The Structural View of Scientific Theories and the Theory of Operant Behavior
Summary: Using a basic theory of operant behavior as an example, the so-called „structural view“ of scientific theories is described and contrasted with the traditional (statement-) view. Within the structural view several difficulties of behavior theory are more easily dissolved, such as the question of its subject-matter, the relationship between theory and practice, the „dynamics“ of theories, the compatibility of different theories (eclecticism) and the problem of behavioral determinism.
Zusammenfassung: Am Beispiel einer elementaren Theorie des Operanten Verhaltens wird die sog. „strukturalistische Auffassung“ wissenschaftlicher Theorien entwickelt und der traditionellen (Aussagen-) Konzeption gegenübergestellt. Es wird gezeigt, dass sich innerhalb dieser Auffassungen einige Schwierigkeiten von Verhaltenstheorien besser erklären und bewältigen lassen, u.a. das Problem ihres Gegenstandes, das Verhältnis von Theorie und Praxis, die „Dynamik“ von Theorien, das Wesen von Erklärungen in den Verhaltenswissenschaften, die Nicht-Falsifizierbarkeit von Theorien, die Vereinbarkeit unterschiedlicher Theorien (Eklektizismus) und das Problem des Verhaltensdeterminismus.
Sommaire: Au moyen de l`example d`une théorie simplifiée du comportmnent operant, l`auteur développe une conception „structurale“ des théories scientifiques et l`oppose au concept traditionnel (de théorèmes). Cette vue permet à l`auteur de mieux résoudre certaines difficultés posées par les théories comportementales telles que: le problème de leur sujet, le rapport entre la théorie et la pratique, le „dynamisme“ des théories, la nature des explications dans les sciences comportementales, l`impossibilité de falsifier les théories, la compatibilité des differentes théories (éclecticisme) et le problème du déterminisme de comportement.
The clinical application of behavior theory, the movement (or flight) from the laboratory to semi-chaotic consulting rooms, homes and institutions, has resulted in the appearance or reappearance of some theoretical puzzles which are well worth considering or reconsidering. Those puzzles are all connected with the problem of the nature of scientific theories in general (and that of behavior theory in particular) and can be traced to an inadequate understanding of such theories. As a starting point it will be useful to outline some important features of the classical, orthodox or „received“ view of scientific theories.
We can assume that at least one of the goals of science is to explain (and/or predict) the phenomena under consideration. For the time being, we shall also assume (although this is a much debated issue) that conformity to the so-called „covering-law-model“ of scientific explanation (Hempel, 1966) is a necessary condition for any proposed explanation to be acceptable. Roughly speaking, this means that the following conditions have to be met.
I. We need an explanandum, a description of the state of affairs we wish to have explained (e.g., „A is afraid of rats“);
II: We need an explanans, consisting of
a) at least one general law or universal statement (e.g., „Whosoever has undergone a conditioning procedure of type M will be afraid of rats“), and
b) at least one singular statement (e.g. „A has indeed undergone a conditioning procedure of type M“), such that
III. the explanandum can be logically deduced from the general law(s) in conjunction with the singular statement(s).
Hence it follows that a theory worth its salt (i.e. able to explain things) must contain at least one (and usually more than one) law of the general form: „For all x: if x has the quality F, then it has the quality G“. Laws are statements or sentences (this is true of deterministic as well as statistical laws), therefore - and this is the orthodox view - scientific theories are sets or systems of statements (sentences) with certain logical relationships among one another. Theories are looked upon as devices into which factual statements are fed with the result that explanations and predictions are ejected (logically deduced).
If, however, we investigate the actual procedure of science we often find something quite different. An instructive example is Teasdale`s attempt (1974) to derive hypotheses about the nature of obsessional-compulsive disorders from experimental studies of animal and human learning. The models he eventually presents are definitely not logically deduced from a given set of general and singular statements. Something ingenious has been done, but it cannot be accounted for by the traditional concept of scientific theories. Furthermore, this view invariably leads to those puzzles already mentioned. I shall list only a few.
1. The theory-practice puzzle. What exactly is the relationship between theory and practice? Can we derive everything we actually do from the theory (which intuively appears to be nonsense) and if not, does this not invalidate the theory (cf. Locke´s critique of Wolpe, 1971)? Is behavior theory an ideology relative to behavior modification (London 1972)?
2. The falsification puzzle. If one of our explanations turns out to be wrong, if a prediction goes astray, if a therapy results in failure, then we should throw away the theory - but we don´t. So, it seems, we are either inconsistent or have to blame the sloppiness of our work.
3. The research puzzle. Why do we need experimentation if our theory tells us what is going to happen in a given case? If it does not do this, why do we keep it? Surely, not every experiment is an experimentum crucis the outcome of which determines whether we shall remain what we are or whether we shall become, say, psychoanalysts.
4. The subject-matter puzzle. If behavior theory deals with behavior, how is it to be distinguished from other disciplines which also deal with behavior such as physiology, ethology, sociology, and physics? If it explains behavior, how can other disciplines explain it as well?
5. The compatibility-incompatibility puzzle. Can we work with different theories (models or paradigms) within a given field without becoming once more logically inconsistent? On a pragmatic level we may also call this the orthodoxy-eclecticism puzzle. Are, for instance, dynamic psychotherapy and behavior therapy really irreconciliable (Marmor, 1971), not to mention countless other psychological approaches?
6. The determinism puzzle. Is the assumption of determinism a necessary requirement for scientific behavior theories (Skinner, 1953) since otherwise we could not find the necessary laws? If so, how are planning, intervention, self-control etc. possible?
If those and other puzzles arise from our inclination to view theories as sets of general statements we might be able to dissolve them by introducing another concept of scientific theories, one which Stegmüller (1975) calls the „structural view“. The basics of this exposition and much of its terminology are derived from the work of Sneed (1971), who first developed this concept in the context of mathematical physics, and from Stegmüller (1973, 1974, 1975), who generalized and clarified Sneed´s results. Applying it to (as yet non-mathematical) behavior theory makes it necessary to change and simplify many of their ideas. For any violation of their letter and/or spirit I bear the sole responsibility.
In order to make this discussion relevant to our problems, I shall reconstruct, as an example, a simplified version of what might be called a basic theory of operant behavior. The reconstruction is not meant to be complete and I am aware that there is more than that to behavior theories; it is merely used to make clear some decisive logical characteristics of such theories in general.
Now, the following lines are going to look truly awful, but I will illustrate them with an example, and then the abstract stuff will become much clearer.
We begin by defining an operant behavior system (OBS) through the introduction of a set-theoretic predicate:
DF-OBS (Definition of an Operant Behavior System)
x is an OBS if and only there exists an organism O, a class B of responses or behaviors of that organism, a class of stimuli S, a class of events K, and their relations c, f, and j such that
1. x consists of the organism O,
2. x consists of elements of the class of stimuli S,
3. x consists of elements of the class of behaviors B,
4. x consists of elements of the class of events K,
5. If an element of S occurs, then (sometimes or always) one or more elements of B
occur, either in the presence of S or shortly afterwards (relation SBc).
6. After each occurance of an element of B, or after a varying number of such occurances,
one or several elements of K occur (relation BKf).
7. The occurance of elements of K influences the probabilty that elements of B occur in
the wake of elements of S (SBKj).
This definition of an OBS may be regarded as an informal axiomatization of the fundamental internal structure of the system, 1. to 7. being the defining axioms. Like any other definition of a system it essentially consists of statements about the constituents of that system and the relationships that obtain between them.
Here comes the example:
There is Charlie, a white rat in cage number five. From time to time a small green light bulb in the cage is lit for about five seconds. As soon as this occurs, Charlie runs to a lever protruding from the wall and presses it a few times. After a varying number of lever presses a food pellet is dropped noisily into a tray. Every time that happens, Charlie runs to the tray and eats the pellet.
Here, the class S (Stimuli) is defined as the green bulb lightening up for five seconds, the class B (Behaviors) is defined as pressing the lever, the class K (consequences) is defined as the dropping of pellets into the tray. Thus, here SBc means that lever presses occur after the little green bulb lights up, and BKf means that after (one or several) lever presses a food pellet is dropped into the tray, and SBKj means that the dropping of the pellets reinforces the pressing of the lever (meaning all kinds of things), with the green light as the discriminative Stimulus. Note that the eating of the pellets is not part of the OBS as just described - it is, so to speak, abstracted out.
The history of operant behavior theory begins with the actual production of operant behavior systems (OBSs), for instance the classical experimental set-ups in operant conditioning with rats and pidgeons. This allows us to do two things.
a) We are now in the position to advance the very fundamental claim „there are indeed OBSs“, pointing at those realizations which we shall call paradigmatic examples of OBSs:
b) we are able to give an approximate explication of what „j“ means, showing for instance how in the presence of elements of S the probability for an occurence of elements of B is increased if in this situation the Bs are followed by Ks, how that probability decreases if we discontinue the presentation of Ks, and so on. Clearly, this expectation of „j“ by giving examples is not exhaustive but leaves much room for further elaboration.
Whereas we thus get some idea about the meaning of „j“, this predicate still differs in an essential way from „c“ and „f“. The latter two are „uncomplicated“ inasmuch as they denote simple temporal and statistical relationships that usually can be determined by mere inspection of an ongoing behavior system. Whether „j“ obtains between the constituents of a behavior system cannot be so determined: in order to do this we must either have extended knowledge about the history or genesis of that system or we must perform some experiments, such as removing and reintroducing the Ks to see what happens. We also must exclude alternative explanations for the effects that may appear. Without giving a detailed justification, I shall just state that „j“ therefore is a complex dispositional predicate and must be regarded as a theoretical expression not explicitly definable in terms of operations or observations. All we can do is provide partial characterizations.
Now, the seventh axiom of DF-OBS, demanding that the constituents of a system are j- related, is obviously the essential one, since only if it is true of a given system, that system is a real OBS. We shall therefore call a system which fulfills all seven axioms of DF-OBS a complete model or realization of an OBS.
A system which fulfills the axioms 1. to 6. - i.e. a system in which axiom 7. may or may not also hold - we shall call a possible model or realization of an OBS, because this is the kind of a behavior system of which it is reasonable to ask whether it is an OBS or not.
Only possible models of OBSs can be complete models of OBSs, that is to say, the class of complete models is a subset of the class of possible models.
Any system which fulfills some of the axioms 1. to 6., but not all of them, for instance a class of behaviors of an organism occuring repeatedly during a given period of time, we shall call a partial possible model of an OBS. They represent, as it were, fragments of possible models.
We have now at our disposal three set-theoretic predicates defining different types of behavior systems related in certain ways to each other. Examining the terms occuring in the defining axioms we find they all designate determinable qualities; each one marks out a possible range of affairs but is not specific about it. Thus, „X is an O“ may be made more precise by saying „X is a white rat“ and even more so by saying „X is Charlie, the white rat sitting in cage No. 5“. However, „X is a stone“ is not a specification of „X is an O“ since the extension of „X is an O“ does not cover objects of that sort. Similarily, „X is a S“ and „X is a K“ can be determined qualitatively (by stating what kinds of events they are, e.g. presentations of lights, sounds, food pellets, removal of E-shocks etc.) as well as, at least to a certain extent, quantitatively (stating e.g. pitch and loudness of sounds, strenghts of E-shocks and so on): Clearly, in any realization of one of the defined systems the relation predicates c, f, and j can also be quantitatively determined. A description of a system in terms more determinate than those used in the defining predicate will be called a specification of that predicate. Thus, „ a is a system consisting of John washing his hands fifty times a day“ is a specification of „a is a system consisting of an organism and a non-empty class of behaviors emitted by this organism“. Most important in this context are specifications with respect to the predicate „j“, since „j“ represents the lawful functional relations between the constituents of an OBS. Such a specification would for instance include statements indicating the effects of various schedules of reinforcement within an otherwise specified OBS or class of OBSs.
We shall now examine and discuss in several steps how all this makes it possible to reconstruct rationally the scientific enterprise, especially the business of working with the theory of operant behavior.
1. As already mentioned, on a very fundamental level and using the most basic predicate we can say „There are complete models of operant behavior systems“ or, formally „There are some X such that X is an element of the class of OBSs.“ We justify this claim by pointing to the paradigmatic examples, those OBSs we have already realized in our laboratories. The class of paradigmatic examples ought not to be considered a closed set. Any specific types of possible models of an OBS may be included in it, provided that it has reliably been shown by repeated experimentation that it is indeed a complete OBS. This last requirement makes it clear that the class of paradigmatic examples is in fact a class of classes. Individual systems will not be included in it but only types of systems, such as lever pushing responses of rats followed by representation of food pellets in the presence of visible light, and so on.
2. The listing of complete models of OBSs requires at least qualitative specifications, indicating the kind of constituents the model consist of (what organism, behavior, stimuli etc.). Next, we try to give quantitative specifications of the variables involved, attributing more or less definite values to the probability or frequency of response, the temporal relationships, the magnitude of stimuli and so forth. Usually, as a first step we define various specific versions of possible models of operant behavior systems. Unfortunately, the concepts used for that purpose (such as positive reinforcement, punishment, free operant or discriminated avoidance) confound procedures and effects and so provide a source of endless quibbles (cf. the hoary debate on the circularity of the law of effect). It is of paramount importance to define procedures (types of possible models) first and determine the effects independently.
3. This is exactly one kind of systematic research whithin the theory of operant behavior. It consists of a) manipulating the values of a specific system or class of systems under investigation and b) finding out what the effects of those manipulations upon other variables are. All studies concerned with the result of different schedules of reinforcement, delay of reinforcement, discriminative learning etc. are of this nature. Restricting our attention to classes of systems the outcome will be twofold:
I. For any specific type of System S we should be able to formulate a set of S-specific laws stating the particular functional relationships holding between the variables of S.
II. Once we have been able to do this with several classes of systems we can put forward a set of application laws stating simply which special or particular laws hold in which classes of behavior systems. If we are able to say what kind of organism will be sucessfully reinforced with what kind of stimulus for some kind of behavior, we have a simple law for this type.
4. There are obviously various and diverse types of OBSs, in other words, the theory is „meant“ not just for one but for several applications (the terms applications and models being used synonymously). Now, the predictive and explanatory power of a theory depends heavily on so-called constraints which „will require that certain relations hold among the values of theoretical functions employed in different applications of the theory“ (Sneed, 1971, p. 66). Using Sneed`s example of classical (Newtonian) particle mechanics, only if we assume that the mass of a given body is a intrinsic property of that body and remains the same whether it orbits around the earth, the moon, the sun, or is used as a pendulum, can we make predictions across different applications of the theory. Otherwise, determining the mass of a body by means of an analytical balance would tell us nothing about the behavior of that body in other circumstances.
Similarily, if we could operationally determine something like „reinforcing power“ of a class of stimuli, for instance by noting the number of contingent presentations needed to raise the probability of a response from p to p´, and if this „reinforcing power“ remained the same across all applications, for all behaviors of all organisms, then we would have a very powerful tool for predictions and explanations. Unfortunately, this does not seem to be the case, and here we have one of the reasons why the theory of operant behavior is relatively weak compared with Newtonian mechanics.
5. Nevertheless, research in operant behavior must be able to propound at least some limited constraints in order to be useful at all. On a very fundamental level we might be able to establish that in a given individual application of the theory a certain stimulus remains reinforcing over time or varies in this property according to known laws, e.g. of satiation and fatigue. Then, hopefully, it can be shown that a stimulus is reinforcing for a well-defined class of behaviors of a definite class of organisms under standard conditions - otherwise S-specific laws and application laws would not be possible. Perhaps we can also demonstrate that a stimulus has controlling properties across several organisms, behaviors and specific procedures. These are just examples, but it should be clear that the search for more and more extensive constraints is one of the central aims of research within a theoretical framework.
6. A theory derives its dynamics from the belief (or conviction or hope) that beyond the class of paradigmatic examples further applications will be found. The attempt to find such applications is the third aspect of research „guided“ by a theory. As it is also one of the most fascinating aspects it will be necessary to give a detailed picture of those operations within the structural view of operant behavior theory.
We start with the suspicion that certain as yet vaguely understood phenomena, such as some kinds of pathological behaviors, might turn out to be constituents of operant behavior systems as defined by the set-theoretic predicate „X is an OBS“.
In order to prove this we must show that they are constituents of possible models of OBSs and this requires beforehand that the phenomena are partial possible models. So, the pathological behaviors under investigation must be describable as (must consist of) a class of behaviors so defined that - as a minimal requirement - their repeated occurance over periods of time can be reliably determined by one or more observers. At this point we see how working with a theory demands submission of certain description-guiding rules. We could of course describe the behavior of a well-conditioned rat in a Skinner box as „working for food“, „gorging on pellets“, „twitching some muscles“, but that would not lead us anywhere. There are numerous ways of describing a complex phenomenon extending over time but only some of those are relevant with respect to a given theory. It now becomes clear what the slogan „description in the light of a theory“ really means. It does not imply that the eyes of the theoretician are endowed with particular powers of sight strangely denied to others. Since description always and of necessity involves abstraction, what it does mean is that the theoretician abstracts from a phenomenon those features that can be described as partial possible models of the structural core (cf. section 8.) of this theory. Out of such partial possible models he hopes to construct the possible models.
Further , the structural core of a theory not only provides description-guiding rules but also search-guiding rules. Once we have described a neurosis as a set (or sets) of behavior(s) occuring every so often, we must look for kinds of events which are requisites for possible models of an OBS. That is, we must look for a set of events that stand in certain antecedent temporal relationships to the behaviors (for only these are elegibile as „discriminant stimuli“) and we must look for another set of events that follow those behaviors in a certain way for only they can be reinforcers. Only when we have found those sets of events and specified their temporal relationships do we have the description af a possible model of an OBS. Then we have to ask the decisive question: is this possible model a complete model, a realization of an operant behavior system? Are the events consequent to the behaviors really reinforcers of that behavior? Are the antecedent events really discriminative stimuli? Is it true that „SBKj“?
At this juncture, the structural core of the theory together with the already existing class of paradigmatic examples provide us with operation-guiding rules. The class of paradigmatic examples is characterized by sets of specific laws which represent a partial interpretation of „j“. They thus delineate a range of interventions which may be applied to certain variables of a possible model under investigation and mark out the kind of effects upon other certain variables which are necessary for the system to qualify as a complete model of an OBS. We may, for instance, prevent the occurance of what are possible reinforcing stimuli and note the effect upon the frerquency of behavior. The operation-guiding rules not only tell us how to determine whether a possible model of an OBS is indeed a complete model (something which in given case may be quite difficult and require sophisticated research techniques), they also tell us how to modify some variables of the system by modifying others.
7. We may not only try to reconstruct partial models as possible models and finally as complete models of an OBS, we may also try to actively construct complete models by supplementing partial possible models (behaviors we find) with those elements that would make them possible models and thereby - if we have been lucky or inspired - complete models. Whenever we attempt to raise the probability or frequency of a response by introducing suitable reinforcers, we do just this. First we choose a certain setting as potential discriminative stimulus, then we select other stimuli which are presented contingent upon the behavior we wish to change. We produce a possible model and if this has been done correctly, it turns out to be a complete model, that is, the behavior changes as intended. Of course, this will not necessarily be the outcome. The choice of potential reinforcers and discriminative stimuli is by and large a matter of trial and error since the constraints available whithin the theory of operant behavior are very limited. But we do have definite knowledge about the temporal relationships called for and we do have some ideas about what might be an effective reinforcer. The search-guiding and operation-guiding rules provided by the theory certainly restrict the range of promising activities.
8. Having come so far, we are now in a position to elucidate the concept of theory. We distinguish between
I. the structural core of the theory, consisting of 1) a number of set-theoretics predicates defining axiomatically the logical (or mathematical) structure of systems, and 2) the description-guiding, search-guiding and operation-guiding rules;
II. the periphery of the theory, consisting of special laws, application laws and constraints, that is, empirical statements which are generated by applying the rules of I. to certain phenomena.
It is important to see that the periphery of the theory (the empirical laws) is not logically deduced from its core. It is created by scientists. The laws of periphery enter into explanations and predictions, they are subjects to confirmation and falsification, but this process does not touch the theory proper, i.e. the structural core. The core is immune to falsification, not because it consists of statements mysteriously invulnerable to contrary empirical evidence, but because it does not consist of factual statements at all. It is just not the kind of thing to which the predicate „falsifiable“ could be meaningfully applied.
9. We can understand this more clearly when we ask what happens if an attempt to reconstruct a class of behaviors as constituents of a complete OBS is not successful. Suppose researcher A wishes to influence the occurence of epileptic convulsions (grand mal) by the subsequent presentation of stimuli which he believes to be punishing. However, no matter what he tries, no results are forthcoming. What does this prove with respect to the underlying theory? Is it thereby falsified? Must we reject it utterly? Obviously, as Stegmüller (1974) put it, the frustrated attempt to find a new application (model) of a given theory first of all throws „discredit“ upon the person who has tried to do so. The researcher has failed, and this is all we can say so far. Only if generations of workers are unseccessful in reconstructing a class of behaviors as an application of the theory, this class of behaviors will be eliminated from the class of intended applications of the theory. Thus, by a process which Stegmüller calls the autodetermination of the range of applications of the theory, that theory will become more specified in the course of its development with respect to the phenomena it can be applied to. Only the (revisable) laws of the periphery can be falsified, the theory as such, i.e. is structural core, cannot. The worst that could happen is that we find no or only a few applications of it. If so, sooner or later there will probably be no one or just very few people working with it. But this does not necessarily have to be the case: there is no conclusive reason to give up hope.
10. We can now characterize more precisely the subject-matter of operant behavior theory. It does not deal with behavior as such - otherwise one could hardly distinguish it from other disciplines such as physiology or ethology. Rather, it deals with behavior systems, that is behavior only insofar as it stands in certain defined relations to other events.
This tells us something about the nature of explanation. Traditionally the relationship between explanans and explanandum has been likened to fatherhood. We have an event, pure and simple, and try to find the true and correct explanation for it, just as we have a child and look for his father - and of course he can have only one. In fact, however, explaining events always means showing how they are dependent on some other variables of a given system, and since events usually are constituents of many different systems, various true and correct explanations can be given. Most or all behaviors can be explained in terms of muscle and sceleton mechanics and/ or physiology and/ or endocrinology; that is, as elements of systems that are not operant behavior systems. „Explain the event!“ is a meaningless injunction. Events can only be explained as dependent on a given set of variables - if they are so dependent. Which kind of explanation one looks for depends entirely on the practical interests one pursues. We know quite well that compulsive hand washing presupposes activities of motoric nerves, but we are not inclined to eliminate this symptom by means of simple surgery. It could be done, but the side effects would be annoying.
11. This analysis, if it is basically correct, leads to important consequences concerning the compatibility of different theories. It has become clear that theories do not advance universal claims about the world; they are applicable to the world, but then only within a limited range. The belief that there are OBSs as defined is of course not imcompatible with the belief that there are other behavior systems, for instance, unconsciously motivated behaviors. The possibility of conflicts arises only if divergent claims are advanced with respect to a definite behavior or class of behaviors, but even in this case the issue is not as unequivocal as it may appear.
Let us give an example. Analysis reveals that the deficient study behavior of a student is due to negative self-evaluation after each attempted performance. The therapist trains his patient in replacing negative self-evaluations by positive ones, pointing out the abilities he actually possesses. He modifies „what the patient says to himself“ (Meichenbaum, 1971) and is succesful. Obviously we were dealing with an operant behavior system.
A psychoanalyst, however, might say this patient´s difficulties were the expression of omnipotence phantasies derived from an unconscious conflict with his father. Now, whether this is true or not, it certainly does not contradict the first statement. The behavior therapist was looking at the negative self-evaluations only insofar as they stood in a certain relationship to the target behavior and he was fortunate enough to be able to influence that variable by therapeutic intervention. But it is certainly legitimate to ask how those evaluations came into existence; quite possibly they were the result of the patient`s unconscious wish to surpass his father in all respects, so that actual performance never lived up to the desired level of archievement. Even if much of what psychoanalysts have said turns out to be gibberish, as a research programme (Lakatos, 1970) psychoanalysis is not incompatible with operant behavior theory.
Generally, when a theoretican looks at systems with certain constituents and defined relations between them there is really nothing that could prevent him from viewing some constituents as elements of other systems. The most obvious case is Mowrer`s two-factor model of phobias (1951) in which operant and Pavlovian behavior systems are conjoined. Furthermore, the occurance and availability of reinforcing events, for instance, depends on many outside factors (relative to the OBS) such as social and economic status, skills etc., so this has to be taken into account as well. Even if we do not look beyond the OBS under investigation, whether the interesting relationships among its constituents actually exist depends on „internal“ and perhaps changing properties of the organism such as deprivation, physiological make-up, genetic endowment, value and belief systems, motivation, past experiences, drugs and so on and so on. One can work with the theory of operant behavior, or for the matter with any other theory, only if one knows a lot besides. If eclecticism means working with different models or approaches, if it means looking at the world as a network of systems differing from each other in essential aspects, then it is not only possible, it is necessary, and has been going on all the time. To let one´s mind stray too much from the staight and narrow path holds many pitfalls, and like any other enterprise, it requires ingenuity and luck. A methodological via regia leading invariably to succes if marched upon with determination does not exist.
12. Finally we shall discuss the issue of determinism which has often been considered a necessary requirement of any behavioral science worth its name (cf. Skinner, 1953). This has led to much criticism and vague feelings of unease. Those who insist upon the freedom of man are confronted with the problem of how freedom can co-exist with well-established laws of behavior, others find it difficult to reconcile the idea of determinism with therapeutic activities such as planning or self-control (Hartig, 1974). The solution I hope to offer proceeds in several steps.
We shall call an S-act any act of a human being that changes the value of a given variable in an existing singular operant behavior system S. That there must be such S-acts is obvious; otherwise we could do no research, no experiments, no behavior modification.
We shall call basic S-act any S-act which changes a variable of the system S directly, that is without first changing another variable of S.
A secondary S-act is any S-act which changes the value of a given variable of S by first changing another variable. This of course requires the existence of functional relationships between the variables of that system.
We also distinguish between S-dependent behaviors, i.e. all behaviors that belong to the class B of behaviors in S as defined, and S-independent behaviors. The latter include all behaviors of organisms which are not in S and also several behaviors of the organism actually belonging to S.
Now, given a class of behaviors B of an organism O that are constituents of a known operant behavior system S, the practical problem is whether anyone has at his disproval an S-independent behavior which is a basic S-act, i.e. which changes at least one variable of S directly. Usually this is done by therapists or investigators who e.g., exercise stimulus control or interfere with reinforcement contingenties in order to change the S-dependent behavior B or O. But it is quite possible that O himself has S-independent behaviors at his disproval which are basic S-acts and can be used to change some variables of S directly, with the result that the S-dependent behavior is modified . The thesis of behavioral determinism can then be stated as follows:
For every behavior C which is independent relative to a behavior system S there exists ar least one other deterministic behavior system S`such that C is dependent relative to S`(that is, a constituent of S`).
This assumption may or may not be true; in any case it can neither be verified nor falsified. Furthermore, it is neither a consequence nor a presupposition of either the structural core of the theory or the laws of its periphery. Therefor, working with the theory of opeant behavior does not require the assumption of behavioral determinism nor does it require the contrary
assumption. The issue of determinism is really a metaphysical problem.
In this paper I have tried to present a view of scientific theories in general and the theory of operant behavior in particular which seems to do more justice to the actual workings of science than the traditional view. Is aims are mainly therapeutic: to solve some puzzles and to show how the acceptance of a theory does not limit the range of activities and beliefs as strongly as is usually maintained. If it encourages communication between proponents of different psychological schools and if it eases the perhaps slightly guilty consciences of those who have never submitted to the relentless rule of the Philosophy of Science, so much the better.
Danto, A. C. (1965). Basic Actions. American Philosophical Quarterly, 2, 141-148.
Hartig, M. (1975). Die Anwendung von Techniken der Selbstkontrolle in der Verhaltenstherapie. In C. Kraiker (Hrsg.), Handbuch der Verhaltenstherapie (2., S. 325-350). München: Kindler.
Hempel, C. G. (1966). Philosophy of Natural Science. Englewood Cliffs, N.J.: Prentice Hall.
Kanfer, F. H. & Phillips, J. S. (1970). Learning foundations of behavior therapy. New York: Wiley & Sons.
Lakatos, I. (1970). Falsification and the Methodology of Scientific Reasearch Programmes. In I. Lakatos, & A. Musgrave (Hrsg.), Criticism and the Growth of Knowledge. London: Cambridge University Press.
Locke, E. A. (1971). Is "Behavior Therapy" behavioristic?. Psychological Bulletin, 76, 318-327.
London, P. (1972). The end of ideology in behavior modification. American Psychologist, 10, 913-920.
Marmor, J. (1971). Dynamic Psychotherapy and Behavior Therapy: Are they irreconcilable?. Archives of General Psychiatry, 24, 22- 28.
Meichenbaum, D. H. (1973). Cognitive Factors in Behavior Modification: Modifying what clients say to themselves. In C. M. Franks, & G. I. Wilson (Hrsg.), Annual Review of Behavior Therapy. New York: Brunner & Mazel.
Mowrer, O. H. (1951). Two-Factor learning theory: summary and comment. Psychological Review, 58, 350-354.
Popper, K. R. (1959). The Logic of Scientific Discovery. London: Hutchinson.
Schulte, D. (1976). Ein Schema für Diagnose und Therapieplanung in der Verhaltenstherapie. In D. Schulte (Hrsg.), Diagnostik in der Verhaltenstherapie. München: Urban und Schwarzenberg.
Skinner, B. F. (1953). Science and Human Behavior. New York: The Free Press/Collier-McMillan.
Sneed, J. D. (1971). The Logical Structure of Mathematical Physics. Dordrecht: Reidel.
Stegmüller, W. (1973). Probleme und Resultate der Wissenschaftstheorie und Analytischen Philosophie (Bd.II: Theorie und Erfahrung, Zweiter Halbband). Berlin: Springer.
Stegmüller, W. (1974). Theoriendynamik und logischen Verständnis. In W. Diederich (Hrsg.), Beiträge zur diachronischen Wissenschaftstheorie. Frankfurt: Suhrkamp.
Stegmüller, W. (1975). Hauptströmungen der Gegenwartsphilosophie (Band II). Stuttgart: A.Kröner.
Teasdale, J. D. (1974). Learning models of obsessional-compulsive disorders. In H. R. Beech (Hrsg.), Obsessional States. London: Methuen.
Wright, G. H. v. (1991). Erklären und Verstehen (3.). Frankfurt: Hain.
 Revised version (2003) of „Behavioural Analysis and the Structural View of Scientific Theories“ (1977). European Journal of Behavioural Analysis and Modification, 1 (4), 203-213. This revision mainly attempted to make the exposition more “user friendly”. No substantial changes appeared necessary, even though in the meantime several books and papers on the “structural view” were published, notably by Hans Westmeyer and Rainer Westermann.
 „behavior theory“ in this paper to be used as short for „theory of operant behavior“
 These ideas are very lucidly presented in Popper (1959), p. 59f.
 All systems for the analysis of the actual behavior such as the „behavioral equation“ of Kanfer and Phillips (1970) or the scheme of Schulte (1976) may be looked upon as devices of generating possible models of the underlying theory.
 The concept of „basic“ acts or actions is more fully discussed in Danto (1968) and von Wright (1991); the contexts, however, are somewhat different.