1. A mechanismic sociology without AS

The period between the 1950s and 1970s constitutes the turning point for those seeking the origins of a conscious, intentional and programmatic mechanism-based movement in sociology. At that time, a first definition of “mechanism” was adumbrated, diverse methodological applications appeared, and explicit epistemological reflection began.Footnote ⁵

As regards the definition, Merton (1957², p. 351; 1967, p. 43) wrote: “mechanisms refer to structures and processes considered in terms of their functional significance for designated requirements of the social organization”; and farther “social processes having designated consequences for designated parts of the social structure”. Although a functionalist connotation persists in these definitions, Merton used them in practice to build middle-range theories that verbally described cumulative processes of various kinds, such as the self-fulfilling prophecy or the Matthew effect (Merton Reference Merton1948, Reference Merton1968).

On a methodological level, however, the link between the concept of mechanism and that of model took shape within nascent mathematical sociology (Coleman Reference Coleman1964), where there arose the idea of studying models of social mechanisms with mathematics or numerical simulations.Footnote ⁶

The first book with the expression “social mechanisms” in its title was a handbook on mathematical sociology (Karlsson Reference Karlsson1958), whose author connected this concept with those of model, middle-range theory, and interdependent system.Footnote ⁷

Fararo (Reference Fararo and Borgatta1969a, p. 225), one of the most active scholars in this new field, introduced the concept of mechanism in the context of analysis of observed non-Markovian processes:

According to Fararo (Reference Fararo1969b, p. 81, pp. 84-85), the goal of mathematical sociology is precisely “to formulate and test specific mechanisms postulated which describe how the phenomenon occurs”. Instead of restricting ourselves to the construction of “models of the data”, claimed Fararo, we should systematically build “model(s) of the process generating the data”.

Sorensen (1976, pp. 71-72, 1977, pp. 965-966) made a similar point in the context of criticism against the status-attainment paradigm: it is necessary to mathematically formalize hypotheses about the mechanisms and empirically estimate parameters which reflect them (Sorensen 1979, pp. 377-383)Footnote ⁸.

Boudon (Reference Boudon, Merton, Coleman and Rossi1979, p. 52) also accepted the concept of a “generating theory or model” and claimed that “a major task for the future will be to learn how to build generating theories to explain statistical structures” (Boudon Reference Boudon1976, p. 1187). According to Boudon, the axioms of such models must contain at least: “a description of the logic to regulate the actions of the individuals” and “a description of the social constraints within which the logic of individual action develops” (ibid., p. 60). However, unlike Fararo and Sorensen, who focused on mathematical models, Boudon studied his generating models of judicial choice, social mobility and relative deprivation by means of numerical simulations (see respectively, Davidovich and Boudon Reference Davidovitch and Boudon1964; Boudon Reference Boudon1974; Boudon Reference Boudon1982, ch. 5).

This opposition between two ways of implementing a generating model also appeared in Schelling (Reference Schelling1971) and Granovetter (Reference Granovetter1978). Apart from the differences between their models, both authors were interested in mechanisms of aggregation of interdependent choices. But while Schelling created this dynamic by iterating local rules of action and interaction (thus anticipating the modern agent-based computational approach), Granovetter studied the evolution and the equilibrium conditions of the aggregate dynamic of his model by means of difference equations.Footnote ⁹

Finally, on the epistemological level, a theory of explanation focusing on the concept of mechanism, as opposed to the covering-law model, was outlined in the context of the realist approach in philosophy.

Harré (Reference Harré1972, p. 116, 121, pp. 136-137) explicitly linked the concept of causality with the concept of mechanism. He opposed a “successionist theory of causality” against a “generative theory of causality” according to which “a scientific explanation of happenings, whether individual happenings or sequences of events, consists in describing the mechanism which produces them” (p. 170). Since these mechanisms are often unobservable – Harré distinguished among “accessible”, “quasi-accessible” and “non-accessible mechanisms” – they need to be modelled. Harré called such models “iconic models” (Harré and Secord Reference Harré and Secord1972, p. 67).

In different terms, similar arguments were put forward by Bunge (Reference Bunge and M.1973, p. 92, 97), who distinguished between “object model” – “a schematic representation of an object” – and “theoretical model”, “a hypothetical-deductive system concerning a model object”. Depending on their degree of detail, these two types of models can assume the form of a “black box”, a “gray box with internal states” or a “translucent box equipped with a mechanism” (ibid., p. 105). Researchers should always tend towards the latter because that kind of model seeks to uncover “the innards of things” (ibid., p. 101) or “the modus operandi of things ” (p. 104). Bunge (ibid., p. 109) termed these models “mechanism models”.

These different strands of literature thus show that, between the 1950s and 1970s, there took shape an approach centred on the concept of mechanism. On a conceptual level, a mechanism began to be defined as a bundle of constraints and interdependent actions able to generate some macro-consequences. On a methodological level, a mechanism was conceived as an input-output function to be modelled. On an epistemological level, a relation of implication was established among the notions of mechanism, causality and explanation. However, although each of these elements would become part of contemporary AS, none of the authors that I have mentioned used the expression AS.

2. The beginning of an AS without mechanisms

The concept of “AS” is not a new one. What is new is its association with the concept of “mechanism”. For several decades, in fact, the idea of building an analytic(al) sociology or analytic(al) theory was pursued with no reference being made to the concept of mechanism.

The oldest case is The Structure of Social Action (1937). In the introductory chapter, Parsons enunciated what one may call the “fundamental analytical principle”: when one wants to explain a given object, “it shall be broken down into simpler elements which shall serve as the units of one or more theoretical system in terms of which it is to be explained” (ibid., p. 31).

According to Parsons, these elements are of two types: on the one hand, the parts or units into which the object can be decomposed (“type-part concepts” refer to these parts); on the other, the properties which can be associated with each of these parts, and which Parsons calls “analytical elements” (ibid., 34) (“analytical concepts” refer to these analytical elements, see also p. 748). Parsons argued that both concepts involve an operation of abstraction (ibid., p. 34). He conceived the explanation of a concrete object as analysis of the relations between the values of the analytical elements and their changes over time.Footnote ¹⁰

Parsons called this perspective “analytical realism”: first “realism” referred to the assumption that some of the concepts, in particular the analytical elements, can “adequately ‘grasp’ aspects of the objective external world” (ibid. p. 730); second “analytical” instead meant that “these concepts correspond, not to concrete phenomena, but to elements in them which are analytically separable from other elements” (ibid., p. 730).

Parsons thus did not use the term AS, although he sometimes adopted the expressions “generalized analytical theory” (ibid., p. 601) or “analytical theory” (ibid., p. 728). However, his commentators, critics and some of his followers have employed it to denote his research program (see, respectively, Camic Reference Camic and Camic1991, pp. liv-lxiv, Burger Reference Burger1977; see also Parsons’s (Reference Parsons1977) reply, Bortolini Reference Bortolini2007 and Pollini Reference Pollini2007).

It seems that Carr (Reference Carr1955) was the first author to use the term AS. Although he did not give an explicit reason for doing so, the first part of his book shows that he understood AS as an exercise in conceptual classification and clarification. Carr claimed that this task was a prerequisite for field observation and beneficial to the solution of concrete social problems.Footnote ¹¹

This opposition between “conceptual” and “concrete” is also apparent in Katz’s (Reference Katz1964) article, which sought to define the characteristics of “sociologists-as-analytic-scientists” and “sociologists-as-applied-scientists”. According to Katz, the task of the former was “the production of conceptual formulations”, while that of the latter was “the harnessing of a body of existing knowledge to concrete problems” (ibid., p. 440). While applied sociologists were constantly subject to the emergency of concrete situations, sociology of analytics (p. 443) was distinguished by introspection, abstraction, explanatory ambitions, and precision in expression; its fundamental principle was “the quest for analytic accuracy” (p. 444).

Gert Mueller’s (Reference Mueller1967, Reference Mueller1969, Reference Mueller1989) work exhibits the greatest tension towards abstraction. In the late 1960s, he defined AS as the approach which should “lay bare the conceptual (or categorical, or formal) framework under which empirical sociology (maps) describes social reality” (1969, p. 86). In an unpublished paper entitled “Analytical Sociology: An Outline”, the goal indeed became that of building a social ontology. Mueller stressed the virtues of abstraction, advocated a strict alliance between philosophy and sociology, sought to demonstrate a convergence among Husserl, Parsons and Giddens, and argued that the basis of the “analytical theory” is formal logic, not mathematics. The idea that “the validity of analytical sociology is based on the coherence of its categories” (p. 36) was defended in another unpublished paper entitled The Logical Foundations of Analytical Sociology (these texts are available on Mueller’s website). The identification of “analytical theory” with a “critical ontology revised” received its final treatment in Sociology and Ontology: The Analytical Foundations of Sociological Theory (1989).Footnote ¹²

Despite their heterogeneity, these contributions thus show that the project of building an “analytical sociology (or theory)” was originally based on a simple idea: using abstraction and decomposition to construct clearly defined conceptual systems with which to interpret social reality. Although this idea is central to contemporary AS, none of the authors mentioned adopted the concept of mechanism which today characterizes this approach.Footnote ¹³

3. The starting point of contemporary AS

Let us now consider how Gambetta justifies the introduction in 1995 of the term “AS” when he was creating a new cycle of seminars at the University of Oxford:

The fourth criterion mentioned by Gambetta shows that the concern for conceptual precision is at the heart of contemporary AS exactly as it was among those who first adopted that expression. The importance of abstraction and decomposition invoked by the “old” AS is also advocated by recent programmatic texts on “new” AS (Hedström and Swedberg Reference Hedström, Swedberg, Hedström and Swedberg1998b, pp. 13-15, 24-25). Elster (Reference Elster2007, p. 455) notes: “What one might call the analytical turn in social science in my view does not rest on the use of quantitative methodology but on a near-obsessive concern with clarity and explicitness”. Hedström (Reference Hedström2005, p. 3) explicitly accepted the concept of “analytical realism” coined by Parsons.Footnote ¹⁵

However, the first three criteria stated by Gambetta show that contemporary AS is filtering the project of the “old” AS through the ideas associated with the first formulations of a mechanismic approach.

On the one hand, analyticity now applies to the decomposition of the internal structure of a mechanism. Conceptual decomposition is therefore a prerequisite for the construction of models whose generative capacity should be empirically testable. On the other hand, the concept of middle-range theory protects contemporary AS against excessive abstraction, the construction of a priori social ontology being the extreme manifestation of such excess (Hedström and Swedberg Reference Hedström, Swedberg, Hedström and Swedberg1998b, pp. 5-6; Hedström and Udéhn Reference Hedström, Udéhn, Bearman and Hedström2009). As noted by Pawson (Reference Pawson2000), AS indeed gives a specific content to the concept of middle-range theory; that is to say, a concatenation of mechanisms that contains at least one bundle of micro-level elements (Gambetta Reference Gambetta, Hedström and Swedberg1998).

However, it would be wrong to deduce that contemporary AS entirely dismisses any attempt to attain a general sociological theory. When some AS proponents express their scepticism about any general theory (Hedström and Bearman Reference Hedström, Bearman, Bearman and Hedström2009b, pp. 6-7), they are questioning more the means usually adopted to achieve such a theory than its legitimacy. In this regard, according to an idea already put forward by Merton (Reference Merton and Merton1967, p. 68), contemporary AS suggests that we should proceed step by step by networking local models, having proven their empirical validity (Bouvier Reference Bouvier2008). From this perspective, a mechanism can legitimately constitute the basis for a general theoretical proposition provided that it is shown to be at work behind different phenomena and, for each of them, appears under various spatial-temporal conditions involving actors with heterogeneous identities, beliefs and action logics.

1. The conceptual level

Since the introduction of the concept of mechanism, it has been clear that a mechanism is not reducible to an intervening variable (in statistical terms). Rather, it is a unit with the capacity to generate the shape and intensity of the association between two or more events (see Pawson Reference Pawson1989, pp. 130-131). However, specifying the internal content of such a unit has proved to be much more difficult. AS now adopts a definition imported from molecular biology and neurobiology (Hedström Reference Hedström2005, p. 25; Hedström and Bearman Reference Hedström, Bearman, Bearman and Hedström2009b, p. 5) and according to which “mechanisms are entities and activities organized such that they are productive of regular changes from start or set-up to finish or termination conditions” (Machamer, Darner and Craver Reference Machamer, Darden and Craver2000, p. 3).Footnote ¹⁶

Within AS, norms, organizations, informal groups, social networks, interdependence structures and actors are entities of equally legitimacy and importance for building models of social mechanisms. On the other hand, reasoning, evaluation, learning, and action occupy a prominent place among the activities that one should consider. Finally, according to what is now customarily called DBO theory (Hedström Reference Hedström2005, ch. 3) (I shall later return to this point at length in paragraph III.b), AS focuses, among actors’ properties, on desires (preferences), beliefs and opportunities.

In order to account for the genesis and the transformation of these micro-level properties, AS emphasizes the circularities that exist:

1. 1) among desires, beliefs, and opportunities (Elster Reference Elster1979, ch. 2, Reference Elster1983a, Reference Elster1989, Reference Elster2007, ch. 3);

2. 2) among desires, beliefs and emotions (Elster Reference Elster1999, ch. 4, Reference Elster2007, pp. 227-229);

3. 3) among desires, beliefs, opportunities, and the interactions (or interdependencies) in which actors are embedded (Hedström Reference Hedström2005, ch. 3).Footnote ¹⁷

AS accepts that actors’ beliefs can be both positive and normative (Boudon Reference Boudon1995), and it is interested in how actors construct their beliefs within a given social context (Boudon Reference Boudon2001a). The instrumental conception of rationality has been extensively and severely criticized (Hedström Reference Hedström2005, ch. 3; Boudon Reference Boudon1998, Reference Boudon2003; Elster Reference Elster1989). Other forms of rationality are often deployed, such as mimetic (Hedström Reference Hedström, Hedström and Swedberg1998), evolutionary (Macy Reference Macy1997; see also Macy and Flache Reference Macy and Flache1995) or cognitive rationality (Boudon Reference Boudon1989, Reference Boudon1993, Reference Boudon1996, Reference Boudon1998).

It is important to stress that AS’s obsession with discovering the micro-level roots of social patterns does not imply any form of reductionism. Upstream, the beliefs, desires and opportunities of each actor are constrained by a plurality of structures. Downstream, the action of each actor influences other actors’ beliefs, desires and opportunities, contributing to the creation of structures which in their turn will constitute the constraints on future actions (see also Abell Reference Abell and Turner1996, p. 261; Boudon Reference Boudon, Alexander, Giesen, Munch and Smelser1987, p. 46; Coleman, Reference Coleman1993, p. 63).

AS’s endeavour to systematically link structures, actions and interactions to explain the genesis of macro-regularities thus induces it to defend a weak form of methodological individualism (Hedström and Swedberg Reference Hedström, Swedberg, Hedström and Swedberg1998b, pp. 12-13; Hedström Reference Hedström2005 p. 5, note 4; Hedström and Bearman Reference Hedström, Bearman, Bearman and Hedström2009b, p. 8). This variant of methodological individualism, which recognizes the centrality of norms and relational structures, has been called “structural individualism” since the late 1970s (Wippler Reference Wippler1978; Raub Reference Raub and Werner1982, and for a synthesis Udhén 2001, ch. 10, 12).

However, any general theory on either the links among structures, interactions and activities or the links among the internal properties of these entities is beyond the scope of the AS research program. AS only sets a constraint on how to proceed: namely, the requirement that a mechanism must be clarified whenever one mobilizes entities and/or activities between which a link is postulated to exist. AS does not exclude the possibility of inserting new entities, properties or activities (for instance, neurons below beliefs). Again, the only constraints that AS imposes are that the introduction of these new entities must be justified, their status clarified, and their link with preexisting ones specified. In short, AS contains nothing more than a principle of representation of the explanation.Footnote ¹⁸

2. The epistemological level

The syntactic rule “to explain an observation is to generate it” implies that there is no causal explanation without empirical evidence of a mechanism. Thus, on an epistemological level, contemporary AS is based on the “generative theory of causality” proposed by Harré in the early 1970s.Footnote ¹⁹

On these bases, AS judges as inadequate the conception of causal explanation developed by Hempel (Reference Hempel1965, ch. 10 and 12, 1966, ch. 5), and according to which explaining is to deduce the instance to be explained from a general law, given certain conditions (Cherkaoui, Reference Cherkaoui2005, ch. 4; Elster Reference Elster1983b, pp. 44-48; Reference Elster1999, ch. 1, Reference Elster2007, ch. 2; Hedström and Swedberg Reference Hedström, Swedberg, Hedström and Swedberg1998b, p. 8; Hedström Reference Hedström2005, pp. 15-20; Hedström and Udhén 2009; see also Fararo Reference Fararo1989, p. 44, 53, 55).

The objection does not concern the deductive character of nomological explanation. Any mechanismic explanation indeed involves a logical deduction (Hedström Reference Hedström2005, p. 30).Footnote ²⁰ Nor do reservations amount to denying the importance of nomic knowledge for the construction of a mechanism. In this regard, if one identifies a law with a probabilistic regularity constructed inductively through statistical analysis (which is one of the options allowed by Hempel himself), AS is perfectly open to the use of “social laws”. As I shall say later, introducing empirical regularities of this type into formal models particularly well suited to the study of social mechanisms is one of the proposals that AS makes in regard to the complex problem of how models of social mechanisms can be empirically tested (Hedström 2005, ch. 6).

Actually, AS is discontent with the covering-law model because it does not require the insertion of mechanisms into the general laws that it demands to mobilize (Hedström Reference Hedström2005, p. 17; among philosophers, see, for instance, Bunge Reference Bunge1997, pp. 411-412; Reference Bunge1998, p. 75).Footnote ²¹

AS makes a similar objection against the probabilistic variant of the covering-law model – which Hempel (Reference Hempel1965, ch. 10, pp. 250-251, 278, ch. 12, pp. 376-412, 1966, ch. 5, pp. 58, 59) himself called “statistical explanation” – especially in its elliptic form “X because Y” obtained when the explanans lacks certain conditions (Hempel Reference Hempel1965, ch. 12, p. 415; 1966, ch. 5, pp. 52, 53).

This form of the covering-law model, which lies at the core of the statistical methods in quantitative sociology, is unsatisfactory, AS argues, because the coefficient of a statistical model quantifies at best the effects of a mechanism; it cannot provide either a representation of its internal structure (and dynamic) or its relations with other mechanisms (Cherkaoui Reference Cherkaoui2005, ch. 4, pp. 98-103; Hedström 2004; Hedström Reference Hedström2005, ch. 5; Hedström and Swedberg Reference Hedström, Swedberg, Hedström and Swedberg1998b, pp. 9-10, 15-17; see also Goldthorpe 1999).Footnote ²²

However, this objection against the variable explanation and statistical causality should not be interpreted as an attack against the importance of empirical description:Footnote ²³

3. The ontological level

The syntactic rule “to explain an observation is to generate it”, on which the AS concept of causality is based, does not require this approach to defend, at the ontological level, the existence of a hierarchy of levels of reality with sui generis nature.

Although AS considers mechanisms to be pieces of the real world, it only draws distinctions between levels of analysis; it does not do so between levels of reality. In this regard, the analytical realism advocated by AS differs from certain variants of contemporary realism in philosophy, such as the “critical realism” developed by Baskar (Hedström Reference Hedström2005, pp. 70-74; Pawson Reference Pawson2000 and Barbera Reference Barbera, Maccarini, Moranti and Prandini2008).

Thinking in terms of levels of analysis means that, firstly, when one is building a model of a mechanism, the attribution of a given entity/activity to a given level of analysis depends on the type of entity/activity that one finally want to generate. Paradoxical as it may seem, this means that the same entity can assume several analytical statuses within a single model. In this sense, Hedström (Reference Hedström2005, p. 74) notes that “notions of levels, as they are used here, are always relative to the mechanism under consideration”.Footnote ²⁴

Thinking by levels of analysis has a second important implication for the ontological status of the causal entities/activities postulated to build models of mechanisms. While each entity mobilized in the construction of a given mechanism may be relevant from a causal point of view – that is to say, it can contribute to producing a given change in a given state of the world – that entity may nevertheless not be sufficient in itself from a generative point of view – that is to say, it needs other entities belonging to other levels of analysis to make effective changes in the state of the world that one is seeking to explain. That is why Hedström and Bearman (Reference Hedström, Bearman, Bearman and Hedström2009b, p. 11) claim that “the micro-to macro relationship is a parts-to-a-whole relationship rather than a cause-to-effect relationship”.

Thus, by distinguishing between “causal relevance” and “generative sufficiency”, AS proposes an elegant and conceptually simple solution to the apparently unsolvable problem of what level comes first and/or matters more.

4. The methodological level

The syntactic rule “to explain an observation is to generate it” raises an important issue. Demonstrating that a concatenation of mechanisms A produces an outcome O is a necessary but insufficient condition for claiming that A explains O. For this purpose, one must eliminate the possibility that other concatenations of mechanisms A_1,[…],n are able to generate O in as satisfactory a manner as the concatenation originally postulated.

This problem – known in philosophy of mind as “multiple realizability” (for a pedagogic introduction, see Bickle Reference Bickle and Edward N2006) – concerns the study of a mechanism once it has been theoretically designed. Studying a mechanism in its turn raises two distinct problems: first, analysis of its behavior (how does it work and what does it produce?); second, empirical validation of the mechanism (are its internal components, the process that it has triggered, as well as the outcome that it has finally generated, empirically grounded?).

AS comprises a formal as well as a qualitative approach to both phases. While the formal approach aims at integrating formal models and quantitative techniques (Hedström 2005, ch. 6; Hedström and Bearman Reference Hedström, Bearman, Bearman and Hedström2009b, p. 16; Manzo Reference Manzo2007a), the qualitative one emphasizes natural logic and qualitative methods (Elster Reference Elster2009, pp. 445-467). However, what is really at stake is finding the way to articulate these methodological options.

As far as the formal approach is concerned, AS considers statistical methods to be inappropriate for analysis of how a mechanism works. This does not mean, however, that such methods are useless. On the contrary, AS gives them an important role in testing a model of a mechanism. First, statistical analysis is necessary to compare the macro patterns generated by a model against empirical data describing the actual macro pattern of interest (Manzo Reference Manzo2009a). Second, statistical analysis is needed to introduce empirical regularities as inputs to formal models that explicitly represent mechanisms (Hedström Reference Hedström2005, ch. 6).

Among such models, AS focuses on mathematical and computational ones in particular.

Mathematical models have been spontaneously associated with a mechanismic approach since the 1960s. However, we now know that these models capture only the surface of a model mimicking a mechanism. Only a form of “systemic causality” has a place in formal models such as differential equations or markovian models (Doreian Reference Doreian1999, p. 88). Here, given some initial conditions, a set of parameters and some equations, only aggregate dynamics are studied. These equations contain some rules on transition between two high-level states, but they do not yield any fine-grained formalization of action or rules on interaction between heterogeneous low-level entities (among philosophers of sciences, see Bunge Reference Bunge1997, p. 423). When this happens, one has switched to a form of “causality” called “algorithmic” (Doreian Reference Doreian1999, p. 99). In this case, however, if the mechanisms to be represented are too complex, it may be difficult to find an appropriate mathematical representation of them; or if a representation exists, the analytical solution may be absent (Axtell Reference Axtell and Axtell2000; Hedström Reference Hedström2005, pp. 87-98).

Computational methods help solve this problem (Fararo and Butts Reference Fararo and Butts1999, p. 35). Although they cannot produce general theorems, they can deductively generate patterns from mechanisms as complex as the theory requires. This applies in particular to the agent-based models which have recently attracted the interest of some AS scholars (for technical aspects, see Ferber 1999; for more general overviews of the literature, see Gilbert 2007, Macy and Willer 2002, Miller and Page 2007, Sawyer Reference Sawyer2003).

This flexibility stems from the specific programming language on which the design of an agent-based model is based. The core of the program consists in a set of “objects”, i.e. computational units defined by certain properties (attributes) and rules of behavior (methods or procedures). According to the bundle of properties and rules defining an object, the latter can be used to model the behavior and interactions of a set of particles, molecules, cells, beliefs, actors, groups (of particles, molecules, etc.), organizations, etc.

There are therefore deep-lying reasons why some analytical sociologists have begun to explore this class of computational models. A structural homology exists between thinking in terms of mechanisms and building an agent-based model. As a mechanism consists of entities, their properties and their activities, so an agent-based model is made up of objects, attributes and procedures which define the behavior of the object. Hence, when one is designing an agent-based model, one is building an artificial mechanism. On the other hand, simulating an agent-based model means iterating the rules that define the objects of which it is made up and which finally enable objects to communicate with each other and to update their attributes over time. Hence when one simulates an agent-based model, one is activating in silico the process that the mechanism potentially contains.

What this process generates is exactly what AS seeks: evidence that a given entities/properties/activities triad is able to generate a given set of associations among certain aggregate events. It is precisely the generative capacity of a mechanism that an agent-based model makes it possible to implement. As noted by Epstein (Reference Epstein2006, p. 8), “Agent-based computational models provide demonstrations that a given micro-specification is in fact sufficient to generate a macrostructure of interest”. Such a demonstration, moreover, has a deductive character: “every realization of an agent model is a strict deduction” (ibid., p. 56).

As powerful as this technique may seem in studying how a mechanism works, it is nevertheless unable by itself to validate the empirical relevance of the mechanism. For this purpose, it is necessary to inject empirical information into both the behavior rules of the entities composing an agent-based model and the relational structures which supposedly impact on them.Footnote ²⁵

Hedström’s above-mentioned proposal concerning empirically calibrated agent-based models identifies such empirical information with previously-highlighted statistical regularities. However, as the literature on the problem of validating an agent-based model shows (Fagiolo, Windrum and Moneta Reference Fagiolo, Windrum and Moneta2007; Moss and Edmonds Reference Moss and Edmonds2005; Moss Reference Moss2008), qualitative materials can also be used to initialize an agent-based model empirically.

This is exactly where the qualitative orientation of AS (Gambetta Reference Udehn1993, Reference Gambetta2009a; Gambetta and Hamill, Reference Gambetta, Hedström and Swedberg2005), as well as its interest in combining survey and experimental method (Bohnet Reference Bohnet, Hedström and Bearman2009; Ermisch and Gambetta Reference Ermisch and Diego2008), can integrate more formal approaches together.

As recent trends in behavioral economics show, only a permanent exchange among formal models, laboratory experiments, survey and qualitative research enables solution to the complex problem of how models dealing with social mechanisms can be constructed and empirically tested (Fehr and Gintis Reference Fehr and Gintis2007). The creation in 2008 of the Center for Experimental Social Sciences (CESS) at Nuffield College (Oxford University) on the initiative of D. Gambetta (with R. Duch and P. Young) demonstrates that AS is seeking to develop a “hard” integrative methodology of this kind.Footnote ²⁶

1. How AS defines the concept of mechanism

To reduce the ambiguity surrounding the concept of mechanism, Gross (Reference Gross2009, pp. 360-362) first considers five definitions and tries to build a synthetic one, which he terms a “working definition”. According to this definition, a (social) mechanism is a:

First to be noted is that AS shares Gross’s concern to overcome the heterogeneity of the meanings associated with the concept of mechanism (see, for instance, the list of definitions compiled by Hedström 2005, p. 25 and Hedström and Bearman Reference Hedström, Bearman, Bearman and Hedström2009b, pp. 5-6).Footnote ²⁸ Then, recent contributions in AS tend to agree on a definition that Gross curiously does not consider:

Compared to this definition, it is unclear whether Gross’s definition enables us to make any real progress.Footnote ²⁹

Firstly, since any “event” or “process” ultimately refers to entities and activities (regardless of their specific content), a definition that focuses on entities/properties/activities triads seems more accurate than a definition which refers generically to such notions as “events” or “processes” – which are at least as unclear as the concept to be defined.

Secondly, because Gross’s definition tends to equate a mechanism with a process, it introduces an ambiguity which instead should be resolved: what differentiates a “mechanism” from a “process”? As has been noted (Vayda et al. Reference Vayda, McCay and Eghenter2001), the concept of process is often itself only a cover concept: as a consequence, it appears that it cannot be usefully employed to specify what a mechanism is. As I pointed out earlier, AS by contrast suggests a criterion: a process is the dynamic side of a mechanism; that is to say, the sequence of changes triggered by the activities (and interactions) of the entities contained in the mechanism.Footnote ³⁰

In the second part of his article, Gross (Reference Gross2009, p. 368) suggests a more specific definition of the concept of mechanism, one based on the alternative theory of action inspired by the pragmatism that he defends. However, to consider the scope of this second definition, it is better first to consider the reasons which induce Gross to refer to this theory of action.

2. How AS conceives the DBO theory

According to Gross (ibid., p. 365), current conceptualizations of social mechanisms have a common shortcoming: they adopt a theory of action which is inadequate. This is because they use assumptions which by and large fall within the domain of rational choice theory or its variants such as DBO theory.

A careful reading of Gross’s article shows that he detects six weaknesses in this theory of action: 1) rationality is an innate ability which is more or less homogeneously distributed in social space (p. 365); 2) beliefs and desires are clear and consistent entities (p. 365); 3) ends distinctively come before means (p. 365); 4) action is intentional (p. 365); 5) action maximizes utility (p. 367); and 6) the intimate cultural nature of the action is ignored (p. 367). On each of these points, the pragmatist approach advocated by Gross admits, by contrast: 1a) a plurality of forms of rationality whose historical origins require explanation (p. 367); 2a) a “psychological postulate of ambiguity” (p. 365); 3a) “means and ends [...] are often emergent from action” (p. 367); 4a) “much action is habitual and typically involves no conscious weighing of means and ends” (p. 367); 5a) “pragmatism does not equate problem solving with maximization of utility” (p. 367); and 6a) “pragmatists insist that problem situations are always interpreted through cultural lenses” (p. 367).

This diagnosis immediately raises two questions. First, to what extent is it correct to claim that AS is basically a new instance of rational choice theory? Second, does DBO theory really have shortcomings similar to those that Gross detects in rational choice theory? My impression is that Gross is simplifying a much more complex picture.

Let us consider the first question. In this regard, the most general feature to be taken into account is that key authors in analytical sociology have expressly distanced themselves from a strict version of rational choice theory. Hedström (Reference Hedström2005, p. 60), on whose theory Gross focuses in particular, says:

By “instrumentalism”, Hedström means the epistemological position according to which unrealistic theoretical premises should be accepted provided that they lead to good predictions and/or to elegant formalizations. According to Hedström, despite the fact that any theory leads to some simplification, a theory of action should be based on assumptions which are only descriptively incomplete but not completely false.Footnote ³¹

Let us also recall that Boudon (Reference Boudon2001a, pp. 451-452) explicitly argued that rational choice theory cannot claim to be a sustainable theory of action for social sciences. According to Boudon, three of its premises – “consequentialism”, “egoism” and “cost-benefit balance” – excessively restrict the theory’s domain of validity.

Last but not least, Elster (Reference Elster2007, p. 5; for an overview of his perplexities, see also pp. 462-463) wrote:

If one separately considers points 1 to 6 above, it seems equally difficult to argue that DBO theory is a variant of the orthodox form of rational choice theory. On the contrary, careful inspection of the AS theoretical and empirical literature even suggests that each of Gross’s counterproposals 1a to 6a already has its place in the way that AS conceives DBO theory.Footnote ³²

Firstly, DBO theory makes no assumption concerning the particular type of rationality which drives actors:

In particular, this perspective does not at all postulate that actors are uniformly and constantly provided with the informational and computational capabilities assumed by the orthodox form of rational choice theory:

On the contrary, rationality takes diverse forms in AS: “cognitive rationality” is probably the most general of them (Boudon Reference Boudon1989, Reference Boudon1993, Reference Boudon1996, Reference Boudon1998, Reference Boudon2003, Reference Boudon, Elster, Gjelsvik, Hylland and Moene2006). In that framework, actors are supposed to theorize the context in which they act and develop systems of reasons which they perceive as strong, given the material, symbolic and cognitive resources to which they have access. “Mimetic rationality” is another form of rationality used by AS to model the actor’s logic of action (Hedström Reference Hedström, Hedström and Swedberg1998). Here, actors build their beliefs by exploiting other actors’ actions as sources of information. Finally, Macy (Reference Macy1997, pp. 435-438, 1998) has introduced into AS a form of rationality which he describes as “emerging” because actors are supposed to choose, not on the basis of the future consequences of their actions, but by thinking about past actions which have shown adaptive capacity.

Thus, within AS, DBO theory is far from being a variant of rational choice theory. Contrary to what Gross claims, the reverse proposition holds: rational choice theory is an instance of DBO theory (Hedström 2005, p. 41).Footnote ³³

The second shortcoming stressed by Gross concerns the exaggerated extent to which DBO theory conceives actors’ beliefs and desires as consistent and clear entities.

In this regard, the first response to Gross’s objections centres on the constant work undertaken by many authors within AS to make actors’ beliefs and desires endogenous.

I have already mentioned, on the one hand, Elster’s analysis of the dynamic interdependency among beliefs, desires and opportunities, and, more recently, between emotions and beliefs, and, on the other, the role that Hedström attributes to dyadic interactions in explaining the genesis of actor’s beliefs and desires. We may now also consider Rydgren’s (Reference Rydgren, Hedström and Bearman2009) chapter on beliefs formation in Hedström and Bearman’s The Handbook of Analytical Sociology, where he studies the role performed by mechanisms of categorization, inference, deduction, network and cognitive dissonance in shaping actors’ beliefs. One of the interests of Rydgren’s treatment (I shall mention a second one later) is that it demonstrates how these mechanisms can give rise to beliefs that are “biased and flawed” (ibid. p. 73). Moreover, Rydgren stresses that these mechanisms often operate unconsciously (ibid., p. 73, 84). As a consequence, he notes, “beliefs are not transparent” (ibid. p. 89).

Elster (Reference Elster2007, p. 212) explicitly makes the same point:

Thus, contrary to what Gross claims, first, unlike standard rational choice theory, DBO theory conceives beliefs and desires as explananda rather than taking them for granted. Second, DBO theory admits that beliefs and desires are often inconsistent and irrational (Elster Reference Elster2007, p. 211). Third AS aims at modeling mechanisms that account for such inconsistency and admits that they tend to operate behind actor intentionality.Footnote ³⁴

Gross then criticizes DBO theory for its linear conception of the ends-means relation. Often, he remarks, actors’ goals, far from preceding their action, are formed during, or even after, such action.

As we have seen earlier, one of the forms assumed by actors’ rationality in the context of DBO theory is what Macy (Reference Macy1997, Reference Macy and Sica1998) calls “emergent rationality”, which implies that actors build the means-ends relation by trial and error on the basis of actions and interactions that they experience over time. As Macy (Reference Macy1997, p. 435-438) puts it:

To be noted is that Macy (Reference Macy1997, p. 437, 444, note 3; 1998, pp. 222-223) explicitly links this conception of rationality with the pragmatist tradition. He also notes that “emergent rationality” is central to an important branch of game theory, i.e. evolutionary games (thus showing that it is incorrect to regard game theory as the prototype of orthodox rational choice theory). Macy (Reference Macy1997, pp. 441-443, 1998, p, 234; Macy and Flache Reference Macy and Flache1995) finally insists on the possibilities that simulation methods open for the formalization of such complex forms of rationality. Macy and Flache’s (Reference Macy, Flache, Hedström and Bearman2009) chapter in The Handbook of AS has recently iterated this argument.

Although Gross (Reference Gross2009, p. 365, note 7) acknowledges that Macy’s contribution is not affected by his objection against DBO theory, he draws no consequences from such a criticism. However, the compatibility between DBO theory and Macy’s conception of rationality is explicitly admitted within AS:

Let us now temporarily set aside the fourth objection that Gross raises against DBO theory – i.e. the intentional nature of the action – and first discuss his two last objections, those regarding the role of maximization and culture in DBO theory.

In regard to the first point, while it is undeniable that the maximization principle is at the heart of standard rational choice theory, one can easily deduce from the three points discussed so far that DBO theory contains no element which suggests that maximization is the driving principle of actors’ action, which is clearly shown by Hedström’s (Reference Hedström2005, p. 61) critical assessment of the assumption of “optimality”.Footnote ³⁵

By contrast, the second point, which concerns the residual place of culture and the simplistic conception of the meaning of individual action within the DBO theory, warrants more detailed discussion.

A first rejoinder to Gross’s objection can be found in Rydgren’s work. His above-mentioned analysis of the socio-cognitive mechanisms of belief formation indeed fits with a more general effort to integrate “cognition” and “culture”. In a study in which Rydgren draws on DBO theory to understand the role played by the past in the genesis and persistence of inter-ethnic conflicts (2007, p. 226), he writes:

Cultural embeddedness is also a core component of the “cognitive rationality” defended by Boudon. One of the problems with this conception of rationality is that it is difficult to define what constitutes a “strong” reason. Boudon proposes two criteria in this regard (Reference Boudon1995, p. 34, 51, 69, 79, 133-134, 189-196; 200, 289, 370; Reference Boudon1996, p. 124, 128, 147; Reference Boudon1999, p. 93, 133, 135). A system of reasons will be strong on the one hand if the actor perceives it as grounded and consistent, and on the other, if he judges that his reasons are able to convince other actors (Boudon Reference Boudon1995, pp. 67-68, calls this second dimension of a system of reasons “trans-subjective”). Of course, the problem here is determining the ultimate source of this subjective conviction in grounds of one’s own system of reasons. Boudon (Reference Boudon1996, p. 148) stresses the relative character of this ultimate foundation, maintaining that it depends on two factors: first, the quality and quantity of the information to which the actor has access; and second, the set of beliefs and symbols taken for granted in a given community. This is why Boudon (Reference Boudon2001a, pp. 455-456) suggests that, in order to emphasize the informational and cultural component of actors’ system of reasons, the concept of “cognitive rationality” should be labeled “contextual rationality”.

Gambetta’s 2009a and b recent work on signal theory also shows that, contrary to Gross’s claims, DBO theory is completely compatible with the assumption that actors always act through “cultural lenses”, to use Gross’s expression. At the heart of the “signaling theory” applied by Gambetta is the following problem: how can ego believe in the trustworthiness of alter’s signal in a context where alter has an interest in concealing the real state of affairs, and where ego’s information about alter is imperfect? Of interest here is that Gambetta (2009b, paragraph 3) stresses that solving this problem requires account to be taken of the norms and symbols which populate actors’ social universes, because, in real social interactions, the credibility and, more fundamentally, the meaning of a signal cannot be separated from these symbolic universes.

Thus, in the light of the elements discussed so far, it seems unsound to argue, as Gross does, that DBO theory has all the shortcomings of standard rational choice theory. In fact, DBO theory gradually builds on multiple theoretical and applied contributions whose common goal is to go beyond the shortcomings of rational choice theory. These contributions show that DBO theory now contains the following elements:

1. 1) it makes no assumptions on the specific form of actors’ rationality;

2. 2) actors’ ends and means may arise from their action rather than precede it;

3. 3) actors’ beliefs and desires are conceived as explananda, not as exogenous elements of the problem;

4. 4) actors’ beliefs and desires are not conceived as being invariably clear, transparent and consistent;

5. 5) the cognitive and relational mechanisms responsible for actors’ beliefs and desires are conceptualized as being able to operate unconsciously;

6. 6) actors’ beliefs and desires are inseparable from actors’ perceptions of the symbolic context in which they act.

One factor which may help explain these misunderstandings is that AS conceives DBO theory as only a specific part of a broader framework within which the analysis of mechanisms should be conducted. As we saw earlier, this framework equates to a complex form of methodological individualism which stresses the existence of recursive loops among context (values, norms, conventions, etc.), structure (organizations, groups), networks and actions (with its internal components). DBO theory is only the starting point for analysis of the micro component of such loops – I say “starting point” because I have sought to show that DBO theory is itself susceptible to very diverse sorts of refinement. Hence DBO theory should not be separated from this larger framework, whose purpose is to conceptualize the circularity between structure and action, not to reduce the former to the latter.

3. Why AS is discontented with the concept of habit

Now it is possible to assess the extent to which Gross’s (2009, p. 368) second definition of the concept of mechanism is able to furnish a conceptualization of social mechanisms more sophisticated than the one based on DBO theory. According to Gross’s second definition:

First to be noted is that Gross’s definition can be easily reframed in terms of the entities/properties/activities triad on which the AS definition of mechanism focuses. In Gross’s framework, indeed, entities equate to actors (individual or collective); the properties of entities equate to actors’ habits (and resources) (Gross insists on resources a few pages after the definition just mentioned); activities equate to actors’ efforts to solve given problems within given contexts. In terms of DBO theory, moreover, the property “habitus” provides the property “belief” with a specific content: without any loss of generality or accuracy, a habit can indeed be thought of as a “latent” belief in the effectiveness of a particular mode of action; the property “resource” falls into the category of “opportunities”; and the action principle according to which actors are problem-solvers is only a specific form of desire – desire or preference for adaptation.

If one then considers the broader context of structural methodological individualism in which AS inserts DBO theory, there is an even more obvious compatibility between Gross’s general conceptualization of social mechanisms and the one which can be built on DBO theory.

In this regard, let us return to Boudon’s (Reference Boudon1986, ch. 2) formulation of this complex (in the sense of “recursive”) form of methodological individualism. Boudon (Reference Boudon1986, p. 51) remarks in his presentation:

Apart from the specific choice of words and the reference to a different philosophical tradition, my impression is that there is only one substantive difference between this principle and Gross’s conceptualization. While Boudon makes it explicit that a complex form of methodological individualism does not contain any specific model of actor Gross starts from exactly the opposite postulate. He believes that the adaptive character of social action must be conceived in terms of habit.Footnote ³⁶

We now have all the elements to go back to Gross’s counter-proposal (4a in my above list) according to which habit, and not intentionality, should constitute the basis for a robust theory of action. This is the crucial step in Gross’s criticism of DBO theory and the main building block of his alternative conceptualization of social mechanisms.

My point here is that the empirical scope and applicability of the concept of habit is narrower than Gross seems to postulate.

What criteria does Gross mobilize to claim that a theory of action which builds on the concept of habit is more “viable” (p. 359), “sophisticated” (p. 359), “solid” (p. 359), “adequate” (p. 365) or “promising” (p. 365) than DBO theory? Since he claims that “few in the theory community agree that rational choice theory or variants such as DBO theory offer empirically or theoretically adequate descriptions of social action” (ibid., p. 365), one may expect that both empirical and theoretical criteria are discussed. However, only the theoretical argument I have discussed so far is part of Gross’s analysis. On an empirical level, by contrast, Gross cites no evidence which demonstrates the realism of his model of actor.

One must concede that Gross (Reference Gross2009, p. 373) notes in the final section: “I hold it to be an empirical question whether the theory of mechanisms laid out here will give sociology more explanatory purchase”. He is clearly referring here to the consequences of his theory. However, evaluating the empirical solidity of its premises is also an important empirical matter. In light of the overall spirit of Gross’s analysis, this aspect is crucial because he himself doubts that “false models” can be useful (Gross Reference Gross2009, p. 365, notes 8 and 9).

In that connection, he might have considered that a rich body of experimental literature has generated a large amount of empirical information which outlines the basic analytic space within which actors’ actions unfold.

In this regard, we should first consider the work accumulated over the past three decades in behavioral economics (Camerer, Loewenstein and Rabin Reference Camerer, Loewenstein, Camerer, Loewenstein and Rabin2004). This experimental research has demonstrated that real actors judge and choose according to principles which are not those postulated by standard rational choice theory: diverse kinds of deviations from rationality occur (for a synthesis, Camerer and Loewenstein Reference Camerer, Loewenstein, Camerer, Loewenstein and Rabin2004, Elster Reference Elster2007, ch. 12 and 20). This conclusion holds both in the case of isolated actors and of actors embedded in interaction structures – behavioural game theory studies this second aspect more in particular (Camerer Reference Camerer2003).

Although Gross does not draw on this literature, his criticism of the orthodox form of rational choice theory is relevant and empirically validated. This literature, however, does not suggest that habit is the alternative mode in which actors act when they are violating the assumptions of the model of homo œconomicus.

On the one hand, although purely selfish behavior occurs less frequently than standard rational choice theory predicts, it regularly arises under certain conditions (Camerer and Fehr Reference Camerer and Ernst2006, Fehr and Gintis Reference Fehr and Gintis2007). On the other hand, experimental studies on the cognitive strategies underlying real actors’ judgments and choices have shown that habit is only one of the problem-solving methods used under uncertainty (Pingle and Day Reference Pingle and Day1996; Camerer, Ho and Chong Reference Camerer2003). Cognitive psychological research on “bounded rationality” has discovered – and modeled – a very wide range of “heuristics” which may account for the shortcuts adopted by real actors in real decision-making processes (see, for a synthesis, Gigerenzer and Selten Reference Gerring2001). Thus, on the basis of this experimental literature, judgment formation, decision-making and actions seem to lie along a continuum whose theoretical extremes are habit on the one hand, and “pure” rational behavior on the other.

The empirical foundation for this representation has been provided by research in social cognitive psychology. According to research on dual-process models two modes of judgment-forming and decision-making co-exist when real actors act (see for a summary, Chaiken and Trope Reference Chaiken and Trope1999, for an application to moral judgments, Greene et al. 2008; Greene Reference Greene2009). Real actors sometimes choose on the basis of automatic and non-reflexive responses (often of an emotional nature), while they sometimes also base their action on a more or less detailed assessment of the situation in which they are acting. Four factors have been reported as being responsible for the activation of one or other of these two operating modes:

1. 1) the extent to which a solution is available to be adopted;

2. 2) the presence of unexpected information or events;

3. 3) motivation (how costly a wrong decision will be);

4. 4) the amount of opportunity to think (the time and cognitive resources available).

Thus, Gross’s choice of focusing with such insistence on the habitual side of action does not seem empirically justifiable. We simply do not have at present sufficient empirical reasons for favoring this component of action when conceptualizing social action. As a consequence, since conceptualizing social mechanisms as “chains or aggregations of actors, problem situations, and habitual responses” (Gross Reference Greene2009, p. 369) restricts rather than enriches the scope of DBO theory, Gross’s proposal seems unable to substitute for it.Footnote ³⁷

A theory of action more realistic than standard rational choice theory usable for a general conceptualization of social mechanisms – which is Gross’s overall intent – should integrate habit, emotion and (cognitive) rationality. Specifying under which conditions one of these logics is more likely to be activated would ultimately provide a general theory of action (DiMaggio Reference DiMaggio and Cerulo2002).

Two items of evidence can be cited to demonstrate the open nature of DBO theory and the AS project to develop a theory of this kind. The first can be found in the section of The Handbook of Analytical Sociology devoted to ego-centered mechanisms. Here, the theory of action at the heart of AS emerges at the intersection of seven chapters, each of them dealing with a specific component and/or sub-component of DBO theory (i.e. emotions, beliefs, preferences, opportunities, heuristics and norms). The second element testifying to the open character of DBO theory is the dialogue in which AS engages with theoretical perspectives that, starting from the experimental evidence just discussed, seek to integrate habit and rationality into a unified formal theory of action.Footnote ³⁸

In conclusion, one may note that Gross’s choice of exclusively focusing on habit seems inconsistent with the philosophical and sociological tradition to which he explicitly refers. As he (p. 365, 368, 372, note 11) himself admits, pragmatism conceives action as lying at the intersection of habit and creativity (see, on this second dimension, Joas Reference Joas1997). The emotional component of individual action is also central to the pragmatist tradition deriving from James’s thought (Barbalet Reference Barbalet, Baert and Turner2007). Moreover, contemporary social psychologists consider James as one of the pioneers of the dual-process models that I have just mentioned (Moskowitz, Skurnik and Galinsky Reference Moskowitz, Skurnik, Galinsky, Chaiken and Trope1999).

So why did Gross’s argument lose this balance? The only justification that he gives is the following one:

One is more surprised than convinced by this argument. Is it not precisely the reflexive, inventive and creative side of social action which enables us to conceive the social world as a probabilistic, rather than deterministic, reality?

Conclusion

The numerous criticisms recently brought against AS raise the crucial question of its specificity. To address this question, I have sought to assemble historical evidence showing that contemporary AS stems from integration and cross-fertilization between two bundles of older ideas. The first concerns the concept of mechanism and explicitly appeared in the 1950s and 1960s. The second set of ideas relates to the concept of “analyticity”, whose roots reach back to Parsons’s intellectual project. I have then tried to show that such integration – which materialized in the mid-1990s – is now being developed in a specific direction.

Without denying the heterogeneity of the authors involved in AS, I have argued that the specificity of the contemporary AS approach must be sought in the combination of a set of conceptual, epistemological, ontological and methodological elements, rather than in each of them taken individually. I have finally sought to show that the most distinctive feature of such an integrative framework is that it only provides a “syntax” for explanation: that is to say, a set of rules on how hypotheses about mechanisms underlying the regularities of social life can be theoretically designed and empirically tested.

This finding has guided my response to the criticisms made of AS to date. Accusing AS of being no more than a particular theoretical orientation (rational choice theory, most often) or, symmetrically, of not taking account of some or other principle (habit, for example) amounts to denying, or failing to recognize, both the intimate meaning and the main objective of the AS research program. In both cases, AS is attributed a limitation that by construction it cannot have. To the extent that AS only provides us with a framework in which to construct and test an explanation, the specific content of each explanation depends on the phenomenon studied, as well as on the empirical evidence that permits exclusion of a particular theoretical option from the outset.

Some of the critics of AS have claimed that it is nothing more than what a scientific sociology should be (Ballarino Reference Ballarino2005). This is a statement everyone may fully agree with. However, if several authors have felt it necessary to collect the options which seem at present best able to make such a scientific sociology effective, this may mean that scientific sociology still hesitates to impose itself. Hence, according to a social dynamic described by Merton (Reference Merton1936, pp. 903-904), i.e. the so-called “suicidal prophecies”, AS paradoxically contains the reason for its death. It will disappear as soon as it has succeeded in moving sociology toward what sociology should be if it were a science like any other.