Reference Kim1998, Reference Kim2005), at least to the extent that we are ultimately interested in the most fine-grained mechanisms underlying any given phenomenon. So it comes as no surprise that mechanism-based approaches to the social sciences, such as the microfoundations program in macroeconomics, are among the most strongly individualistic ones. When causation is understood as difference making, by contrast, it is an empirical question whether the most systematic regularities in which some effect E in a social system stands can be found at the lower level or at the higher level of description.

As just noted, scientists usually try to find evidence for causal relations by looking for robust correlations that persist even in the presence of careful experimental or statistical controls. Suppose we find a robust correlation between a higher-level “cause” property (e.g., the interest rate set by the central bank) and some “effect” property (e.g., inflation) in a system (here the economy). We can then ask whether the effect would continue to occur across variations in the lower-level realization of its putative higher-level cause. If the effect continues to occur under at least some such variations (other things being equal), we call the higher-level causal relation “robust to changes in its microrealization,” for short “microrealization-robust.” While causal relations that lack such robustness are reducible to a lower level of description, robust ones are not. In the latter case, the two “tracking conditionals” for difference-making causation are satisfied by the higher-level properties, not by the underlying lower-level properties (as shown in List and Menzies Reference List and Menzies2009; their term for “microrealization-robustness” is “realization-insensitivity”).

Figure 1 (adapted from List and Menzies Reference List and Menzies2009) gives a schematic illustration. Here each point in the two-dimensional space represents one possible world (a “possible scenario”), with the central dot standing for the actual world (the “actual scenario”). The concentric circles around it include increasingly distant possible worlds, with fewer and fewer ceteris paribus conditions met. Suppose we want to explain what caused some effect E that actually occurred and would have occurred throughout the innermost circle of possible worlds (the shaded region). One candidate cause of E is the higher-level property C_H, which we find in all possible worlds in the large shaded region on the left. The two tracking conditionals introduced above are satisfied by C_H with respect to E. Positive tracking is satisfied because in the closest possible worlds (relative to the actual world) in which C_H is present, E is present too; and negative tracking is satisfied because in the closest possible worlds in which C_H is absent (moving towards the right), E is absent too. Suppose now that the lower-level realization of the higher-level property C_H in the actual world is the lower-level property C_L. Since this is only one of many possible realizations of C_H, it is not present throughout the entire shaded region corresponding to C_H, but only in the smaller half-oval inside it. If one were to accept the logic of the exclusion argument, one would have to identify C_L as the cause of E and regard C_H as merely epiphenomenal. But only one of the two tracking conditionals—the positive one—is satisfied by C_L with respect to E: E is present in all closest possible worlds in which C_L is present. Negative tracking is violated, because it is not true that in the closest possible worlds in which C_L is absent, E is absent too: rather, E continues to occur even if the higher-level property C_H has a different lower-level realization. Accordingly, we have an instance of microrealization-robust causation.

FIGURE 1. Microrealization-robust Higher-level Causation

As a stylized real-world example, consider the failed Copenhagen climate summit in 2010. Leaked audio recordings give a detailed picture of how the negotiations faltered in the final hours. An offended Wen Jiabao had withdrawn to his hotel room, directing his chief negotiator by mobile phone. Nicolas Sarkozy lost his temper. Angela Merkel was sidelined. Finally, Mammohan Singh, Lula da Silva, and others struck a minimal deal with Barack Obama (Rapp, Schwägerl, and Traufetter Reference Rapp, Schwägerl and Traufetter2010). Fascinating as this individual-level account is, it is of limited use as a causal explanation. As many analysts have observed, the summit failed for structural reasons, such as the large number of parties, the lack of a common interest, the arguably unhelpful legal framework of the UNFCCC process, and a U.S. president without a domestic mandate for any substantial concessions (see, e.g., Victor Reference Victor2011, chap. 7). The suggested higher-level causal relations are microrealization-robust, because the summit would have been likely to fail even if Wen Jiabao had not sent his chief negotiator, Nicolas Sarkozy had reined himself in, or a different group of politicians altogether had negotiated the same subject matter in broadly similar conditions.

The bottom line is that, once causation is understood as difference making, the causal-closure and exclusion principles are by no means conceptual truths about causation, but rather contingent principles that may apply to some causal systems but not to others. On which side of this divide a given system falls depends on whether it exhibits microrealization-robust higher-level regularities (List and Menzies Reference List and Menzies2009). We return to this point when we spell out criteria for identifying systems that require causal-explanatory holism. In sum, supervenience individualism does not imply causal-explanatory individualism, but is fully consistent with causal-explanatory holism.

INDIVIDUALISM AND HOLISM IN POLITICAL SCIENCE

The debate between individualists and holists in political science often suffers from a lack of clarity, due to a conflation of the supervenience, token, type, and causal-explanatory dimensions. We now revisit some salient areas of that debate, beginning with the heated arguments for and against rational-choice approaches.

Rational Choice Theory and Political Economy

As Laver (Reference Laver1997, vii) recalls,

But what exactly is the accusation levelled against Laver? Our taxonomy allows us to ask more specific questions: is Laver accused of subscribing to supervenience individualism, token individualism, type individualism, causal-explanatory individualism, or a combination of these positions? Laver does not make this precise, but the charge of “reductionism” gives a hint. First, one can set aside the supervenience thesis because, as we have seen, supervenience individualism does not entail the reducibility of all social-level to individual-level explanations and even Durkheimian type holists could accept it. Second, token individualism by itself would not imply any interesting form of explanatory reducibility either: if all particular objects in our social ontology—for example, particular events—could be redescribed as individual-level objects, we would certainly be able to perform a reduction of tokens, but this would tell us little about social regularities in general. What is at stake between Laver and his critics is the reduction of types and causal explanations. The charge is either that Laver is a type individualist, who holds that all social properties are identical with individual-level properties; or, more plausibly (since type individualism is implausible), that he is a causal-explanatory individualist, who holds that good explanations should be stated in terms of individuals alone. This is a common position among rational choice theorists. But as we have argued, causal-explanatory individualism fails when the most robust regularities can be found at a higher level of description.

The individualist focus of rational choice theory goes back to some of its founders. Buchanan and Tullock (Reference Buchanan and Tullock1962) describe a methodological and normative black-and-white choice between an “organic conception” of the state on the one hand, and taking the individual as “the primary philosophical entity” on the other (p. 11), warning that organic conceptions “postulate the emergence of a mystical general will that is derived independently of the decision making process” among individuals (p. 12). The unpalatable mysticism of social-fact holism is Buchanan and Tullock's argument for methodological individualism. But the black-and-white choice is too stark: as we have seen, the falsity of social-fact holism does not entail the falsity of other forms of holism.

The view that the economic analysis of politics requires purely individualistic explanations is not shared by everyone, even on the rational-choice side. Satz and Ferejohn (Reference Satz and Ferejohn1994), for example, argue that there are different kinds of explanations, some of which operate at a microlevel, offering concrete psychological accounts of action, while others operate at a macrolevel, focusing on aggregate behavioral patterns (see also Scharpf Reference Scharpf1997). The upshot is that rational choice theory, broadly conceived, is not wedded to causal-explanatory individualism. An example of a broadly rational-choice-theoretic approach that embraces type and causal-explanatory holism is the emerging field of network analysis (e.g., Christakis and Fowler Reference Christakis and Fowler2009), to which we return later.

Contexts and Institutions

There are many areas of political science in which reference to higher-level aggregates, institutions, parties, cultures, groups, and social contexts is natural to the subject matter at hand, entailing an implicit commitment to causal-explanatory holism. For example, “[t]he central tenet of modern comparative politics is. . .that context—structural, cultural, institutional, and strategic; social, economic, and political; international, domestic, and local—matters” (Franzese Reference Franzese, Stokes and Boix2009, 29). All the explanatory variables mentioned here refer to social properties. To see the ontological commitments of this approach, suppose for a moment that type individualism were true. This would imply that all social properties are identical to individual-level properties and thus that all those “context” properties could be reduced to individual-level properties. Clearly, that is not what most scholars of comparative politics believe. For many, it is precisely the irreducibility of structural, social-level properties that makes comparative analysis a worthwhile mode of investigation. Such a commitment to irreducible “context” variables requires type holism. If one also assumes that irreducible social-level properties feature in the best explanations (as Franzese suggests), then one arrives at causal-explanatory holism.

Similarly, the turn towards a “new institutionalism” has revived the analysis of political phenomena in terms of the “relative autonomy of institutions” rather than the “aggregate consequences of individual behaviour” (March and Olsen Reference March and Olsen1984, 734; cf. Peters Reference Peters2012). Researchers using social properties as independent variables do not, however, need to assume the existence of mysterious “wholes” (and many new institutionalists are concerned with providing individual-level foundations; see, e.g., North Reference North1993). As noted above, type and causal-explanatory holism are compatible with supervenience individualism and the more plausible, weak forms of token individualism.

States and Other Collective Entities

In another area of political science, international-relations theory, the question of whether one should take only individuals as the ultimate unit of analysis is particularly pressing, because a common unit of explanation is the state. A type individualist with regard to states believes that all properties of a state are nothing but properties of the individuals constituting it, and a causal-explanatory individualist believes that all explanations in international-relations theory can be given in terms of individuals alone, without treating states as actors in their own right. If type and causal-explanatory individualism were true, the subdiscipline of international-relations theory could be reduced to theories of individual behavior. Resisting this view, and drawing, like us, on nonreductive physicalism in the philosophy of mind, Wendt (Reference Wendt2004) argues that a commitment to the individual as the basic unit of our ontology does not imply that a reduction from state-level to individual-level explanations is possible. This is in line with our proposition that supervenience individualism does not imply any of the other forms of individualism.

Questions about the ontological status, and explanatory significance, of higher-level units can also be asked about other collective entities in politics, such as legislatures, committees, parties, interest groups, and nongovernmental organizations. Hay (Reference Hay, Goodin and Tilly2006, 88–9), for example, writes the following:

Recently, there has been a growing interest in the thesis that the most parsimonious explanations of the behavior of such collectivities often involve modelling them as purposive, rational agents in their own right, with preferences and judgments that need not be a simple function of the underlying individual preferences and judgments, although they supervene on them (e.g., List and Pettit Reference List and Pettit2002, Reference List and Pettit2011; Pettit Reference Pettit2001, chap. 5, Reference Pettit and Schmitt2003; Tollefsen Reference Tollefsen2002). A central idea is that the higher-level regularities in the behavior of collective entities may sometimes warrant taking what Dennett (Reference Dennett1987) calls an “intentional stance” towards them. Even micro-economists take this stance towards some collectives, for example when they model firms as unified rational actors in the theory of the firm. Many, of course, would argue that the ascription of agency to firms is nothing but an instrumentally useful modelling simplification, and the notion of group agency remains controversial (for a critical discussion, see, e.g., Kornhauser Reference Kornhauser2008).

A New Consensus?

Hay (Reference Hay, Goodin and Tilly2006, 90) suggests that in recent years a “commonsense ground” has emerged within political science. It supposedly consists of the claims that (i) there are irreducible social wholes, (ii) these wholes have properties of their own, which cannot sensibly be seen as properties of their constituent parts, but (iii) the wholes are ultimately constituted only by their parts, so that there is no mysterious additional ontological ingredient. In our terms, Hay accepts supervenience individualism, type holism, and causal-explanatory holism, and does not take a stand on the token dimension. The conjunction of (i), (ii), and (iii) also echoes Sawyer's “non-reductive individualism” (Reference Sawyer2002, Reference Sawyer2003).

We are not sure whether this “commonsense ground” is as entrenched in the discipline as Hay suggests; the short survey in this section raises some doubts. Perhaps a general consensus on the matter is not even desirable, as different phenomena may require different methodological approaches. We now suggest criteria to determine which systems require “holist” methodologies and which require “individualist” ones.

Three Conditions for Causal-explanatory Holism

A social system requires explanatory holism if and only if three jointly necessary and sufficient conditions are met:

The first condition is almost always met by social systems or phenomena. The lower level of description typically refers to individuals and their properties, while the higher level refers to the properties of social aggregates.

The second condition demands that although higher-level facts supervene on lower-level facts, many configurations of lower-level properties can instantiate the same higher-level properties. This multiple realizability implies that higher-level properties are equivalent at most to complicated enumerations of lower-level properties (“wild disjunctions,” as discussed in Sawyer Reference Sawyer2002, Reference Sawyer2003). Evidently, many social-level properties (e.g., an unemployment rate of 8%, a single-peaked profile of preferences, a conservative majority in parliament, etc.) can be realized by many different configurations of individual-level properties. Think of all the different possible distributions of jobs and job seekers that would correspond to an unemployment rate of 8%, the different possible preference profiles that are single-peaked, and the different possible conservative winning coalitions of legislators.

With regard to the third condition, recall that when a higher-level property is a difference-making cause of another property (satisfying positive and negative tracking), this causal relation is “microrealization-robust” if the effect property would continue to occur under at least some variations in the lower-level realization of its putative cause. As we have seen, while causal relations that lack this robustness can be reduced to the lower level of description, this is not the case for microrealization-robust causal relations (List and Menzies Reference List and Menzies2009). Hence higher-level descriptions are indispensable in causal explanations of systems involving microrealization-robust causal relations.

The Study of Ethnic Conflicts and Civil War

To illustrate these conditions, consider first one of the most cited articles in the American Political Science Review, Fearon and Laitin's “Ethnicity, Insurgency, and Civil War” (Reference Fearon and Laitin2003). Their central claim is that

Note first that the phenomena of interest clearly admit multiple levels of description. We can describe them either at the aggregate level of countries or groups, or at the level of individuals. Although Fearon and Laitin occasionally offer rough sketches of what a lower-level mechanism might look like (e.g., p. 79), they are interested in identifying causal relations between higher-level properties. Second, multiple realizability of higher-level properties is also satisfied. While facts about state weakness or insurgency supervene on certain individual-level facts (the totality of all facts about individuals determines the aggregate facts about states, populations, etc.), these aggregate facts can nevertheless be realized in so many different ways that it may be hard to enumerate all their possible realizing conditions. For example, an aggregate property such as “state weakness” can be instantiated by myriads of different individual constellations, whose enumeration is neither practically feasible nor useful. Third, Fearon and Laitin plausibly discover microrealization-robust causal relations, since the identified higher-level regularities are most likely to be robust to changes in individual-level realization. Just as the Copenhagen climate summit would probably still have failed with different political actors in place, state weakness will cause insurgency independently of how exactly it is realized. If Fearon and Laitin are right, the causal relation between state weakness and insurgency should hold independently of the details of the realizing individuals, since the causal relation is supposed to hold not only for past but also for future cases, for which individual realizations are unknowable.

Social-network Theory

As a second example, consider the newly emerging field of network analysis. Christakis and Fowler (Reference Christakis and Fowler2009, 32) suggest that

Assume, for instance, that a property of the network structure (e.g., average path length or the property of being “scale-free”) determines how quickly political information is disseminated. There are multiple levels of description, namely individuals (network nodes) on the one hand, and aggregate network properties (the structure of links) on the other. Identical network properties can be realized by many different constellations of individuals, which implies that some higher-level properties are multiply realizable. Most importantly, higher-level causal relations (based, for instance, on a correlation between average path length and the spread of information) are plausibly microrealization robust. The point of the network-theoretic research program is to identify structural properties of networks that feature in certain causal relations, even though the specific networks instantiating them can differ dramatically. For instance, two networks can have the same aggregate property (such as being “scale-free”) while having very different local topologies and nodes. It is the property of being scale-free that is robustly correlated with certain effects, and this correlation does not depend on the precise local topology or the properties of individual nodes.

States in International Relations

As already noted, international-relations scholars face the question of what the appropriate unit of analysis in explanations of international politics should be: individuals, governments, states, or even larger units? Here again, our three conditions for causal-explanatory holism are typically met. First, there are multiple levels of description, because one can describe phenomena in international relations at the state level, individual level, or various sub- or supranational levels. Second, multiple realizability of higher-level properties holds because the same behavior of a state can be realized in many different ways at the individual level. Third, if international relations theory ever uncovers genuine causal relations that go beyond single cases, these must almost by definition be microrealization robust. Consider for instance the “democratic peace hypothesis,” according to which “democracies do not fight each other because norms of compromise and cooperation prevent their conflicts of interest from escalating into violent clashes” (e.g., Maoz and Russett Reference Maoz and Russett1993, 624). The higher-level property of “being a democracy” supervenes on the properties of individuals; i.e., it must ultimately be instantiated by a complex mix of individual attitudes, beliefs, behaviors, and so on. However, listing all possible individual-level constellations that can instantiate “being a democracy” is practically impossible. Furthermore, the point of the democratic peace hypothesis is that there are structural features of democracies that lead them to avoid wars with one another, independently of the individual political details of each case (e.g., the personalities of the relevant politicians, the specifics of the public debate on the issue, the voting pattern in the legislature, and even the details of the democratic constitution itself). For this reason, if there is a causal relation between “being a democracy” and “avoiding wars with other democracies,” it must be microrealization robust.