Cairn - The role of rational choice and narrative action theories in sociological theory

Mon compte |

Mon panier d'achats |

Ma bibliographie

Contacts |

Aide

· Accès distant · Mot de passe oublié

· Recherche avancée

Revue française de sociologie
Ophrys

I.S.B.N.2708010530
212 pages

p. 255 à 273
doi: en cours

Veille sur la revue
Veille sur l'auteur

Acquérir ce numéro

Vous consultez

Volume 44 2003/2

Discipline > Revue > Sommaire > Article

Télécharger cet article en PDF

The role of rational choice and narrative action theories in sociological theory

The legacy of Coleman’s Foundations

Peter Abell Copenhagen Business School Solbjerg Plads 3 2000 Frederiksberg - Denmark London School of Economics Houghton Street London WC 2A 2AE

Résumé de l'article

Les Foundations de Coleman sont fondées sur une conception de l’action rationnelle qui est à la fois spécifique et générale. L’article défend l’idée selon laquelle cette dernière dimension sera sans doute la plus influente des deux ; mais telle qu’elle est formulée, elle demeure trop générale. La théorie narrative de l’action se présente comme une voie possible pour dépasser la difficulté. Coleman’sFoundations isbaseduponbothaspecificandmore general conceptionof rational action.Thepaperargues thelatter mayturnouttobemoreinfluential. Asformulated, however,it is too broad. Narrative action theory is a possible refinement. Colemans Foundations sind begründet auf einem Konzept des rationalen Handelns, das sowohlspezifischalsauchallgemeinist.DerArtikelverteidigt denGedankenwonachdieletztere Dimension die einflußreichere ist. Sowie er jedoch formuliert wird, bleibt der Gedanke zu allgemein. Die narrative Theorie der Aktion ist hiermit ein möglicher Weg, diese Schwierigkeit zu umgehen. Las Foundations de Coleman están basadas sobre una concepción de la acción racional que simultáneamente es específica y general. El artículo defiende la idea mediante la cual esta última dimensión será sin duda la mas influyente de las dos, pero aparece demasiado general de la manera como esta formulada. La teoría narrativa de la acción se presenta como una posible vía para superar esta dificultad.

Plan de l'article

• Methodological individualism

• Optimality/Consistency

• Assumptions of narrative action theory

— Comment

— Homo sociologicus interventions/action rules

— Homo economicus interventions/action rules

• RÉFÉRENCES

A theory is a more or less complex conjecture about a latent mechanism which may account for a puzzling empirical relationship (fact !). Empirical relationships themselves are, however, properly constructed within the framework of a theoretical description. That is to say, they do not inhabit our senses in an entirely innocent form. Some have proposed that what we might naturally describe as “descriptive theory” should be the abiding and exclusive concern of social theorists (ie interesting conceptualisations of the social world). This viewpoint was, however, rejected by Coleman in his Foundations (1990) where he advocated giving rational action or choice theory pride of place in formulating conjectural theories. In the latter part of his book he makes use of a rather specific model of social exchange where actors are assumed to have Cobb-Douglas preferences and arrive at parametric exchange “prices”. [1] My own hunch is that, despite the elegance of this model, the earlier, less formal part, of Coleman’s book will have a more abiding influence. There Coleman propounds a much less rigorous formulation of rational action and it is this upon which I wish to concentrate in this essay. Early on in Foundations (p. 13) he says, “for some purposes in the theory of this book, nothing more than a common sense notion of purposive action is necessary”. Furthermore, later (p. 18) he continues “[…] the theoretical aim of social science must be to conceive of action in a way that makes it rational from the point of view of the actor. Or to put in another way, much of what is ordinarily described as non-rational or irrational is merely so because the observers have not discovered the point of view of the actor, from whence the action is rational”. Although I am sympathetically inclined to this viewpoint it does, I believe need careful treatment. To describe actions as rational from an actors point of view hardly amounts to a predictive (explanatory) theory. It is difficult to see, for instance, what would amount to a refutation, unless the action under review is, on reflection, deemed not to be a choice (ie action) at all. For rational choice theory to claim anything beyond post hoc descriptive power, some additional constraints must be built in, rendering the theory internally refutable. How this might be achieved whilst remaining faithful to Coleman’s precepts is the subject matter of this essay.

There are essentially five types of puzzling empirical relationships for which a sociologist may wish to provide a conjectural explanation. A useful way of indicating these is by adapting a diagram introduced by Coleman in his Foundations (p. 10).

FIGURE I.

An adapted Coleman diagram

If we initially suppose that all four points in the diagram stand for (theoretically) described empirical states then each arrow depicts an empirical relationship all of which, in turn, may invite a conjectural theory claiming to show how the states become related. [2] The five types are :

Social system to system (type 4). An example might be, how the size of a community influences its social identity.
System to individual interaction (type 1). For example how the size of the community influences the pattern of interactions (ie who interacts with whom).
System to individual action (type 1a). For example, how the size of a community influences individual actions (independently of interactions).
Individual interaction to individual action (type 2). For example, how the pattern of interaction influences individual actions.
Individual (actions) to system (type 3). For example, how individual actions influence the system level of social identity.

I have deliberately used the imprecise word “influence” in the above list though the relationships between empirical states can be variously construed as causal, definitional or aggregation (type 3) and disaggregation (type 1) or even, in the case of type 3 as “emergent”. I shall start, however, by assuming causality for no other reason than it expedites the argument.

A few initial observations about the diagram are probably in order. First, it is normally the objective of sociologists, according to Coleman, to explain the system level outcome (ie the top right hand corner of the diagram). I shall assume so. Thus, it appears that three rather different sorts of (time distributed) causal account can be offered of the system outcome :

Solely in terms of system level causes (type 4);
The conjunction of types 1(a) and 3;
The conjunction of types 1,2 and 3.

The key question, of course, is whether type 4 causal relationships (ie system or social causality) can in any sense be deemed to operate independently of types 1 (or 1[a]), 2 and 3. [3] A reductionist will construe type 4 as the transitive closure of the other types. Similarly, the same query can be applied to type 1(a) in respect of types 1 and 2.

If, for the moment, we adopt a reductionist standpoint, then any puzzling type 4 empirical relationships will find their conjectural theoretical explanations in terms of what we might term the Weberian explication of mechanisms which account for the other three types of causal relationships. If any of the constituent empirical relationships are, however, not puzzling then they will enter the explanation in a theoretically unadorned form. It is, thus, worth noting that with this formulation a distinction can be drawn between, on the one hand, explanations whereby (ceteris paribus) one empirical state (eg size in the above example) causes another one, the latter comprising the explanandum (eg social identity) and conjectural theoretical explanations which take the connection itself as the explanandum. In the former sense we can envisage a number of intervening variables appertaining to any casual link.

It is of some moment that the bottom right hand corner of the diagram is designated in terms of “individual actions”. [4] It follows, accordingly that, if type 4 relationships (system causality) can be conceived as in some sense operating independently of the other types, then they do so without reference to individual actions (ie in a non reductionist manner).

If one is prepared provisionally to accept the five types of puzzling (causal) relationships as a starting point for any theoretical analysis, then two overriding questions arise. First, which model of individual actions and, second, which model of interactions should be adopted in order to explain the system outcome ?

In the context of the second question, one needs to distinguish between two very different approaches, namely parametric and strategic. The distinction should be central to social theory but unfortunately has not proven to be so (Abell, 1996). Parametric interactions arise when focal individuals’ (social) actions can be modelled as in some manner responsive to (caused by) other’s actions conceived as a fixed environment. The interactive environment provides, as it were, a stable point of reference for each focal actor in turn. Such models take on an appearance of “stimulus and response” where the stimulus is provided by the actions of others. Where, however, the interactions are strategic, the focal actor must reason about how others will react to his/her own actions (and assume the other actors are doing likewise) in selecting how to act. Strategic interactions are, of course, often best conceived in game-theoretical terms. When one might wish to interpret the causal relationships in Figure I (especially type 2) as strategic or as parametric is a key theoretical issue (Abell, 2001).

When adopting a theoretical model of individual actions, the issues become substantially cloudier. As my title suggests I wish, following Coleman, to make a case for reasoned/rational action theory (RAT). My argument will be that such a theory comprises the least bad starting point given the present development of theoretical sociology. In this respect I am, I think, on common ground with Coleman. Rational action assumptions will, however need some adjustment so that a more general approach narrative action theory (NAT), can be fashioned.

The problem when attaching rationality to the idea of action is that we possess no universally accepted concept of action, in the first place. I do not, however, wish to rehearse here the fine grained philosophical debates about the conception (Danto, 1985). From a practical sociological standpoint, as I suggested above (footnote 5) actions are distinguished from behaviours in terms of intentionality. The extent to which intentionality should embrace sub-conscious motives, I leave entirely open. But in addition I shall say that when an action is ascribed to an actor then it normally implies the actor also forbears (more or less consciously) to act in other available, (feasible) ways. This being the case, it is appropriate to speak of an actors preferences as between the action chosen and those forgone opportunities. Both RAT and NAT are preference driven or at least model the individual as if this is the case.

In moving from the broader concept of action to the more restricted conception of rational actions, three major additional assumptions are usually involved :

Methodological individualism
Optimality/Consistency
Self regard.

Each assumption has been subject to strong criticism by many sociologists. I shall deal with the first two only, here, noting that self-regard is, indeed, not intrinsic to RAT or NAT (Abell, 1992). Coleman (1990) deals with other-regarding sentiments (utilities) at a number of places in Foundations. I shall give most attention to optimality and consistency as its these assumptions which gives rational choice its distinctive capacity to predict actions.

Methodological individualism

I shall interpret methodological individualism to imply that type 4 (social causality) can always, in principle, be reduced to/deduced from either :

the conjunction of (causal) relationships of types 1,2, and 3;
the conjunction of (causal) relationships of types 1(a) and 3;
both (a) and (b).

On this reading type 4 causality is merely the transitive closure of the other types from which it is derived. [5] There is, correspondingly no sense in which type 4 social causality can be said to operate independently (or in addition to) the other types. The substantive point is that the motor energy of individual actions is necessary for the creation of any pattern of social causality. Individualism, therefore, implies that reduction of this sort can always in principle–even if not completely in practice– be made. This is, I think, Coleman’s standpoint.

It is perhaps important to separate this conception of individualism from others, some of which have been strongly criticised by various “social theorists”. The viewpoint does not imply anything like the “priority” of the individual over the collective. Indeed as Figure I should convey “social causes” (eg the impact of social norms or institutions can be accorded exogenous priority in explaining both individual actions [type 1 or 1(a) and 2 causality] and collective outcomes [type 3 in addition]). Similarly, nor does the individualism, as proposed here, (and by Coleman) obviate the need to postulate the “existence” of collective entities of one sort or another.

It may be instructive to quickly connect the present conception of individualism with a couple of alternative viewpoints, namely, “structuralism” and “emergence”. Although the word structuralism has been put to many purposes I shall interpret it to mean that level 2 mechanisms in Figure I are essentially random and further I shall call this Durkheimian. As such structuralism is a special case of “individualism”. “Emergence” I shall interpret to imply that mechanism 3 allows that the map from individual actions to a macro (state) outcome may be one to many (eg bifurcation in micro dynamics). Needless to say, we normally expect the map to be many to one (Abell, 2002).

Although “methodological individualism”, “reductionism” and so on have traditionally caused some considerable concern in sociological circles and have, in so for as they underpin rational choice, called the approach into question, the way Coleman formulated individualism implicit in Figure I should nevertheless not cause any major conceptual problems. It is not here that RAT is vulnerable but elsewhere.

Finally, even though individualism presses in the direction of reducing type 4 relationships to the others in the diagram, in practice it will often be either not possible or perhaps even uninteresting to effect the reduction. If, for instance, a relationship is found between societal characteristics (eg openness to trade and equality) than it is difficult to see how a micro reduction would prove worthwhile in anything but the most schematic form. In addition, as has long been recognised, system (level 4) regularities may show much more stability than individual ones. Indeed, we need to be open to the idea that random individual interactions can, with systematic aggregation, produce social level Durkheimian propensities (Hannan, 1992). I shall pursue this point in some detail in a moment.

It is, of course, possible to conceive of a “conjectural theory” operating solely at level 4, claiming to account for a puzzling relationship at this level. Such “non reductionist theory” may treat collective entities (eg social groups) as agents (ie as acting and forebearing) but it is clear that, at least in principle, such theoretical locutions must be reducible to individual level statements – only individuals can ultimately act. Alternatively, system level relationships may be seen as spuriously related in terms of a common (latent) theoretical state of the system (eg small countries, are more open to trade and more equal) or as refinable by the introduction of intervening variables. But once again, in both cases the ardent reductionist would want to construct an account of the impact of these variables at the individual level (Type 1 and 1[a] causality).

So where does this leave the individualism inherent in Coleman’s conception of RAT ? I think it would be wrong to always demand a reduction of type 4 relationships. But it is precisely when they strike us as not puzzling that they seem most acceptable. All that the individualism of RAT requires is that level 4 relationships are not formulated in a way such that if they were to be reduced they would be inconsistent with the other individual (micro) level assumptions of RAT. Although Coleman is not explicit in these matters I believe he would have agreed with this conclusion

Although the Coleman diagram (Figure I) was not Coleman’s invention it in effect provides an organising framework for a great deal of both his theoretical and empirical work. He was from early on in his career interested in cross level contextual and endogenous effects (Manski, 1995). Furthermore, he was rightly impatient with unrigorous social theory which merely coins terms without locating them in a technical understanding of what is implied. Perhaps his most lasting legacy will prove to be the example he set in this respect. Although the Coleman diagram looks, at first sight, rather simple and can certainly be used in a qualitative manner (eg p. 10 in Foundations), it, nevertheless, ultimately directs our attention towards the theoretical-technical underpinnings of the whole sociological enterprise itself. Indeed, it provides one way of reconciling respectively the Weberian and Durkheimian precepts about sociological inquiry. That is to say, the interpretive understanding of social action in order thereby to arrive at a causal explanation of its cause and effects, with an emphasis on the causal impact of “social facts, on other social facts” (ie level 4 and social causality). It is probably useful to see how this may be achieved.

In general we may interpret Coleman’s diagram in terms of hierarchical models (Raudenbush and Bryk, 2002) involving shall we say (macro) groups (j) and (micro) individuals (i). Let the action of individual i in group j be given by Y_ij and the “intensity of interaction” be X_ij (the bottom left-hand corner). For the sake of expository convenience I shall, for the moment, drop any mention of time and construe the “causal” relationships as cross-sectional (in practice this would not prove sensible). So at level 2 in Figure I we postulate a (linear) relationship:

(Again in practice there may be a vector of X variables). b_0j and b_1j vary across j and for convenience X_ij is measured as the departure from the mean of group j. Assume, in standard fashion, r_ij is a random variable.

Now allow b_0j and b_1j to vary across groups as follows:

where G_j and H_j are (macro) group level variables (eg size of group or some other macro variables in our earlier example about social identity). The g are fixed coefficients and m_0j and m_1j are, once again, random variables with the standard properties.

Equation (2a) now represents effect 1(a) and equation 2(b) effect 1, both in Figure I. G_j impacts the mean of Y_ij in group j and H_j the marginal impact of (X_ij – X_.j ) upon Y_ij. It is possible, of course that G_j = H_j. Combining equations (1) with (2a) and (2b) gives:

So, this represents the combined effect of 1(a) and 1 and 2, in Figure I on action Y_ij. If, as Coleman originally conceived, g₀₁ is null (ie no arrow [1a] in Figure I) then the term (g₀₁ G_j ) accordingly disappears in equation (3). Although equation (3) has a complex error structure standard iterative estimation methods exist (Raudenbush and Bryk, 2002). Effect 4, in Figure I, is also implicit in equation (3) as it will give j estimates of Y_.j the mean level of Y in group j (see below). Thus, as mentioned above, effect 3 in Figure I is then a matter of aggregation (of Y) within groups. Alternatively, however, a group outcome, as a function of Y_0j, or indeed, the distribution of Y scores may be postulated. The reader might be excused at this point in feeling all this is a little laboured and only of interest in so far as the correct estimation techniques is chosen in equation (3) which could, after all be set down directly. Note also, that if m_0j and m_1j, macro level random effects, are set at zero, equation (3) becomes a standard regression equation with an interactive effect and is then open to standard least squares estimation. However, deriving equation (3) in the way I have, does I believe bring into the open Coleman’s urgings about interweaving of theory and technique.

For example, Durkheim’s invocation of social facts becomes a direct implication, by virtue of certain reasonable assumptions, if we assume that Y_ij (ie individual actions) are purely random at the micro-level (effect 2 in Figure I) then from equation (1):

Here, there are no variables at the micro-level that have a generalisable impact on Y_ij. Now, under the assumption b_1j = 0, we are left with equation (2a). So:

If in addition as is sometimes assumed, equation (2a) is construed as deterministic then µ_0j disappears from equation (5). The random variation is then confined to the micro-level. This could I suppose be regarded as an ingredient of the Durkheimian approach. A special case of equations (5) is a “contextual effect” studied by Coleman where Gj = X_.j. Although equation (5) has an endogenous variable at the micro-level (Y_ij ) this will determine Y_.j at the (aggregate) macro level (g₀₀, is of course the grand mean). Note, however, we can easily mistake a situation with a non-random micro effect as Durkheimian (ie involving equation [4]). If equation (1) is recomputed with mean Y_.j as the outcome (ie the top right hand corner of Figure I) then:

If this is combined with equation (2a) with G_j = X_.j then:

This is a genuine effect of type 4 in Figure I, which does not, however take the micro impact of X, into account. The lesson here is that in our theoretical deliberations about the macro and the micro we need to be very careful how we specify our ideas. The Coleman diagram enjoins us to make our assumptions clear at both levels and in their joint interconnection. Unadorned type 4 relationships Figure I (ie Durkheimian) always require explicit assumptions about the mechanisms of type 2. Note, that in all those “Weberian models” where actions are (at least partially) determined by non random individual interactions, the macro variable H_j modifies the impact of X_ij upon Y_ij. Indeed it may be more faithful to these ideas to depict arrow 1 in Figure I as impacting arrow 2 (ie modifying relation 2).

If g₁₀ = 0 our equation (2b) then in the deterministic version when H_j = 0, b_ij = 0. So, in the absence of the macro variable the micro mechanism is null. This is yet another way of realising Durkheimian precepts.

In sum then the Coleman diagram gives a powerful way of conceiving the sociological enterprise itself. Indeed, it is perhaps more suggestive then even he realised.

Optimality/Consistency

Individuals do not in, fact, act optimally (rationally) according to many sociologists and this is undoubtedly sometimes–even perhaps often– true. So it is apparently here that RAT stumbles; but, then, how can Coleman be so insistent about the adoption of RAT ? Optimality implies that actions are chosen which are in some sense of the word the “best in the circumstances as conceived by the actor concerned”. Thus, on this reading, to say that actions are not optimally chosen is to imply that they are not deemed by the actor to be the best available. It would be strange (as Coleman notes) if this were to be predominately the case. But much rests upon how it is we interpret the phrase “conceived as best”. Furthermore, how, if at all, is the preference driven “rational” model related to the major alternatives often proposed, namely, rule (norm) following and emulation (copying). It is probably helpful to briefly review, informally, the standard approach to optimality in rational choice or action theory (ie essentially the derivation of subjective expected utility theory):

The theory runs somewhat as follows :

Actors have a complete transitive preference relation (ordering) over a grand-set A of alternative courses of action. Particular decisions involve a choice between the elements of feasible sub-sets of A.
The preference relation is in turn derived from a further preference structure over the consequences of each alternative and the (exogenous) mutually exclusive “states of the world” which are not under the control of the actor. It is important to recognise that uncertainty in this model derives from lack of surity about the latter state not about the consequences, once the state is fixed. It is assumed actors deterministically understand the causal consequences of the courses of action, given a state.
Actors (if they behave/act optimally) choose the best alternative from a (currently feasible) sub-set of alternative courses of action, given some prior on the state of the world.
A consistency condition is at the hub of the theory : it protects it against any sort of ad hocery across differing decisions/actions (pace Coleman’s earlier, commented upon, “less rigourous” interpretation of RCT). It states that, if a particular course of action is chosen from any feasible sub-set of A, then it will also be chosen from any sub-set of that feasible set which contains the particular course of action. This assumption although apparently innocuous, guarantees a fundamental feature of rational choice theory. That is to say, whatever determines the feasible courses of action available, at a particular juncture, does not also influence the preference structure (Abell, 1992). RCT thus, in practice, envisages the actor as in possession of a preference structure (over A) as he/she faces each particular decision (choices from sub-sets of A). Actors can update their knowledge of the world but this is not influenced by the feasible set they currently face. It should be noted though that the consistency criteria is compatible with a wider set of choice models than “choose the best”. For instance, it is easy to show a “satisficing model” (Simon, 1982), where choice is “at least as good as” an independently constituted threshold (target) value, is consistent with the criterion. It is not, however, easy to see how Coleman’s conception of “rationality” from the “point of view of the actor” is always consistent in the above sense of the term.

It is, however, in my opinion essential, if Coleman’s conception of rationality from the “actors point of view” is to retain any theoretical credibility (and refutability) that it should not only select the best action but also in a consistent manner from situation to situation. Optimality and consistency are both conjointly necessary even to the less rigorous version of RCT. If the best course of action were to fluctuate from one feasible set to another (inclusive) one then the theory would fail. And, indeed, it may do so and this is a pertinent strength. In learning situations, where the information available to the actor evolves, the inter-temporal consistency might not be observed. Though this invalidates a simple use of RCT/RAT it directs our attention to understanding of the impact of the changing information upon the preference structure. RCT has, thus, one of the characteristics of all good theories–namely that in its failure (or partial applicability) it suggests a deeper theory (in this case perhaps an evolutionary endogenisation of preferences). At the risk of pedantry–RCT/RAT, in either its rigorous or less rigorous forms– is refutable if an actor with fixed information is inconsistent in choosing a (best) course of action.

Although, sociologists, like Coleman, have urged the adoption of a RCT/RAT framework upon their colleagues, they have not made it explicit how much of the above outlined standard model of optimality/consistency they would wish us to embrace. Elster (1989), though to a degree critical of Coleman, has informally set out the lineaments of what we might term a more sociological friendly version. A simple diagram (Figure II) adapted from Elster might prove helpful in organising our thoughts.

FIGURE II.

Explaining actions

Therein actions are chosen in accordance with preferences which in turn are derived from (the conjunction of) beliefs and affects (or values). Further-more, beliefs are inferences form the state of information (about the world). In spirit, the diagram is largely consistent with the standard RCT model where the beliefs range over the (probabilistic) states of the world and the (deterministic) consequences of actions, but where affects are exogenous. [6]

We may now pose three questions in connection with the applicability of both optimality and consistency :

Given a preference structure are the actions optimally and consistently chosen ?
Are the preferences (which range over actions) optimally formed from beliefs and affects ?
Are beliefs optimally formed from the information (available or gathered)?

There are those, whom I above called structuralists, who wish to deny that a preference driven model is appropriate in the first place. It is not, for them, a matter of whether or not (a) in Figure II is optimal, but rather that the whole conception of preferences (and associated ideas like choice utility and so on) is misconceived. I will follow Coleman in assuming that preference driven actions and, thus, “best” actions are a central, even if not exclusive, concern of sociological theory. The problems which arise in connection with the selection of the “best” action are, however, many. The issue is, what does Coleman want us to understand, when choosing a “best” action, by his phrase –“from the point of view of the actor”. The answer to this question is not entirely clear in Foundations though we may perhaps capture Coleman’s intentions by assuming that we should model actors as choosing a “best” action from an individually (subjectively) constructed feasible set and preference/belief structure. This, without further constraints, may, however, condemn the theory to a purely post hoc descriptive role. To avoid the “subjectivity” making each decision unique it is still necessary to preserve the independence of the processes which determine, on the one hand, the feasibility set and, on the other, the preference structure. This is true, however, partial the picture is that the actor holds in an “objective” sense. So, if, the “subjectivity” could, in some way, be systematically endogenised we could then ask under what conditions the process are or are not compatible with rationality (ie consistency in the choice of “best” in the RCT sense). To achieve this objective it is helpful to think in terms of combining Figures I and II so that the “interactions” in mechanism 2 in Figure I determine “information” in Figure II which then becomes mechanism 2 terminating in individual actions at the bottom right hand corner of Figure I.

Figure I would suggest further that both preferences and feasibility sets should be derived from (explained by) independent specifications of mechanisms 2 and perhaps 1 (or 1[a]). To put it succinctly preferences and feasibilities are directly shaped either by interactions (mechanism 2) and, indirectly, by a macro level factor (mechanism1) or, directly, by the macro level factor (mechanism 1[a]). Recall also that mechanism 2 may be either parametric or strategic.

In order to address these issues a rather different (from RAT) model of actors (individuals) and their interactions which is rather closer to the socio-logical insights about how “best” actions are, in fact, reasoned is required. I call this Narrative Action Theory (NAT). I believe that most of its basic assumptions are consistent with Coleman’s views about RAT.

First, however any “theory of action” should have a number of desirable general characteristics :

A minimal (simple) cognitive structure; minimal in the sense that it will provide adequate explanations (by mechanism 3) of the macro outcome (Figure I).
Flexible assumptions; so that situations ranging from the qualitative to the quantitative and small to large N can be accommodated (recall Coleman specifies a minimum of N=2).
Be compatible with (a) expected utility theory (b) rule/norm following and (c) pure copying; in each case under additional assumptions.
Capable of embracing both strategic and parametric interactions (mechanism 2, Figure I).
Provide clear a solution as to the selection of a “best” action in accordance with the consistency condition.

In the spirit of Figure II preferences are fashioned from beliefs and (exogenous) affects (or utilities). The standard model of rational choice normally requires that beliefs be formed in a coherent way from the information available. Uncertainty, about the consequences of actions arise (as we saw earlier) entirely from uncertainty about the state of the world. The model requires the actor to hold in her mind a complete picture of possible states of the world and actions (set A) and their joint consequences. One interpretation of Coleman’s phrase –“rational in their own terms”– centres upon limited information. And, indeed the information may be constrained by other actors giving a starting point for theories of inter-actor power.

Many sociologists believe, however, that things go deeper. Preference formation they aver is almost invariably based upon past experience rather than upon hypothetically reasoned prospective consequences with limited information about the world. Furthermore, the experience arises, at least in part, from past social interactions with others (Figure I). So, actors form preferences over actions and their consequences of which they have previous experience, either personally or by observing the agency of others (ie their reference group). It is the experience of the connection between a situation (as we shall see below conceived as a narrative) and the evaluation of the consequences of an intervention (action) which guides an actor’s understanding of the causal implications of their actions; net of experience, these implications are unknown to the actor. As with expected utility we can envisage a super set of possible actions (A) but, when faced with the current situation (narrative) the actor draws upon her experience of a number of similar situations (narratives) each of which have lead to specific actions and outcomes. The set of actions deriving from similar narratives defines the current feasibility set in the particular situation. This set is, in turn, determined by patterns of social interaction. It might be useful to think of structurally equivalent actors (Wasserman and Faust, 1994) as having the same feasibility set given the current situation. So let us look a little more closely at NAT.

Assumptions of narrative action theory

(i) Individuals take actions within the framework of a narrative understanding of their current action environment. Narrative structures are defined formally in Abell (1987). Broadly speaking they can be depicted as a di-graph of causal relations (ie actions) on a set of situations.

Comment

The assumption envisages actors as selecting a course of action on the context of a more or less comprehensive (ie evolving) local “story” of how the actor sees herself as socially located. The story may have little or much history and little or much detail but, whatever, the actor conceives her prospective action as an intervention in the continuing story. The question facing the actor is, “what should I do now” (to include forbearance), in the current situation (narrative). That is, how should I make the next move in the narrative. Following, amongst others, Ricœur (1985), we assume that individuals are to be conceived as conceptualising their action environment in narrative terms whilst seeking a subjectively “best” and consistent intervention. The implicit claim here is that this conceptualisation is close to the way in which individuals do, in fact, think (Sarbin, 1986).

It does appear that individual actors often draw analogies with past (historical) experiences of their own and others in deciding how to act now. It is from such experiences that they construct mental models of “how the world works” (ie the likely consequences of interventions). This is equally true of the social world (where the consequences will often be the action of others) and the physical world. Even in the latter case, and where the “laws of nature” are known, every day actors view their options in terms of an intervention in an ongoing narrative process. This viewpoint is also easily reconcilable with the parametric/strategic distinction made above. In the latter case the intervention may be regarded as a move in a game in extensive form (either on the equilibrium path or off, in the case of learning). NAT is also consistent with the non-Humean concept of causality which regards actions as “observable singular causes” connecting states of the world not dependent upon generalisation and comparative method (Abell, 2001). Thus, action interventions in a particular experienced narrative deterministically yield an outcome and probability issues only arise in so far as in the context of similar narratives identical actions have in the past led to different outcomes. So in terms of NAT, Coleman’s exhortation to view actors as rational in their own terms, reduces to an understanding of how actors through either social interaction (link 2 or 1[a] in Figure I) acquire beliefs about consequences of interventions in narratively described situations. I believe Coleman would have been sympathetically inclined to this standpoint, as one possible route to realising his broad theoretical objectives. It remains, however, to see whether it can in principle yield a picture of optimality and consistency across actions/decisions protecting it against decision by decision ad hocery ?

(ii) Individuals select feasible actions (from a feasibility set) in terms of (a) their experience of past interventions (actions) by themselves and others (reference group) in similar (to include identical) narratives and/or (b) the evaluation of the consequences (outcomes) of those interventions.

Comment

The assumption implies that in any course of action, actors pay attention to their own past and to the past of a reference group each placed in similar situations (narratives) to their current ones. Furthermore, they may also take account of the value of the consequences of these past interventions. In this respect, interventions (actions) contribute to a path in a narrative (di-graph) connecting the action environment to the consequences. Thus, actors in some way form preferences over feasible actions in terms of a narrative understanding of past interventions and their consequences. Interesting questions arise concerning the boundaries of the reference group, the weighting of recency of experience and so on, but these matters fall beyond the scope of the present paper. It should be noted that uncertainty arises in this model because past interventions in subjectivity identical/similar narrative situations have been experienced as leading to different consequences. The consistency condition (which recall guarantees the independence of the process of generating the feasibility (ie similar set and preferences) may now be formulated over historical patterns of interaction (either directly or in terms of a reference group). So, if any course of action (ie a narrative interaction) is chosen from a given historical pattern of similar narratives then it should be chosen from any subset of that pattern containing the course of action. This is, I believe, one way of legitimating Coleman’s general conception of rational action.

The technical details as to how all this may be achieved need not detain us here (see Abell, 1987,1993). The essentials can be appreciated by considering the two very simple chain narratives depicted in Figure III.

The upper narrative (di-graph), comprising a sequence of “situations” which are transformed one to the other by actions, is mapped to the lower narrative comprising of two “abstract situations”, once again transformed by action. In general the mapping (technically a homomorphisms) will create equivalence classes on the nodes (situations) of the upper narrative which both preserve the connectivity whilst not creating connectivity when it does not exist. For connectivity, read a path of actions. This procedure captures the idea whereby actors compare narratives in order to detect any similarity/analogy. In so doing they pay attention to both the constituent situations and how human agency transforms them (ie causes change). In the example we may, for illustrative purposes, regard the upper narrative as a description of how the current situation has evolved. The focal actor shall we say is at situation 5 and poses the question should I now act in a certain way, recognising the mapping to the lower narratives and its outcome situation. The set of such mappings to other narratives and the actions taken, at the appropriate situation, defines the feasible set of actions. These are the narratives of which the actor has either directly or indirectly had experience.

FIGURE III.

Two simple narratives

It is probably not unreasonable to assume that, as in Figure III, the current narrative will contain more detail than any of its historical comparators. The latter, one would expect, because of inevitable cognitive limitations, to have relatively schematic structures. Whilst I pen these lines arguments, which find some similarity between Bush’s confrontation with Iraq and Churchills with nazi-Germany, are becoming common currency. I suspect most commentators will have little detailed knowledge of Churchills situation. Note that the current situation could be at 2,3 or 4, rather than 5, in the upper narrative. The lower narrative is, of course, complete in the actors memory. Technically, if a mapping exists between any pair of narratives then one exists between all sub-narratives. Substantially, if the actor now finds herself at situation 3 then she may by analogy, with the lower narrative, surmise that situations 4 and 5 will eventuate if actions 3 and 4 are realised. These could be commissioned by the focal actor or by others. Thus, actors when choosing how to act in terms of NAT will use similar/analogous narratives experienced in the past as a guide in making the next more in an anticipated sequence of (inter)actions.

It would be helpful if a measure of similarity of some sort could be constructed from the path-conserving mappings. If such could be found this would reconcile NAT with the many possible versions of case-based decision making (Gilboa and Schmeidler, 2001). I am a little sceptical about the feasibility of such an exercise and it may be that in the spirit of Coleman we should conceive of NAT in an entirely qualitative manner. Be this as it may, we might expect a binary similarity relation between narratives to be reflexive, symmetric, but not always transitive. The lack of transitivity arises because if narrative 1 is similar to narrative 2, which, in turn, is similar to narrative 3, it does not always follow that narrative 1 is similar to narrative 3. Path conserving mappings and their inverses are reflexive, symmetric and transitive. Given a set of narratives, path conserving mappings will generate a set of equivalence classes (Abell, 1993). Additional constraints are needed to weight “nearness”. The intuitive idea here must be : how much distortion/alteration does a mapping of one narrative into another introduce. The greater the distortion the less the similarity. I leave open here how this might be achieved.

(iii) Preference formation over feasible actions (interventions), thus, take place in the context of :

The current narrative,
Reference narratives (homomorphically similar to [a]) and interventions,
Outcomes of the interventions (b),
Some affect/evaluation over (c).

Comment

The technical details of how (a) to (d) might be combined into a ranking over feasible actions are not my main concern here. Differing additional details concerning both the metric (or otherwise) properties of similarity and value allow us to think in terms of a family of models running along a continuum from homo sociologicus to what we must term past referencing homo economicus (Abell, 2003,2004). [7]

Homo sociologicus interventions/action rules

Select as the best action, the most commonly chosen action in the feasibility set (ie ignore the degree of similarity and evaluation). A probabilistic version would be to choose the action in proportion to the relative frequency of actions taken in the reference set (Bahr and Passerini, 1998). Homo sociologicus is, thus, a pure density dependent actor. Cognitivity, given the current narrative situation, the actor surveys, all those past narratives which are to any degree similar and selects the action (intervention) most frequently used, independently of the degree of similarity. The decision rule is also independent of any evaluation of the expected outcome. Norm following will take this form as will density dependent emulation.

Consistency–which it will be recalled is required of any theory which is to avoid ad hocery– amongst the best actions chosen across different situations will now run as follows : if an action is chosen as the best one from a given reference set of narratives then it will be chosen from all subsets of the reference set where the action is feasible.

Homo economicus interventions/action rules

Homo economicus will incorporate an evaluation of outcomes (ie utility) and a similarity measure into his calculations. So :

Select as the best action the one with the maximum similarity weighted value/utility in the reference set.
Select as the best action the one with the maximum similarity weighted average value/utility in the reference set.

The consistency condition will be identical to the one appropriate to homo sociologicus. Thus, the consistency condition will vary along the above mentioned continuum.

Coleman’s Foundations will probably prove to be one of the most significant contributions to sociological theory written in the last decades of the twentieth century. His model of the interplay of the micro and macro levels of analyses provides a general framework for sociological analysis and in so doing clarifies the assumptions behind Weberian and Durkheimian conceptions. Coleman’s analyses are based upon both a specific and a more general conception of rational action. I have argued that the latter may turn out to be the more influential. As formulated, however, it is too broad. This being the case I have outlined a model of narrative action theory which, may, I hope, begin to contribute to a more focussed and predictive theory.

BIBLIOGRAPHIE

· Abell P., 1987. – The Syntax of Social Life : the Theory and Method of Comparative Narratives, Oxford, Oxford University Press.

· — 1992. – “Is Rational Choice Theory a Rational Choice Theory ?” in J. S. Coleman, T. Fararo, Rational Choice Theory. Advocacy and Critique, Newbury Park, Sage Publications.

· — 1993. – “Narrative Method : a Reply”, Journal of Mathematical Sociology, 16, pp. 253-266.

· — 1996. – “Sociological Theory and Rational Choice Theory” in B. S. Turner (ed.), Social Theory, Oxford, Blackwell.

· — 2001. – “Causality and Low Frequency Complex Events”, Sociological Methods and Research, 30,5, pp. 57-80.

· — 2002. – Narrative Action Theory, London, London School of Economics [Mimeo].

· — 2003. – “On the Prospects of a Unified Social Science : Economics and Sociology”, Socio-Economic Review, 1, pp. 1-27.

· — 2004. – “Narrative Explanation : an Alternative to Variable Centred Explanation ?” Annual Review of Sociology [Forthcoming].

· Bahr D. B., Passerini E., 1998. – “Statistical Mechanics of Opinion Formation and Collective Behaviour : Micro Sociology”, Journal of Mathematical Sociology, 23,1, pp. 1-27.

· Coleman J. S., 1990. – Foundations of Social Theory, Cambridge (Mas), Belknap Press of Harvard University Press.

· Danto A. C., 1985. – Narration and Knowledge, Morningside, Columbia University Press.

· Elster J., 1989. – The Cement of Society, Cambridge, Cambridge University Press.

· Gilboa I., Schmeidler D., 2001. – A Theory of Case Based Decisions, Cambridge Cambridge University Press.

· Hannan M. T., 1992. – “Rationality and Robustness in Multilevel Systems” in J. S. Coleman, T. Fararo, Rational Choice Theory. Advocacy and Critique, Newbury Park, Sage Publications.

· Manski C. F., 1995. – Identification Problems in the Social Sciences, Cambridge (Mas), Harvard University Press.

· Raudenbush S. W., Bryk A. S., 2002. – Hierarchical Linear Models, Applications and Data Analysis Methods, Thousand Oaks, Sage Publications.

· Ricœur P., 1985. – Time and Narrative, Chicago, Chicago University Press.

· Rubinstein A., 1998. – Modelling Bounded Rationality, Cambridge (Mas), MIT Press.

· Sarbin T. R., 1986. – “The Narrative as a Root Metaphor for Psychology” in T. R. Sarbin (ed.), Narrative Psychology. The Storied Nature of Human Conduct, New York, Praeger.

· Simon H., 1982. – Models of Bounded Rationality, Cambridge (Mas), MIT Press.

· Wasserman S., Faust K., 1994. – Social Network Analysis, Cambridge, Cambridge University Press.

NOTES

[(1)]

This part of Coleman’s book comprises an ambition attempt to bring something like the full rigour of general equilibrium theory into the ambit of social exchange.

[(2)]

Figure I, without the 1(a) arrow is often named the Coleman diagram. I have modified it by including 1(a) and limiting the micro level to the relationship between interaction and actions (Abell, 1992,1996,2002).

[(3)]

Coleman using a rather more general when two individual actors are involved in model than the one depicted in Figure I insists generating the system outcome. that sociological issues arise, at a minimum,

[(4)]

Type 4 could be the lower level in a “higher level” (causal) diagram. Thus, communities could be pictured as in a micro/macro relation with “societies”. Then communities would in a sense be at a “meso-level”. Hannan (1992) argues for the systematic analysis of multi-level systems.

[(5)]

I use the term action in the standard way as “intentional behaviour”. It is usual these days to interpret this phase as covering unreflective behaviour that can ex post be given an intentional purpose. Such an approach does of course require us to demarcate ex post rationalisation from ex post explication. Coleman does not exclusively designate the bottom right hand corner of his diagram in terms of actions.

[(6)]

With deterministic models it is possible to speak of deduction, with probabilistic models derivation is the more appropriate word.

[(7)]

Past-references homo economious because the actor does not indulge in forward-looking hypothetical reasoning.