2.1 A brief indication of the distinction

In this chapter I shall deal with an epistemological distinction which will form the basis of the later analysis. In order to make it easier for the reader to grasp this distinction, I shall begin with an illustration drawn from familiar material.

Let us consider what rising price indices tell us about inflation. Do they tell us why prices are rising or do they tell us that prices have risen? Do they convey the sort of information we would need to devise a policy to counter inflation, that is, do they provide us with a guide to action in the future? Perhaps in some cases they would, but it is far more likely that we should have to regard them (with due allowance for the index number problem) merely as a record of the course of inflation over a particular period in a particular area. It hardly seems likely that a mere scanning of statistical time series of prices would bring to mind some kind of laws governing price increases which would enable us to control price changes in the future. Common sense would tell us to distinguish between facts that merely record past events and facts that somehow relate to the structure of the economy. If we want to recommend an anti-inflationary policy we may well look at price indices to see what actually has taken place, but we would know that this in itself would not be enough. We also would need knowledge of certain structural features of the economy. We would turn, according to our inclinations, to the quantity theory of money, the inflationary gap analysis, the degree of price competition in markets, especially in labour markets, and so on.

Let us suppose that we had not only price indices but also other information ancillary to them. Let us suppose that we had a copious description of the circumstances and intentions of each and every seller at the time when he changed his price and of each and every buyer at the time when he offered to pay a higher price. (The practical difficulty of gathering such information is of no importance here.) We should now be able to satisfy ourselves that we have a detailed knowledge of how it was that a particular bout of inflation came about. However, even all this information would not necessarily tell us anything about what we could do to prevent inflation in the future. If we were very lucky we might notice certain common elements in the situations in which prices were raised and so have a clue to some hitherto unknown structural feature of the economy. However, the history of economics shows that the record of such things simply springing to mind is very poor indeed. It is far more likely that we should notice, on looking back over the course of inflation, so many events that seem to be of a purely coincidental and fortuitous nature that we should not expect them to recur, and so we would be left without a guide to possible action.

I have chosen this rather extensive illustration because I hope that what I have said will seem obvious to the reader. The point was to show that it is possible to distinguish between two entirely different orders of facts. I shall try to explain the distinction at some length in a later section. To facilitate the discussion in the meantime, I shall provisionally label ‘ex post’ those facts which owe the meaning they have for us to their position in a possibly unique and fortuitous or stochastic course of events. The other order of facts, which relates to structures, that is, to not altogether transient patterns, and which may, therefore serve as a guide to our purposeful action, I shall label ‘ex ante’. Why these terms, which are, of course, in current use in economics, seemed to me appropriate will, I hope, become apparent as I proceed.

A few words on the purpose of distinguishing between ex post and ex ante facts may here be in place. The distinction is by no means always as easy to perceive as in the illustration provided. In fact, I shall attempt to show that economists frequently mistake ex post facts for ex ante facts; so much so that I want to suggest that not very many serious attempts are made to isolate true ex ante facts, at least in a useful form. If this could indeed be established, it would follow that economists cannot have much to contribute when others turn to them for advice on action to be taken. I believe one could show, for instance, that even the inflationary gap analysis, cited earlier as apparently concerned with an occasional structural feature of the economy, really dresses up ex post facts as ex ante facts, and indeed only has meaning in an ex post context. In this work, however, I shall confine myself to trying to show how a confusion between ex post and ex ante facts has manifested itself in micro-economics.

2.2 Plan of the chapter

If one were to ask a number of economists whether they believed in a universal determinism, the majority would perhaps take up an agnostic position. In any case, whatever their opinions, they are likely to feel that they can carry on their work as economists quite well without deciding the issue. One should not, however, underestimate the surreptitious and pervading influence on academic work of subconsciously held philosophical presuppositions. I want to suggest in sections 2.3 to 2.7 that a widely held view of the vocation of science is largely responsible for, and makes entirely understandable, the frequent failure to distinguish between what I have called ex post and ex ante facts. In sections 2.5 and 2.6, I shall also deal with two types of objections, raised by economists, against the application to economics or to the social sciences in general of this metaphysical view of what are the proper aspirations of science. I shall do all this in order to prepare the ground for an outline of a conceptual framework based on the distinction I have described. I shall give this outline in section 2.8. In section 2.9 I shall discuss the selection and testing of hypotheses in so far as this pertains to the distinction here made. Finally, in section 2.10, I shall make a few brief remarks on the potential role of a philosophy of science applied to economics.

2.3 The deterministic presupposition

Laplace went on to argue that the ever-present gap between the human and the ‘vast’ intelligence makes a theory of probability essential. However, the ideal that science should ever aspire to this vast intelligence is still deeply rooted in our thinking. Given the governing laws and the initial state, or in mathematical language, given the relevant set of differential equations and the boundary conditions, we can trace the evolution of any system. The genesis of this view perhaps owes something to man’s early scientific preoccupation with astronomy. In its own special sphere, astronomy has indeed come close to that vast intelligence, as Laplace pointed out. Can we not go to a planetarium and ask to be shown a representation of the night sky as seen from any point on the surface of the earth and at any date, in the past or in the future, that we care to name? And where else in science is this kind of thing possible? It was natural perhaps, in the excitement of the profound scientific discoveries of the 17th and 18th centuries, that people should have expected the same progress to be possible in other spheres. ‘The regularity which astronomy shows us in the movements of the comets doubtless exists also in all phenomena’, exclaimed Laplace.² However, astronomy and the related classical mechanics have proved to be special cases. Here, man was somehow able to discern structures which, in the form of theoretical models, could be thought of as viewed from no particular place or time, so that man in turn could be thought of as an entirely passive and unimportant observer whose actual constraints of place and time did not really matter.³

I do not know whether all present-day physicists would agree with Eddington’s assertions. He was known for his belief in the a priori character of physical laws, and this does not find general favour. Perhaps there are still differences of opinion. The point is that for the most part it does not really matter in the physical sciences. Scientists other than astronomers (and other than economists, one could add) hardly ever have to see themselves as passive observers and predictors of the movements of whole systems. The ordinary scientists in physics and chemistry laboratories, and the men who develop the knowledge that is applied in technology, are engaged in a much more humble task. Prediction in the physical sciences, and the knowledge needed for it, are in the most usual cases conditional on human agency. The sort of knowledge that most scientists are concerned with can be characterized as follows: If one performs the operation x in a situation which contains the elements y, then there will appear, possibly after a number of distinguishable intermediate stages, a new situation which contains the elements z, or contains, according to calculable probabilities, either z' or z'', z''', and so on. He is concerned, as we shall see, with one of the two types of what I have called ex ante facts – facts which in principle can be used as guides to action.

2.4 The deterministic presupposition in economics

The position is different in economics. By tradition, economists very often concern themselves, as passive observers, with the movements of whole systems, if for ‘systems’ one reads ‘economies’. In the case of the crude extrapolations of some business economists, the analogy to the methods of the old astronomy is obvious. No doubt one cannot accuse the majority of economists of such naivety. However, even in far more sophisticated work there is very often the implication that we are placing ourselves in the position of passive observers of a deterministic system. There appears to be such an implication, for instance, whenever we speak of a determinate equilibrium solution. We may say (1) that we are here using the word ‘determinate’ in the mathematical sense, (2) that usually we are interested only in the implications of certain parameters and not in tracing a course of events from some specified initial state, and (3) that the models we use for this purpose are not meant to represent, but only to approximate very loosely, not all, but only certain aspects of an actual economy. Nevertheless, such qualifications do not remove the deterministic implication.⁷

We have seen that determinism consists not merely of the notion that we may find regularities by empirical means – ex ante facts in my terminology – but rather of the further notion that these facts, which we gather piecemeal, may be built up according to Laplace’s vision into a vast model that could explain and predict the entire course of events, given only the state of things at any instant of time. Within the confines of what was probably its original context, namely, classical mechanics, this notion seemed quite well-founded. Problems arise, however, when the mechanical analogy is taken into other fields. In economics it has to be admitted that even so basic a question as how a person will act in a given environment, or react to given stimuli, cannot be answered with anything like the exactness of mechanics, if indeed it can be answered at all. Explanations of the failure of the mechanical analogy in economics have usually followed either one of two themes, namely, that the human will is undetermined or that the great complexity of social phenomena makes a strictly deterministic scientific programme unworkable.⁸

2.5 The free-will argument

Determinism does not fit in with what we think we know of the voluntary nature of our own actions. Those who argue in terms of free will replace the belief in a universal determinism by a dichotomy of a determinate ‘dead’ nature and a purposefully acting, freely choosing human spirit. (Animals, curiously, are often explicitly excluded from this status.) The long chains of causal reactions stretching through time, which are necessarily a part of the deterministic outlook, are seen as broken whenever they come into contact with a freely acting human being. Often the matter is put in teleological terms. While physical events are determined causally by antecedent events, human action is determined teleologically by imagined events in the future and imagined events do not necessarily materialize. It is not denied that man is influenced by his environment, but his reactions depend on an independent and free will, on his intentions or plans and on his expectations of future events, which in turn depend on his necessarily limited knowledge at the time he acts. No two human beings can therefore be expected to react identically in identical physical conditions.

The allusions to Laplace’s scheme are here obvious. It should be noted that Shackle does not deny determinism in physical events (nature’s process) and that he must suspect his audience of extending this view also to human affairs. Otherwise, what would be the point of his message?

The dichotomy of nature and free will also leads to the view that economics as one of the sciences of human action requires an approach different from that of the physical sciences. The approach that has often been advocated is teleological and hermeneutic. Economic as well as social phenomena in general must be seen as either the intended or the unintended consequences of the purposeful conduct of human beings. Any understanding of such phenomena must then be gained by constant reference to the apparent intentions of acting individuals, or, in the case of unintended consequences of actions, an investigation must at least start with such a reference. Such an approach is sometimes characterized by the term methodological individualism. In so far as it is used to deal with empirical facts, it thus tries to make past events intelligible to us in the ordinary human terms of wishes, beliefs and intentions, and so on. What emerges is not unlike the telling of a story.

An insistence that economic analysis should constantly refer to the meaning which individual actors attach to their actions is a feature of the Austrian approach to economics. In a context in which human action is the subject of an empirical investigation, which is the context with which I shall be largely concerned, the seminal ideas of Menger and von Mises may be associated with the methodological implications of the free-will argument, though it would not be strictly correct to associate them with the free-will argument itself. I shall deal with certain aspects of the approach of the Austrian school of economics in Chapters 4 and 5. Here it may be noted that Menger actually said that economics is unaffected by the question of free will, but he understood pure economics to be a set of what he called laws of thought or, in effect, logical propositions. He seemed to take it for granted that human action is unpredictable, and the application of pure economics to empirical work which he had in mind was not a deterministic one. The main theme running through von Mises’s work is the logic of choice (or praxeology) as he called it, and to him the question of free will also appears to have little relevance. His views on free will were rather complicated. In effect he rejected the free-will argument but nevertheless regarded human action to be quite unpredictable because of the complexity of the mind. The conclusion drawn from this ‘complexity’ argument is often that one should make the best of things by adopting watered-down deterministic methods. Von Mises, however, did not come to this conclusion; in fact he was violently opposed to it. He advocated a strict epistemological division between the physical and the social sciences. Deterministic methods were appropriate in the former but not in the latter. The view that human action is completely unpredictable is to all intents and purposes equivalent to a belief in the freedom of the will.¹¹

It is a feature then of the explicit and implicit free-will arguments not only that human action is completely unpredictable but also that deterministic methods are entirely valid in the physical sciences. (In the case of the explicit free-will argument, a juxtaposition of determinate nature and free will seems to be implied by the term free will itself.) Neither assumption is of course essential for the method of making past events intelligible to us by an enquiry into the intentions and beliefs that prompted acting individuals. One may say, and this is in fact sometimes said, that this method is simply interesting in itself and obviously not available to the physical scientist because he deals with things which we do not believe to have intentions and beliefs. This view, however, gives us no a priori reason for supposing that the physical-sciences type of search for guides to action could not also bear fruit in the social sciences. Furthermore, one could regard determinism as merely a particular feature of celestial mechanics which also gives us no a priori reason for believing that all physical processes may be explained and predicted in the same way. The fact that we can think of innumerable instances in which we could not predict how a person would act would not then be particularly significant because we would then also expect to find many other cases in which we cannot predict what will happen next. It seems to me that this would be a far less constricting outlook. Perhaps the proponents of the free-will argument do not proceed to such an outlook because the case for determinism just seems too plausible or because the fundamental distinction between the social and the physical sciences (which has deep philosophical roots) has to be maintained as a matter of principle.¹² Having seen that the deterministic presupposition introduces difficulties into economics, they reduce its scope, but they are not prepared to abandon it altogether.¹³

2.6 The complexity argument

In other words, the deterministic scientific programme is impracticable for us because our subject matter is too complex. Hayek does not commit himself on the question of determinism. He does not have to, because the question can be decided even within the confines of this presupposition (p 37): ‘There may well be valid and more grave philosophical objections to the claim that science can demonstrate a universal determinism; but for all practical purposes the limits created by the impossibility of ascertaining all the particular data required to derive detailed conclusions from our theories are probably much narrower.’

Not many economists would object to the way Hayek has put this. Difficulties which the deterministic presupposition introduces into economics have been seen and, as in the case of the free-will protagonists, certain adjustments have had to be made to methods and aims.

I should like to take the analysis a little further so that the nature of the complexity can be seen more clearly. By what criterion are we to judge complexity? Hayek says (p 25): ‘The minimum number of elements of which an instance of the pattern must consist in order to exhibit all the characteristic attributes of the class of patterns in question appears to provide an unambiguous criterion.’ Let us put this a little differently. Let us say that we are interested in an entity y and let us assume – and it is important to remember that this is an assumption – that we can define y in functional form such that y = f(x₁, x₂, x₃, …, x_n). The degree of complexity of y is now directly proportional to n, that is, to the number of variables that determine it. Here we have taken into account only the ‘number of elements’ and not their permutations, which could conceivably make y vastly more complex if we thought in terms of a time order, that is, if we were interested, as we often are in economics, in an entity or pattern that ‘takes place’ rather than ‘exists’. (But, I think, the point can be made quite well without these additional complications.)

If y were the orbital motion of a planet or the radioactive decay of a substance (which is a function of time) then one could judge it to be a very simple phenomenon. In comparison, the complexity of the Walrasian general equilibrium system is indeed awe-inspiring. Now, we can say that physical scientists have been lucky to find such simple entities (Hayek even maintains that we classify as ‘physical’ anything that is simple), and that we may as well face the fact that economists have not. But scientists have not been content just to count on their luck. I suggested at the end of section 2.3 that the sort of fact that a scientist is most usually concerned with is conditional on his own agency. The orbital motion of a planet and the radioactive decay of a substance are among the exceptions to this, for they can be observed without the intervention of the scientist. (This is where the luck lies.) However, suppose a scientist inherits from everyday life a concept y₀ such that it seems that y₀ = f(x₁, x₂, x₃ … x_n) with n fairly large, that is, with y₀ a complex phenomenon. Not knowing what to do with y₀, he says: ‘Let me discard the concept y₀ and form another concept y₁ which focuses attention on what happens when I deliberately change x₁.’ He now has y₁ = g(x₁), which is a simple concept. One may remonstrate that this is all very well for a scientist or an engineer who can also deliberately keep x₂, x₃ … x_n constant. Economists could not. It is the old ceteris paribus/no-experimentation problem. However, this problem is not necessarily insurmountable in all cases. For instance, if x₂, x₃ … x_n are independent variables, none of which by itself has a marked effect on y₀, and n is large, then, as we vary x₁ deliberately a large number of times, x₂, x₃ … x_n will vary on their own accord in a way which from our point of view is random. It may then be possible with the aid of the central limit theorem to state y₁ = f(x₁) with an acceptable level of significance, or it may be possible to give confidence intervals of y₁ for various values of x₁.¹⁵ I do not want to underrate the problems involved, and in any case either y₁ or x₁ may not be measurable so that statistical methods could not be used. However, a host of everyday business decisions must be based on rough estimates which are in principle of this nature, since business people also are interested in the consequences of their own actions.

The point to be brought out is that we have not chosen to form simple concepts in this way. We have chosen to define our entities such that y = f(x₁, x₂, x₃ … x_n) with a large n, that is, we have chosen to deal with complex phenomena. The complexity is not due to the inherent qualities of the material, but to the definitions we have chosen. The fault does not lie in the material, but with us.¹⁶

Economists may take comfort from the fact that when the applied physical sciences choose to deal with a complex phenomenon, they are also unable to predict specific events. Applied science cannot even predict the weather very well (and worse, it cannot blame free will for this). The fact that an organizer of an outdoor function cannot be told whether it will rain on a particular night a month hence is in sharp contrast to the fine predictions needed in many industrial processes. However, to blame complexity for this is misleading. Engineers and technicians manipulate the material they deal with, and predict only the consequences of their own actions, not a whole course of events. The meteorologist, on the other hand, is just a helpless bystander taking measurements all over the place; and in this the meteorologist is not unlike the economist turning to the Bulletin of Statistics to do his forecast.

It may be said that we are not always free to choose our own concepts. Just as people take an interest in the weather, so they take an interest in, for example, inflation, and the economist can be expected to say something about it. Even if we had controlled and simple entities at our disposal, we should still have to be able to deal with uncontrolled and complex phenomena. I believe that the approach economists have adopted, as expounded by Hayek, is the correct one, but, as with free will, it does not go far enough. Economists have taken the normal deterministic formula of initial state and governing equations and we have said: ‘Let us dispense with the initial state, let us dispense with the actual numerical values of the variables in the equations, and let us simply say something about the form of the equations.’ It is an adjustment, but it is the minimum conceivable under the circumstances. We are left with a truncated determinism and this truncated determinism debars us from taking our approach to its logical conclusions.

2.7 The impersonal perspective in the deterministic presupposition

2.8 Outline of a sceptic conceptual framework

Let us now leave the notions of determinism and of the dichotomy of dead nature and free will. Let us instead try to think within a different conceptual framework. We shall take the point of view of an observer bound to an eternal here and now. In this situation the observer experiences the continuous flux of events that has intrigued man back to the time of Heraclitus.¹⁸ In an effort to gain empirical knowledge (there is other knowledge), the observer does two things: (1) trying to discern in, or to impose on, the flux of events an order that conforms to the observer’s beliefs or presuppositions and thus makes events intelligible to the observer; (2) trying to find in the flux of events some things which seem to endure, or at least to last for some time, and are therefore available to the observer as guides to purposeful action.

I shall enlarge on each in turn. In order to make sense of the flux around them, people draw up an account of the orderly progression of events. This orderly progression conforms to some belief they hold about the nature of things, and in this sense the order is imposed on the events. The account is made up of a number of elements which owe their meaning entirely to their interrelations within the account, that is, the meaning of each element is given by the unique role it plays in the progression of events. These elements are what I have called ex post facts. In many contexts it may be necessary to focus attention on the activity of making the past intelligible, rather than on the ex post facts which arise from it. I shall refer to this as the genetic understanding, not only because it is an adequate description, but because the term genetic is already used in this sense by writers in the philosophy of science.¹⁹ When it is important to emphasize the fact that this understanding is closely related to the human capacity for telling and following a (true) story, I shall refer to it as the narrative understanding.²⁰

The coherence of the genetic or narrative understanding, as I have said, depends on a set of beliefs. To most educated modern minds an intelligible account of the past must no doubt conform to what appears to be known about causal relations. The further view that everything must have a cause even if the relevant causal relations are unknown, is somewhat different. It makes the task of finding an intelligible account rather easy, since any event that cannot be explained by a known causal relation may then be explained by an imputed one. The advent of a deterministic outlook has probably made the genetic understanding less of a problem than it was at one time. However, it must be emphasized that causal explanations are not essential to this understanding. There may be other arbiters of what is reasonable. Many peoples at many different times, for example, seem to have been quite content to explain what we would call natural events by the behind-the-scenes machinations of all sorts of mythical creatures. Then there are those who, though well aware of the deterministic outlook, would yet maintain that a more profound view of the course of events is attained when one sees Divine intervention in it. Others again may prefer the Marxian dialectical materialism. But if the account that satisfies should have to involve Fate or even gremlins, then this also would do. The important point is that the nature of the belief colours the meaning of the ex post facts that arise from the account.

There are not necessarily any lessons to be learnt from the genetic or narrative understanding. This would be so only when an intelligible account cannot be found, so that beliefs and knowledge of structures have to be adjusted to make an intelligible account possible. Such induction, however, is very difficult and does not appear to be the usual reason for trying to make the past intelligible. To the question why therefore people seek a genetic or narrative understanding, I think one must answer that it is vain to look for a reason, as vain as to ask why people seek the approbation of their fellow human beings. It must be accepted as a propensity of the mind, a love for order.

The meaning of ex ante facts is easier to grasp because we are familiar with them from the physical sciences. It remains to be shown that when we speak of guides to action, structural facts and causal relations we are essentially speaking of the same thing. I have said that one of the ways in which we try to gain empirical knowledge is to try to discern in the flux of events some things which endure, at least for some time. I want to use a mechanical analogy to illustrate the significance of such an enduring thing. Suppose that I have found a rigid lever which rests somewhere near the middle of its length on a rigid fulcrum, so that end A of the lever rests on the ground and end B is in the air. The lever and the fulcrum, their spatial relationship, my knowledge of the nature of rigid bodies, of gravity, and so on, all this I regard as an indication of a structure, a very particular structure in this case. I also believe that this structure will remain as it is, that is, I regard as unlikely that someone or something will knock the lever over, or that the nature of rigid bodies will change, and so on. The structure is therefore an enduring thing. If now for some reason I wish end A (which is on the ground) to be up in the air, I can press down on end B, and my wish will be fulfilled. I have now used the structure as a guide to action. By pressing down one end of the lever, I cause the other end to go up – there is a causal relation between the going down of one end and the going up of the other. Moreover, I can predict the consequences of my action without actually engaging in it. It may seem trivial when put this way, but I believe that this limited sense of the words cause and predict is the most useful sense, and that it is in this sense that prediction and causation are most usually thought of in the applied physical sciences.

The following are explanatory notes:

1. 1.In the definition of a situation I purposefully avoided using the idea of points in time and space because then, strictly speaking, we could not observe movement. In economics at least we may want to identify, say, a person working, or buying something, and this we must do by judging the intentions which in turn we must infer from the person’s actions which involve movement. The notion of here and now also has other more profound advantages over points in time and space (that is, in this context) which I shall not go into here.
2. 2.By defining procedure, I mean the way in which we subsume aspects of situations under classes, that is, how we identify something as, say, water, a dog, the barking of a dog, houses, a person offering something for sale, exchange transactions, and so on.
3. 3.It is important that the elements observed and the elements predicted be identified by quite different defining procedures. Such is the case with, say, the aural impact of a dog’s bark and the visual impression of a dog. The statement that there must be at least one buyer and one seller whenever a sale takes place, is not a structural fact. It is a purely analytical statement based on the meaning of the words used. Only one set of defining procedures is used and the statement merely explores its implications. Many statements in economics are of this type.
4. 4.Following David Hume, who was of course a pioneer of this kind of conceptual framework, one would expect relations to be either spatial or temporal. Relating procedures (which were not part of Hume’s set-up) would then be of a type such as look around the corner, or wait ten minutes, and so on. In economics, however, one may require far more intricate relating procedures. Note that relating procedures relate classes of elements and not unique elements in individual situations.
5. 5.Structural facts may be expressed as mathematical functions or relations, provided two additional requirements are met. First, it must be possible not only to subsume elements of situations under classes by defining procedures but also to relate these elements uniquely to all possible members of their respective classes by measurement procedures. Second, it must be possible to state the relating procedure mathematically. The definition of a structural fact is therefore more comprehensive and may be used in contexts where there are only generic concepts.
6. 6.The relation between a set of elements observed and the set of elements predicted may be asymmetrical. If x is a sufficient condition for y, y is a necessary condition for x and the relation is not reversible unless y is also a sufficient condition for x. The definition of a structural fact is for the case where the observed elements are a sufficient condition for the predicted ones. (For example, if I stand next to the wall of a house and hear the bark of a dog without seeing a dog, I shall believe that if I look around the corner, I shall see a dog.)

   One could rephrase the definition to serve other cases. For instance, let x stand for a sufficient condition for y, and let y stand for a necessary but not sufficient condition for x. Then the observed absence of y is a sufficient condition for the absence of x. The observed presence of y, on the other hand, creates a certain potential for x (where potential is used so that probability may be reserved for something that can be given a numerical value) and y may here be called a potential condition for x. For instance, if I see that there is no dog in the hotel room next to mine, I shall believe it unlikely that I shall be disturbed by barking from that quarter (that is, absence of necessary condition). Whereas, if I see that there is a dog there, I shall believe that there is a certain potential for being so disturbed (that is, presence of potential condition). The relation between sets of elements observed and predicted may also be symmetrical, where each is a sufficient and necessary condition for the other, or even where each is merely a potential condition for the other.

7. 7.It must be emphasized that the definition of a structural fact is not a definition of a structure. It would be difficult to define structures. They must be inferred from structural facts and many such facts may be stated about what we feel intuitively is a single structure. By following up various structural facts we may be able to pinpoint where a structure must lie.

We must now make the transition from structural knowledge to the closely related knowledge of causal relations. Let us consider two cases of purposeful action. (a) The captain of a ship waits for high tide before leaving a harbour. (b) An industrial chemist produces chlorine from rock salt or brine in order to manufacture a household bleach. In both cases a knowledge of apparent structures is used as a guide to action. The two cases differ in so far as in (a) there is no question of the captain having made the tide rise so that he could leave the harbour – the tide would have risen in any case – while in (b) the chemist had to make the salt yield chlorine so that he could manufacture the bleach. The structural fact of the succession of tides is of the type I defined earlier. It can be seen from this definition that we may come by such knowledge as purely passive observers. We assert a structural fact when we simply notice some regularity like the succession of tides or like the constant conjunction (to use Hume’s terminology) of barking noises and dogs. The sort of knowledge applied by the chemist in (b), on the other hand, is of a type that is unlikely ever to spring to the mind of a passive observer. Salt does not break up into its constituent elements of its own accord; we must make it happen in order to see that it is possible.

The similarity between this definition and the one for a structural fact is obvious, as in that case, one could elaborate the definition by distinguishing between sufficient and potential conditions. This would introduce the idea of a potential cause and in some cases of a measurable probability. I indicated this in the shorter version of the definition at the end of section 2.3, I also indicated there that one could allow for a number of distinguishable stages between the operation X and the appearance of the set of elements D. This would introduce the idea of a closed system, or of a dynamic model of a process. It is the deterministic dream that such models may be expanded to include all events – a dream which does not seem to allow for anyone outside of the closed system to activate the whole thing. These deterministic ideas, it seems, also lead us to speak of lines of causation that somehow enter the model through stochastic events or even to suppose that causation is vaguely represented by the notion of equilibrating forces. What this overlooks is that the outcome D (in my definition) depends as much on the set of elements A as on the operation X. The line of causation is established purely by what we take an interest in. It is a convention of thought to say that there is a causal relation between the element introduced by the operation X and the outcome D. Of course, I am not saying that structures are established by our deliberate operations. Whatever structures there are, presumably would be there whether we acted or not. ‘But to think of a relation between events as causal is to think of it under the aspect of (possible) action.’²¹

Since the structural and causal facts that I have defined are so similar, I shall also use the more comprehensive term ex ante facts to cover them both.

With the definitions out of the way, let us make sure that it is well understood that ex ante and ex post facts are simply quite different. Because an ex ante fact seems to retain its character in a variety of situations, we may rightly regard it as a distinct entity with some kind of separate existence. An ex post fact, on the other hand, is not an entity distinct from the context in which it appears and cannot be said to have had a separate existence at any time. It is merely part of an intelligible account of the course of events.²²

Let us take stock now in order to see what sort of picture emerges from this conceptual framework. At the same time let us bear in mind the free-will and complexity arguments that I discussed in sections 2.5 and 2.6. Confronted by incessantly changing events, we try to bring order to these events by finding an account of their progression that is intelligible to us because it conforms to our beliefs. We also seek knowledge that indicates the existence of structures that are permanent or at least last for some time. We may seek such knowledge of structures merely to satisfy our curiosity, or we may do so to use it as a guide to action. But when we use it as a guide to action, we use it for the very limited purpose of predicting some of the consequences of our own actions, that is, we want to make our purposeful actions effective. We may seek to frame our structural and causal knowledge in ever more general and comprehensive terms, but we do not build it up into a vast model that could predict the entire course of events. We may occasionally mistake our intelligible accounts of the past for such vast models, but all the same, as far as the human intellect is concerned, the entire course of events remains ever unpredictable. And if this is all that some of us want to show (and there may be more to it), then we do not need the notion of free will; all we have to do is to recognize the humble nature of the human intellect.

2.9 The selection and testing of hypotheses: two aspects of the problem of induction

The sources for our empirical work can only be the present situation and our records of the past. It may therefore be argued that a genetic understanding of events that can only be achieved if we make adjustments to our beliefs is exactly that mental activity of induction by which we ascertain ex ante facts, and that our direct experience therefore always relates to ex post facts. It is a profound truth that we cannot have a direct experience of ex ante facts, and it is this which is usually referred to as the problem of induction.

There is a vast literature on this problem and any study that sets out to suggest means by which ex ante facts may be established in economics would have to tackle this literature. In this work, however, I am concerned only with what I regard as a confusion between ex post and ex ante facts in economics, and I have therefore steered clear of the problem of induction and shall continue to do so. For reasons that I shall give in what follows, there are, however, a few remarks about the problem of induction that I should like to make here.

One may distinguish two aspects of the problem of induction. First, there is the question of validity of ex ante facts, or the justification we may have for believing in a hypothesis. The first clear statement of this aspect of the problem was given by David Hume.²³ Second, there is the question of a practicable method of inductive inference where there is not much prior knowledge, that is, the question of the selection of hypotheses to be tested when there are not many leads to go by. I shall refer to these two aspects as the question of validity and of approach respectively.

I mention this because I think it is relevant to an objection that may be raised against my suggestion of a possible confusion between ex post and ex ante facts in economics, and especially against the further suggestion that as a result of this confusion no really serious attempts appear to have been made to isolate true ex ante facts. Am I not aware, it may be asked, that empirical work in economics is dominated by regression analysis and the testing of hypotheses, that is, by clear attempts to isolate structural and causal knowledge? Now, I have no doubt that regressions and significance tests are a very valuable aid for finding constant conjunctions and causal relations when these are not obvious to the eye, provided that there is some prior reason for believing that they may be there. The proviso is important, as I shall try to explain.

Opinions may differ on the value of the results achieved by regression analysis in economics. It would be futile to argue about what is of any consequence and what is not. But if anything is amiss, then one could have a shrewd idea that the fault does not lie in the mathematical part of the statistical procedures but rather in the selection and collection of the data before they reach the mathematical stage. In other words, it seems more likely that the shortcomings of our inductive methods, if any, hinge on the question of approach rather than on the question of validity.

These remarks apply equally to the wider issue of falsifying hypotheses, however the tests are conducted. The falsifiability criterion, in the form given to it by Karl Popper, has attained what one could call a measure of popular appeal among economists. But economists appear to have adopted it with the idea that it is a complete method of induction, one that provides an answer to both aspects of the problem of induction. Popper himself, however, expressly disclaims any interest whatever in the question of approach and concerns himself purely with the question of validity. I shall quote rather fully from the work in which he develops the falsifiability idea, so that his attitude may be seen clearly.²⁴

The initial stage, the act of conceiving or inventing a theory, seems to me neither to call for logical analysis nor to be susceptible of it. The question how it happens that a new idea occurs to a man – whether it is a musical theme, a dramatic conflict, or a scientific theory – may be of great interest to empirical psychology, but it is irrelevant to the logical analysis of scientific knowledge. The latter is concerned not with questions of fact, but only with questions of justification or validity. Its questions are of the following kind. Can a statement be justified? And if so, how? Is it testable? … In order that a statement may be logically examined in this way, it must already have been presented to us. Someone must have formulated it and submitted it to logical examination.

Accordingly, I shall distinguish sharply between the process of conceiving a new idea, and the methods and results of examining it logically.

Some might object that it would be more to the purpose to regard it as the business of epistemology to produce what has been called a ‘rational reconstruction’ of the steps that have led the scientist to a discovery – to the finding of some new truth. But the question is: what, precisely, do we want to reconstruct? If it is the processes involved in the stimulation and release of an inspiration which are to be reconstructed, then I should refuse to take it as the task of the logic of knowledge. Such processes are the concern of empirical psychology but hardly of logic. It is another matter if we want to reconstruct rationally the subsequent tests whereby the inspiration may be discovered to be a discovery or become known to be knowledge.

However, my view of the matter, for what it is worth, is that there is no such thing as a logical method of having new ideas, or a logical reconstruction of this process. My view may be expressed by saying that every discovery contains ‘an irrational element’, or a ‘creative intuition’, in Bergson’s sense.

Popper is of course entitled to delimit his field of interest as he pleases. But it must not be forgotten that ‘falsifying hypotheses’ is at best only half an inductive method, and the second half at that. If one were to take it seriously as a whole method, then even so simple a problem as diagnosing why the air in my office seems rather stale would be a gigantic task. If my ‘creative intuition’ was not very sound, I would have to take hypotheses at random from that infinite number that it would be possible to devise. I would have to try relating the staleness of the air to everything from the phases of the moon to the state of the stock market, and I would not finish the job in a lifetime. There must be some way of narrowing down the field of likely hypotheses, and in practice, of course, there is.

That many economists do not really take the full implications of falsifiability to heart was brought out by the controversy that followed the publication of Professor Friedman’s paper on ‘The Methodology of Positive Economics’.²⁵ Friedman maintained, quite correctly in the context of falsifiability, that the validity of hypotheses must be judged by their implications, that is, their predictive capacity, and that ‘to suppose that hypotheses have not only “implications” but also “assumptions” … is fundamentally … wrong and productive of much mischief’. In the ensuing controversy there was much discussion of the ‘realism’ of assumptions and of the need that theories should pass this test too. But this steps out of the ambit of the principles behind the falsifiability criterion completely. After all, what are the assumptions of a hypothesis? Surely, only the hypothesis itself. And how do we judge the realism of the assumptions? The test is whether the implications remain unfalsified. The matter really is quite simple if one takes falsifiability to heart. However, many economists found it irksome (not to mention a certain suspicion that Friedman had an axe to grind) to have to accept hypotheses, however absurd they may otherwise seem, simply because they have not been falsified when put to the test. They wanted to apply to hypotheses the kind of once-over by which we judge whether the fellow across the dinner table is talking sense or not. And this is not merely a minor addition to the falsifying procedure. It establishes a completely different principle. The additional criteria for selecting hypotheses presumably cannot be established by taking hypotheses out of the blue and putting them to the falsifying test, for these surely would also have to pass the realism test, and one would be back at the same problem, multiplied a few times over. What Friedman’s critics seemed to be getting at is that we need not only a method that reasons from the unknown general to the known particular, as Popper has described the testing of hypotheses, but also a method that starts from what we already know.

In everyday problem-solving we do not pick hypotheses out at random. We use a host of consciously known or subconsciously recognized structural and causal facts to narrow down the field of possible hypotheses, before ever we put a hypothesis to the test. When I open my office window to remove the stale air around me, I may not be successful. The stale air may be outside, or whatever. But I have good reasons for preferring the hypothesis that relates the stale air to the position of my window over one that relates it to the phases of the moon or the state of the stock market. Again, one may diagnose an engine failure within minutes if one knows how car engines work; if one does not, one probably would not care to posit any hypotheses at all. The precedence that the question of validity has taken over the question of approach in much methodological discussion is probably due to the fact that the physical sciences were seen as the exemplar, and here so much is already known that the question of approach takes care of itself. But the physical sciences presumably also had to be built up from fundamental facts that no one would question. Who could fail to notice, for instance, the regular alternation of day and night or of winter and summer?²⁶ But what of fields where constant conjunctions are not so common. When one is confronted by a bewildering flow of events that no ‘creative intuition’ seems able to explain, then surely the question of approach should take precedence over that of validity. After all, falsifying hypotheses is all very well when one hardly ever succeeds, but not much fun when one succeeds every time. Whatever Popper may think, we do need a logical reconstruction of the way knowledge is built upon knowledge and a criterion for what sort of knowledge we can begin with.

In economics we naturally do have certain conceptions of the sort of knowledge we should begin with, and we frame our theories accordingly. When empirical work calls for hypotheses, we do not pick them at random but select them according to these conceptions and theories, and then judge their validity by well-known methods. However, when one considers questions such as the inflation of our time or the transfusion of a more production-oriented culture into subsistence economies, then perhaps one may wonder whether we really know where to begin. When everyone has had their say, do these questions not remain as enigmatic as ever? At least those who think they do may consider it worthwhile to examine again the approach aspect of the problem of induction, as distinct from the aspect of logical justification, that is, the selection as distinct from the testing of hypotheses. It is towards such a possible examination that I hope the distinction between ex post and ex ante facts may make some small contribution.

2.10 The philosophy of science applied to economics

I realize that exercises in the philosophy of science applied to economics are not to everyone’s taste. Also, it may be held against me that I could quite easily have outlined the conceptual framework I am putting forward, and have had my say on hypotheses, without ever mentioning the word determinism. However, not only was it necessary to give some indication of why a different conceptual framework may be justified, but it will be necessary in the following chapters to refer back to the nature of determinism. When I consider certain aspects of the history of micro-economic theory in the light of the concepts developed in this chapter, it hardly would be a promising tack to assume that by some strange coincidence the eminent thinkers of the past were not able to muster what may seem like common sense to the person in the street. It will be necessary to view their work in terms of what most likely were their philosophical presuppositions.