6
JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR WHAT IS THE EXPERIMENTAL ANALYSIS OF BEHAVIOR? B. F. SKINNER HARVARD UNIVERSITY A new division of the American Psychologi- cal Association calls for some effort to define the enterprise that has come to be known as the Experimental Analysis of Behavior. It is sometimes easier to place a piece of research correctly within that field or to exclude what seems to be similar work than to give one's reasons for doing so. Although sharp bounda- ries can seldom be drawn in defining any sci- entific area, certain distinguishing features are worth pointing out. The Dependent Variable A natural datum in a science of behavior is the probability that a given bit of behavior will occur at a given time. An experimental analysis deals with that probability in terms of frequency or rate of responding. Like prob- ability, rate of responding would be a mean- ingless concept if it were not possible to specify topography of response in such a way that separate instances of an operant can be counted. The specification is usually made with the help of a part of the apparatus-the "operandum"-which senses occurrences of a response. In practice, responses so defined show a considerable uniformity as the organism moves about in a framework defined by its own anatomy and the immediate environ- ment. Changes in rate are usually recorded and inspected in the ubiquitous cumulative record, although distributions of interresponse times and on-line computer analyses of rates and changes in rate are increasingly used. An lIntroductory remarks at a Symposium on the Appli- cation of Operant Conditioning, Annual Meeting of the American Psychological Association, Los Angeles, September 6, 1964. Preparation of this manuscript has been supported by Grant K6-MH-21,775 of the Na- tional Institute of Mental Health of the U. S. Public Health Service, and by the Human Ecology Fund. Re- prints may be obtained from the author, Dept. of Psy- chology, Harvard University, 33 Kirkland St., William James Hall, Cambridge, Mass. 02138. emphasis on rate of occurrence of repeated in- stances of an operant distinguishes the experi- mental analysis of behavior from kinds of psy- chology which proceed in one or more of the following ways. (1) Behavior is taken merely as the sign or symptom of inner activities, mental or physi- ological, which are regarded as the principal subject matter. Rate of responding is signifi- cant only because it permits us to follow a process (such as learning or maturation) or to determine a state or condition (such as an ex- citatory tendency or alertness or wakefulness) or to detect available psychic energy or the strength of a drive or emotion, and so on. The observed behavior is not expected to be very orderly because it is only a rather noisy "per- formance", from which presumably more sta- ble states and processes are to be inferred with the help of statistical procedures. These prac- tices have discouraged a careful specification of behavior, and the data obtained with them are seldom helpful in evaluating probability of response as such. (2) Behavior is held to be significant only in meeting certain standards or criteria. An organism is described as "adjusting to a situa- tion", "solving a problem", "adapting to the environment", and so on. With respect to nor- mative criteria its behavior may improve or deteriorate, with respect to developmental cri- teria it may be arrested or accelerated, and so on. In reporting these aspects of behavior the experimenter may not specify what the organ- ism is actually doing, and a rate of responding cannot be satisfactorily inferred. (3) Changes in probability of response are treated as if they were responses or acts. The organism is said to "discriminate", to "form concepts", to "remember", to "learn what to do" and, as a result, "know what to do", and so on. These are not, however, modes of re- sponse. To discriminate is not to respond but 213 VOLUME 9, NUMBEP, 3 MAY, 1966

What is the experimental analysis of behavior

Embed Size (px)

DESCRIPTION

B. F. Skinner

Citation preview

  • JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

    WHAT IS THE EXPERIMENTAL ANALYSISOF BEHAVIOR?

    B. F. SKINNERHARVARD UNIVERSITY

    A new division of the American Psychologi-cal Association calls for some effort to definethe enterprise that has come to be known asthe Experimental Analysis of Behavior. It issometimes easier to place a piece of researchcorrectly within that field or to exclude whatseems to be similar work than to give one'sreasons for doing so. Although sharp bounda-ries can seldom be drawn in defining any sci-entific area, certain distinguishing features areworth pointing out.

    The Dependent VariableA natural datum in a science of behavior is

    the probability that a given bit of behaviorwill occur at a given time. An experimentalanalysis deals with that probability in termsof frequency or rate of responding. Like prob-ability, rate of responding would be a mean-ingless concept if it were not possible to specifytopography of response in such a way thatseparate instances of an operant can becounted. The specification is usually madewith the help of a part of the apparatus-the"operandum"-which senses occurrences of aresponse. In practice, responses so defined showa considerable uniformity as the organismmoves about in a framework defined by itsown anatomy and the immediate environ-ment. Changes in rate are usually recordedand inspected in the ubiquitous cumulativerecord, although distributions of interresponsetimes and on-line computer analyses of ratesand changes in rate are increasingly used. An

    lIntroductory remarks at a Symposium on the Appli-cation of Operant Conditioning, Annual Meeting ofthe American Psychological Association, Los Angeles,September 6, 1964. Preparation of this manuscript hasbeen supported by Grant K6-MH-21,775 of the Na-tional Institute of Mental Health of the U. S. PublicHealth Service, and by the Human Ecology Fund. Re-prints may be obtained from the author, Dept. of Psy-chology, Harvard University, 33 Kirkland St., WilliamJames Hall, Cambridge, Mass. 02138.

    emphasis on rate of occurrence of repeated in-stances of an operant distinguishes the experi-mental analysis of behavior from kinds of psy-chology which proceed in one or more of thefollowing ways.

    (1) Behavior is taken merely as the sign orsymptom of inner activities, mental or physi-ological, which are regarded as the principalsubject matter. Rate of responding is signifi-cant only because it permits us to follow aprocess (such as learning or maturation) or todetermine a state or condition (such as an ex-citatory tendency or alertness or wakefulness)or to detect available psychic energy or thestrength of a drive or emotion, and so on. Theobserved behavior is not expected to be veryorderly because it is only a rather noisy "per-formance", from which presumably more sta-ble states and processes are to be inferred withthe help of statistical procedures. These prac-tices have discouraged a careful specificationof behavior, and the data obtained with themare seldom helpful in evaluating probabilityof response as such.

    (2) Behavior is held to be significant onlyin meeting certain standards or criteria. Anorganism is described as "adjusting to a situa-tion", "solving a problem", "adapting to theenvironment", and so on. With respect to nor-mative criteria its behavior may improve ordeteriorate, with respect to developmental cri-teria it may be arrested or accelerated, and soon. In reporting these aspects of behavior theexperimenter may not specify what the organ-ism is actually doing, and a rate of respondingcannot be satisfactorily inferred.

    (3) Changes in probability of response aretreated as if they were responses or acts. Theorganism is said to "discriminate", to "formconcepts", to "remember", to "learn what todo" and, as a result, "know what to do", andso on. These are not, however, modes of re-sponse. To discriminate is not to respond but

    213

    VOLUME 9, NUMBEP, 3 MAY, 1966

  • B. F. SKINNER

    to respond differently to two or more stimuli.To say that an organism has learned to dis-criminate between two stimuli is to report apossibly useful fact, but it is not to say whatthe organism is actually doing.

    (4) The dimensions studied, though quanti-fiable, are not related in any simple way toprobability of response. The force with whicha response is executed and the time whichelapses between stimulus and response-called,often inaccurately, latency or reaction time-are popular measures. When they change un-der differential reinforcement, they are rele-vant to an experimental analysis, but they maynot throw much light on probability. Othercommon measures, such as the time requiredto complete a task-to get through a maze, tosolve a problem, or to cross out all letters of agiven kind on a page-or the number of errorsmade or the number of trials taken in meetinga criterion, are still less useful. "Amount re-membered," an aspect of behavior first empha-sized by Ebbinghaus, has recently enjoyed arenewed popularity. The experimenter maywant to know, for example, how a set of re-sponses comes under the control of a corre-sponding set of stimuli, but instead of follow-ing the change in probability he measures thenumber of responses "correctly emitted in re-call" at a later time.An experiment is often designed so that the

    important result is a ratio between two suchmeasures, when the arbitrariness or irrele-vance of the aspects measured seems to cancelout. A ratio is still of little help in an experi-mental analysis. Such measures are chosen pri-marily because they are quantifiable-force ofresponse can be accurately recorded, numberof trials exactly counted, and elapsed timemeasured on the most accurate of clocks-butquantifiability is not enough. Rate of respond-ing is a basic dimension, not simply becauseresponses can be accurately counted, but be-cause rate is relevant to the central concernof a science of behavior.

    (5) The inner entities of which behavior issaid to be a sign or symptom include the traits,abilities, attitudes, faculties, and so on, forwhich various techniques of psychologicalmeasurement have been designed. But eventhe most impeccable statistical techniques andthe most cautious operational definitions willnot alter the facts that the "tests" from whichthe data are obtained are very loosely con-

    trolled experimental spaces and that the"'scores" taken as measures have some of thearbitrary features just mentioned. The impor-tant issues to which these techniques havebeen directed-for example, the covariation inprobability of groups of responses-must bestudied in other ways before the results willbe useful in an experimental analysis.

    (6) Instead of observing behavior, the ex-perimenter records and studies a subject'sstatement of what he would do under a givenset of circumstances, or his estimate of hischances of success, or his impression of a pre-vailing set of contingencies of reinforcement,or his evaluation of the magnitude of currentvariables. Observation of behavior cannot becircumvented in this way, because a subjectcannot correctly describe either the probabil-ity that he will respond or the variables affect-ing such a probability. If he could, he coulddraw a cumulative record appropriate to agiven set of circumstances, but this appears tobe out of the question.

    The Independent VariablesThe task of an experimental analysis is to

    discover all the variables of which probabilityof response is a function. It is not an easy as-signment, but it is at least an explicit one. Itdistinguishes an experimental analysis of be-havior from other approaches at many points.

    (1) The stimulus is, of course, an importantindependent variable. An early associationwith the concept of the reflex gave it the char-acter of a goad, something which forced anorganism to respond. This was perhaps aswrong as the traditional view that the organ-ism forced the environment to stimulate-tobecome visible, audible, and so on. The posi-tion of an experimental analysis differs fromthat of traditional stimulus-response psycholo-gies or conditioned reflex formulations inwhich the stimulus retains the character of aninexorable force. It does not follow, however,that the organism acts upon the environmentin the manner suggested by terms like detect,identify, perceive, experience, classify, andjudge, or by terms which appear to describelater responses to stimuli, such as recall howsomething looked or remember what hap-pened. Such terms, like expressions borrowedfrom computer technology which describe theorganism as processing information, do notspecify what the organism is actually doing.

    214

  • EXPERIMENTAL ANALYSIS OF BEHAVIOR

    The concept of the discriminative stimulus(the well known "SD") and the related notionof stimulus control assign to stimuli a morereasonable role as independent variables.An experimental analysis describes stimuli

    in the language of physics. The experimenterdoes not ask whether a stimulus looks the sameto the organism as it does to him. In studyinga generalization gradient with respect to wavelength of light, for example, lights are some-times matched for brightness, so that the gradi-ent will represent a reaction to color only; butthis is an unwarranted intrusion into the data.To guess what an organism sees when a stim-ulus is presented and to suppose that what isguessed is what is being presented would beto abandon all that physics has to offer by wayof specifying environmental events. The im-portance of certain classical problems is notthereby denied. Stimuli are often difficult tospecify in physical terms. Different stimulimay appear to have the same effect and thesame stimulus different effects under differentconditions. But it is no solution to fall backupon the response of an experimenter toachieve some sort of invariance. Similarly, anyreference to "parameters relating to the com-plexity of a task" or to "frustrating" or "anxi-ety-generating" properties of a situation is alsoobjectionable, whether the subject or the ex-perimenter serves as indicator of the complex-ity or the emotion.

    (2) Other independent variables are foundin the classical fields of motivation and emo-tion. The experimental analyst does not ma-nipulate inner states as such. He manipulates,not hunger, but the intake of food; not fear asan acquired drive, but aversive stimuli; notanxiety, but preaversive stimuli. He adminis-ters a drug, not the physiological effects of adrug. He takes the age of an organism, notsome level of maturation, as a variable. Hesometimes uses a collateral dependent variable-but not as a measure. He may use body-weight, for example, in lieu of a history ofdeprivation, but it is simply another effect ofdeprivation, not a measure of hunger.

    (3) The so-called "contingencies of rein-forcement" are an important feature of theindependent variables studied in an experi-mental analysis. A few contingencies, such asconditioning, extinction, and delay of rein-forcement are familiar. Somewhat more com-plex contingencies, such as those responsible

    for stimulus discrimination and response dif-ferentiation, are also fairly well known. Butmany psychologists are unaware of the com-plexity of the contingencies now commonlystudied. In addition to many standard sched-ules of reinforcement, reinforcement may becontingent on rate of responding, rate ofchange in rate, or specific patterns of ratechanges detected by on-line computer analyses.Contingencies may involve several stimuli andresponses interrelated in various ways. Consid-erable skill may be needed to design programsof instructional contingencies which will bringbehavior under the control of complex termi-nal contingencies of this sort. The importanceof programming is, indeed, often completelyoverlooked. For example, the statement that agiven type of organism or an organism of agiven age "cannot solve a given kind of prob-lem" is meaningless until the speaker has spec-ified the programs which have been tried andconsidered the possibility that better ones maybe designed.

    Describing a set of contingencies in instruc-tions to the subject is no substitute for ex-posing the subject to the contingencies, par-ticularly when they need to be programmed.Instructions have effects, of course, dependingin part on the verbal history of the subject, butthe behavior of a subject to whom an experi-menter has explained how a piece of appa-ratus works will not necessarily resemble onewho has come under the control of the termi-nal contingencies established by that appa-ratus.

    Contingencies of reinforcement have beenanalyzed formally in theories of probability,decision-making, and games, but the theoristoften has no way of knowing, aside from ob-servation of his own behavior, what effects agiven set of contingencies will have or whatkind of program may be needed to make iteffective. Certain assumptions-for example,that an organism will behave rationally-aresometimes used in lieu of observations to com-plete a statement of contingencies. Formalstatements of contingencies, like instructions,have their effects and if detailed enough maysupply rules which function as prior stimulito control behavior resembling that whichwould be generated by prolonged exposure tothe contingencies themselves. The two casesmust, however, be clearly distinguished. Whenan organism is brought under the control of

    215

  • B. F. SKINNER

    complex contingencies, it is not necessarily"applying the rule" which describes them.The increasing power of an experimental

    analysis has made it possible to examine theeffects of complex contingencies to which anorganism has traditionally been assumed toadjust only by exercising certain cognitiveprocesses. It is sometimes obvious that suchprocesses have been invented simply to ac-count for the behavior in the absence of anybetter information as to how the contingenciescould generate it. The experimenter has notbeen able to relate the behavior to the contin-gencies, and he is forced to conclude that theorganism has somehow done so mentally. Sup-posed cognitive processes of this sort may bedisregarded. Others, however, may be a sortof internalized version of precurrent behavior-behavior maintained by its effects in maxi-mizing the reinforcement of subsequent re-sponses. Precurrent behavior is part of the sub-ject matter of an experimental analysis. It isusually studied ip overt form though it mayeventually drop to the covert level. In eithercase it is defined as behavior which affects be-havior rather than as mental activity.

    Treatment of Relationships among VariablesThe behavioral processes studied in an ex-

    perimental analysis usually consist of changesin probability (or rate of response) as a func-tion of manipulated variables. The changesare followed in real time rather than from"trial to trial"-a practice derived from acci-dental features of early psychological research.An emphasis on real time is another reasonwhy cumulative records are useful. (A cumula-tive record is sometimes used to "smooth"other kinds of data-for example, the errorsmade during repeated trials in learning a mazeor in solving a problem-and it is often im-plied that a cumulative record of responses intime also gains an unwarranted smoothness ofthe same sort. The important difference is thatthe slope of a cumulative curve in real timerepresents a meaningful state of behavior.)

    Relations among dependent and independ-ent variables are seldom explored according toa prior "experimental design", as R. A. Fisherused that term. The null hypothesis finds itselfin the null class. Research which is not de-signed to test hypotheses-physiological, men-talistic, or conceptual-may seem puzzling tothose who identify statistics with scientific

    method, though it appears perfectly reason-able to physicists, chemists, and most biolo-gists. The usual practice is to construct an ex-perimental space in which stimuli, responses,and reinforcements are interrelated in a set ofcontingencies. The contingencies depend inpart on the behavior which the organismbrings to the experiment. Provision is usuallymade for changing the apparatus as the be-havior changes, but seldom according to a pre-determined plan. The experimental control ofvariables is emphasized rather than a laterevaluation of their presumed importancethrough statistical analyses. The number of or-ganisms studied is usually much smaller thanin statistical designs, but the length of timeduring which any one organism is observed isusually much greater.

    It is often said to be impossible --to distin-guish between significant and insignificantfacts without a hypothesis or theory, but theexperimental analysis of behavior does notseem to bear this out. It has progressed bybuilding upon its past. Improved formulationsand techniques have led to more precise andreproducible data over a much greater range,but not to the outright rejection of earlierwork. (For one thing, few data have becomeuseless because a theory they were designedto test has been discarded.) In retrospect thereappears to have been little random or aimlessexploration. Such a field as the systematicanalysis of contingencies of reinforcement, forexample, does not require a theory. In ourstudy of schedules of reinforcement Fersterand I proceeded in a rather Baconian fashion,fil.ling in a table of the possibilities generatedby combinations of clocks, counters, and speed-ometers, fixed and variable sequences, and soon. Most of the contingencies examined intheories of probability, decision-making, andgames are generated in a similar way-the"theory", if any, being concerned with whatorganisms will do under the contingenciesanalyzed. The experimental analysis of behav-ior dispenses with theories of that sort by pro-ceeding to find out.

    In addition to the systematic manipulationof contingencies, the interpretation of humanaffairs is a rich source of suggestions for ex-periments. Do conditions detected in some epi-sode in daily life actually have the effectsobserved when more carefully controlled? Cana certain history of reinforcement be shown to

    216

  • EXPERIMENTAL ANALYSIS OF BEHAVIOR

    be responsible for a current performance?What changes in contingencies will have dif-ferent and possibly more acceptable results?And so on. The guesses and hunches withwhich the experimenter proceeds to answerquestions of this sort are not the formal hy-potheses of scientific method; they are simplytentative statements for which further supportis sought. The philosopher of science may stillwant to reconstruct the behavior so that it fitsa hypothetico-deductive model, but efforts inthat direction grow less impressive-particu-larly as an alternative formulation of the be-havior of Man Thinking is glimpsed as one ofthe more distant reaches of an experimentalanalysis (Skinner, 1957).Research which enlarges an established cor-

    pus of facts or simplifies an effective formula-tion is usually less dramatic than researchwhich topples hypotheses or confirms broadtheories, but it has its compensations. Forthose so inclined, theoretical activities are byno means ruled out, even though scientificmethodologists have usually been hesitant inaccepting the position often adopted in an ex-perimental analysis. Quite aside from testinghypotheses, one may look for simplifying uni-formities. For example, one may develop atheory as to why schedules of reinforcementhave the effects they have, seeking certain sim-plifying relations among the many perform-ances generated by different schedules. Fersterand I hazarded some informal guesses alongthis line, arguing for the importance of theconditions which prevail at the precise mo-ment of reinforcement, but a better theory inthis sense is no doubt possible and desirable.

    In representing the relationships discoveredby an experimental analysis of behavior, littleuse is made of metaphors or analogies drawnfrom other sciences. Reports seldom containexpressions like encode, read out from storage,reverberating circuits, overloaded channels,gating, pressure, flow, drainage, networks, cen-ters, or cell assemblies. Little use is made ofmaps or schemata, such as Tolman's sow-bug,Lewin's fields and vectors, or block diagramsrepresenting organisms as adaptive machines.The advantage in representing processes with-out the use of metaphor, map, or hypotheticalstructure is that one is not misled by a spuri-ous sense of order or rigor. Early in his careerFreud wrote to Fliess that he had put psychol-ogy on a firm neurological basis. The theory

    permitted him "to see the details of neurosisall the way to the very conditioning of con-sciousness" .(quoted in Fine, 1962). His letteremphasized number, structure, and terms bor-rowed from neurology, biology, and physics.He spoke of "the three systems of neurones,the 'free' and 'bound' states of quantity, theprimary and secondary processes, the maintrend and the compromise trend of the nerv-ous system, the two biological rules of atten-tion and defense." Terms of this sort encour-age euphoria, and Freud was vulnerable; inhis first report he was "wildly enthusiastic".Within a month or so he had abandoned thetheory. He had the insight to tell Fliess thatit seemed to him in retrospect "a kind ofaberration".

    Attitudes toward ResearchThe experimental analysis of behavior is

    also generally characterized by an unhurriedattitude toward the as-yet-unanalyzed or theas-yet-unexplained. Criticism often takes theline that the analysis is over-simplified, that itignores important facts, that a few obvious ex-ceptions demonstrate that its formulationscannot possibly be adequate, and so on (for ex-ample, Miller, Galanter, and Pribram, 1960).An understandable reaction might be tostretch the available facts and principles in aneffort to cover more ground, but the generalplan of the research suggests another strategy.Unlike hypotheses, theories, and models, to-gether with the statistical manipulations ofdata which support them, a smooth curveshowing a change in probability of responseas a function of a controlled variable is a factin the bag, and there is no need to worry aboutit as one goes in search of others. The short-comings and exceptions will be accounted forin time. The strategy is supported by the his-tory of early criticisms of the Behavior of Or-ganisms. It was said that the book was notabout organisms but about the rat, and verysmall groups of rats at that. How could onebe sure that other rats, let alone animals ofother species, would behave in the same way?Only food and water were used as reinforcers,social reinforcers being conspicuously lacking.The stimuli-lights and buzzers-were crudeand poorly controlled. Two levers should havebeen used so that the data would throw lighton behavior at a choice point. And, after all,could we be sure that the rat was not pressing

    217

  • 218 B. F. SKINNER

    the lever simply because it had nothing else todo? These criticisms have all been answeredwithout effort in the course of time simply aspart of the normal development of the anal-ysis.

    Patience with respect to unexplored partsof a field is particularly important in a scienceof behavior because, as part of our own sub-ject matter, we may be overwhelmed by thefacts which remain to be explained. Subtleillusions, tricks of memory, the flashes whichsolve problems-these are fascinating phenom-ena, but it may be that genuine explanationswithin the framework of a science of behavior,as distinguished from verbal principles or"laws" or neurological hypotheses, are out ofreach at the present time. To insist that a sci-ence of behavior give a rigorous account ofsuch phenomena in its present state of knowl-edge is like asking the Gilbert of 1600 to ex-plain a magnetic amplifier or the Faraday of1840 to explain superconductivity. Early phys-ical scientists enjoyed a natural simplificationof their subject matters. Many of the mostsubtle phenomena were to come into existenceonly through technical advances in the sci-ences themselves. Others, though occurring innature, were not recognized as parts of theirfields. The behavioral scientist enjoys no such

    natural protection. He is faced with the fullrange of the phenomena he studies. He musttherefore more explicitly resolve to put firstthings first, moving on to more difficult thingsonly when the power of his analysis permits.A final distinction. Those who engage in the

    experimental analysis of behavior are usuallyconspicuous for their enthusiasm. In a recentarticle Bixenstine (1964) attributes an unwar-ranted optimism in all behavioral science tothe methodological position taken by experi-mental analysts. This is perhaps to overesti-mate their influence. In any case, he points tothe wrong cause. He suggests that the op-timism springs from release from the anxietyof theory construction. There is a more ob-vious explanation: the analysis works.

    REFERENCESBixenstine, V. Edward. Science, 1964, 145, 464.Fine, Reuben. Freud: A Critical Re-evaluation of his

    Theories. New York: McKay, 1962.Miller, George A., Galanter, Eugene, and Pribram,

    Karl H. Plans and the Structure of Behavior. NewYork: Henry Holt, 1960.

    Skinner, B. F. Verbal Behavior. New York: Appleton-Century-Crofts, 1957.

    Received December 30, 1965