William S. Verplanck
“A purely descriptive system is never popular”
So far as scientific method is concerned, the system set up . . . may be characterized as follows. It is positivistic. It confines itself to description rather than explanation. Its concepts are defined in terms of immediate observations and are not given local or physiological properties. A reflex is not an arc, a drive is not the state of a center, extinction is not the exhaustion of a physiological substance or state. Terms of this sort are used merely to bring together groups of observations, to state uniformities, and to express properties of behavior which transcend single instances. They are not hypotheses, in the sense of things to be proved or disproved, but convenient representations of things already known. As to hypotheses, the system does not require them—at least in the usual sense.
It is often objected that a positivistic system offers no incentive to experimentation. The hypothesis, even the bad hypothesis, is said to be justified by its effect in producing research . . . and it is held or implied that some such device is usually needed. This is an historical question about the motivation of human behavior. There are doubtless many men whose curiosity about nature is less than their curiosity about the accuracy of their guesses, but it may be noted that science does in fact progress without the aid of this kind of explanatory prophecy. Much can be claimed for the greater efficiency of the descriptive system, when it is once motivated.
Granted, however, that such a system does possess the requisite moving force, it may still be insisted that a merely descriptive science must be lacking in direction. A fact is a fact; and the positivistic system does not seem to prefer one to another. Hypotheses are declared to solve this problem by directing the choice of facts (what directs the choice of hypotheses is not often discussed), and without them a distinction between the useful and the useless fact is said to be impossible. This is a narrow view of a descriptive science. The mere accumulation of uniformities is not a science at all. It is necessary to organize facts in such a way that a simple and convenient description can be given, and for this purpose a structure or system is required. The exigencies of a satisfactory system provide all the direction in the acquisition of facts that can be desired. Although natural history has set the pattern for the collection of isolated bits of curious behavior, there is no danger that a science of behavior will reach that level.
—B. F. Skinner, The Behavior of Organisms, pp. 44-45
I. The System
In dealing with Skinner, we are concerned with a theorist who now espouses no theory, a systematist whose system is still developing, and a constructive thinker some of whose most important contributions have been those of a critic.
In the course of his writings, Skinner has presented the results of a comprehensive experimental program, and elaborated a theory of behavior based upon it. Since its publication in comprehensive form in The Behavior of Organisms, he has, one may infer from more recent writings, modified it greatly by eliminating several central concepts without substituting others. These publications are not sufficient to enable us to analyze the system in its current status, so that we will restrict ourselves to its earlier form.
From an examination of this theory, we may learn something of the reasons for its alteration, and perhaps reveal some relationships between the adequacy of the theory as it was stated and the procedures which were followed in its construction. That portions of the theory as it was presented in 1938 no longer find complete acceptance is not relevant to our purpose; much may be learned from autopsies.
The revision of Sinner’s theoretical views has not extended downward to his basic assumptions with respect to the nature of psychological theory, nor to the elementary statements of much of his data language and of the basic laws of behavior. The systematic position is unchanged. It is largely at the level at which complex concepts are introduced that revisions have been made.
General Systematic Position
Skinner sees as the problem of modern psychology the development of a comprehensive system, or theory of behavior (44, 57) designed to predict and control (and hence to explain) the ongoing activities, motor and verbal, of living organisms. The system must grow from the raw data of behavior, without bias from preconceptions based on self-observation, on ill-defined concepts with theological connotations and capricious properties, or on the limited physiological knowledge of our day. Such a theory must be objective, in that it must take as its subject matter intersubjective events, descriptive and positivistic, in that it must be purely empirical, and it must refrain from elaborating mediating concepts borrowed from unrelated fields. It must be analytic, isolating significant and lawful relationships between behavioral dependent and environmental independent variables. Ultimately, it will be statable quantitatively, and consist in a set of empirical mathematical laws. Its vocabulary will include terms referring to the basic data, i.e., to the analytic units which are teased out experimentally and their relations to one another. The theory will have, as Skinner states at one place, “nothing to do with the presence or absence of experimental confirmation. . . . They are all statements about facts, and with proper operational care they need be nothing more than that” (57, p. 28). To speak of verifiability of theories, according to this view, is meaningless, since a theory will contain no statement which was not verified before it was placed, in conjunction with others, in the theory.
The unique characteristics of the system may be found implicit in this pre-systematic orientation. The emphasis on positivistic1 description is reflected in a definite rejection of formal theory with its explicitly stated postulates, axioms or hypotheses, and subsequent formal derivation of logical consequences which may be verified by experimental test.
Some logicians argue that the hypothetico-deductive is the unique method of science. They do so on the basis of their own analysis. But that is not what Skinner finds that most scientists do—they experiment on the basis of “intuition” or “hunch, ” elaborated from daily experience in and out of the laboratory. They arrive inductively at general laws and put these together into systems. The general laws come after the experimental facts are in (61). The great majority do not formally postulate, deduce, test—only the minority who have studied the logicians. There need be no formal check of a system for consistency and independence of its laws, for these are empirical assertions, and need not meet the requirements, at any given time, of a logic-tight system. The Maxwells and the Bohrs follow the Faradays and the Rutherfords. Laws are subject to modification or rejection as new experimental data come in, not with demonstrations of logical inadequacy. Experimental failure tells.
The positivistic position is interpreted as one that deals with behavior at its own level. The lawfulness, the orderliness to be found in observation of an organism interacting with its environment is statable without reference to events primarily dealt with by other sciences. To be sure, physiological, chemical, and physical events may uniquely correlate with that behavior, but statements about these, however lawful, however well worked out, add nothing to the basic data and laws with which the psychologist must concern himself. They are interesting, to be sure, and certainly the psychologist will not wish to formulate laws or to develop concepts which will be in any obvious conflict with those laws, but that is all. The psychologist, in dealing with his subject matter, may (and should) keep half an eye on the contents of the physiological journals; he should not feel it necessary to bind his own investigations and concepts to information found there. Presumably a comprehensive set of physiological correlates of behavior may be found, but there is nothing in behavioral data which renders this logically necessary or even possible. Psychologists, concerned with the investigation of behavior, lose time when they engage in physiological diversions.
The psychologist should equally be careful to refrain from the use of concepts derived from his “common-sense” thinking about his own activities, for the vernacular is deeply contaminated with older, usually animistic, systems and theories of the causation of behavior. The task of the psychologist is to relate behavior to the environment within which it occurs, and the laws of behavior will be reducible to statements of relationships between the two. Although it may be convenient for the present to make inferences with respect to “non-behavioral” intervening concepts (e.g., drive), these should be developed only when they are completely necessary, and they must then be assumed to be completely lawful, that is, never to act capriciously. They must, moreover, be stated only in terms of explicit behavioral operations. Formulations involving mental states, whether termed ideas, beliefs, or expectancies, or derived from colloquial verbal behavior with respect to “minds” cannot be readily freed, however redefined, from an element of capriciousness and subjectivity.
Skinner’s views on other current theories of behavior may be summarized— “A science of behavior must eventually deal with behavior in its relation to certain manipulable variables” (61). It will not be developed so long as theories are constituted of “explanation[s] of . . . observed fact[s] which appeal to events taking place somewhere else, at some other level of observation, described in different terms, and measured, if at all, in different dimensions . . .” (61). Such theories serve only to create a false sense of security and to produce research which is essentially wasteful. Skinner’s positivism looks for no models, whether in physiology, physics, or in the social behavior of little men housed in the cranium.
How can one pass from the laws derived from a restricted set of data to laws adequate to the whole area with which they purport to deal? For some theorists, there is no problem. A theory is set up to deal with a restricted set of data, and that alone; no attempt is made to extend the theory. Such miniature theories are readily exemplified. But Skinner’s specified area of interest is all the (lawful) behavior of all organisms in all environment. The inductive basis of the system is limited to rather restricted performances of members of a few species. When formal theory is lacking, one may expect that a mechanism for orderly extrapolation, as distinguished from analogizing, will be absent as well. And, indeed, it is. A basic terminology for behavior, derived from studies of the dog and of the white rat, is applied to the activities of other organisms, and so too, are certain experimental procedures. When this terminology and these laws are established, and when similar variables may be identified in more complex instances of behavior, extrapolation follows directly. But the statement of precise laws, and the detailed application of high-order concepts must await, for Skinner, the labor of experimentation in each new field. Such experimentation, it is assumed, will verify the extrapolation by analogy that must serve in the interim.
For Skinner, then, the work of the behavioral theorist is largely performing experiments, of the functional rather than the correlational type. Theory grows concomitantly in three explicit steps.2 One must identify the data which are dealt with and develop a consistent terminology for dealing with them, state laws relating classes of data to one another, and, when the number and precision of laws justify it, develop high-order integrative concepts which summarize the laws on a broad basis.
The implications of this position are too often overlooked. It is, in a sense, nihilistic. It proposes that all the conventional modes of thought in psychology, phenomenalistic, mentalistic, physiological, be rejected. It insists that psychologists begin their labors over again, that they develop their concepts from the ground up, and base them on the characteristics of the data themselves, and not on the language habits and intellectual biases of the theoretician. Earlier data may, where they meet the criteria of experimental control and orderliness of result, be salvaged, but earlier concepts may not.
General. The system, or, as we shall see, theory, of behavior developed by Skinner finds its intellectual forebears in the experimental tradition of the physiological investigation of the reflex, and its immediate empirical basis in the behavior of dogs in the Pavlovian stock, of rats (and, latterly, pigeons) in the Skinner-box,3 and of humans in their use of words and in guessing. Its theoretical constructs are remarkably free of the products of other trends in psychological thought, and the data accepted for treatment include no maze studies, no jumping-stand discriminations—indeed, the work of few psychologists working outside the framework of the system. Skinner is not interested in the experimental problems with which most learning theorists concern themselves, unless they are formulated as he might have done. The reasons for this parochialism are perhaps many, but two can be recognized as most important. Many psychological experiments are unsystematic and complex, that is, they involve scattered values of too many parameters of behavior, with the result that the relative control exerted by each cannot be evaluated. Second, they make use of measures of behavior that are considered limited in significance and treacherous to interpret.
The Skinner-box, which permits precise control of the environment, sharply limits the variety of behavior which can be manifested, and easily provides measures of rate of response, was designed early (23) in the course of the system’s development. It proved to be especially apt for the determination of behavioral laws not already found in the work of Sherrington (19) and Pavlov (17). The only other apparatus employed to any extent in the experimental phase of the program has been the activity wheel, used in certain studies of drive (26, 44). To be sure, Skinner has occasionally employed other kinds of apparatus, but these have been turned to the demonstration of the efficacy of the system’s concepts in their application to the control of behavior engineering-wise,4 and not to scientific research.
The system of behavior which has developed from these bases is essentially a simple one that has shown itself sensitive to the demands of new empirical data. The “finished” or relatively complete form of the theory has been fully presented in the book, The Behavior of Organisms (44).5 The changes which have been made since that time are not generally familiar to the psychological profession, for they are largely unpublished, and were not at all considered (although they were available) in a recent summary of learning theories (14) widely considered authoritative. The earlier form is that with which we are concerned, and it will be briefly outlined to serve as a point of departure for an analysis of behavioral theory construction as it is currently practiced.
Specific treatments. The basic products of analysis of the flow of behavior (“what [an organism] is observed by another organism to be doing”) (p. 6)6 in a changing environment are stimulus and response, defined, respectively, as “a part, or modification of a part, of the environment” (p. 9) and a part of behavior. A second product of the analysis of the environment may also be distinguished (although it is not explicitly stated as such in the system). This is the class of events called operations (e.g., deprivation and reinforcement): manipulations of the environment of the organism according to specified rules. The laws of behavior state orderly dependent relationships between properties of stimuli and of responses, and of operations and of responses. Two classes of reflexes, relationships between stimulus and response, are distinguished, according to the laws, static and dynamic, that pertain.
The kind of behavior that is correlated with specific eliciting stimuli may be called respondent behavior, and a given correlation a respondent. . . . Such behavior as is not under this kind of control I shall call operant. . . . The term reflex will be used to include both respondent and operant even though in its original meaning it applied to respondents only. A single term for both is convenient because both are topographical units of behavior and because an operant may and usually does acquire a relation to prior stimulation (pp. 20-21).
A respondent, then, regarded as a correlation of a stimulus and a response and an operant regarded as a functional part of behavior are defined at levels of specification marked by the orderliness of dynamic [vide infra] changes (p. 40).7
Neither stimulus nor response, then, may be defined independently of each other and both are classes of events, whose defining properties must be determined by experiment. Thus the identification of stimulus, response, and reflex require more than that the experimenter, on the one hand, present an object to the organism or otherwise produce a change in the environment, and that he observe, on the other, a correlated movement or act. He must demonstrate the repeatability of the observation, and he must also show by a series of experiments that this stimulus-response, or operation-response relationship behaves as do others already shown to obey the laws of the respondents and operants. These usages of the terms “stimulus” and “response” do not correspond to those of physiologists.8
Given this analysis of behavior and environment, Skinner has developed a series of laws, in part restated from Sherrington’s reflex physiology, in part from Pavlovian conditioning (“conditioning of type S”), and the remainder derived from his own researches. It is with these latter that we are concerned, since Skinner has not expanded a treatment, theoretical or experimental, of respondents. The system developed in The Behavior of Organisms, then, deals principally with operant behavior.
Although it appears that operant behavior includes all behavior that is not elicited by specific stimuli, in detailed treatment only those responses are termed “operants” that can be brought under experimental control through the operation of “reinforcement” (so that their rate of occurrence becomes high, and changes in an orderly and predictable way as a function of schedules of reinforcement, of “drive,” and, in some cases, of the presence of specific “discriminative stimuli”). They are spoken of as “emitted” rather than elicited since their first occurrence is neither predictable nor controllable except to a crude approximation; that is to say “spontaneous” (p. 20).
The presentation of reinforcing stimulus following the occurrence of an operant response is found to have two effects: first, that of increasing the strength of the response (as measured by its rate of occurrence), and secondly, of increasing the “reflex reserve,” a store of potential responses that may occur without further reinforcement and that show themselves in extinction. Thus if reinforcing stimuli are withdrawn, the animal continues to respond, at a declining rate, “until the reserve is emptied,” at which time the response reaches its initial strength and is no more predictable or controllable than it was before reinforcement was introduced.
Two quantifiable variables, aspects of the identified response, underlie these concepts: the rate at which the response occurs in the situation, and the total number of times the response is given after reinforcement is withdrawn. Both are typically obtained from response curves generated by the organism in the Skinner-box, which show the course of changes in rate and total number of responses, and hence in strength and reserve, and whose form approximates an “envelope,” a smooth and ideal extinction curve.
Skinner’s earlier laws relate these two dependent variables, through their set of correlated constructs, “reflex strength” and “reflex reserve,” to environmental variables. Additional concepts, some closely related to these (e.g., strength/reserve ratio, immediate reserve) and others, correlated with complexes of operations, (drive, emotion) enter. Further sets of operations permit the operant response to come under the control of discriminative stimuli whose presence “set the occasion” for (p. 22) the emission of (n.b., they do not elicit) a response. In these terms, Skinner is able to handle almost all the conventional data-set problems of conditioning and learning. This treatment may be briefly stated.
Conditioning. The acquisition of an operant C. R. is shown by an increase in the rate of response following reinforcement. “A C. R. is said to be conditioned in the sense of being dependent for its existence or state upon the occurrence of a certain kind of event, having to do with the presentation of a reinforcing stimulus (33).
[It] may be identified as such by showing . . . that it did not exist until the operation of reinforcement had been performed. It may also be distinguished by showing that through elicitation without reinforcement it is removed from the repertory of the organism. (p. 61). “The change in strength called conditioning is distinguished . . . by the specific operation that brings it about (p. 62).
One reinforcement (the presentation of a reinforcing stimulus) in the absence of complicating variables such as emotional stimuli, raises the (momentary) strength of the response to a maximum. As the number of reinforcements increases beyond one, the size of the reflex reserve increases, but the strength does not change. The reflex reserve has the “dimension” of a number of responses, which will occur without reinforcement. Reinforcing stimuli, such as food and water, must be presented immediately following the response if conditioning is to occur. As the presentation of a reinforcing stimulus is delayed, its effectiveness in changing response strength and in increasing the reflex reserve is decreased.
Stimuli that regularly produce behavior that has been reinforced by food (23) become reinforcing stimuli themselves.9
Extinction. As responses are made without reinforcement, the reflex reserve is depleted until it is emptied. The strength of response is a function of the size of the remaining reserve (the number of responses available to be made after extinction is complete), so that the rate of response drops off gradually to its unconditioned value (“operant level”). The process reverses the process of conditioning (24). It is dependent upon the occurrence of the response extinguished: “In a chain of reflexes not ultimately reinforced only the members actually elicited undergo extinction” (30).
Stimulus generalization (or equivalence). The strength acquired by an operant response through reinforcement is a function of the stimuli present at the time of acquisition, and will vary as a function of the degree of similarity of the stimuli present at the time it is emitted to those under which it was conditioned. The size of the reserve, however, is independent of these stimuli. This behavioral phenomena is called “induction.”
Response equivalence. If two parts of behavior, however different topographically, may be interchanged without altering the form of functions obtained when both may and do in fact occur, they fall into the same response class; i.e., they are both instances of the same response. Thus, a movement of either the right or left paw may constitute a “barpress.” Further analysis may be left to the physiologist; this is not the psychologist’s immediate concern.
Discrimination. “The strength acquired by an operant through reinforcement is not independent of the stimuli affecting the organism at the moment, and two operants having the same form of response may be given widely different strengths through differential reinforcement with respect to such stimuli” (p. 228).
Response decrement (not associated with fatigue). Other learning theorists deal with problems developing from experiments where the spacing of trials is varied in terms of concepts such as “reactive inhibition.” In the Skinner-box, when rate of response is the dependent variable, such effects cannot occur, and correspondingly, Skinner gives no treatment of this and allied problems.
Drive. “[Drive] presents itself simply as a class of dynamic changes in strength” (p. 23). Certain operations (withholding food from the animal) define “states” of the organism, of which the proportionality between response strength and reflex reserve is a function. These do not affect the size of the reserve. The learning-performance distinction is handled in terms of this ratio. It should be noted that reinforcing stimuli are most effective in conditioning when they are “appropriate” to the drive (p. 376). Thus: “Whenever we present a state of affairs which is known to be reinforcing at a given drive, we must suppose that conditioning takes place” (59), and “the reinforcing effect of a pellet of food varies linearly with the amount previously eaten” (38), and “the reconditioning effect of a single reinforcement is a function of the drive” (p. 402). Again, “in a conditioned operant, the drive governing the strength is determined by the reinforcement ” (p. 372). No theoretical account of this relationship between reinforcement and drive is given. Drive reduction theory is avoided as follows: “behavior that is strengthened during the heightened state of a drive usually leads to an operation affecting that drive” (p. 373).
Higher mental processes. No full accounts are given of the “higher mental processes,” which presumably include such behavior as matching, choosing, and so on. It is proposed, however, that they may be fully understood in terms of simple modes of behavior: “The data in the field of the higher mental processes transcend single responses or simple S-R relationships. But they appear to be susceptible to formulation in terms of the differentiation of concurrent responses, the discrimination of stimuli, the establishment of various sequences of responses, and so on. There seems to be no a priori reason why a complete account is not possible without appeal to theoretical processes in other dimensional systems” (61).10 Thus of choosing it is said, “choosing (like discriminating, matching, etc.) is not a particular piece of behavior. It is not a response, or an act with specified topography. The term characterizes a larger segment of behavior in relation to other variables or events” (61).
Such , briefly stated, is Skinner’s theory of behavior as it was outlined in 1938. We may note that no quantitative laws have been stated. There are a few equations which may have this status, one of which relates the number of responses made during the extinction to the time since reinforcing stimuli were withdrawn (N = k log t) (p. 88). Time, it should be noted, is the independent variable given in almost all of Skinner’s data; the other independent variables appear as parameters.
Since 1938, several changes have been made in the formulations presented in The Behavior of Organisms. In one case, an explicit change in the statement of the properties of a higher-order concept (reflex reserve) is made (48); thereafter the concept is termed “not particularly useful” and presumably dropped (61). In others, new concepts and terms are introduced in the place of those used earlier; these are formulated without the rigor needed to make it possible to restate the system. They are, however, explicit enough to indicate that certain older concepts are no longer advocated. Thus, “novelty” (61) of the environment displaces the concepts of reflex-reserve and reflex-strength in some contexts, and the notion of “probability of response” (61)11 in others.
It may be inferred that, since the publication of The Behavior of Organisms, both new results and the inadequacies of the system presented here have led Skinner to give up for the time being attempts to systematize the experimental data as extensively as in that book, and to devote himself to the gathering of experimental data that may serve to suggest further concepts, equally positivistic, to replace or extend the old (61).
Since this seems to be the case, there is little point in attempting to bring Skinner’s more recent work into the scope of the present critique.
II. A Structural Analysis of the System
The structure of Skinner’s theory, its special usages and systematic conceptions prove to be determined in many respects by the pre-theoretical point of view of its writer. Certain difficulties in the derivation of concepts, and in the relating of laws and primitive terms to their experimental bases appear. Similarities in terminology often obscure gross differences between this theory and some others; these exhibit themselves when the structure of the theory is extracted from its experimental and polemic context.
Among behavior theorists, Skinner has been concerned most explicitly with the problem of the general terminology and data language of psychology, a language whose function is unequivocal communication between individuals on the subject matter of the science, whose referents are the intersubjectively observable, and whose terms are the undefined primitives of theory.
The data language which Skinner employs is composed of several parts. The first is ordinary English, that is, the vernacular, with the very important restriction that all terms or grammatical constructions that imply conceptual schemes be barred. As Skinner puts it: “The sole criterion for the rejection of a popular term is the implication of a system or of a formulation extending beyond immediate observations” (p. 7). Thus, “here,” “feel, “try,” “need,” “in order to,” and “intention” cannot be included in the data language of a science of behavior, although in many cases it might be possible, by a Watsonian process of translation or by the recently fashionable and perhaps over-worked “operational definition”12 to give these terms a sharply restricted meaning within the data language, and so to introduce them into it.
Data language. In this first component, Skinner is in excellent agreement with other behavior theorists, and his usage meets all the criteria for an adequate data language for a science of behavior. Examples of this data language may be found in Chapter II, “Scope and Method,” of The Behavior of Organisms, e.g.:
The principle precaution that must be taken is in the handling of the animals at the beginning of an experiment. The effects of handling may be minimized by confining the rat behind a release door when it is put into the experimental box and allowing it to remain there for a minute or two after the box is closed and before the experiment proper begins. The release door should be reasonably silent in operation and out of reach of the rat when open. The drawing in Figure 1 shows such a door in place. It is operated by a projection of the shaft upon which it is mounted and is held against the ceiling when open (p. 57).
The entire behavior of lifting up the forepart of the body, pressing and releasing the lever, reaching into the tray, seizing the pellet of food, withdrawing from the tray, and eating the pellet is, of course, an extremely complex act (p. 51).
The movement of the lever is recorded electrically as a graph of the total number of responses plotted against time. The required apparatus consists of a slow kymograph and a vertically moving writing point. At each response the point is moved a uniform distance by an electrically operated ratchet. A step-like line is obtained, the slope of which is proportional to the rate of responding. The speed of the kymograph and the height of the step are chosen to give a convenient slope at the more frequent rates of responding. In Figure 2, some representative slopes are given for the coordinate values used in the greater part of the following account. The step-like character is not shown in the figure. [The movement of the lever operates the recorder by closing a mercury switch on the other side of the panel bearing the lever. In the first experiments with this method a needle attached to the lever-arm dipped into a small cup of mercury. When the lever was moved slowly there was a tendency for the contact to chatter, and this was corrected by inserting into the circuit to the recorder a device which made it impossible for a second contact to be recorded within, say, one second. It has been found that a commercial mercury tube switch does not require this precaution.] (Pp. 59-60.)
Doubtful data language. A second component of Skinner’s data language is in quite different status. Terms that function as data terms are introduced into the system without definition. Theoretical concepts are derived from them, but not in such a way that the derivations can be said to constitute implicit definition. These words do not meet one very important criterion for data language. Specifically, they are not such that agreement upon usage can be obtained from workers in the field regardless of theoretical biases and which are free of any reference to theory (that is, to the theory for which the set of terms in question functions as data language).13 Since these terms often appear in conjunction with theoretical concepts, and in the same statement of quasi-empirical laws, they introduce serious problems to the analyst. A simple example is the following:
A few special aspects of the records may be noted. The last horizontal line in each case might have been extended considerably to the right, for with the last recorded pellet the rat ceased eating altogether. The curves, therefore, end abruptly. This is typical of records obtained with either procedure. An example of another characteristic, which has already been discussed, is offered by Figure 3. At about the fiftieth piece of food the rat stops eating for a short period and consequently falls slightly behind the schedule set by the equation. The delay is followed, however, by a sharp acceleration, which brings the curve back to its proper position. After the first step, which is exceptional, the recovery curve is convex upward and remarkably uniform in its curvature. Note that its initial rate is greater than that of the beginning of the main record (23).
More difficult terms appear in further statements about cumulative response curves, and they are used to derive several higher-order constructs such as the “immediate reserve.” They include the following: “scallops” on the curves, “compensatory increase,” “grain,” “wave-like character,” “nearly maximal value,” “depressions,” and, especially, “envelop.” It is impractical to assemble an exhaustive list. We shall, however, analyze one such term in this context.
“Envelop” is introduced in this wise:
In Figure 8 we may inspect more easily the theoretical curves that have been fitted to the data. They are logarithmic and are drawn as envelops, upon the assumption that the deviations are depressions (p. 77).
A logarithmic envelop may be drawn above the curve (the broken line is for the equation N = K log t, where N is the number of responses at time t and K is a constant), although the actual contacts with the experimental curve are no justification for such a form in this particular case (p. 88).
The difficulty is this: the “envelop” refers to a line drawn above empirical curves. These lines seem to have been drawn so that they touch a number of points on the curve where the rate of response drops off sharply for a time, and so that they are approximately logarithmic. When equations are given for them, as in the example in the preceding paragraph, they are referred to as theoretical. Irregularities, departures from orderliness in empirical extinction curves are thus taken to be deviations below a “real” curve. Thus the term envelop is introduced into and used as part of the data language, drawn, perhaps, from mathematical English. It is not, however, used in the mathematicians’ sense, and it communicates little to the reader.
This use of the term disturbs the theory-analyst, since it plays a large role in the “explanation” of data and in the definition of the higher-order constructs already mentioned. Such data-language usage of the term envelop, with its consequent ambiguities, could be avoided if the writer had included among his defining laws one stating a specific form for the “extinction curve.” If this had been done, the family of concepts based on the envelop would have an explicit basis. One is, in fact, forced to conclude that just such a defining law (postulate) is involved in the theorist’s use of “envelop” but that it has been suppressed.
The third class of data language. Members of the third class of data-language terms are perhaps the most difficult to evaluate. As we shall see later, the theory incorporates a set of concepts that are defined in terms of the primitive laws and concepts of the theory. These concepts are reducible to the data language, through the chain of laws, formal definitions, and the usual paraphernalia of theory. These concepts are also linked to the data through a second series of steps. Such a dual linkage of concepts to data is not unusual: since the concepts are thus able to integrate two sets of laws, one might prefer them. Indeed, some would consider it essential to a concept that such explicit duality be demonstrable.
In the present instance, however, the second series of reduction chains are not clearly stated as such. The chain is omitted, and the theoretical terms are endowed with data-language status. An equivocation occurs. The analyst, interpreting what has been done after the presentation, may state that the theorist is giving the experimental symptoms of the concept introduced. Skinner, however, often writes as if he were making assertions in the data language about observable events. Examples of these usages which are pertinent to the present discussion include the following:
So defined a reflex is not, of course, a theory. It is a fact. It is an analytical unit, which makes an investigation of behavior possible (p. 9).
The reflex as an analytical unit is actually obtained in practice. The unit is a fact, and its validity, and the validity of the laws . . . do not depend upon the correctness of the analytical assumptions or the possibility of a later synthesis of more complex behavior (p. 29).
As we shall see later, a “reflex” is indeed a term that is formally defined, and is only eventually coordinated with statements in the data language.
These words do not have data-language status, but the theorist seems to be stating that they do. Not everyone can identify a reflex.
The data language employed, then, is basically sound, but it is contaminated by data-language usage of two other sets of terms. Tolman, we shall see is the section of this book dealing with his theory, also employs such a quasi-data language. The difficulties Skinner encounters in his data language are quite different from Tolman’s; they come from an uncritical introduction into the data language of terms that cannot be successfully used by many others, and from a tendency to treat the names of higher-order concepts as if they had data-language status, that is, as if they referred to events directly observable and atheoretical.
Basic Concepts and Laws
The technique followed by Skinner in introducing the primitive terms of the theory is straightforward. When the embedding material is stripped away, his procedure is this; the primitive terms, such as “stimulus,” “response,” “reflex,” “operant,” “respondent,” and “reflex strength” are implicitly defined in a set of formally stated laws. At this stage, the terms have no empirical content, and are not subject to any kind of independent definition. When the laws serving as rules for the use of the terms within the system have been stated, definitions coordinating these terms with data-language statements, and hence asserting an empirical status for the laws, are explicitly given. The whole operation, then, serves to define the subject matter of the theory and to make explicit those kinds of events with which the system deals. Given this system, still further laws are stated, which have been arrived at inductively through experimental procedures. It is necessary, then, to consider the status of these laws, and of the primitive concepts together.
The formal laws.14 The first twenty laws, which include five “Static Laws,” eight “Dynamic Laws,” and seven “Laws of Interaction,” are purely formal and are those that yield the definitions referred to above. “We are [however] in a position to demonstrate [them]” (p. 12).15 The balance have the character of empirical laws, since they have been established independently of both the other laws and of the definitional system, although once stated they play a formal role as well. Each of these formal laws involves a statement relating a dependent variable to an independent variable. These variables are specified as some property of the terms “stimulus,” “response,” “reflex,” “strength of a reflex”; they are brought into relationships with one another through the medium of a verbal structure which has the character of data language, descriptive of some manipulation. The defining laws are approximations to empirical laws in that they are statements based upon classical experimentation on the spinal reflex, and the salivary conditioned reflex. They are not empirical laws in the absence of independent definition, formal and coordinating, of the terms that appear in their statement.
The terms appearing in the laws are next brought into relationship with one another formally, in further statements:
Such a part, or modification of a part, of the environment is traditionally called a stimulus and the correlated part of the behavior a response. Neither term may be defined as to its essential properties without the other. For the observed relation between them I shall use the term reflex. . . . Only one property of the relation is usually invoked in the use of the term—the close coincidence of occurrence of stimulus and response—but there are other important properties to be noted shortly (p. 9).
A respondent [reflex], then, regarded as a correlation of a stimulus and a response and an operant regarded as functional part of behavior are defined at levels of specification marked by the orderliness of dynamic changes [i.e., obedience to the dynamic laws]” (p. 40).
In these formal terms, full definitions within the system of “operant” and “respondent” behavior are achieved, as well as of “reflex strength.” When this self-contained system is complete, then and only then are coordinating definitions that relate these terms to data-language statements and that assert empirical relationships, achieved. This resolution proceeds in an orderly way, beginning with the word “response,” which is coordinated, in The Behavior of Organisms, with “pressing a bar.” “Reflex strength” is coordinated with the frequency of occurrence of closures of a bar-operated relay. “Reflex strength,” incidentally, acts as both a primitive concept, and a higher-order concept, as we shall see later.
The empirical laws. Many empirical laws are formally stated; these appear in the appendix of this chapter. Others are not, and their generality remains questionable. For example, consider: “it may be concluded from the high frequency of occurrence of the instantaneous change that a single reinforcement is capable of raising the strength of the operant to essentially a maximal value” (p. 69). The informal statement suggests that this is not a very important generalization; the phrase “is capable,” that it is a tentative statement. But the next few pages of the text give five reasons that account for observed failures of the law. These are treated in some detail. The status of this law, like that of many others, is doubtful.
The Basic Constructs
Stimulus. The specification of those parts of the environment which are coordinated with the term “stimulus” is determined empirically. “Stimulus” refers to any part of the environment that is related to some specified operant or respondent according to the laws of the system. Thus, any specifiable part of change in a part of the environment may prove to be identifiable as a stimulus. Doors and bars may be stimuli, lights, buzzers, may be stimuli, and further specification need not be given. Any physical event or object, however, produced by and controlled by the experimenter is not a stimulus; it must be shown empirically to play a role in the lawful relationships that have been stated.
Stimuli are not, then, independent variables in the strict sense of the term, although once they have been identified they can perhaps be manipulated as such. The result is that, within the system, stimuli cannot be identified or specified in advance of an investigation. They must be discovered in the course of it. Parts of the environment may be manipulated by an experimenter as if they were stimuli, but it falls upon the experimenter to demonstrate that they do indeed control the dependent variable, frequency, or rate of a specified operant or magnitude or strength of a specified respondent according to the laws. It may prove impossible to specify stimuli verbally other than in such terms as “the bar,” “the lower right angle of a triangle,” and even “anything red.”
This treatment of “stimulus” is extended to human verbal behavior. Words are “stimuli” only if they act as such in lawful behavior. In an experiment in which abstract words were manipulated as stimuli, response distributions deviated from those obtained with other words, whose status as “stimuli” was unequivocal. In Skinner’s words, “This may mean, not that the relation is invalid, but that words of this sort are not properly to be regarded as simple units in the dynamics of verbal behavior” (42a).
The specified independent variables of behavior—stimuli—are then, in fact, “chosen by the organism,” i.e., they are inferred from the organism’s behavior just as are Lewin’s “behavior fields.” Once identified, however, they may be manipulated independently of it. In practice, it should be noted that this restriction of the word stimulus is overlooked, and “stimulus” is used much as others use it. If Skinner’s experimental interest in “stimuli” were greater and the use of the term in describing behavior more rigorous, the difference might be more evident. But so far, experiments concerned with determining the defining properties of a stimulus class16 are seldom encountered, and the striking difference between Skinner’s concept of “stimulus” and Hull’s is submerged.
Some remarks on “stimulus.” If one looks into the various definitions and usages of the word “stimulus” by psychologists, one finds at least four different ways in which the term is used.
The first usage, which we shall call “Stimulus I,” placed “stimulus” unequivocally in the data language, and corresponds closely to the verbal definition given by Skinner: a part, or a change in a part of environment. This simple definition is often amplified by the inclusion of statements about states or changes in states, of physical energy. In ordinary use, this amplification is not made, either because it is considered unnecessary or because it is not feasible. Stimulus I appears in the experimental learning literature as a data-language term (e.g., “the positive stimulus was a single white circle eight inches in diameter”), in psychophysiological experimentation (“the intensity of the stimuli presented varied in steps of .10 log units over a range of 1.40 log units, which made it possible to obtain the complete frequency of seeing function’), and is perhaps implicit in Hull’s position. The most explicit treatment of this use of the term is that of Bergmann and Spence (1).
“Stimulus II” corresponds with the concept of stimulus found in dictionaries of psychology, in physiology, and in textbooks. Both Harriman’s definition (9), “any form of energy which elicits a response,” and Warren’s (64), “an energy external to a receptor, which excites the receptor,” accept the basic definition of Stimulus I, but they limit the kinds of environmental events that will be termed stimuli to those classes that have been found empirically to produce some kind of response (unspecified) in some kind of organism (unspecified) under some kind of conditions (again, unspecified). Thus, a tone of 40,000 cps is now a stimulus if we are talking of bats, but not if we are concerned with men, and an illuminated area of a specified physical characteristics whose brightness is 3.50 log micro-micro-lamberts is a stimulus for Doakes, who is normal, but not for Oakes, who is night-blind. This definition places a limitation on the use of the term which is perhaps based on the fact that there is seldom any occasion, in psychology, to speak of energy changes to which no known organism ever responds (and hence which appear in no law). It leads to a shift of stress in the specification of environmental independent variables to the organism. This may have rather peculiar consequences; for example, one becomes rather uneasy in speaking of “sub-threshold stimuli.” It is no wonder that this definition of stimulus, supposed by many to be employed by all psychologists, is more honored in the breach than in the observance.
A third usage, Stimulus III, is that followed by Skinner. Here, “stimulus” refers to a class of environmental events that cannot be identified independently of observations of a specified activity of the organism and that must control that activity according to a specified set of laws. A red triangle of specified physical characteristics may not be termed a stimulus when it is repeatedly presented in association with food to, say, a dog, until the dog comes to salivate regularly in response to it according to the laws of behavior. Thereafter, it need be specified only insofar as it can be seen to control the specified behavior systematically. But if we find, upon further experimentation, that any red object controls the response, according to precisely the same laws, and also that triangles that are not red do not, then, by this usage, the red triangle can no longer be termed the stimulus, and something else, presumably “anything red,” is the stimulus. And so, although we may empirically identify manipulable objects and events that we may call stimuli, we do this on the basis of a construct, in Skinner’s case, the reflex—and the term stimulus is stripped of all data-language status. It is a quasi-independent variable, and when the term is used rigorously must be carefully stated as a stimulus-for-knee-jerk, stimulus-for-bar-press, and so on. A parallel concept of stimulation, the “releaser” is found in the work of the ethological school of students of instinctive behavior (62).
The fourth common usage of the term in psychology corresponds to the second usage, with the very important modification that hypothetical or inferential classes of physical events (usually intraorganismic) are also referred to as stimuli. This important extension of the concept “stimulus” is based upon physiological laws. Given these laws, and the observation of some of the events the physiological laws describe, the occurrence of appropriate stimuli is assumed. Stimulus IV, then, refers to purely hypothetical events, inferred from the behavior of the organism, much as Skinner’s “stimulus” is a logical construct derived from the observation of orderly behavior. Thus, Guthrie (8) speaks of “movement-produced-stimuli,” and Miller (16) of “response-produced-stimuli.” There is a difference between these two, however, since Guthrie uses the word indifference between these two, however, since Guthrie uses the word indifferently and miller seems to restrict himself to the use of stimulus for the hypothetical concept, and to use the word “cue” for what we have termed Stimulus I.
In neither of these last two usages, however, can the word “stimulus” be considered as referring to an experimentally manipulable environmental variable.
It is unfortunate that a single term is employed for such different concepts; theoretical difficulties will necessarily ensue, as they have for Skinner. By his indifferent use of the term—data-language-wise, and construct-wise, he produces a store of future difficulties for the system, although they are not immediately apparent in The Behavior of Organisms itself. In this work both usages I and III of the term are applied to the same environmental events, and the inconsistency seldom becomes apparent. In large part the superficial success of the ambiguity is an artifact of the particular experimental problems with which Skinner has concerned himself, and with his lack of interest in the stimulus control of behavior as such. If he had studied problems such as transposition in discrimination learning, the equivocation would quickly be evident17 and perhaps a solution found.
In Skinner’s system, then, the term stimulus does not refer to an entirely independent variable of behavior. Stimuli are not identifiable, except on the basis of the organism’s behavior; they are not manipulable,18 except by inference through the laws of the reflex.
This use of a term ordinarily employed in referring to the classical independent variable of behavior is a significant one. The restricted range of experiments on which the theory is based, and the extreme positivistic point of view both seem to have facilitated the easy intrusion of an equivocation and the evasion of its consequences, experimental and theoretical. The difficulty suggests, for all learning theorists, the need for a terminology that will not permit such an ambiguity to arise in the first place.
Such inference-backward to quasi-independent variables in behavior seems to be characteristic of the work of many behavior theorists. Guthrie, we have noted, speaks of “movement-produced-stimuli,” Lewin of behavior fields and forces, and others of perceptions. Quasi-variables are almost characteristic of psychological theory. It is not impertinent to ask whether, since this is so regularly the case, it will ever be possible to develop a science of behavior in which laws relating data-language-stimuli to data-language-responses can be found. Is it necessarily the case that “stimuli” become response-inferred concepts, bearing no necessary relationship to what is put in front of the organism?
One possible solution to this problem can be found in psychophysics. In dealing with the response to electromagnetic radiation, a parallel problem was encountered. The solution achieved required considerable experimental work, and was not easily arrived at. But by the introduction of a hypothetical construct— “retinal sensitivity to wave length” on the basis of empirical functions (the visibility curves), which were systematically related, on the one hand to the physical dimensions of the stimulus, and on the other to a simple response—a system has been constructed which provides for the rescaling of physical dimensions of the stimulus, and on the other to a simple response—a system has been constructed which provides for the rescaling of physical energies for convenient prediction of response, such that the manipulation of electromagnetic energy as a relatively simple independent variable, and the elaboration of laws relating various values of this variable to behavior become possible.
Few behavior psychologists have recognized the existence of this problem, and where they have (as Tolman may have, 63), the labors of developing the necessary constructs have perhaps led them to hold in abeyance work on this critical problem.
Response. The term response presents analogous difficulties, which remain unresolved within the theory. It is used indifferently in referring to a broad concept, the class of responses that produce the cumulative curves, which is “generically” defined (32), and to the single observed event. In the latter instance the term is used as if it were part of the data language. Operant responses are typically defined in terms of the “response,” and yet it is possible, it appears, to refer to “two operants with the same form of response,” (the 23d law, page 313 of this article, even before “two” operants have been distinguished on the basis of differences in discriminative stimuli.
The testability of the laws of behavior. It is clear from the foregoing that it is impossible to choose responses, or to specify stimuli or reflexes, as one wishes. The procedures for relating the terms of the theory to the data language require that in each new case where it is desired to use the terms, empirical investigations first show that the specific coordination leads to another verification of the laws of behavior. If it does not, then the behavioral and environmental variables were not “responses” and “stimuli.” Thus many of Skinner’s laws of behavior are exhibited as untestable. Those twenty that define “stimulus” and “response” are formal, definitional laws, which are not subject to disproof. They state the conventions for the application of the terms stimulus and response. They serve, in empirical application, to define that segment of the activities of an organism, those parts of his environment with which the theory will deal. Thus the theory only deals with that part of the activities of organisms that obeys its laws; this is behavior (cf. p. 273 above). Anything else that the organisms observed to do is presumably not behavior, for it is asserted that operants and respondents together include all behavior. To test the empirical content of this last statement is difficult. It is not at all impossible, or improbable, that much of what others call behavior may lie quite outside Skinner’s system. This is especially interesting when we consider the concept “reinforcement.” If a change in behavior which acts like conditioning can be shown to occur without the administration of “reinforcing stimuli,” then it simply does not fall within the scope of Skinner’s system. “Reinforcing stimuli” might have to be hypothesized if this were the case, even though they were unobservable. They might be termed “unidentifiable,” or “unspecifiable,” as is the case with the stimulus components presumed for operant reflexes before they are conditioned.
This state of affairs does not lead to difficulties as serious as it might, since in practice the defining laws can be restated if they prove to exclude too much of an organism’s activities from consideration. In effect, they can be treated as if they were verifiable.
The “operant-respondent” distinction. The system, it will be remembered, distinguishes between two broad classes of reflexes, “respondent” and “operant,” which are distinguished on the basis of the reflex laws which each follows. Respondents are defined as those reflexes which obey the “static” laws of the reflex; operants are those that do not.
Respondent behavior includes all responses elicited by stimuli. Operant behavior is “spontaneous,” in that its “originating forces . . . are not located in the environment. . . . It might be said to be emitted” (p. 20).19 “An operant is an identifiable part of behavior of which it may be said, not that no stimulus can be found that will elicit it (there may be a respondent the response of which has the same topography), but that no correlated stimulus can be detected upon occasions when it is observed to occur” (p. 21). The application of the term reflex to both “operant” and “respondent” behavior necessarily ensues from the preliminary formal development of the reflex concept.
[But] the term reflex [is] used to include both respondent and operant even though in its original meaning it applied to respondent only. A single term for both is convenient because both are topographical units of behavior and because an operant may and usually does acquire a relation to prior stimulation (pp. 20-21).
Respondent behavior varies as a function of stimulus intensity; operant does not. Operant behavior varies as a function of drive manipulation; respondent behavior, in most cases, does not. The latencies, magnitudes, and duration of respondents are significant dependent variables; only rate of response and total number of responses in extinction are legitimate measures of operant behavior. These differences follow from the differing set of laws20 entering into their definitions. No attempt is made to reduce this differentiation experimentally. On the contrary, there is an insistence on maintaining it. Although many experiments (e.g., 7) have shown that behavior which Skinner terms “operant” obeys the same laws as the static laws which define “respondents” when the appropriate experimental conditions (control of presentation of the stimuli, discrete trials, and so forth) are introduced, these are rejected because measures deemed inappropriate to operant behavior are introduced. Such experiments define their observables outside Skinner’s system. For reasons that are not clear to many psychologists concerned with learning theories, only “rate of response” is an acceptable measure of behavior.
The independent and dependent variables of the system. At the conceptual level the primary independent variables are stimuli, and the dependent variables responses, or properties of them. But this is only as it seems. In practice, and indeed in theory, the independent variables are quite otherwise. Consider the case of operant behavior.
The dependent variables measured in operant behavior are two. The first is the rate of response, and the second is cumulative number of responses. Both of these measures of behavior are directly related to theoretical concepts: strength and reflex reserve, respectively. Both are significant only after the “response” has been quantitatively defined. Typically, both are measured by the use of the cumulative response curves obtained from the operation of the Skinner-box. The whole curve is often reproduced, and numerical values are not taken from the records.
The independent variables are more complex and provide new theoretical problems. Presumably the basic independent variables of behavior are stimuli. But we have already seen that, as Skinner develops this concept, stimuli are not independently specifiable, as independent variables ordinarily are, nor do they appear in the experimental procedures of the program as independent variables. Rather, stimuli (together with time and the animal’s behavior) appear in the specification of operations, and it is these operations that act as the independent variables of the system. But even operations do not appear very often as independent variables in experimental work. They are the experimental parameters of particular experiments, and the actual independent variable is time: time in the box, time since beginning of extinction, and so on. In a given experiment nothing is varies, but time passes with the stated operations remaining constant as the data accrue.
Taken as a parameter of an experiment (and hence, after rearrangement of the data treated as an independent variable), an operation may lead to theoretical and practical difficulties.
Consider the most important of the operations, reinforcement: “the presentation of a certain kind of stimulus in a temporal relation with either a stimulus or a response” (p. 62). For operants, the “certain kind of stimulus,” or “reinforcing stimulus,” is any that has been shown, experimentally, to produce an increase in the strength of a response that it has followed at some time in the past. When a stimulus has been shown to have this property under a given drive, it is henceforth referred to as reinforcing. It is not the reinforcing stimulus21 that causes trouble, but the “temporal relation with . . . a response,” into which it may enter. The experimenter may manipulate the interval between response and reinforcement. He may select the number of unreinforced responses preceding a reinforced one (fixed-ratio). He may reinforce a response, on the average, every thirty seconds (a periodic reinforcement). But in all these cases, it is the responding animal that determines the delivery of the reinforcement, and hence that determines the values of many temporal parameters of reinforcement. In a very literal sense, the independent variables of Skinner’s system are under the direct control of the subject of the experiment, and not of the experimenter.22 Much the same difficulties, of course, can arise in other treatments of behavior.
This ambiguity of experimental control serves to reveal new and orderly phenomena of behavior. That all these phenomena will lead themselves easily to quantitative theoretical treatment is questioned, since it does not seem possible to specify accurately all the experimental conditions under which they appear. What is perhaps a more serious difficulty is that this self-imposed limitation on experimental control and the restriction of measurement to rates of response and to total numbers of responses given over stated periods of time automatically prevent those working within the system from finding certain behavioral effects. These effects occur when the temporal interval between successive presentations of given environmental changes (either “discriminative” or “reinforcing” stimuli) is varied. They have led to Hull’s concept of “reactive inhibition,” among many others. Although such laws do have a place in respondent behavior, they cannot be found in operant behavior by the methods described in The Behavior of Organism.
Higher-order constructs and relations among them. Tastes differ considerably in the specification of higher-order concepts. Variables related in a central construct by some differ from those related by others. From time to time, a particular theorist rearranges his order, seeking simpler and more effective schemata.
There are many constructs in Skinner’s system. They differ widely in the empirical laws that they relate to one another, in the rigor of their definition and treatment, and in the range of data that they integrate. It is neither possible, nor necessary, to treat with all of them exhaustively, to track down each usage to which each construct is put, or indeed, to specify all of them. Our present purpose, critical evaluation, can be adequately served by indicating, with some examples, the procedures followed in the formulation of several of Skinner’s higher-order constructs.
Skinner’s system incorporates two identifiable types of construct. The first type includes such concepts as “reflex strength,” “reflex reserve,” and “immediate reserve.” The first two of these are in part defined explicitly in the laws of the reflex; their definition is extended in a series of statements that may be found largely in the opening chapters of The Behavior of Organisms. All three are constructs that relate in a complex way the dependent variables of behavior (latency, rate and magnitude of response for respondents, and rate of response and number of responses to extinction for operants) as a function of independent variables (number of reinforcements and number of elicitations for respondents, and preceding rate and number of responses for operants). Although, as we have already noted, these are clearly constructs that integrate empirical laws, they are considered by Skinner to be “very near to being directly treated experimentally” (p. 26). They are treated as if they were basics of the system and not conceptualizations pulling together many properties of behavior not directly observable.
The second class of constructs includes those that relate various operations to changes in reflex strength, reflex reserve, and other members of the first class. They include “drive” and “emotion.”
Reflex strength. This term is most interesting analytically. Indeed, one term refers to two quite different concepts, one for respondents, and the other for operants. Respondent reflex strength is understood to “describe the sate of the reflex with respect to all its static properties [i.e., with respect to the static laws] at once. . . . The value of the strength of a reflex is arbitrarily assigned to it from the values of the static properties and is never measured directly. . . . [It] is not to be confused with the magnitude of the response” (p. 15). This concept is an intervening variable, defined in terms of a number of operations, and measured in terms of a number of measures of response. In practice, no attempt is made to develop the concept of respondent reflex strength beyond these statements.
Operant strength is defined otherwise. Since the operant is distinguished on the basis, among other things, of the failure of the static laws to apply, the former definition cannot be applied. We find that operant strength is defined as “proportional to its [a response’s] frequency of occurrence” (p. 21), and that “the dynamic laws describe the changes in the rate of occurrence that are brought about by various operations performed upon the organism” (p. 21). Thus where respondent strength is at once directly coordinated with the primary dependent variable of the experimental program, although, at the same time, it is treated as an intervening state variable in relationship to the concept of “reflex reserve.”
Reflex reserve. We have already indicated that this concept is a construct, which is occasionally spoken of as if it were a data-language term. Its hypothetical status is clear. It is defined as “available activity, which is exhausted during the process of repeated elicitation of which the strength of the reflex is at any moment a function” (p. 26). For respondents, the reflex reserve is assigned the property of constantly being restored spontaneously when it is not already at maximum. It is, then, related to the experimental facts of recovery from fatigue and adaptation for the respondent.
For the operant, the reserve is defined in direct relationship to the previously defined concept of strength, and also to the total number of responses given in extinction. Thus:
Since the strength of a reflex is proportional to its reserve, it may be altered in two ways. Either the size of the reserve or the proportionality between it an the strength may be changed. All operations that involve elicitation affect the reserve directly, either to increase or to decrease it. Conditioning increases it; extinction and fatigue decrease it. The other operations [which are not unique in their action and affect groups of reflexes] change the proportionality between the reserve and the strength. Facilitation and certain kinds of emotion increase the strength, while inhibition and certain other kinds of emotion decreases it without modifying the reserve. The operations that control the drive also affect the proportionality factor. Without altering the total number of available responses, a change in drive may alter the rate of elicitation of an operant from a minimal to a maximal value. Several demonstrations of the distinction between altering the reserve and altering the proportionality will appear later (p. 27).
The immediate reserve is a subordinate concept, which does not receive extensive theoretical treatment. The concept is introduced in this way:
The phenomenon of compensation, like that of recovery, requires the notion of an immediate reserve distinct from the total reserve which determines the rate in the absence of interruption. The process is catenary. The rate is proportional to the immediate reserve, which is contributed to from the total reserve. When elicitation is continuous, the total reserve controls the process. When elicitation is interrupted, the immediate reserve is built up; and a period of increased activity is made possible when responding is resumed, until the total reserve again becomes the controlling factor. The period of time during which responding may be suspended without making the original envelop inaccessible will depend upon the size of the immediate reserve (p. 85).
The “immediate reserve” is inferred from the phenomena of spontaneous recovery and from sets of data where a “depression” is followed by “compensation.” It is nowhere related systematically to another concept, based upon the opposite effect—where a high rate of response is followed by a low rate—which is called “strain on the reflex reserve.”23
The concept of “strain on the reserve” is not developed, experimentally or otherwise, although a set of statements such as the following may be assembled: “. . . we may assume that the effect of the accumulation of SDPR is to strain the reserve by bringing out responses which under normal discriminative stimulation would have remained within the reserve for some time” (p. 296) and “the recovery of the reserve from the strain imposed by the preceding run of responses” (p. 293). An attempt to tease out a set of rigorously stated relationships among these concepts meets with little success. Suffice it to say that both “immediate reserve” and “strain” are derived from the “envelop,” and hence from the suppressed postulate with respect to the form of the extinction curve which has been mentioned already. Both are used in the a posteriori explanation of results, but neither is formulated so that it can be manipulated directly or through control of the environment. They suggest that some kind of a model is involved.
Drive. Skinner’s concept of drive is perhaps one of the best specified and “purest” examples of what have been termed “intervening variables” that can be found in psychological theory. It is defined in terms of sets of operations and of response changes. For the drive “hunger,” the operations relate to various procedures involving food deprivation, and the response changes are alteration in the rate of occurrence of a variety of response (those that have to do with the ingestion of food). Drive is also related, through a series of laws, to the concept of the reflex reserve: the reserve/strength ratio is a function of drive strength. Similarly drive is related to the effectiveness of certain stimuli in reinforcing behavior. If food is reinforcing, then the animal is hungry. If an animal can become conditioned and shows extinction with the use of water as a reinforcing stimulus after a period of water deprivation, then it may be stated that thirst is a drive (32). The technique followed in conceptually developing a state variable in terms of a set of antecedent inducing operations, and a correlated set of related consequences is impeccable. But it is not quantitative.
Emotion. Emotion is also given the status of a concept of the “intervening variable” class, analogous to drive in that it must be defined in terms of a set of operations and correlated changes in reflex strength. “The important thing is the recognition of a change in strength as a primary datum and the determination of the functional relationship between the strength and some operation” (p. 409). The operations, in the case of emotion, remain only vaguely specified. One seems to be the withholding of reinforcing stimuli (extinction), and the other the presentation of emotional stimuli, which are not independently defined. Unfortunately, the concept of emotion appears frequently in the role of a deus ex machina, in that it is used to account for changes in strength which are observed when the conceptual system, taken with the experimental procedures, would otherwise lead to the expectation (but not prediction) that no change in strength would occur. Thus “when the lever has not been present prior to the day of conditioning, its movement may have an emotional effect, one result of which is a depression in rate” (p. 70).
A final set of higher-order constructs that has not been extensively treated includes those relating to human behavior. In most instances, the effect of this redefinition is to sharpen the content of the concept considerably. An example is this: “A simple way to state the fact of perseveration is in terms of the covariation in strength of groups of related responses” (40). Again, “Anticipation must be defined as a reaction to a current stimulus S1 which arises from the fact that S1 has in the past been followed by S2, where the reaction is not necessarily that which was originally made to S2” (6). “Guessing is a special kind of (usually verbal) behavior in which two or more responses are about equally likely to be emitted” (51). Not very much has been done with such concepts, either in the direction of redefining the whole vocabulary of psychology, or of pursing the experimental and theoretical implications of the redefinitions proposed.
Unclassifiable concepts. A few concepts employed in the system are not easily analyzed from the viewpoint of theory construction. Each is used in explanation. Two, at least, bear names that suggest that they are related to the laws of the reflex and hence that they represent no more than the application of these laws to the explanation of behavior. When the statement of the laws is examined, however, this hypothesis proves to be untenable. Another is unrelated to anything else in the system; it seems a purely ad hoc explanatory device.
An example of the first sort is “adaptation.” In some contexts the term is used as might be expected. In others, it is applied where the strength of an operant fails to reach its maximum after one reinforcement. Thus one reads that such failure may occur because “. . . administration of any stimulus to which adaptation has not taken place will normally depress the rate of eating” (24). This depression in rate is elsewhere refereed to as an “emotional effect” (pp. 70, 80). Unfortunately, adaptation is undefined with respect to the concept of emotion.
A more interesting case is this: In the course of his writings, Skinner has developed a powerful case against a concept of “inhibition” (cf. p. 290 above). All instances where such a process is inferred by others are shown to fit descriptively into a system that uses no such concept. In one experiment (42), however, benzedrine was shown to have the effect predicted by inhibition theorists. The concept of “total energy-output” was then introduced to account for the data. This concept finally appears as a “state of general excitability in which a response characteristic of the situation is emitted at a high rate” (p. 415). It is otherwise undefined, and is used only with reference to the one experiment.
Of the principal higher-order constructs of the system, then, some are defined in terms of the laws, and sometimes are given the status of being “directly observable.” Some are conceptually undeveloped and are introduced unsystematically on the basis of independent definition (envelop) or of as yet unspecified operations (emotions). A final one, drive, is clearly defined, and is employed in explanation in a way quite in accord with its definition.
Our analysis has been made, and it becomes possible to characterize the theory in fairly general terms. First highly formal, but not a highly formalized, theory. Its formal structure, so far as it goes, contrasts with its “empirical,” “inductive,” “descriptive,” “non-postulational” intentions. Rather than being a set of empirical laws embodying statements that represent inductive generalizations based on a set of terms initially defined in a data language, it is a set of formally defined terms, and defining laws, which are only coordinated with data-language statements after they have been fully stated. Stimuli and responses cannot be identified independently of the theory; they are defined by the theory for the theory. Similarly, the central variable of the system, with which the experimental program has been preoccupied, the operation “reinforcement,” rather than being inductively achieved as a central principle for the explanation of behavior, turns out to be a principle that serves, with some others, to define the area with which the theory deals. The actual independent variables of the system are different both from those of other systems and from those stated for the system.
The interesting result of this procedure, with its expository stress on empiricism and its structural stress on formality, is that despite very great differences in the meanings of many terms used in the theory from meanings of the same terms as used by other people, very few psychologists seem to be aware of the divergence. In fact, most SR psychologists if not students of learning in general, find its tenets familiar, and customarily employ many of its data-language usages. This is doubtless a tribute to the broad empirical program associated with the system and suggests that most behavior can indeed be successfully classified as operant or respondent. In practice, the terms seem theoretically neutral, and not only Skinner, but such divergent theorists and experimentalists as Schiller (18) and Graham (7) can refer to “operant responses” without introducing possible confusion or indicating the kind of theory they are suggesting. Indeed, an examination of recent publication reveals that Skinner’s data language and primitive concepts are in practice almost identical with those used by experimental followers of Tolman, Hull, and Spence, even though these men do not seem to have sought explicitly to develop or use a single data language. The careful empiricism of the coordinating definitions must account for these agreements on usage in the face of very real differences in the underlying concepts.
III. General Methodological Summary
On the basis of a survey of the mode in which skinner’s systematic treatment of behavior works out in practice and of a careful analysis of such explicit statements as may be found, it is possible to examine the position taken by Skinner on certain broad methodological issues that have become generally accepted as critical, if not crucial, in the evaluation of such systems and to indicate how the position is reflected in the structure of the theory.
The system of behavior expounded by Skinner in The Behavior of Organisms is explicitly a quantitative one. It has been developed with the intent of deriving empirical quantitative laws, presumably stated mathematically; the basic experimental procedures encountered are selected so as to yield physically quantifiable measures of behavior. Both the dependent and independent variables of behavior are defined in terms that should render easy the ultimate development of a fully quantitative system.
Yet in practice, in the actual gathering and analysis of data, quantification, i.e., measurement, the assignment of number, plays a small role. The data presented are, in almost all cases, quantifiable. The reader may extract from the data presented (which are largely copies of cumulative response curves) relatively precise measures of rates of response, of numbers of responses to extinction or to satiation and so on. Yet, these quantities, determinable though they may be, are not evaluated and play little or no role in the procedure by which general propositions are derived from the data. In practice Skinner is satisfied with statements of the type “more than,” “equal to,” and “less than.” He is more interested in statements of the sort “The retardation has been greatly reduced by the procedure. . . .” than in statements such as “Operation X reduces rate of response from 20 per minute at t = 0, to 5 per minute at t = 10 minutes.” In graph after graph, the independent variable is specified as “time in minutes,” “daily period,” or “days,” with no numerical scale. The intent toward quantification is not everywhere reflected in action.
Much the same gap between intent and execution is encountered in the statement of quantitative laws. On the one hand, it is possible to find verbal laws stated precisely enough to permit their translation into straightforward and unequivocal mathematical terms, but the statement is not made. On the other hand, quantitative laws are sometimes stated [in some contexts referred to as “empirical” (p. 189), and in others as “theoretical” (p. 88)] as relating variables within the system, when analysis shows that the mathematical statements are not in accord with the observations they are intended to describe.
The most interesting example of verbal laws stated in such a way that they may be translated into mathematical terms is found in the series of statements relating to extinction curves and the reflex reserve collected by Ellson (5). Ellson summarized these statements in the equation
(1) r = R (1 – e-ct)
where r is the number of responses at time t, R is the asymptotic number of responses (which has been defined as the reflex reserve), t is the time since the first response in extinction, and c is a constant. Although this equation does not itself appear in Skinner’s writings, Ellson states:
This is apparently the form used by Skinner to plot his theoretical curves; it permitted the duplication of the curves which he presents graphically within the limits of error of measurement.
In any event, this expression may be directly derived from explicit statements with respect to the proportionality obtaining between the magnitude of the reflex reserve, and the rate of response at any time. Experimental tests of this function led Ellson to conclude that the concept of the reflex reserve required rather drastic modification (a view which Skinner later adopted on the basis of his own experimental work); the function did not fit any large set of data. The empirical concept, “reflex reserve,” proved to be not empirical, but theoretical.
The second type of difficulty with the statement of quantitative laws is found in another law (p. 88), which describes the relationship of the number of responses to the time since the beginning of the extinction operation. This is of the form
(2) N = k log t,
where N is the number of responses in extinction (the reserve), t the time since the beginning of extinction, and k a constant. This function, of course, does not reach an asymptote, and N will increase logarithmically with t throughout the animal’s life in the Skinner-box, a result which is an implication contrary to the verbal statement of the concept of the reflex reserve and is unsupported by data. Again, the function describing “the rate at which a white rat eats a daily ration of a standard food” (20, 22),
(3) N = ktn
where N is the amount of food eaten in unit time, t is the time since the beginning of eating, and n and k are empirical constants, is not well adapted to describing the cyclical free-eating behavior of the rat and is especially incompatible with the phenomenon of satiation.
Skinner is quite explicit, however, in giving the reasons for his rather casual treatment of quantification. Two quotations will seem to clarify this:
The need for quantification in the study of behavior is fairly widely understood, but is has frequently led to a sort of opportunism. The experimenter takes his measures where he can find them and is satisfied if they are quantitative, even if they are trivial or irrelevant (p. 58). . . . there are many relevant variables, and until their importance has been experimentally determined, an equation that allows for them will have so many arbitrary constants that a good fit will be a matter of course and a cause for very little satisfaction (61).
And elsewhere, the science of behavior is characterized as not ready for
formal representations of the data reduced to a minimal number of terms (61).
Other theorists have argued, of course, that quantification serves quite different purposes from those implicit in Skinner’s statements—e.g., that it makes it possible to test the adequacy of general laws that have been stated verbally, and hence that it facilitates the progressive growth of both experimental data and theoretical constructions based on them. Ellson’s analysis is an interesting case in point, for the failure of the reserve concept exhibited by him foreshadows Skinner’s own rejection of it, on much the same grounds, but at a later date. The rigorous statement of the reserve in quantitative terms exhibited its emptiness.
There is another role in which numbers may appear in a science.
Statistics is a class of mathematics that may play a role in a theoretical system, and it represents a type of quantification that Skinner rejects. He is willing (pp. 442-443) that the system be termed “statistical,” if by statistical we mean correlational, and if by correlational we mean the explication of relations between dependent and independent variables.24 He distinguishes, however, two other usages of the term. One relates to the number of measurements made. It is assumed by some experimenters that only through the use of large numbers can smooth and orderly functions be obtained. The other concerns the relatively imprecise and unrefined measures obtained in some classes of experiment where statistics is emphasized. In these senses, the system is characterized as not a statistical one, and indeed it is not.
There are two other usages of “statistical” that we might examine. One relates to the use of probability theory models for the development of a theory of behavior. No such model appears in our basic reference, although more recent publications suggest that future formulations will rely in part upon some such formulation. The exact use to which the concept “probability” will be put remains unclear.
The last usage of the term which must be examined is perhaps the most common. Statistics is a field of mathematics which has provided many of the sciences with a method of relating conclusions, general statements and laws, to the data on which they are based. It embodies “rules of inference” which permit one to state whether the changing values of supposed independent variables do indeed relate to changing values of a supposed dependent variable. In the sense that a system or set of data is “statistical” if these rules of inference are used, the present system is not a statistical one. Small groups of animals are used, and orderly results are obtained. But it is left to the reader’s discretion whether he is to agree with the interpretations given and conclusions drawn. He is offered no statistical props, no mean values, no indices of variability, or tests of significance to lean upon.25 The argument is everywhere based upon the smooth and reproducible data presented. Statistics, like other forms of mathematics, is never employed for the verification of generalizations.
Reasons for this failure to use statistics in the manner typical of psychologists are given. With adequate experimental techniques, and proper measures of behavior, statistical procedures are obviated; exact rather than statistical laws, exact rather than approximate predictions, may be made of the behavior of individuals rather than populations of them.
We may summarize Skinner’s treatment of the problem of quantification in the following way: the system accepts implicitly the desirability and inevitability of quantitative laws in psychology; the data are quantifiable, and in some cases quantified. But quantification within the system is not achieved.
Degree of Axiomatization, Consistency, and Independence of Axioms
A scientist, in building a systematic treatment of his subject matter, necessarily accepts certain axioms with respect to his observations which are acceptable to all scientists; nature is orderly, and empirical science can reveal that order. The game is worth the candle. Skinner differs not at all from other scientists with respect to these axioms; he accepts them. A second class of axioms includes postulates of the explicitly stated type most characteristic of Hull (and perhaps essential to the writing of a finished scientific theory). Skinner has expressed himself clearly on the status of such axioms. A third set of axioms are assumptions peculiar to the general approach of a given systematist. They are seldom explicitly recognized. It is possible to find, implicit in Skinner’s writings, many unstated assumptions upon which much of his systematic thinking seems to be based.
The second class of axioms, the “postulates” of Hull’s system, have the character of general laws of behavior, which have been derived inductively. At the same time, they are treated as statements appearing from “nowhere” that constitute the skeletal structure of a theory. If we are to follow most analyses of theory construction, such axioms need not be derivable by any formal procedure of inference from an empirical foundation. It is asked of them only that they generate verifiable theorems. This characteristic seems typical of the application of mathematical theory to the data of the physicist.
Skinner, on the contrary, is preoccupied with the problem of arriving at empirical laws, by induction, from the data. He explicitly rejects theory construction by the axiomatic method.26 To quote, “For the man whose curiosity about nature is not equal to his interest in the accuracy of his guesses, the hypothesis is the very life-blood of science” (p. 426). The fundamental variables of a system must be experimentally isolated: there are too many possible consequences to make it probable for an experimental scientist to use a hypothetico-deductive procedure fruitfully.
Deduction and the testing of hypotheses are actually subordinate processes in a descriptive science, which proceeds largely or wholly without hypotheses to the quantitative determination of the properties of behavior and through induction to the establishment of laws (p. 437).
Hull, he states, has “failed to set up a system of behavior as distinct from a method of verification” (p. 436).
In developing a system, Skinner has, however, inadvertently written a theory and has formulated as empirical law statements of the sort that serve as the postulates of a system. These have been presented at some length elsewhere in this section. Such laws presumably have many of the properties of the axioms of a hypothetico-deductive system.
With his basic assumption that behavior is orderly, Skinner accepts the view that his laws must have logical consequences that are presumably not distinguishable from theorems. These consequences should be experimentally verifiable, else the inductive law is not adequately stated. They must, by the same argument, also be internally consistent. To quote Skinner,
The virtue of Hull’s work lies in an insistence upon the experimental validation of statements about behavior and upon the necessity of confining oneself to statements that are internally consistent and may be experimentally verified (p. 436).
The rejection of postulational technique is, then not an unqualified one: many “postulates,” or “assumptions” are acceptable for a scientific system. Thus we read “. . . these ‘assumptions’ are actually nothing but descriptive statements. . . .” and “. . . it is possible that any example of postulational method in the empirical sciences may be interpreted in the same way and that ‘predicting’ a fact from a set of assumptions is never intended to mean more than describing it. But if this is the case, the pretense of deduction should be abandoned, at least by those who are interested in a descriptive science of thought and who wish, therefore, to see the number of thought processes reduced to a minimum” (51).
It is not, then, unfair to inquire whether the laws of Skinner’s theory are consistent, independent, and verifiable. The evidence shows that many of them are not verifiable, but definitional. Whether or not they are independent cannot be determined in their present statement. Inconsistencies, it is submitted, have already been shown in the statements that define the principle constructs of the system. This is our finding, and not that of Skinner, for he has made little or no effort to examine either the logical status of his theory or the verifiability of the laws found therein.
Where Hull’s technique has led him and his followers to explore in detail the experimental status of many theorems derived from one or two postulates and to make a more or less conscientious effort to avoid inconsistencies within the system and between theory and data, Skinner’s approach has led him from generalization to generalization. Having arrived at a general law experimentally, he has not preoccupied himself with exploring its every implication, or with demonstrating that it is not inconsistent with some earlier law. He has rather gone ahead, to manipulate further variables in a search for further laws. The experimental consequences are evident. Skinner spends no time studying “latent learning.”
Skinner has uncovered a wide new range of phenomena, involving variables not at all considered by others and has obtained results beyond the scope of other theorist-experimenters. He has done so by foregoing careful tests of many of the implications of his system. If, in the course of experimentation not designed to test hypotheses, one of the implications of a law fails of verification, then modification occurs. Thus the experimental data that led to the abandonment of the concept of the reflex reserve as it was originally formulated were not obtained in an experiment designed to test the law, but rather in an experiment on drive in which the magnitudes of reflex reserves were relatively incidental measurements. “Rectification” occurs in Skinner’s system, but not only by the mechanism of explicit effort to test the theory nor by any formal procedure.
No attempt is made to demonstrate that the laws are consistent with one another. That they are is inferred only from their empirical status and the orderliness of nature.
Similarly, we cannot find any activity by Skinner calculated to test the independence of his laws. By his procedure, he needs not do so, since he does not propose them as independent postulates, even though he may use them as such. In any event, it is clear from other chapters in this book that the process of axiomatization, or postulation, is not necessarily an efficient or even foolproof method of avoiding inconsistencies in the statement of laws.
The third class of axioms that Skinner accepts are different from those we have just discussed. They are nowhere clearly stated as such. These axioms have the nature of prescriptive and proscriptive statements; they have served to mold the system in a variety of ways and to give it some of the special flavor which distinguishes it sharply from other systems. They are not easily teased out of the system. They are not stated explicitly, but lie embedded at the foundation of the whole approach to the problem of behavior. Nor can it be said how many of them there are. It is possible, however, to exhibit certain of the more important ones and to demonstrate their status as axioms from which both theoretical constructions and experimental procedures are developed.
The first of these is this: The behavior of an organism may be fruitfully analyzed into correlations of stimulus and response. Overt events, physicalistically described,27 are the sole object matter of the system. The apparent converse of this axiom, however, is not stated. Skinner does not axiomatically rule out of the reach of science “psychic” or “internal events”; they are simply not included in his subject matter.
A second such axiom may be stated this way: “A science of behavior which will permit the adduction of laws permitting the prediction and control of behavior can be developed without reference to the internal physiological events which may accompany behavior.” This statement has probably led to more misunderstanding of Skinner than any other feature of his systematic structure. It is not a denial of the occurrence of events that are systematically correlated with behavior. It is not a statement that laws of behavior cannot be developed from such events. It is not a statement prohibiting the statement of laws relating behavior to such events. It is, above all, not a dictum prohibiting psychophysiological investigation to the student of behavior, although it has been interpreted by one or another psychologist as any or all of these. It is a simple axiom which permits Skinner to proceed with his experimental investigations and system-building without reference to the physiology of the organism or to the structure and function of the neuro-skeletal system. The absence of physiological constructions and laws is attributable, not to the axiom in question, but to the quite different consideration that, strategically, it is more expedient to work at the behavioral level, since the science of physiology, where it is itself not based upon behavioral data, is not sufficiently well developed to permit its useful application to behavioral problems.
“Adequate experimental control of the independent variables of behavior and appropriate measurement of the dependent variables are achieved when orderly changes in the dependent variable occur, i.e., when smooth and reproducible curves appear on the recording paper. . . .” (p. 442) is an axiom that has not attracted the critical evaluation that it unquestionably deserves. Yet the definitions of the terms “stimulus,” “response,” and “reflex” derive from it. It has determined the level of specification at which these terms are coordinated with the subject matter, as well as the mechanical features of the recording apparatus. In both cases, the uncharitable have not hesitated to point out that the orderliness found in Skinner-box data is in part attributable to the use of crude cumulative measures of behavior. This same axiom is related to the rejection, in deriving general laws, of the use of any of the statistical tests that most psychologists find necessary to apply to their data before stating their conclusions. The smooth and orderly function is enough. Lastly, the axiom has produced, as corollaries, the concept of the “envelop of the extinction curve” (a smooth and orderly function from which the experimental data present departures), and several other terms and concepts developed for such departures. These have been discussed elsewhere, and we need not go over them again. Here is the suppressed axiom noted earlier.
When the statement of this axiom of orderliness is examined, the more extraordinary it becomes that it is taken as axiomatic and the more questionable the experimental and theoretical practices that have derived from it. It is not evident that adequate control and measurement necessarily yield “smooth and reproducible curves” (p. 442), nor that such smooth curves, such “orderliness of dynamic changes” (p. 40), can be obtained only by adequate “uniquely determined” (ibid.) control and measurement, nor that they mark a “unique point in the progressive restriction of a preparation” (p. 40). Neither statement is true in the physical sciences [and Skinner’s system seems modeled after physical chemistry (p. 434) ], nor need they be true in the behavioral sciences. Without denying the esthetic merits of smooth curves, whether they appear in treatises on analytical geometry or on the recording paper of a Skinner-box, one can argue that smoothness is an inadequate criterion for determining the care with which an experimenter will control his variables and for rejecting the statistical procedures for arriving at inductive generalizations that have earned wide acceptance in all the sciences.
Many scientific laws embody gross discontinuities. The “triple point” of water is a case in point. The occurrence of discontinuities in empirical data is not taken to be a problem for more or less careful experimentation, but rather as a problem for the theorist, who finds himself forced to talk about “states,” and to bring in new hypotheses, new principles which may account for the discontinuities. Since it is not Skinner’s goal to present, at this time, such a theory, it would seem incumbent upon him to accept discontinuities as something other than the inevitable and undesirable consequences of either too careless or too careful experimentation.
On the other hand, it is possible to get smooth and orderly functions where experimental manipulations are inadequate, although it is not necessarily easy to do so. Fortuitous combinations of variables and crude techniques of measurement by no means assure that the data will display discontinuities or that they will be unreproducible. For reproducibility all we need is to reproduce the same combination of variables. Let us take an example in psychology. We bring a subject into the laboratory for an hour a day and measure his reaction time, now to auditory, now to cutaneous, now to visual stimuli, both with and without ready signals. Let us measure, on any one day, the reaction time of anywhere from 200 to 2000 responses to such stimuli. On some days, our subject is fresh, on others fatigued, on some days hungry, on others stuffed from overeating. Let us use Marietta, Standard Electric, Hipp, and Dunlap, chronoscopes, all uncalibrated. Let us, furthermore, continue this procedure until 40, 0000 reaction times have been taken, and present the data in the form of a curve, in which “cumulative reaction time” is plotted as the function of “time in the laboratory’ or “number of stimulations.” We will get a smooth curve for the one subject that “looks like” a curve obtained by the same means from another subject. To be sure, if we introduce the obvious controls (derived from experiments where data are accepted as variable and necessitate the calculation of means and statistical tests of significance), the curves will be “smoother.” But how smooth is smooth enough? Given this axiom, on what criterion, except the perception of smoothness, are we to conclude that the curve we have obtained reveals, or does not reveal, inadequate control of stimulation, or adequate measurement of behavior? There is none, and our axiom is a dubious one indeed.
It can be asserted then, that something more than “smooth and reproducible” curves must serve as the basis for a definitive system of behavior, even though, within limits, this criterion may prove temporarily useful in some instances.
We shall not take time here to present an exhaustive set of inexplicit axioms such as, “There is no set of data adequate to define a concept ‘inhibition.'” There seem to be many, that vary widely in content. We shall rest with having pointed out that such a class of axioms can be distinguished in Skinner’s system.
The Use of Models
“Model” is a fashionable term, but of variable referent. If we restrict ourselves to the use of the term as it appears in the physical sciences (e.g., Riemannian space or the Newtonian mechanics of perfectly elastic spherules of the older gas theory), no models appear in Skinner’s system, nor could there be any in such a nonpostulational system.
If we use the term loosely, in the fashion of psychologists, we could defend the position that the spinal-reflex is taken as the model of all behavior but that the reflex reserve is based on a cash register, water-bucket,28 or reservoir, model, into which we may place determinate quantities of response-stuff and withdraw it in rather different units but with the total “stuff” available invariant. We will not defend it; “models,” in this sense, are insignificant analogies. It would be otherwise, of course, if these “models” were employed hypothetico-deductively.
Techniques of Derivation
It is evident that Skinner’s technique of derivation is entirely informal. Data are presented and graphs displayed. Statements describing them in the data language and the primitive terms of the system are added. The final description, in these terms, has the status of a law. After a number of such statements have accrued within a section dealing with a defined set of variables, a summarizing statement is found which states a law broadly embedded in collateral textual material.
In other cases, where general laws are presented, their formulation and presentation may be more formal. Thus, “The Law of the Discrimination of the Stimulus in Type S” is stated after a preliminary treatment which states the problem of discrimination in the terms of the system, and which is largely concerned with the rejection of alternative treatments.
Our analysis has been relatively detailed. It has been based on a well-established view of the role of theory in the development of science and of the structural characteristics of useful theory. Such analyses almost inevitably develop into listings of the sins of omission and commission of particular theorists, especially when one treats with a theorist whose views on theory diverge from the analyst’s.
Hull, Tolman, and even Guthrie and Lewin share with us much of our view of theory. It is, then, appropriate to point out gaps between many of their stated objectives and their performances. But it is not clear that Skinner is dealing with the same subject matter as these others. He often seems to be developing a system that not only differs from the others in particulars of theoretical detail, but also represents a re-orientation towards the science.
We have indicated certain shortcomings in the theoretical and systematic developments offered by Skinner. Many of his proponents will sharply dispute these judgments. Criteria highly suited to the examination of such highly axiomatized and deductive systems as Hull’s may be irrelevant to Skinner’s, since he has explicitly disavowed any intention of writing such a theory, and as we have seen, has used formal laws solely to define his subject matter. In its constructional aspects, however, Skinner’s work falls within the scope of the term “theory,” and can be so examined. Skinner has written a theory. On the other hand, it remains necessary to examine the status of the structure as a “system.”
Skinner states that he is concerned with the setting forth of a system. A system “consists of an aggregation of related variables, singled out for the sake of convenient investigation and description from all the various phenomena presented by a given subject matter. In the case of behavior, a system in this sense can be arrived at only through the kind of experimental analysis to which this book is devoted, in which the parts or aspects of behavior which undergo orderly changes are identified and their mutual relations established” (pp. 434-435).
Skinner, then, since he has defined his subject matter, considers that his work is solely one of empirical investigation, the assembling of an orderly body of experimentally established facts of the sort that must exist before a succinctly stated group of lawful relationships can summarize varied sets of experimental data within the area. The area is formally defined; only afterwards is the procedure empirical, descriptive, inductive, nonpostulational. A system, in this view, is not so much a “theory” as it is a Weltanschauung.
How successful has Skinner been in setting forth such as system? Certainly, the body of experimental fact is there and secure. Certainly, the most important variables that are manipulated, and the dependent ones that are measured, are established. The experimental substrate of a system has been developed. Much of the vocabulary and definitional system needed to handle the results of the program is given within the system. But what is available is not entirely adequate to define the kind of system Skinner is concerned with, and we must turn to the embedding material, what the systematist has to say about what he is doing, rather than to the formal statements themselves, for further illumination. When we do so, we find that the kind of system conceived is quite different from what the hasty reader thinks is there, and that it is indeed based on premises with respect to the nature of things which will not be universally accepted.
That Skinner’s concepts have often been misunderstood and misinterpreted probably stems from his choice of a set of terms. Implicative and associational values turn up frequently in the selection of a theorist’s terminology. To the “Tolmanite,” conditioned responses are mere, or mechanical. To the “Hullian,” expectancy and cognition carry the suggestion of the capricious intervention of entities extraneous to behavior. Skinner has attempted to avoid such considerations, and to eliminate the preconceptions (about what organisms ought or ought not to do) that may flow from the use of terms with extensive connotations. He wishes to find out how animals behave and seeks a vocabulary that will let him talk about how they behave. Because of the existence in Sherrington and Pavlov of sets of data of the kind he believes are needed, he has adopted many of their terms and applied some of heir laws in defining his area. As a consequence, he has been misinterpreted. In his choice of terminology, Skinner has assured that his works and those of his fellows will be read easily by the followers of Hull and Guthrie and only with emotion, if not with difficulty, by those who have selected the organismic-field-Gestalt-force family of words to work with. Skinner’s conditioned responses seem to many readers just as mere as those of Pavlov or Hull, with the extraordinary result that he has been classed with Hull rather than with Tolman, with Guthrie rather than Lewin, in his general position. Skinner’s work has, in fact, very little in common with that of any of these men. Controversies, such as those over “latent learning,” and “continuity” and “discontinuity” interpretations are pointless within the Skinnerian framework.
Skinner, as we have shown, does not endow his concept of stimulus with the physical-energy specification given to the word by Hull; nor “drive.” Hull uses the terms in rough correspondence with the usages of physiology, and they are stated in a physically reduced way. He employs them, not physicalistically, but physically. Skinner, on the contrary, does not attempt to write a set of statements about behavior which may be readily related to statements about protons, molecules, quanta, nerve impulses, or brain structures. Such a treatment, by its analytical omissions, produces disparities between physical world and psychological world, and so leads to a necessity for phenomenological investigations such as those of the Gestalt psychologist, that seek to put the “two” back together again. These problems are meaningful only when we accept as basic data for a point of departure the analytic products of other sciences. It may ultimately become desirable to incorporate reductive statements of some sort, as a more and more complete description of behavior develops. But it is not clear that this procedure is logically necessary or pragmatically desirable for a science of behavior that aims to predict and describe. The experimental analyses forced by an acceptance of reductionism of ten displaces scientific inquiries into behavior from the observation and quantification of the organism’s activities, to the elaboration of detail on relatively insignificant phenomena.
Skinner’s approach, then, bears no more than a terminological resemblance to Hull’s or to Pavlov’s, but it is at least first cousin to Kantor’s system (15), which explicitly rather than implicitly accepts a metaphysical position, naive realism, and rejects even the logical possibility of a reductionism. His approach has affinities to Tolman’s. Tolman postulates that the so-called laws of perception, derived from phenomenological studies, apply to the rat; Skinner does what amounts to the same thing implicitly, by starting with what comes to him, to all the other experimenters, and until proven otherwise, to the experimental animal. Hull, on the contrary, seems to wish (where he does not evade the issue by speaking of “stimulus interactions”) to derive “perceptual” laws on the basis of his reductively stated postulates. Skinner wants to start with a point-at-able world, with point-at-able operations, and to carry on from there. He accepts as his point of departure the world of things and activities and leaves to others, who start reductively, the fields of “perception” and “sensation.” It is often with surprise that persons most familiar with earlier frames of references in psychology recognize that this is true of other current behaviorists. Physiological elementarism, in the style of Watson, is not a necessary characteristic of today’s behaviorists.
As a systematist, then, Skinner attempts a clean break with much of traditional psychology, a fact that has escaped many psychologists.
How does Skinner’s work stand if it is evaluated in terms of its own stated objectives? One must say that it survives examination more successfully than when it is viewed as just another theory. Its very faults seem to accent its fundamental merit; they come from the fact that, even with the best of intentions, Skinner has not been able to eschew theory as he wished. Theoretical constructs have appeared within the system, and have assigned properties that do not appear in the data. Data-language terms have accrued surplus meanings. Statements incorporating operationally defined terms assert relations that have not yet been experimentally demonstrated. If Skinner had followed the critical analysis of theory construction that we have expounded, and if the highly formal nature of much of the system had been recognized, perhaps some of these difficulties might have been circumvented. Skinner, too, would object to the very things to which we have objected.
The experimental program, on the other hand, has developed in a satisfactory manner. The experimental data that have accumulated over the years do not show the result of a preoccupation with “critical” (and complex) experiments, whose outcome becomes a matter of indifference to all, as soon as the theoretical point to which they were addressed is re-evaluated. In any event, as a system, Skinner’s is still growing, and data-gathering has outstripped theory.
We may finally test the system against those evaluative criteria that were proposed in an earlier section of this monograph.
Clarification of events. Does the system “clarify the description of the world which is possible in ordinary language?” There is little doubt that in eliminating, in an explicit way, terms of loose referent and vague connotation, Skinner has enabled many theorists to find for themselves suppressed premises in their thinking. Beyond this consideration, the terms employed in the system are such that many phenomena may be restated to reveal relationships that might otherwise be overlooked. These results, in turn, necessarily lead to experimental investigation. The same terms, we have noted, however, have led to confusion about the content of the system.
Summarization of existing data. How effectively does the system summarize existing knowledge? Skinner has sharply delineated the classes of existing scientific knowledge that he is willing to incorporate within his system. While doubtless accepting as highly defensible the proposition that any set of scientific psychological data may be interpreted in terms of his variables, he explicitly refuses to admit, as a basis for scientific induction, the results of most psychological experiments. The reason for this is that the selection, manipulation, and control of the independent variables is so erratic and incomplete that significant sets of data are lacking. Further, the measures of behavior that have been used are meaningless. Many experimental designs are, finally, too complex for fruitful analysis. To a remarkable degree, the theory is applied only to behavioral experimentation in its defined area, so that it “fails” to handle many data for the simple reason that it does not attempt to do so.
For the same reasons that most existing psychological data are ignored in developing the system, the system is not responsive to data currently obtained in non-Skinnerian experiments. Only when the variables have been adequately selected and manipulated according to the standards of the system, and when appropriate measures of behavior have been made does the system take account of experimental data. Then it shows itself highly sensitive to experimental results. Concepts well established in the system such as “reflex reserve”29 have been abandoned since their original statement on the basis of experimental results (48) without any particular attempt to reformulate them or otherwise to salvage them. Systematic problems are likewise faced, and new hypothetical constructs have been, or are being, developed to account for new classes of data. The experimental program proceeds, and no particular attempt is made to “tidy up” the theorizing that goes with it. In short, the system is highly sensitive to data deemed appropriate. Almost no effort is made to reinterpret, in terms of the system, data gathered in other experimental contexts or in any other way to make use of them or to synthesize them within the system. The data may be valid (i.e., reproducible), but they are taken as inappropriate for the systematic science of behavior toward which Skinner is striving. The system, them, summarizes such knowledge as has been obtained by the specified experimental techniques.
Prediction. Does the system mediate the application of knowledge to new situations? Does it predict?
Some systems or theories of behavior lay great stress on their ability to predict the outcome of planned experiments—often taken to be experimenta crucis. It is not surprising, however, that a theory of restricted empirical basis, “informally stated,” and “inductively” developed does not generate rigorous predictions about the behavior to be observed in novel situations. In fact, such statements as these must be qualified. Several aspects of Skinner’s view of the problem of prediction and extrapolation must be treated individually. Although one of our more conservative theorists when he is making statements about prediction, Skinner is more willing than most to extrapolate his concepts from the situations in which they have been developed to some of the more intricate cases of human behavior. This willingness is clearly indicated in the title of the treatise, The Behavior of Organisms, which deals with the white rat in the Skinner-box.
Two situations may be distinguished in which a systematist may wish to make predictions. The first is that in which a logical or operational analysis shows that the “same” variables that have already been isolated and studied in the laboratory are operative elsewhere in the “same” relationships that have been investigated. The data are in, and the theorist simply asserts the genotypicality of the situation, applies his theory, and “predicts” the course of events. In the other situation, familiar variables may be encountered, but in novel configurations, and the theorist is required to generate statements that go beyond those he has already made. Or, again, new variables may be encountered, and, again, prediction may be called for. While Skinner will predict, or rather again extrapolate freely where a logical analysis reveals familiar variables acting in familiar ways, he will not predict at all under other circumstances. As a consequence, it is possible to find no predictions at all of the behavior of rats, or of pigeons, when novel combinations of stimuli are presented to them in the Skinner-box, and many predictions among Skinner’s writings with respect to human behavior in a social environment. The great difficulty is that these predictions are usually unverifiable, because of the complexity of the situation and the consequent impossibility of meaningful experimental test. Prediction, then, is represented by extrapolation, by analogy; its use for the generation of propositions that may be put to experimental test is avoided.
Does the system lead to fruitful lines of experimental inquiry? There can be no doubt that the theory has been associated with a very active experimental program. There can be no doubt that its atheoretical orientation dictates that experimentation shall proceed at all times, with full exploration of the behavioral consequences of varying the many experimental variables that the system discovers and treats. Many experimenters have argued that such parametric approaches to experimentation are fruitless, and that only experimentation directed toward particular theoretical points have the merit of fertility. If anything, the development of the Skinnerian system argues against this view. The theory does not predict, and experimentation is not designed to check, theoretical statements. Yet the experimental program has been an exceedingly productive one. The experimental pages of The Behavior of Organisms, speak for themselves. Most particularly, the area of research on schedules of reinforcement was first explored and developed within the system. This set of results was not anticipated in other theoretical frameworks and still represents a difficulty for them. It constitutes a body of data that very likely will have considerable influence on theory for years to come. The experimental stress of the theory has freed the experimental program from restrictions forced by respect for the integrity of theoretical constructs. It is difficult to find a match in either quantity or variety in the works of other theorists with whom we have been concerned for the experimental program associated with this system. That this reflects much with respect to the system may be doubted. It is certainly a tribute to the systematist, and it demonstrates the force of his strongly empirical position.
The Formalized Laws of Behavior
(The Behavior of Organisms)
1. The Law of Threshold. The intensity of the stimulus must reach or exceed a certain critical value (called the threshold) in order to elicit a response.
2. The Law of Latency. An interval of time (called the latency) elapses between the beginning of the stimulus and the beginning of the response.
3. The Law of the Magnitude of the Response. The magnitude of the response is a function of the intensity of the stimulus.
4. The Law of After-Discharge. The response may persist for some time after the cessation of the stimulus.
5. The Law of Temporal Summation. Prolongation of a stimulus or repetitive presentation within certain limiting rates has the same effect as increasing the intensity.
6. The Law of the Refractory Phase. Immediately after elicitation the strength of some reflexes exists at a low, perhaps zero, value. It returns to its former state during subsequent inactivity.
7. The Law of Reflex Fatigue. The strength of a reflex declines during repeated elicitation and returns to its former value during subsequent inactivity.
8. The Law of Facilitation. The strength of reflex may be increased through presentation of a second stimulus which does not itself elicit the response.
9. The Law of Inhibition. The strength of a reflex may be decreased through presentation of a second stimulus which has no other relation to the effector involved.
10. The Law of Conditioning of Type S. The approximately simultaneous presentation of two stimuli, one of which (the “reinforcing” stimulus) belongs to a reflex existing at the moment at some strength, may produce an increase in the strength of a third reflex composed of the response of the reinforcing reflex and the other stimulus.
11. The Law of Extinction of Type S. If the reflex strengthened through conditioning of Type S is elicited without presentation of the reinforcing stimulus, its strength decreases.
12. The Law of Conditioning of Type R. If the occurrence of an operant is followed by presentation of a reinforcing stimulus, the strength is increased.
13. The Law of Extinction of Type R. If the occurrence of an operant already strengthened through conditioning is not followed by the reinforcing stimulus, the strength is decreased.
Laws of Interaction
14. The Law of Compatibility. Two or more responses which do not overlap topographically may occur simultaneously without interference.
15. The Law of Prepotency. When two reflexes overlap topographically and the responses are incompatible, one response may occur to the exclusion of the other.
16. The Law of Algebraic Summation. The simultaneous elicitation of two responses utilizing the same effectors but in opposite directions produces a response the extent of which is an algebraic resultant.
17. The Law of Blending. Two responses showing some topographical overlap may be elicited together but in necessarily modified forms.
18. The Law of Spatial Summation. When two reflexes have the same form of response, the response to both stimuli in combination has a greater magnitude and a shorter latency.
19. The Law of Chaining. The response of one reflex may constitute or produce the eliciting or discriminative stimulus of another.
20. The Law of Induction. A dynamic change in the strength of a reflex may be accompanied by a similar but not so extensive change in a related reflex, where the relation is due to the possession of common properties of stimulus or response.
21. The Law of the Extinction of Chained Reflexes. In a chain of reflexes not ultimately reinforced only the members actually elicited undergo extinction.
22. The Law of the Discrimination of the Stimulus in Type S. A reflex strengthened by induction from the reinforcement of a reflex possessing a similar but not identical stimulus may be separately extinguished if the difference in stimuli is supraliminal for the organism.
23. The Law of the Discrimination of the Stimulus in Type R. The strength acquired by an operant through reinforcement is not independent of the stimuli affecting the organism at the moment, and two operants having the same form of response may be given widely different strengths through differential reinforcement with respect to such stimuli.
24. The Laws of the Operant Reserve. The reinforcement of an operant creates a single reserve, the size of which is independent of the stimulating field but which is differentially accessible under different fields.
1. Bergmann, G., & Spence, K. W. The logic of psychophysical measurements. Psychological Review, 1944, 51, 1-24.
2. Breland, Keller, & Breland, Marion. A field of applied animal psychology. American Psychologist, 1951, 6, 202-204.
3. Carnap, R. Logical foundations of the unity of science. In H. Feigl & W. Sellars. Readings in Philosophical Analysis. New York: Appleton-Century-Crofts, 1949. pp. 408-423.
4. Cook, S. W., & Skinner, B. F. Some factors influencing the distribution of associated words. Psychological Record, 1939, 3, 178-184.
5. Ellson, Douglas G. The concept of reflex reserve. Psychological Review, 1939, 46, 566-575.
6. Estes, W. K., & Skinner, B. F. Some quantitative properties of anxiety. Journal of Experimental Psychology, 1941, 29, 390-400.
7. Graham, C. H., & Gagné, R. M. The acquisition, extinction, and spontaneous recovery of a conditioned operant response. Journal of Experimental Psychology, 1940, 26, 251-280.
8. Guthrie, E. R. The Psychology of Learning. New York: Harpers, 1935.
9. Harriman, P. L. The New Dictionary of Psychology. New York: Philosophical Library, 1947.
10. Hempel, C. G. The logical analysis of psychology. In H. Feigl & W. Sellars. Readings in Philosophical Analysis. New York: Appleton-Century-Crofts, 1949, pp. 373-384.
11. Heron, W. T., & Skinner, B. F. Changes in Hunger during starvation. Psychological Record, 1947, 1, 51-60.
12. ——. The rate of extinction in maze-bright and maze-dull rats. Psychological Bulletin, 1939, 36, 520. (Abstr.)
13. ——. The rate of extinction in maze-bright and maze-dull rats. Psychological Bulletin, 1940, 4, 11-18.
14. Hilgard, E. R. Theories of Learning. New York: Appleton-Century-Crofts, 1948.
15. Kantor, J. R. Principles of Psychology. New York: Knopf, 1924.
16. Miller, N. E. Studies of fear as an acquirable drive: I. Fear as motivation and fear-reduction as reinforcement in the learning of new responses. Journal of Experimental Psychology, 1948, 38, 89-101.
17. Pavlov, I. P. Conditioned reflexes. London: Oxford University Press, 1927. (Transl. G. V. Anrep.)
18. Schiller, P. S. Analysis of detour behavior. I. Learning of round-about pathways in fish. Journal of Comparative and Physiological Psychology, 1949, 42, 463-475.
19. Sherrington, C. S. The Integrative Action of the Nervous System. New York: Scribner, 1906.
20. Skinner, B. F. On the conditions of elicitation of certain eating reflexes. Proceedings of the National Academy of Sciences, 1930, 16, 433-438.
21. ——-. The concept of the reflex in the description of behavior. Journal of General Psychology, 1931, 5, 427-458.
22. ——-. Drive and reflex strength. Journal of General Psychology, 1932, 6, 22-37.
23. ——-. Drive and reflex strength. Journal of General Psychology, 1932, 6, 38-48.
24. ——-. On the rate of formation of a conditioned reflex. Journal of General Psychology, 1932, 7, 274-286.
25. Skinner, B. F. On the rate of extinction of a conditioned reflex. Journal of General Psychology, 1933, 8, 114-129.
26. ——-. The measurement of “spontaneous activity,” Journal of General Psychology, 1933, 9, 3-23.
27. ——-. The rate of establishment of a discrimination. Journal of General Psychology, 1933, 9, 302-350.
28. ——-. “Resistance to extinction” in the process of conditioning. Journal of General Psychology, 1933, 9, 420-429.
29. ——-. The abolishment of a discrimination. Proceedings of the National Academy of Sciences, 1933, 19, 825-828.
30. ——-. The extinction of chained reflexes. Proceedings of the National Academy of Sciences, 1934, 20, 234-237.
31. ——-. A discrimination without previous conditioning. Proceedings of the National Academy of Sciences, 1934, 20, 532-536.
32. ——-. The generic nature of the concepts of stimulus and response. Journal of General Psychology, 1935, 12, 40-65.
33. ——-. Two types of conditioned reflex and a pseudotype. Journal of General Psychology, 1935, 12, 66-77.
34. ——-. A discrimination based upon a change in the properties of a stimulus. Journal of General Psychology, 1935, 12, 313-336.
35. ——-. A failure to obtain “disinhibition.” Journal of General Psychology, 1936, 14, 127-135.
36. ——-. The reinforcing effect of a differentiating stimulus. Journal of General Psychology, 1936, 14, 263-278.
37. ——-. The effect of the amount of conditioning of an interval of time before reinforcement. Journal of General Psychology, 1936, 14, 279-295.
38. ——-. Conditioning and extinction and their relation to drive. Journal of General Psychology, 1936, 14, 296-317.
39. ——-. Thirst as an arbitrary drive. Journal of General Psychology, 1936, 15, 205-210.
40. ——-. The verbal summator and a method for the study of latent speech. Journal of Psychology, 1936, 2, 71-107.
41. ——-. The extinction ratio and its modification by a temporal discrimination. Psychological Bulletin, 1936, 33, 784. (Abstr.)
42. ——-, & Heron, W. T. Effects of caffeine and benzedrine upon conditioning and extinction. Psychological Record, 1937, 1, 340-346.
42a. Skinner, B. F. The distribution of associated words. Psychological Record, 1937, 1, 71-76.
43. ——-. Two types of conditioned reflex: a reply to Konorski and Miller. Journal of General Psychology, 1937, 16, 272-279.
44. ——-. The behavior of organisms: an experimental analysis. New York: Appleton-Century-Crofts, 1938.
45. ——-. The frequencies of occurrences of associated words. Psychological Bulletin, 1938, 35, 675. (Abstr.)
46. ——-. The alliteration in Shakespeare’s sonnets: a study of literary behavior. Psychological Record, 1939, 3, 186-192.
47. ——-. A method of maintaining an arbitrary degree of hunger. Journal of Comparative Psychology, 1940, 30, 139-145.
48. ——-. The nature of the operant reserve. Psychological Bulletin, 1940, 37, 423. (Abstr.)
49. ——-. A quantitative estimate of certain types of sound patterning in poetry. American Journal of Psychology, 1941, 54, 64-79.
50. ——-. Some quantitative properties of anxiety. Psychological Bulletin, 1941, 38, 539. (Abstr.)
51. ——-. The processes involved in the repeated guessing of alternatives. Journal of Experimental Psychology, 1942, 30, 495-503.
52. ——-. Reply to Dr. Yacorzynsky. Journal of Experimental Psychology, 1943, 32, 93-94.
53. ——-. Review of Hull’s principles of behavior. American Journal of Psychology, 1944, 57, 276-281.
54. ——-. The operational analysis of psychological terms. Psychological Review, 1945, 52, 270-277.
55. ——-. Differential reinforcement with respect to time. American Psychologist, 1946, 1, 274-275.
56. ——-. The effect of the difficulty of a response upon its rate of emission. American Psychologist, 1946, 1, 462. (Abstr.)
57. ——-. Experimental Psychology. In W. Dennis (ed.), Current trends in psychology. Pittsburgh: University of Pittsburgh Press, 1947.
58. ——-. “Superstition” in the pigeon. American Psychologist, 1947, 2, 426. (Abstr.)
59. ——-. “Superstition” in the pigeon. Journal of Experimental Psychology, 1948, 38, 168-172.
60. ——-. Concurrent operants. American Psychologist, 1948, 3, 359. (Abstr.)
61. ——-. Are theories of learning necessary? Psychological Review, 1950, 57, 193-216.
62. Tinbergen, N. The hierarchical organization of nervous mechanisms underlying instinctive behavior. In Physiological Mechanisms in Animal Behavior. New York: Academic Press, 1950.
63. Tolman, E. C. Purpose Behavior in Animals and Men. New York: Appleton-Century-Crofts, 1932.
64. Warren, H. C. Dictionary of Psychology. New York: Houghton-Mifflin, 1934.
1 It is apparent that Skinner’s positivism is closer to that of Mach and Pearson than to that of the more recent logical positivists and scientific empiricists.
2 These steps, described by Skinner in 1947 (57, pp. 34-38), parallel closely his own procedure, experimental and theoretical.
3 Skinner did not so name this instrument.
4 And very effectively, too. Those who have observed work with animal behavior in different laboratories are often struck by the remarkable degree of control which the experimental technique of Skinner and his students enables them to exert over rats, pigeons, pigs and people (e.g., 2). While it may be argued that the group has avoided problems or situations yielding poorer control, it must be pointed out that many who have tried to duplicate the procedures cannot always do so until they have had an opportunity of observe them in action. These experimental skills should not be mistaken for tests of the adequacy of the theory.
5 Later references to this work will give only page citations.
6 Also, “that part of the functioning of an organism which is engaged in acting upon, or having commerce with the outside world” (ibid.), and “the movement of an organism or of its parts, in a frame of reference provided by the organism itself or by various external objects or fields of force . . . it is often desirable to deal with an effect rather than with the movement itself, as in the case of the production of sounds” (ibid.).
7 This definition, taken with the tenth and twelfth of the laws of behavior, implies that both operants and respondents can be conditioned, i.e., that they are “docile.” The first of these two quotations further clearly suggest, then, that all behavior obeys one or the other of two sets of laws, and hence can be conditioned.
8 The restrictions placed on the use of the terms “stimulus” and “response” in Skinner’s system are often overlooked; most writers consider that Skinner’s responses are “acts,” i.e., effects upon the environment produced by the organism and not the simple, reproducible movements which are the reflex response of the physiologists. As Skinner has defined his terms, a response may be of either sort; it is the organism’s behavior that determines whether an “act” or a “muscle-twitch” is a response. Similarly, whether stimuli are “mazes of a particular pattern,” “chairs,” “bars,” or a “light of a specified wave-length, intensity, size and duration, emitted by a patch of such and such size, pattern, and distance from the organism,” will depend on the organism’s behavior, rather than the predilection of the theorist. The cant terms “molar” and “molecular” cannot be applied in an intelligible way to this construction.
9 Skinner states it, “The sound of the magazine now acquires reinforcing power. . . .” (p. 53); and again, “the sound of the magazine thus becomes the S [reinforcing stimulus] of the second formula for conditioning” (23).
10 Skinner uses the term “dimensions” in the sense of sets of independent and dependent variables associated together at a given level of observation.
11 This term is not well-defined in Skinner’s usage; it cannot easily be interpreted in terms of its usual mathematical meaning. It might be defined as “the distribution of responses in time,” which is not very different from the previously employed rate of response.
12 Or, more often, operational redefinition.
13 We should perhaps entertain the suggestion that if one were to work with Skinner, and read his records with him, he would find himself able to make the same discriminations as does Skinner and hence eventually give some of them at least data-language status.
14 These laws are given in the appendix to this section.
15 This is perhaps another instance of Skinner’s use of a third class of data language.
16 Despite the exhaustive and powerful theoretical treatment of the word “stimulus,” Skinner may not have completely clarified the meaning of the term to himself. Thus the failure of an animal to increase his rate of response maximally after one reinforcement (one-trial conditioning) is attributed in one paper to conditioning of the response to “elements of the stimulus” (23) in Guthrian fashion. This statement is meaningless within the systematic treatment given to “stimulus” in other places (32, 44).
17 In Skinner’s most recent publication (61) the difficulty does arise. In this, an account of the shape of the extinction curve obtained under specified conditions is given in terms of “novelty”—presumably alterations in the stimulus field which affect the organism during extinction. But the physical environment is not varied in these experiments.
18 Skinner is not entirely clear on the question whether stimuli, once identified, can be manipulated except in the sense of being presented or not to the organism. In the latter sense, they are indeed manipulable. But any radical changes in responding associated with changes in, say, the intensity of a buzzer sounded as a discriminative stimulus could force the conclusion that the loud buzzer was not a stimulus. In this sense, stimuli are not manipulable.
19 At a later date, and in other treatments, these are indeed located in the environment but are considered unidentifiable or unspecifiable.
20 It is possible to introduce several other classes of behavior at this level of analysis; defining operations can be stated for avoidance and escape conditioning that clearly distinguish them from “operant” and “respondent” conditioning.
21 The reinforcing stimulus usually encountered in the experimental program is the food pellet—certainly relatively easy to specify. We have already noted that Skinner has made no attempt to relate this function of stimuli to the concept of drive theoretically, except as follows: “In a conditioned operant the drive governing the strength is determined by the reinforcement” (p. 372).
22 At Columbia University, undergraduates are well aware of the Skinner-box. It was the Columbia Jester in which a cartoon appeared that showed two rats at the bar of a Skinner-box. Said one rat to the other: “Oh boy, have I got this guy conditioned. Every time I press the bar, he gives me a pellet. . . .”
23 This concept may be related by some analysts to “reactive inhibition.”
24 Elsewhere (57) Skinner elaborates on his views towards the “correlational” approach (used in the Pearsonian sense) and suggests that it is one of the means by which applied psychology may be distinguished from experimental.
25 An especially interesting case (37) gives conclusions based on four groups of three rats each; no means, no measures of variability, are given, although conclusions are based upon inter-group differences.
26 Cf. Kantor’s striction on “imposing theories on the data” (15).
27 That is to say, stated in the same language as that which serves as the substratum of physical theory (3, 10). No implication should be drawn as to the reducibility of behavioral events to statement in terms of physical concepts.
28 Skinner acknowledges this model in a recent writing (61).
29 This is now (61) characterized as “not a particularly useful concept, nor does [it] . . . add much to the observed fact that extinction curves are curved in a certain way.” It is “a defensible description at the level of behavior,” and not “a theory, . . . not assigned to a different dimensional system. . . .” It “could be operationally defined as a predicted extinction-curve, even though, linguistically, it makes a statement about the momentary condition of a response.”