Edmund Chattoe
Department of Sociology, University of
Surrey
Guildford, GU2 5XH.
econec@vax.oxford.ac.uk
28 February 1995
This research is part of Project L
122-251-013 funded
by the ESRC under their Economic Beliefs and Behaviour Programme.
This paper addresses three linked difficulties in using economic and
sociological theories of consumer decision-making as the basis for a
computational model. The first difficulty is the non-operational
nature of many of the theories. Their explanatory power cannot be
assessed using data that can actually be obtained. The second difficulty
is that of grounding, of what a given theory rests upon by way of
lower level constructs and explanations. This gives rise to the final
difficulty, that of reconciling both the aims and methods of economic
and sociological theory. In each case, the computational perspective
provides a measure of clarification and potential for development.
This paper arose as a result of a literature survey on the economic and sociological theories of consumer decision-making, in the early stages of a project to construct a computer simulation of the budgetary decision process. The original plan was to choose a suitable theory, implement it as a computer simulation model and use interviews and diary research to provide a rich source of data and suggestions for theory development. In fact, both the prevailing economic theory, and a large amount of sociological material proved unsuitable for this purpose. The process of understanding why much of this theory was unsuitable suggested an important role for computational simulation and cast some interesting highlights on the methodological differences between economics and sociology. As an economist recently arrived in a sociology department, I realised that my attempts to explain myself and understand others were themselves a suitable object of social research. This paper is both a result of that process and a contribution to it.
The paper is divided into three parts. In the first, the economic possibilities for consumer theory are considered, and the deep difficulties of operationalisation and grounding are addressesed. In the second part, the contributions of sociology are considered. It is argued that these are currently non-theoretical, at least by economic criteria, but nonetheless extremely important. (Economics has understressed their importance because they cannot readily be fitted into its theoretical framework, but this reflects neither their true importance, nor an accurate assessment of the defensibility of the resulting economic theory.) Finally, the possibility for a reconciliation of both views is suggested within the (non partisan) computational framework.
Methodological prescriptions are often unsatisfactory because they simply descend into rhetoric. It is almost impossible to decide on the correct way to do science that has yet to be done. Instead, I shall consider science that has already been done using the traditional methodology and explain how it has led to the paradoxes that now exist. In suggesting an alternative and illustrating the way in which it avoids these paradoxes, I shall argue that it is therefore a better way to proceed.
The economic theory of consumer choice (1) posits a preference ordering over a specified set of goods, a set of ``axioms of rationality'' and a budget constraint. The axioms of rationality serve two purposes within the theory. Firstly, it is suggested that they have behavioral content as descriptions of consumers. For example, they are often ``supported by argument'' in economics textbooks where everyday examples are used to suggest that the choice of axioms is based on ``common sense'' or plausibility rather than the demands of the theory. Secondly, they serve a theoretical and pragmatic function, allowing any preference ordering satisfying them to be expressed as a ``utility function''. When tastes can be expressed as a well-behaved function, decision problems become suitable for solution using the framework of the calculus.
This model, which dominates economic consumer theory, raises a number of problems that can be divided into limitations, obscurities and paradoxes. The limitations are the least damaging to the theory, they are simply under-developed aspects, like the treatment of goods which are durable or public (2). Removal of the limitations involves straightforward generalisation of the theory and constitutes a large part of the ``normal science'' practised by consumer theorists (3). By contrast, obscurities and paradoxes should raise genuine concern, not simply with the state of development of the theory, but with its suitability as a description of real phenomena.
Obscurities are more difficult to address, since by their nature they are obscure. This is not a truism, part of their obscurity is that they have not even been identified in the description of the theory. The term obscurity is here used to refer to different interpretations, left unresolved by the theory, which hamper the process of conversion into an algorithm suitable for computer simulation. (Building a computer simulation is an extremely acid test for the completeness of a theory. The vocabulary of computer languages is sufficiently expressive that a failure to develop an algorithm can safely be attributed to incompleteness in the theory, rather than an inability to specify it computationally.)
Paradoxes are the most damaging type of difficulty, since they suggest that the theory may actually be incoherent, rather than simply incomplete or unclear. A paradox consists of a pair of interpretations that cannot be true at the same time. Obvious paradoxes seldom survive long within a theory, especially in fields like economics where formal representation is common. Instead paradox is often associated with one or more obscurities about the theory.
Plainly, these classifications overlap and different classes of difficulty are often found together. Obscurities often conceal paradoxes and the attempt to remove limitations can often reveal, or even cause, further obscurities. Nevertheless, the classification performs a useful task and in the next three sections, it will be used to structure a discussion of the major difficulties in economic consumer theory.
The fundamental limitation, that still shows little sign of being resolved, is that the theory is static in two important respects. Firstly, it says nothing about the way that decisions change dynamically as prices and other determinants of demand change. In a complex and dynamic environment, this seems to be a major weakness unless some convincing argument is brought forward to justify it. (It is true that the transition from one static state to another can be represented on a consumer choice diagram, but the theory itself does not determine the path which the transition will take.) Secondly, while admitting that preferences themselves can change, the theory has nothing to say about the determinants of this change. Even when modelling changes that are ostensibly ``intertemporal'', one of the routine extensions of the model, the ``time'' so described has none of the features associated with the usual meaning of ``time'' such as irreversibility and genuine uncertainty, rather than risk. The economic notion of time could rather be described as an ``extended present'' (4). These limitations appear to arise from the positivist tradition in economics. What goes on ``inside the heads'' of agents is simply not the province of economists. Unfortunately, as we shall see the ramifications of this positivism are wide ranging and, indirectly, extremely damaging to the integrity of the theory.
For example, because the theory is static, and excludes any mechanism of preference change, while admitting that such change is possible (and indeed observed in econometric studies), it is impossible to distinguish a violation of any number of axioms of rationality from a simple change in preference (ref 5, page 69). This is a problem, not simply of inadequate data, but of genuine non-identification, that is to say that no amount of additional data could resolve it. (The ceteris paribus assumption is not an acceptable defence because the method for establishing when that assumption is valid is not included in the theory either, thus introducing another degree of freedom.)
The most general obscurity of economic consumer theory is in the relations of theory constructs to objects outside the theory, which they are presumably intended to represent. An example would be the status of the axioms of rationality. Are they supposed to represent the process by which individuals actually make decisions, or are they supposed to be some sort of summary of other, unspecified, processes by individual consumers which, in aggregate, result in the observed ``rational'' behaviour? This is popularly referred to as the ``as if'' argument. It is not that agent carry out the explicit calculations required to optimise, but it is somehow as if they do (5). If the latter, then what confidence do we have that the unspecified processes will invariably achieve the desired result? If the former, then what admissable evidence can we provide that the mechanism is the one actually used by consumers? For each interpretation, the theory is equally silent and the simulation representation will have to differ. In fact, the ``as if'' argument is not typically an additional argument at all, merely a reiteration.
A similar problem exists for the specification of a ``bundle of goods''. For heuristic purposes numbers of oranges and apples are often used, but it is commonly asserted that the theory can be applied to more abstract objects like ``lifestyles'' and that a single good can be set against ``all other goods'' on the choice diagram without loss of generality. The plausibility of the axioms of rationality is seldom reconsidered after this generalisation, suggesting a return to the ``axiomatic interpretation'', yet it would seem that in certain circumstances, for example the moral preference ordering suggested by Sen (ref 20) or the theory of needs developed by Maslow (refs 16, 17, 18), the generalisations make the axioms seem even more implausible. (A lexicographic hierarchy seems almost constitutive of moral judgements, where, for example, in law, no amount of money can justify a murder, though an assault might.)
Generally, it is not clear how we should identify goods or budget constraints in the positivist manner implied by the theory. Certainly if identification is taken in the strictly econometric sense, in terms of stability of parameters, out of sample prediction and so forth, current attempts to identify the determinants of consumer demand fail rather badly. We can certainly observe what people do, with some difficulty, but that gives us no insight into the descriptions under which it is done and it is these categories, even in the basic economic theory, that will determine the choices of the agent. If the agent allocates a budget to ``food'' or separately to ``starters'', ``main courses'' and ``desserts'', the resulting behaviour will be different in each case, for example if the price of mince increase (6). Prawns and custard may be substitutes as food, but not as dessert. To express the difficulty in strictly positivist terms, the classifications used by agents provide additional information that is being thrown away by the theory. (One might go further and question the inferior status of the agents' classification as data on the grounds that it lacks a justification. The difficulty of measuring something accurately does not reveal anything about its importance.)
In fact, the problem of identifying goods has been made worse by the economic attempt to defend itself against accusations of naive positivism. Acknowledging that it is not true that a rose is a rose is a rose, economists suggest that goods should be ``indexed'' by time and place. Thus a cup of coffee or a glass of lemonade on hot and cold days are actually different goods. (Many of the objections to the axioms of rationality tell stories where the nature of the goods is said to be ``incompletely specified'', though this rather implies that complete specification is a feasible operation.) In fact, this indexation solution actually heightens the problem, for what is it that now relates demand for a glass of lemonade on a hot day to demand on a cold day? Now that they are separate goods by assumption, what do they still have in common? Plainly it is not their ``lemonadeness'' and ``coffeeness'', that would simply be a return to positivism, but without some linkage, we again have a non-falsifiable theory with a vengeance, as no two goods are ever the same (7).
The problem of objective identification also applies to the position of the budget constraint. Theory often begins by asserting that this is objective and can in principle be observed directly. This assumption is then relaxed to allow for the agents' perception of the constraint. It is not clear that this perception can be deduced objectively, for example in the case of credit facilities, since the agents' beliefs about the credit facilities available may be complicated and mistaken in a wide variety of ways. Furthermore, the concept of a constraint is itself not a robust one, in that the increasing cost of credit does not represent a hard distinction but an increasingly blurred area where confidence and social interactions play an increasingly dominant role (8). Yet no guidance is given concerning how this subjective perception is to be assessed within the terms of a theory based on observed purchases. In fact, the acceptance of a subjective budget constraint seems to destroy the theory altogether. With both utility and budget subjective and incapable of independent measurement, the theory is unidentified and therefore non-falsifiable using behavioral data alone. (If it makes sense to talk of a theory being even more unidentified.)
In the last section, I described the ambiguous status of the rationality axioms as an obscurity, but it also involves a paradox. If the axioms are regarded as behavioral assertions, and defended as such, then they could simply be false, and much empirical evidence suggests that they are commonly violated (ref 11). Unfortunately, the axioms must also be treated as theoretical constructs that cannot be false, else the theory will typically fail to produce any prediction. It is possible to observe the rhetorical status of the axioms changing depending on the criticisms levelled at the theory. If it is accused of a lack of realism, the theoretical need for the axioms will be emphasised, while if the potential emptiness of the theory is remarked, support for the behavioral plausibility of the axioms will be provided. It does not appear that both these interpretations can be justified at the same time (9). This observation also makes clearer the distinction between obscurity and paradox. The different possible interpretations of the rationality axioms do not simply constitute an obscurity, it is not just that it is not clear from the theory which interpretation is the correct one, rather that the theory seems to require both interpretations to be correct at the same time in order to remain credible.
Lest it be thought that all this is mere hair-splitting, it does not appear that the economic theory even meets its own rather distinctive criteria particularly well. Empirical work translates consumer theory into a form whence it can be empirically tested by the estimation of demand functions (ref 4). A demand function relates the quantity of each good purchased to the prices and quantities of other goods, though in principle they can include other determinants. The three minimal requirements, adding up, that the quantity of all goods purchased should satisfy the budget constraint, homogeneity, that a doubling in all prices and income should result in the same consumption pattern and symmetry (10) are all rejected in almost all econometric models where they are not simply imposed. Whether these requirements are sensible ones is, in a sense, almost beside the point, they are the ones by which economics judges the achievements of its consumer theory.
These achievements, or their absence, also cast some light on the goals of consumer theory. These are to produce rigorous, typically mathematical models that will explain the response of consumers purchases of goods to changes in such variables as price. The quality of these theories is to be tested using an econometric approach according to such criteria as out-of-sample prediction, parameter stability, noncorrelation of errors and so forth. The possibility of satisfying these objectives by computational means will be discussed in the last section.
There are at least two possible objections to this description of economic consumer theory that would argue that I have simply misinterpreted or misrepresented it. The first suggests that such a model cannot be taken to apply generally, but remains useful where it does, in fact, apply. Apart from the casual observation that the tendency to expand the alleged applicability of consumer theory is far greater than any contrary tendency, the difficulties I have suggested actually makes it impossible to decide, empirically at least, where the model applies and where it does not. That is the point about models that cannot be falsified. A second objection comes from those who take a sort of ``empirical axiomatic'' approach. This views each set of axioms as a hypothesis without any privileged logical status, though that is often forgotten in the rhetorical mode, and seeks a `plausible' set of behavioural rules to restrict preferences in certain ways. This approach can answer the objection to ungrounded theoretical terms by making use of concepts that can be demonstrated in other behaviours than the ones they explain, but unless it does so, the issue of non falsifiability still arises, as do the questions of ungrounded preference and reciprocity. (There are also pragmatic considerations. Once it is appreciated that these `axioms' actually have no special status, but are merely hypotheses, even if they are plausible ones, the axiomatic method seems a pretty thin basis for a theory of rationality in dynamic decision making. What other advantages remain to justify the claim that the research exercise should take the form of specifying sets of axioms?)
All the difficulties discussed so far can be summed up as problems resulting from an attempt to be positivist at any cost, about a process that plainly involves a cognitive actor (11) as its central and crucial feature. Indeed consumption is inextricably linked to the presence of a living creature who ``consumes'', whether by something as concrete as ingestion or as abstract as perception or appreciation. When we talk of the sea ``consuming'' a rock, we are being deliberately metaphorical. To attempt to strip preferences (and their relationships) away from the person in whom they inhere, and to attempt to define goods without reference to the dynamic cognitive process of that person, is simply to abstract these words from any relevant meaning, to reduce them to a formal system. When science involves the study of persons, naive positivism is not only not objective but frankly totalitarian. (If the model requires a conscious agent, but method forbids you to consult other agents, you simply tend to model yourself unthinkingly. (12))
The side effects of this positivism have already been discussed in the critique of the economic theory, but the issue of reduction must also be raised. The economic model not only abstracts from the entities it models, but removes important causal links between them, for example the link between action and preference mirroring the link between preference and action. From a computational point of view, causality and ontology are equally important in creating a working model, but economics appears to hold the view that removing causal linkages from a model is on a par with the process of abstraction. In fact, it appears to change the whole nature of the model. For example, without a link from action to preference, where do preferences originate? The insistence on linear causality leaves the ends of the causal chain unsupported. This is the problem of grounding in economic theory. The economic attempt to answer this question illuminates its relationship to sociology, discussed in the following sections.
This reductive approach to causality explains not only the problem of groundedness, but that of non-falsifiability. The model of consumer theory is not closed by any causal link back from choice to preference. There is a unidirectional causality postulated from primitive preference, restricted by axioms, to choice, which is the only thing actually observed. Since both axioms and preferences may be varied without constraint from any other part of the theory, including observation and previous choice, how can a falsifying choice ever be made? This is another way of pointing out the absence of a mechanism for dynamic preference, the `reduction' of an interactive mechanism to a linear causality destroys falsifiability because the model cannot be `projected forward' into the future to generate predictions that could be falsified (13). All data is recorded in the ``extended present''.
One interesting side effect of this reductive attitude to causality is the tendency to build linear or hierarchical theories (14). (It is not clear whether the appeal of hierarchy results from other factors or whether, once removal of causal links begins, hierarchy is the most satisfying place to stop.) For example, dissatisfaction with the `thinness' of the preference ordering has led, not to attempts to build a substantively different theory of preference (though there are a few exceptions (refs 2 and 19), but to the postulation of `higher order' preferences that have been equated with morality, community or longer term goals (ref 20). As has already been remarked, enriching a theory in this way, without the acceptance of additional data, weakens rather than strengthens it. The method is reminiscent of the cosmology of a world supported by an elephant on the back of the tortoise. Under the new scheme, we must not only identify preferences and axioms using behavioral data alone, but the preferences must be divided into two sets, one of which overarches the other in a specified way (15).
All these causal simplifications and methdological assumptions in the economic theory of consumer behaviour seem to be abstracting out from what is unique about intentional systems, such as persons. (This may be an unconscious attempt to preserve the operation of physical law in social systems.) By removing reciprocal causality, we prevent the possibility of feedback and learning (16). By describing the self solely through preferences grounded exogenously to the model, we ignore the possibility of a permanent cognitive `self' which initiates, plans and possesses goals. (A self is also affected both by its preferences and actions. Only with a permanent self can we make any sense of morality as other than extended self-interest. This is another way of expressing concern with the non-falsifiability of changeable preferences. Without a self as a stable and independently observable object, there is just no possibility of doing what you hate, because there is nowhere for what you feel about what you do to reside separated from what you do. We are not simply our preferences, they are things that arise within us and cannot be separated from their point of origin.) By disregarding the distinction between intentions and actions, we also dispose of the agents' model of the world as a determining factor, either supposing it to be non-existent, or invariably correct. Many of these disregarded features of the agent also have significant social components, and it is to these we turn in the next section.
After the last section, the average sociologist may be feeling as if they have just listened to an extensive astrological explanation of the reasons why the earth is flat and wondering why they bothered, except as an interesting piece of academic pathology. While admitting that both the methodology and the theories described are deeply flawed, I nonetheless believe that the allocation of resources in consumption is a worthwhile topic for social science as a whole and that it is not simply an artifact of the economic theory (17). Furthermore, although there isn't adequate space for more than a sketch of the reasoning here, I believe that an attempt at scientific objectivity including building models that predict and may be falsified, need not involve these fallacies of naive positivism.
The major approach to the (largely self-inflicted) difficulties of groundedness and non-falsifiability has been to take preferences as `primitive' within the economic theory. (This may also explain why no feedbacks to preference are considered.) This assumption is sometimes expressed by the claim that the explanation of preference is not the proper province of economics, but should be allocated to psychology or sociology. This approach seems to offer a considerable hostage to fortune. There is no reason to suppose that sociologists will be able, even if they are willing (18), to continue the causal chain by providing explanations of preference that are, in turn, dependent only on the findings of, say, psychologists or physiologists (19). (Such a strategy does not really solve the problem of grounding in any event, it merely postpones it, becoming more and more implausibly reductive, until the actions of persons have been reduced to particle physics.) This approach suggests an implicit belief in the hierarchy of the sciences (another reductive and positivist concept) much the same as the postulated hierarchy of preference orderings. (Such a belief might also go some way to explaining economic imperialism, though not why economics is presented as the `top' science. We might seek sociological and psychological explanations for that claim!) It would indeed be fortunate if the sciences had `ordered' the phenomena they chose to study in this way before constructing their boundaries. Typically, however, the boundaries of a science, which are the most strongly defended, are their most contingent feature. More generally, sociology strongly rejects the sort of positivism implied by this ``division of labour'' in stressing the interpersonal construction of reality and the socialisation of persons. (Sociology has also struggled extensively with the apparent conflict between individual action and social structure.) On a less theoretical level, it may be extremely hard even to find a useful common language for discussions across this disciplinary ``boundary''. There is no method by which the discussion of primitives like ``preference'', defined in the economic fashion, can be made ultimately theory free. In this view, economists seem to be determining not only where the boundary lies between disciplines, but which methods must be used to communicate across that boundary, and what (potentially subordinate) role sociologists should play in the discovery process. This ability to set the rules, an exercise of power, appears to be based on the assumption that the economic theory and methods are more or less correct and therefore that further consultation is unnecessary. The situation of sociology, in the view of economists, becomes that of a ``science of leftovers'' (ref 3, page 34).
Rather surprisingly, this situation appears to have been accepted by sociologists, at least until very recently (20). The discipline of ``economic sociology''appeared to accept the claim that the role of sociology was to act at a different level to economic theory in providing explanations for tastes and other economic ``primitives'', perhaps in terms of class, gender or socialisation (ref 15, pages 273-284). Even now, there is a strong tendency to separate attitudes and patterns of consumption from the general process of consumption itself. Descriptions of concepts like ``consumer culture" (ref 6) often operate at an extremely high level of description, at least relative to the corresponding economic concepts.
This distinction is not surprising. Apart from the obvious advantages of a division of labour, sociological observations typically cannot be incorporated into the mathematical models favoured by economists. This is in part precisely because they stress the complex and interactive nature of social forces. By contrast, and rather worryingly, economics has shown a tendency to encroach on sociology, substituting rigour for a full appreciation of social richness (ref 1). Nonetheless, it is perhaps surprising that there is no ``separate take'' on consumption that could be called a sociological theory with the same level of generality as the economic one.
Despite the absence of a single theory of consumption, the sociological contributions to the creation of such a theory would be indispensable. The process of socialisation, by which an agents model of the world is constructed through education and interaction, is fundamental to the understanding of any cognitive tasks subsequently performed. The modelling of situations where the actions of others are to be interpreted, via models of their models, based on their behaviour, requires precisely the notion of interpersonal construction of reality urged by sociology. The need for a theory of this type includes the possibility for explaining such fundamental behaviours as communication and collective action. The complex relationship between independent action and social structure is also addressed almost exclusively by sociologists. On this view, sociological insight is not subordinate, or even at a different level, but impacts directly on the choice decisions of consumers and must be part of the model of that behaviour. All these social issues have been consistently underplayed by the economic theory of consumption, without adequate support for the claim that they are unimportant. What arises instead, is an impression that they are disregarded because the theory cannot contain them. The reasons for this will be discussed in the next section. (Nonetheless, the existence of this same ``rigorous'' theory is urged as one of the main strengths that economics has over the other social sciences, though it disregards precisely those features that illustrate its limitations.)
The same situation exists in method as in theory. A practical example is provided by the process of unstructured or semi-structured interviewing, common in sociology, but extremely rare in economics (21). Interviews reveal significant diversity, not just in goals and information, but in attitudes and methods. The amount of variation encountered, and the fact that the resulting descriptions are coherent and not simply ``irrational'', suggests the complete implausibility of models that postulate a single type of axiomatic rationality for all agents, whether bounded or not. (This is not to say the models may not be constructed that attempt to draw out and validate underlying structure in these divergent models, merely that a single model constructed without reference to the full range of possibilities will almost certainly not be correct.)
I would argue the absence of a usable theory in economics and a single theory in sociology reflect the impossibility of either science creating a theory alone, or in isolation from psychology and cognitive science. The difference appears to be that sociology is more aware of the difficulties of creating a rigorous social theory, and is more content to draw attention to a multitude of causal links and patterns that can nonetheless be supported empirically. This is in keeping with acceptance of the subjective and complex nature of social action. By contrast, economics as a reductive and positivist science tends to argue that the features that it chooses to model are the only important ones, and to aspire to test that claim using the methods of falsification. The dangers of each strategy are apparent and observed. In economics, the removal of diversity and the increase of dogmatism. In sociology, the difficulty of comparing one well told tale with another and the profusion of partial ``theories'' that are hard to integrate into a conceptual whole. (Integration is not the same as reduction.) In turning to the last section, on the computational approach, I will argue that economics confuses its intuitions about social action, which are valid but complementary to those of sociology, with the structure it uses to support them. Furthermore, not only does this structure encourage the rejection of variety and complexity, but it is not actually necessary to rigorous theorising, even as defined by economists.
Having rejected an arbitrary separatist approach to social science, and considered the differing attitudes and problems of sociology and economics, I will now develop the idea that the two subjects are actually on an equal footing in the analysis of any given phenomenon, provided they are seen within the framework of building computational simulations of cognitive models (22). This involves a mentalist view that may appear incredibly naive, simply that social practices, on a par with reasoning processes, should be seen as residing predominantly, possibly exclusively, inside the heads of one or more individuals. Social actions result from the models that agents have of the world and these actions give rise to new models, yet the models are prior because it is agents that appear to possess the independent power to initiate action and modify the world. Indeed, it is this power that we use in identifying agents. Of course, there are many artifacts that can also be said to embody social practices and modes of reasoning, everything from computers to frescoes. Nonetheless, it is still coherent to maintain that these derive their significance, though not their existence necessarily, only from the activities of those who recall or perceive them (23). Furthermore, this view refers only to social practices, it does not imply that there is anything at all insubstantial about physical actions or states of the world that may arise from them. We can still talk of the power of a man with a gun and not imply that his power is merely a mental construct. On the other hand a socialisation that denied a role to some social group would be an example of a disempowering social structure. This structure could, in fact, be backed up ultimately by the gun, but it need not be so.
This approach seems both very much in keeping with the idea of inter-personal construction of reality, and of the fundamental uniqueness of individuals, without denying the possibility of social action and convergence through any number of channels: obedience, affection, coercion, imitation and so on. Nonetheless, it also recognises the large number of social behaviours that can arise when models of the models of others are incorrect or incomplete. The tension between the uniqueness of individual models of the world and the processes of convergence implied by social action is a key feature of the computational approach. There are systems by which certain constraints become perceived as fixed, and others as amenable to change, and it is within this framework that actions take place. At the same time, the resulting actions may change other agents perceptions of what is possible and impossible. The idea of socialisation can be seen as the process by which new agents are provided with models of the world suitable (or unsuitable) for the society in which they live. Agents models of the world interact both with the world and with each other through visible actions and we may aspire to model this process and, possibly, even to predict it. If there are any ``natural breaks'' in the structure of the cognitive model of the world, and the limitations on human processing capabilities suggest that there may be, we would be very bold to suppose they would coincide with disciplinary boundaries. This view of social science is simultaneously daunting and inspiring. On one hand, it instils a healthy humility about the magnitude of the task and reveals the naivety of the models that have gone before. Conversely, it places social scientists firmly within their society and our ability to interpret and function in the world in an everyday mode allows us to hope that we might do the same in an analytical mode. It also suggests that the distinctions between disciplines are largely self enforced.
Having discussed the relatively straightforward way in which a model centred view addresses important sociological concerns, it can also be observed that it addresses major difficulties in economic theory. What is plainly missing from the economic theory is a model of a cognitive self that may vary systematically from individual to individual as a result of any number of influences, but nonetheless displays crucial similarities as a result of numerous channels of socialisation and interaction.
The difficulties in economic theory result from a refusal to centre such notions as preferences and rationality on the properties and information processing capacities of cognitive agents, rather than equating them with ``external'' objects like actions or goods. Preferences reflect the adjustments of agents mental models to various sorts of consumption experience, but the paths these adjustments take are dictated by the structure of these models. They are also dictated as much by social structures, such as the models of others, as they are by individual rationality, or the examination of your own view of the world. Indeed the two processes can be seen as richly complementary. The regularities of demand therefore reflect regularities in cognitive models, and cannot be explained by reference to the regularities of goods alone. Such a process obviously proliferates the number of variables to be modelled, but, importantly, increases the number of sources of information at the same time. For example, the demand for cold drinks is not evaluated solely by purchase, but by measurements of the ambient temperature and the increasing anger of people where soft drink machines are defective. In the latter case, there is no economic demand, but a very obvious suppressed social demand. Furthermore, not only are more channels of information opened, but there are more methods for checking and cross-validating sources of data such as observation, measurement, interviews and so on.
This view of the task of modelling also suggests that the notion of absolute truth must be replaced by a concept of coherence, though this is still subject to the possibility of falsifiability. The various sources of data must be integrated into a model that co-determines not just the correct ontology, but also the correct causal links between objects. This model can be compared with actual behaviour using simulation, the measure of coherence arising where features that were not designed parts of the simulation match the corresponding parts of the social system. This is the same process that other agents perform when they develop mental models of the world that satisfy them and allow them to act with competence (24). The complexity and size of models required to actually demonstrate coherence is an open question, but we can derive some comfort from the fact that much social competence is, in fact, observed. However, if the level of complexity were to turn out to be intractable, such a fact would not justify the continued building of simple models that did not work. It would suggest that a social science was not a viable objective (25). Because the process of forming models is a holistic one, based only the ultimate distinction between self and other, actions can legitimately said to be grounded in those models without question begging, even if psychological and physiological factors do play a part in determining the process by which the models change. When explaining actions in social science, therefore, we cannot refer solely to any set of prior or ``lower level'' determinants of those models, such as the ``Laws of Nature'' but must always refer to the models themselves and their continuing interaction with the world and with each other. (This is perhaps what we mean by an intentional system, one that cannot be reductively treated as a non intentional system. Another way of saying the same thing is to refer to the cognitive models of agents as emergent which is again to say that they cannot be treated reductively but have an independent status as explanatory factors.) In a sense, this point is just carrying the idea of coherence and holism a stage further. There is no reason to suppose a linear causal chain from physiology to economics, though the subject matter of these different sciences must often be investigated by different methods. (Another satisfying feature of this methodology is its self-referential nature. Not only should we not construct hierarchical theories, but we should not carry out our science hierarchically.)
To sum up, this discussion is not intended to be a specification of an actual simulation for consumer decision, that work is to follow. Instead, it is an illustration of the problems resulting from positivism and reduction and a set of suggestions about why sociology and economics consequently find it hard to communicate. In a nutshell the difficulty appears to be the economic inclination to confuse the mathematical representation of its intuitions with a genuine formal framework such as that provided by simulation. As a result, it not only downplays the importance of corresponding sociological intuitions, and disregards these in its modelling, but overstates the essential nature of its chosen representation for the existence of such scientific qualities as falsifiability. (I believe that the bad reputation of the scientific method rests not on any intrinsic defects, but rather on a history of clumsy positivism.) The brief discussion of simulation attempts to show that it possesses many of the features economists (correctly) require for rigourous theory without suffering the limitations of mathematical theory or being obliged to choose between social and individualistic modes of explanation. (It also possesses one or two intriguing methodological side effects like the absence of priviliged status for scientists and a sort of self-referential observance of its own prescriptions.)
In suggesting such a sweeping and general description of social action, the sheer scale of the task of social science is revealed anew. Potentially, it involves a model that includes the model of every person in the world. The process of generalising the model to its limits plumbs the depths of scepticism. But from there, the only way to proceed is up. Not only do all the social processes that cause convergence imply a dramatic reduction of the task, certainly to the point where social competence is widely observed in practice, but the difficulty of doing social science is limited to the positive and genuine difficulty of the world and not the far more pernicious and negative difficulty of escaping the paradoxes of contorted theory.