Interbehavioral Methodology: Lessons From Simulation

Roger D. Ray

Rollins College

 

In the most comprehensive bibliography on basic and applied systems research, Klir and Rogers (1977) reported that only 17 authors, or 1.6% of the total bibliographic listing, had 10 or more publications. This diversity in authorship makes the field difficult for the uninitiated to access and understand. Likewise, in his seminal paper critiquing experimental psychology, Kantor (1970) suggested that readers of psychology literature would find it lacking a sensitivity to, and a recognition of, the field system properties of psychological phenomena. Some 15-20 years after these two works were presented, little has changed. The need is as great as it was in 1970 for researchers to apply a more comprehensive and sensitive methodology to behavioral problems, and the researcher still finds learning how to access such a method confusing and difficult at best. Within this context, Delprato and I (Ray & Delprato, 1989) attempted to illustrate the fundamentals of a field systems approach to behavioral science.

In the present paper, I assume that readers already have some background in the field systems literature in general and are familiar with the strategic and tactical issues. First, I establish the working concepts, with all their attendant specialized vocabulary, of psychological fields and interbehavioral systems. Second, I present one new contribution: a preliminary version of a working simulation model that demonstrates how one integrates and applies the fundamentals of a field systems approach.


Concepts and Vocabulary

J.R. Kantor wrote volumes articulating his conception of psychology's place in the historical evolution of natural science. Foremost among his contributions are the concepts of interbehavior, psychological events, setting factors, and psychological fields. I have chosen to rely upon many of his terms because (a) they have such a rich literature to illuminate them, and (b) I believe they describe the task of psychological researchers so precisely.

First let me emphasize my use of the term research process. I am a strong advocate of theory as inductive integration of observations. Psychology as a science has a history of attempting deductively driven theoretical research. This tradition was epitomized both by Pavlovian "neurologizing" of the classical conditioning mechanisms and by Hullian learning theorists who spent more time on theory construction and the empirical evaluation of hypotheses generated from these theories than on elevating the base of significant applicable knowledge in psychology. I am more inclined to the "atheoretical" constructions championed during that same era by B.F. Skinner (1950, 1972). The type of theory I offer here begins with research, builds toward inductive integration, and only then becomes directive to more research, not so much because it derives hypotheses but because it illuminates where our current knowledge is insufficient. This being said, let me highlight a bigger picture of the phenomena to be charted.

Kantor depicted the fundamental process underlying all psychological phenomena as the psychological event. In many respects, the psychological event is a quantal unit derived from phenomenological properties of the irreducible psychological instance of perceived time. Central to this work, I must assert (a) that there is such a perceptual moment, and (b) that it is attended by several properties available to external observers of those events as well as properties available only to the subject him/herself. Among these attendant properties are changing and unchanging elements within the external environment (E), changing and unchanging elements within the organism (0), and changes in the interface between E and 0, which Kantor referred to as interbehavioral transactions (113).

This emphasis on interbehavior, instead of the more traditional term behavior, was Kantor's attempt to emphasize that he was not just describing the organism's reaction (R) to simple isolated stimuli (S). True enough, there are S and R elements in the event, but there is more: "Account must also be taken of what is done by the stimulus object in connection with organismic acts, and still further of many setting factors, that is, enabling and impeding conditions" (Kantor, 1970, p. 105).

There are fundamental, as well as structural, aspects to these elements that give them a psychological interpretation. I see a man with a pink sport coat and yellow pants and react to him either as a man with poor taste or as an exhibitionist. I see him functionally, in that all of my interactions with him will now take place within the context of my interpretations. In the poor-taste interpretation, I alter how I interact with him with respect to his collective behavior regarding other objects having artistic or social standards; in the case of the exhibitionist interpretation, I might alter social and interpersonal distances, alliances, and so forth.

As Kantor pointed out, there are also attendant "setting" factors that play a significant role in defining the functional character of this interaction. Given the contextual social setting of a Mardi Gras party, this man is now interpreted as being imaginative, depending upon the contextual specifications for how one was to dress for the party. Likewise, organismic and historical setting factors will have functional significance.

The reason one is concerned with these additional elements is the same reason Kantor's approach may be aptly called a big-picture orientation with concern for detail. Kantor, in the tradition of many other scientists and scientific philosophers dealing with the revolutionary changes in physics at that time, referred to their orientation as field theory. I think that description is appropriate and continue their tradition. However, fields tend to be treated somewhat as static phenomena. I am as interested in the dynamics of field changes as in any other topic. Thus, I also continue a well- established tradition in biology and ecology by relying on the term system to describe this interest in how fields not only participate in events but also participate reciprocally to alter field conditions. To make it clear that both are of interest, I have occasionally used both terms: a field systems approach.

A system is, literally, anything comprised of two or more elements that share varying degrees of mutual implication (enabling or constraining influence). As used here, an interbehavioral system is a temporal, and thus sequential, array of psychological events that are composed of Kantor's psychological events. There are momentary stimulus elements that may be defined both physically (structurally) and psychologically (functionally) by researchers: "While discussing the analysis of stimuli I mentioned that only by distinguishing between stimulus objects and stimulus functions could a thoroughly naturalistic psychology be constructed." (Kantor, 1970, p. 107). There are momentary responsive elements involving the organism that may also be defined both physically (structural anatomical articulations or biophysical/kinesiological measurement) and psychologically (functional outcome or linguistic/reflective interpretation). There is the interactive (interbehavioral) implication, or influence, mutually felt by environment and organism, which results in a change in one, the other, or both. And there are the attendant longer durational and historical setting factors in the configuration of both the environment and the organism.

Taken collectively in a given moment, this temporal/spatial array defines a psychological event. A series of functionally related events occurring within a common setting defines a psychological field. Considered sequentially and organizationally across time and space, these collective psychological fields show a recurrence and a regularity that reflect a systemic quality to psychological events. What is required is some way to describe these various elemental events, their mutual implications, and their organizational/systemic dynamic characteristics across variations in space and time-that is, their regularity in the context of time and space as field circumstances.


 Descriptive Methodological Requirements

This field systems conception of psychology places certain demands on the research strategies that are appropriate to the task. Therefore, some highlights are appropriate to review. First, a physical detailing of environmental events and responsive actions is required to understand the physical field. Second, a functional detailing of goals accomplished, purposes projected, and alterations created in the physical field conditions is required. In both structural and functional specifications, the most relied-upon tool is linguistic description. Other forms of representative measurement are possible, including (a) the temporal/spatial displacement measures commonly used in kinesiology, choreography, and graphic animations; (b) reliance upon pictorial, artistic renderings intended to demonstrate temporal/spatial details and arrangements; and (c) mathematical or formal representation via equations or complex algorithms, whereby precise quantitative, if not qualitative (cf. the Q Methodology; Stephenson, 1953, 1980), specification of both element and elementary relations are made specific.

These methods articulate how informal and formal linguistic descriptions of both structural and functional elements may be accomplished. Accounts of how such data may be analyzed to obtain summary descriptions of underlying regularities in pattern (sequence of change), coherence and variability in pattern (predictability and simplicity of sequence of change), and velocity dynamics (rate of change or patterns in the rate of change) among the descriptive elements are also available. Collectively, such analyses are called kinematic analyses. What I offer here is a glimpse of what might be learned about the adequacy of those kinematic descriptions by adding yet another descriptive strategy to our methodological arsenal: computer simulation.


 Contributions of Computer Simulation

In suggesting that computer simulation should be used for interbehavioral field systems research, I advocate a convergence of several descriptive research strategies. The simulation I demonstrate in this paper begins with the formal verbal categories and a subsequent kinematic analysis. I am not advocating simulation be conducted in lieu of categorical research; categorical research is a necessary precursor to the computer simulation I propose. But the field factors attending such descriptions will be required as well, and herein lies the value of simulation.

Assume we are simulating a monkey in a cage. For purposes of simplicity, assume that this monkey has the following behavioral repertory: sit, stand, walk, turn, eat, and climb. Whether adding to the behavioral repertory adds more, but similar, elements or the need for fundamental change will be ignored for now. We already have enough complexity to demonstrate the value of simulation and the need for specificity in describing attendant enabling and limiting field factors.

To begin this simulation, we must assume some array of interdependent probabilities for each behavior. These are the familiar unconditional and conditional probabilities calculated by empirical kinematic analyses. For sake of simplicity, we will assume a random array of behavioral interdependencies, which translates into a kinematic matrix with equal unconditional probabilities in the totals column and equal conditional probabilities in the behavioral -sequence cells. Thus, in a computerized simulation, a random generator will choose between I and 100, and the selected probability number will fall within our class of sit (1-25), stand (26-50), walk (51-75), and turn (76-100). But what happened to climb and eat? This is our first lesson on enabling field factors.

The simulation must begin with not only a selected behavior but also a selected environmental field. The monkey stands, sits, and walks in a specified location. But which needs specification first, the behavior or the location? To answer this, we must determine the mutual enabling functions of each. Walls limit walking (can't go through them) but enable climbing (can climb up them). Open space enables walking, standing, or sitting but disables climbing. Thus we should first specify which environmental field conditions exist, and this will, in turn, access quite a different array of behavioral categories defining the kinematic matrix. Open floor space will define stand, sit, and walk possibilities. Walls will define stand, sit, and climb.

But how does the monkey ever leave a wall field without walking? He doesn't. Therefore we discover our second specification problem: When walking, animals most often walk in the direction they are facing, but sometimes they turn (change directions) while continuing or initiating a walk. Open spaces imply some probability that this will occur; walls imply a relatively high probability of turning (although the animal may also just stop and stand, sit, or climb). Thus there are actually two determinant kinematic matrices and one new behavior required to simulate our monkey in his cage: (a) an open-space matrix constructed around walk, turn, sit, and stand; and (b) a wall matrix constructed around turn, sit, stand, climb, and walk (if facing away from the wall).

What happened to eating? Here we discover our second environmental enabling/disabling field condition. To eat, an animal must interact with food as a stimulating object. Although he can carry food with him about the cage, he typically finds food at singular locations within his environment. Food is thus a more temporally fleeting environmental element than open spaces and walls because animals consume it and environments do not typically offer ad libitum supplies of it. If we want our monkey to eat, he will have to be fed! And as soon as we arrange to present food on some temporal schedule, whether fixed or random, whether contingent on some specific behavior or not, we find ourselves establishing a feeding history, which defines yet another field condition: The probabilities in our open-space and wall matrices begin to change, thereby creating a dynamic historical setting factor.

Now, assume we want to simulate not just behavior but this learning activity as well. That is, we want to simulate an adaptive system, not a static one. Our first problem is how specific behavioral conditions enable or constrain the monkey's response to food delivery. Assume that food will be delivered in the left comer. If the monkey is sitting facing the wall in the right comer, delivery of food in the left comer is likely to have virtually no effect on behavior at all; but if the monkey begins walking in the direction of the left wall, continues until he reaches that comer, and finally confronts food directly, then he is very likely to begin eating. But first, he must stop walking, which means he sequences into standing or sitting, which will enable eating. If he sequences into climbing the wall, this disables eating because he is withdrawing from physical proximity with the food.

Let's make it simple and assume that all eating takes place while standing. Thus we have the following field elements: right comer, food present, and the monkey is standing (not to mention the organismic field conditions of the monkey being food deprived and thus hungry). Imagine the kinematic matrix that describes this condition versus the one where the monkey is in the right comer and food is not present. The conditional probabilities are completely different in the two matrices, and they will change differentially as a result of the delivery of food, which itself defines a field change. As food is delivered, it must also be accompanied by some attendant sounds. The contiguous pairing of food with such sounds as a part of this field condition creates what operant psychologists refer to as magazine training." Pavlov called it conditioning.

Thus, the first adaptational process we must model is the change in probability that those sounds will elicit behavior change. As all good animal trainers know, such processes begin with the animal in the immediate vicinity of the food. In our case, the monkey is in the left comer. Thus we have a second and third enabling component added to our previous one of wall/comer. These are food and food sounds. But our simulation must take account of the fact that different behavioral states may change their elicitational probabilities at different rates, given equivalent numbers of exposures (Ray & Brown, 1976). That is, an animal engaged in turning around is probably less likely to increase his responsiveness to the sound than one that is standing and looking directly at the food accompanying that sound. The contiguity effect is weakened because turning as a behavior does not enable the adaptation process as much as standing. This phenomenon can be accounted for by incorporating a variable, which Ray and Brown (1976) called interbehavioral sensitivity, into the adaptational algorithms. The change in this sensitivity is a function of both behavioral condition and the number of pairings between stimuli and that behavioral condition.

Over time, this sensitivity increases for each behavior, thus assuring that when food is delivered, the monkey quickly engages in whatever functional sequence of structural behaviors will most efficiently and quickly result in his eating the food. If engaged in a turn, he turns back and begins eating. Likewise, there is a generalization between like behavioral sequences in similar field conditions. As a result, the monkey also begins to respond with a behavioral change when in the open space. If walking, he will turn, if necessary, and continue walking directly to the food. A successful elicitational adaptation has been acquired such that, no matter where he is within the cage's space, the sound enables a functional response to obtain food and consume it. A pattern of responses emerges that always bridges the sound and eating in the shortest space and time. The monkey may now be described as engaging in a higher level functional pattern that we might call obtaining food.

But the condition we have modeled is not empowered to create food's presence. The monkey is relying upon the good graces of another field factor not yet explored: the process by which food and its associated delivery sounds are regulated. Food must be transported into this artificial environment, and that transport process, like the cage itself, is an expression of social interaction between experimenter/trainer and subject/adaptor. Now we have a real opportunity for testing the realism of our simulation. If we require a real experimenter to interact with our simulated monkey, we may (a) use the simulation to train the trainer in his ability to train, (b) evaluate the realism of our model by initiating a true Turing test (cf. Biermann, 1972), and (c) compare the simulated output of data to real data from real subjects.

Thus, we add the additional element of an interactive experimenter capable of controlling food delivery. We will soon find the experimenter building social/behavioral rules for delivering food. Uninspired by watching animals first acquire, and subsequently engage, the functional behavior of "obtaining the food" each time she presents it, our experimenter will begin to use selective rules based on what the monkey is doing to determine that food will be delivered. She might decide that she wants the monkey to s ' it before being fed. Now we have a true social/interbehavioral situation where the monkey's behavior is controlling the experimenter's delivery of food at the same time that the experimenter's food delivery is impacting the probabilities of what the monkey will do next.

 Let me illustrate these conditions with a computer-animated model. The model presents a monkey in a cage, as described. The model includes algorithms for randomly sequencing behaviors based on assumedly equal probabilities. Real data could supply more accurate initial-condition data to prime the model, but are not necessary. There is a role for an experimenter to press a computer entry key that immediately delivers food and an accompanying sound. That food always appears in the left comer of the cage and diminishes in quantity with eating behavior until it is all gone.

There are three layers of kinematic matrices (i.e., three major field conditions) required to model the operant conditioning of any selected behavior from the monkey's simple repertory of walk, sit, stand, turn, climb, and eat. The first layer is open space, which enables walk, sit, stand, and turn. Walking enables the second field condition by bringing the monkey into contact with wall/comers, which in turn adds climb to the repertory available in open space. Experimenter interaction is limited to food/sound (i.e., reinforcement) delivery into the left comer. Thus the left comer also enables eat when food is present. In addition to these field conditions, we must allow algorithms to incorporate a sensitivity to the occurrence of the presentation, but not the continued presence, of food. Representation parameters include conditional and unconditional probabilities for each behavior, the preparedness of each behavior's probabilities for alteration by reinforcement, the mean duration of each behavior, and the initial behaviorally conditional sensitivity of behavior to stimulus-elicited change.

Modeling adaptation in this system (i.e., operant conditioning) requires that probabilities change as a result of experimenter reinforcements. This is accomplished by tracking the behavior engaged at the time of reinforcement (the operant response) and using it as a " look-up table" entry guide into that kinematic matrix associated with the operant's current field conditions (open space, comer, or food + comer) and current stimulus conditions (sound or no sound). Whether and how much these probabilities also need to be changed in the other field matrices requires empirical study. It is conceivable that there is a spreading effect on probabilities that diminishes along some spatial or stimulus generalization continuum of deviation from the reinforced field conditions.

Further, it is not sufficient to model behavior based exclusively on initiation probabilities. It is obvious that some behaviors persist longer than others before alternative behavioral states take their place. Thus, each modeling algorithm must account for (a) a behavior's conditional probability, based upon the type of behavior preceding it; (b) a behavior's sensitivity to elicitational change when intercepted by the sound of food delivery; (c) the field conditions attending this particular matrix, and thus the enabling/disabling of various behaviors in the repertory; and (d) a variable representing the power of reinforcement to change a given conditional probability (or what Seligman, 1970, called the "preparedness" of this behavior for learning). Thus, each behavior must have a table of associated interbehavioral sensitivity, durations, and reinforcement values to accompany its various field matrices before the simulator can function.

The simulation model's algorithms are currently realized as HyperTalk scripts, and HyperCard button Icons, which give animated realism to behavioral loops that, once initiated, continue for the prescribed duration or until intercepted by sound, indicating food delivery. Upon that sound's occurrence, the sensitivity of the behavior, measured in probability of elicitation-induced change, is checked to determine if the next behavior should be engaged. If so, the monkey has a functional behavior routine that brings him to turn, if required, walk, if necessary, and then eat the food that accompanied the sound. The monkey eats until the food is gone. Being disabled for further eating until more food is presented, the monkey returns to the comer and behaves accordingly.

Both the probability of remaining in the reinforced field conditions and the associated conditional probabilities of that field's matrix are altered by various degrees, dependent upon the organismic setting factors (sensitivity and preparedness) associated with that behavior. Likewise, until data are obtained instructing us on the parametrics of spreading effects of increased sensitivity and operant probabilities into other field conditions, we are assuming equal increases for equivalent behaviors and sequences under all field conditions (i.e., all three matrices are altered simultaneously, regardless of the field associated with reinforcement).

Nevertheless, it is also necessary to assign independent probabilities to the field conditions. These probabilities are also altered, but selectively to that field, by reinforcement. Otherwise, the monkey would never learn the function of staying in the comer to enable his climbing, if that is what is reinforced. As such, each field takes on a higher order unitary character that begins to replace the behavior-only elements. Ultimately, these field units require a kinematic analysis of their own, thereby accounting for sequential dependence (conditional probabilities) among the array of field conditions as well as simple (unconditional) probabilities.


Conclusions Concerning Interbehavioral Methodology

A number of features of this model were not anticipated before I began its construction. Even after 20 years of research on behavioral systems and kinematic analysis, I did not fully anticipate the need for field-specific matrix construction or for reinforcing the field probabilities directly. Nor did I think the implications of stimulus setting factors versus stimulating events would be so different. The persistent presence of food must be handled differently from the momentary presence of sounds accompanying food delivery. Delayed enablements that persist across time will be the eventual embodiment of memory in the model.

Nevertheless, the model works relatively well when proper parametrics are supplied for its variables. I am now looking forward to submitting it to a Turing test: Will students know when the animations are coupled in real time to a real monkey versus when they are self-generated? If not, and if the model generates training data comparable to those derived from real animals, then the model can be expanded to incorporate motivational and memory field factors as well.

Regardless, there is already enough realism in the model to bolster our belief in Kantor's (1970) assertions that:

Of extreme importance for the appreciation of behavior fields is their uniqueness and individuality. There is no fixed or universal type. Implied in the field construct is the principle that each class of behavior events must be analyzed according to its intrinsic factors. Certainly complex fields yield upon analysis a larger inventory of factors and very different ones from simpler fields. And it is imperative to be alive to the greater complexity of non-reflex behavior, especially the interpersonal aspects of human performances. (p. 105)

Unfortunately, Kantor's vision did not include methodological prescriptions and demonstrations suited to his understanding in principle. Today, we not only know how to generate and analyze field-sensitive data, we also are beginning to understand how the generation of a virtual-reality simulation aides in providing feedback on the accuracy and completeness of our empirical efforts. We already recognize the need for further research on how generalization affects probabilities of common behavioral elements in uncommon field conditions. We also see the need to develop more precise parametric specifications for each type of behavior with respect to its changing sensitivity to classical stimulus pairings and to its rate of change in conditional and unconditional probability upon reinforcement. Much must be accomplished before even the simplest simulation will approximate reality.

This is not the first time I have been taught this lesson by attempted simulations. Ray and Delprato (1989) provided a brief summary of an unpublished paper by Ray and Wruble (1986) on the reconstructive linguistic simulation of an observer describing an animal's behavior based upon a coded data base. This simulation attempted to give a clear and descriptive English transliteration of the coded data by simulating the grammatical rules that had been eliminated by the coding strategies. Within the first attempts, we discovered serious problems in the code structure by being shown that our codes were, without our intent, shifting focal domains. For example, we had described an animal as sitting under a shelf; then, as he began self-contact by scratching himself, found ourselves describing the animal as sitting under himself! This was not obvious from a casual reading of the coded record, but the simulation made it quite apparent.

It thus takes attempted simulation to show us much of what we have been missing, or only assuming, in our pursuit of systems method development and application. I would encourage all field systems researchers to seriously consider simulation as a corrective second stage to their own research efforts.


Conclusions Concerning Pedagogical Research

Having learned these lessons for myself, what advice do I have for those who aspire to study the special expertise demonstrated by teachers? First, I suspect that expert teachers are more expert at teaching some things than others. This may seem a self-evident and trivial statement, but it is important to remember that selecting what and when to describe is as important as choosing how to describe it. From a simulation point of view, this means that any given simulation effort will probably have its application and validity restricted by the specific expertise one is attempting to understand. One should probably look at a variety of experts doing a variety of things before emulating any singular strategy or approach. Don't forget, too, that variations between individual students help to define what does, or does not, work. Expert teachers not only fail when they try to teach all things, they will also fail on more than one occasion in teaching their best thing to all students.

Second, teaching is interactive on several levels. One level involves the overt behavior of students and teachers- what each is doing to, and in response to, the other. This is the type of behavior I have been assuming in my discussion of simulators. Another level is more qualitative and interpretative in character: the level at which emotional communication and attachment take place. This level is conveyed by more subtle, but still potentially objective, events that are more difficult to capture in descriptions: verbal intonations and inflections, rates of speaking and pausing, choices of words for their connotative (rather than denotative) meanings, nonverbal communicative gestures and body language, and even social distances. These are the elements that challenge us most in describing how an expert teacher works.

The challenge lies not only in recognizing the task before us in trying to describe good teaching but also in recognizing the tools we already have for accomplishing that task. Artistic teaching doesn't have to remain the exclusive domain of the artist; it should be amenable to scientific description and assessment. Likewise, scientific description is of little use if an artist cannot use it to duplicate the events being described. Computer simulation is, after all, merely an artist's test of making the event appear real to those observing and describing the simulation. This is a form of replication.

Good teaching is also a problem of replication. It involves teaching (describing for) new teachers what expert teachers already know. The final test is whether the students of this second generation of teachers can tell the difference between original and second-generation teaching. In the simulation business, one always evaluates simulation outcomes with the Turing test. In essence, this test goes something like this: If the events we describe to others (the simulator) generate results that fully replicate the original events, then we are adequately describing. If they can't, we aren't. If the teachers we teach from our analyses of good teaching can teach well, then we are adequately teaching (and adequately describing our teaching). If they can't, we aren't either. To take on the task of scientifically analyzing the process of expert teaching is to take on this ultimate task of simulation. We won't know whether we are simulating good teaching or actually teaching well because the results will finally be the same. That is truly an expert system. Unfortunately, we have a ways to go yet in its full development. Hopefully, a field systems approach to expert teaching will facilitate that journey, and simulation from that analysis will help us quality-check our progress.


 References

Biermann, A.W. (1972). On the inference of Turing machines from sample computations. Artificial Intelligence, 3, 181-198.

Kantor, J.R. (1970). An analysis of the experimental analysis of behavior (TEAB). Journal of the Experimental Analysis of Behavior, 13, 101- 105.

Klir, G.J., & Rogers, G. (1977). Basic and applied general systems research: A bibliography. Binghamton, NY: The University Center at Binghamton, State University of New York.

Ray, R.D., & Brown, D.A. (1976). The behavioral specificity of stimulation: A systems approach to procedural distinctions of classical and instrumental conditioning. Pavlovian Journal of Biological Sciences, 11, 3-23.

Ray, RD, & Delprato, D.J. (1989). Behavioral systems analysis: Methodological strategies and tactics. Behavioral Science, 34, 81-127.

Ray, RD, & Wruble, M. (1986, October). Computer-based psycholinguistic programs as quality control devices for descriptive research systems. Paper presented at Pavlovian Society Meetings, St. Louis.

Seligman, M.E.P. (1970). On the generality of the laws of learning. Psychological Review, 77, 406-418.

Skinner, B.F. (1950). Are theories of learning necessary? Psychological Review, 57, 193-216.

 Skinner, B.F. (1972). Cumulative record: A selection of papers (3rd Ed.). New York: Appleton-Century-Crofts.

Stephenson, W. (1953). The study of behavior. Chicago: University of Chicago Press.

Stephenson, W. (1980). Newton's fifth rule and Q methodology: Application to educational psychology. American Psychologist, 35, 882-889.


 Acknowledgment
Portions of the research and program development reported in this article have been funded in part by NSF-ILI Grant USE-8952419.