Modeling Complex Systems (Nebraska Symposium on Motivation, Vol 52)

Modeling Complex Systems Volume 52 of the Nebraska Symposium on Motivation University of Nebraska Press Lincoln and L...

Author: Nebraska Symposium

30 downloads 1306 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Modeling Complex Systems

Volume 52 of the Nebraska Symposium on Motivation

University of Nebraska Press Lincoln and London

Volume 52 of the Nebraska Symposium on Motivation Richard A. Dienstbier Bill Shuart Will Spaulding Jeffrey Poland

Modeling Complex Systems Series Editor Volume Editors

Presenters Bill Shuart

Madonna Rehabilitation Hospital

Will Spaulding

University of Nebraska–Lincoln

Jeffrey Poland

Rhode Island School of Design / Brown University

Richard W. J. Neufeld

University of Western Ontario

Wolfgang Tschacher

University of Bern

Zeno Kupper

University of Bern

Susanne P. Lajoie

McGill University

Mark A. Musen

Stanford University

Eduardo Salas

University of Central Florida

Kevin C. Stagl

University of Central Florida

C. Shawn Burke

University of Central Florida

Gerald F. Goodwin

U.S. Army Research Institute

Michael J. Mahoney

University of North Texas

Modeling Complex Systems is Volume 52 in the series CURRENT THEORY AND RESEARCH IN MOTIVATION © 2007 by the University of Nebraska Press All rights reserved Manufactured in the United States of America International Standard Book Number ISBN: 978-0-8032-1387-6 (Clothbound)

The Library of Congress has cataloged this serial publication as follows: Nebraska Symposium on Motivation. Nebraska Symposium on Motivation. [Papers] v. [1]–1953– Lincoln, University of Nebraska Press. v. illus., diagrs. 22cm. annual. Vol. 1 issued by the symposium under its earlier name: Current Theory and Research in Motivation. Symposia sponsored by the Dept. of Psychology of the University of Nebraska. 1. Motivation (Psychology) BF683.N4 159.4082 53-11655 Library of Congress

Preface

The volume editors for this 52nd volume of the Nebraska Symposium on Motivation are Bill Shuart, Will Spaulding, and Jeffrey Poland. The volume editors coordinated the symposium that led to this volume, including selecting and inviting the contributors and coordinating all aspects of editing. My thanks to our contributors for excellent presentations and chapters. This symposium series is supported by funds provided by the chancellor of the University of Nebraska–Lincoln, Harvey Perlman, and by funds donated in memory of Professor Harry K. Wolfe to the University of Nebraska Foundation by the late Professor Cora L. Friedline. We are extremely grateful for the chancellor’s generous support of the symposium series and for the University of Nebraska Foundation’s support via the Friedline bequest. This symposium volume, like those in the recent past, is dedicated to the memory of Professor Wolfe, who brought psychology to the University of Nebraska. Richard A. Dienstbier Series Editor

Contents Bill Shuart, Will Spaulding, and Jeffrey Poland

Introduction

Richard W. J. Neufeld

Composition and Uses of Formal Clinical Cognitive Science

Wolfgang Tschacher and Zeno Kupper

A Dynamics-Oriented Approach to Psychopathology

123

Susanne P. Lajoie

Developing Computer-Based Learning Environments Based on Complex Performance Models

145

Mark A. Musen

Technology for Building Intelligent Systems: From Psychology to Engineering

185

Eduardo Salas, Kevin C. Stagl, C. Shawn Burke, and Gerald F. Goodwin

Fostering Team Effectiveness in Organizations: Toward an Integrative Theoretical Framework

245

Michael J. Mahoney

Constructive Complexity and Human Change Processes

ix

1 85

Editors’ Postscript: Modeling Complex Processes in a Rehabilitation Application

275

287

Contributors

293

Subject Index

307

Author Index

Introduction Bill Shuart, Will Spaulding, and Jeffrey Poland Madonna Rehabilitation Hospital; University of Nebraska–Lincoln; Rhode Island School of Design / Brown University Capturing the complexity of human behavior has been a recurring theme in the Nebraska Symposium on Motivation: We expect behavior to be patterned or integrated, and to make biological sense; and so patterning and biological utility are what we see. And of course what we see is actually there—behavior in general is not chaotic; it is organized. (Nissen, 1954, p. 314) When fundamental psychologists do make excursions into the human motivational world . . . it is rare that they survey the requirements for theory or pre-theory by intensive descriptive analysis of behavior related to such motives as produced by concrete human beings. More remote still is the chance that anyone will select for illustration, let alone analysis, behavior or experience relevant to man in his most characteristically human performances: man as he creates or loves or plays or responds to the aesthetic surfaces of the human and natural environment. Such matters are threateningly complex. (Koch, 1956, pp. 64–65) I have tried, ﬁrst, to show that it is possible to formulate a meaningful theory of complex motivation by analyzing the

x modeling complex systems sorts of variables involved, together with their interactions. I have sought, second, to show that we possess, now, many sound and useful concepts and techniques to translate these complexities into productive experimental research. I suppose the moral is: be not afraid of complexity. If motivation is indeed complex, then let us ﬁnd the means to cope with it. (Vinacke, 1962, pp. 42–43) Through the representation of a few very simple psychological concepts, in a rudimentary mathematical way, a good deal of complexity can be generated. . . . Let us now proceed to generate complexity from simplicity. (Burke, 1966, pp. 49–50) Although disorder may be experienced and expressed in highly patterned processes of human activity, it is diverse, individually unique, and systemic; we shall advance in our attempts at conceptualization and classiﬁcation only as we are willing to embrace the limits of symbol systems to capture human uniqueness and the ultimate ineffability of complex system dynamics. (Mahoney, in this volume, pp. 265–266) The contributions to this volume of the Symposium describe contemporary approaches to the modeling of complex psychological and behavioral processes, ranging from molecular to molar phenomena. Although the contributions reﬂect a range of theoretical and epistemic perspectives, they all explicitly or implicitly incorporate complex frameworks of dynamic, system-like relations involving perception, learning, concept formation, emotion, motivation, intention, behavior, and the social context in which behavior occurs. One special feature of all the contributions from this particularly distinguished group of theorist-practitioners is an emphasis on practical applications of the conceptual frameworks in which they work. This reﬂects an important idea in the zeitgeist of the contemporary scientiﬁc community, that of translational research. Translational research is a process of translating the principles and truths that emerge from basic science into practical applications. The complexity of the processes captured in the contributors’ models enhances the models’ applicability to the complexities of clinical practice, industry, and education. To consolidate the relevance of application

xi Introduction and translational research, this volume ends with a volume editors’ postscript, describing a practical model for the complex processes of rehabilitation, as manifest in rehabilitation services currently evolving in Nebraska. Translational research demands, not just practical application, but continuity with theory and basic science. This converges with the historic role of the Nebraska Symposium as a prominent (and now the oldest sustained) forum for psychological theory. All the contributions in this volume emphasize the theoretical basis of application and the necessity of logical and conceptual continuity in understanding complex processes. In the ﬁrst contribution Richard W. J. Neufeld discusses the advantages of formal mathematical theory for illuminating relations between variables as they interact in experimental science. He applies these advantages to the clinical practice of assessing cognitive impairments. Decrying a continuing overreliance in much psychological research on statistical analyses associated with Fisher and Pearson, Dr. Neufeld asserts that formal mathematical modeling of cognitive processes will, ultimately, lead to greater theoretical clarity about normal and abnormal cognition and better clinical-assessment techniques. It is noteworthy that, while the tradition of mathematical modeling in psychology has a long and honored past, the increasing availability of powerful computational tools (e.g., computers and analytic software) supports the kind of sophisticated modeling in the hospital or clinic that was impractical in earlier decades. At a more general level, Dr. Neufeld characterizes his approach as a novel form of construct validity, one based on the inherent mathematical properties of the cognitive processes he studies. In this sense his contribution is a sophisticated exemplar of the use of complex modeling to achieve traditional theoretical goals of experimental and clinical psychology, as articulated by such historical ﬁgures as Lee Cronbach and Paul Meehl. In the next contribution, Wolfgang Tschacher and Zeno Kupper provide a synthesis of dynamic systems theory and current cognitive science. Inspired by the historic role of Gestalt psychology in the evolution of cognitive science, their discussion invites us into the heart of psychology’s theoretical legacy. Drs. Tschacher and Kupper then apply their perspective and methods to the complex realm of psychopathology. They present data sets and analyses from recent

xii modeling complex systems research with people diagnosed with schizophrenia and demonstrate the importance of tracking individuals with multiple measurements over time in order to detect oscillations or trajectories in rehabilitation and recovery that would be missed in the typical cross-sectional approach. Using a time-series analysis, they identify unique patterns or dimensions of intrasubject characteristics that have complex but meaningful interrelations. Returning to theoretical principles, they show how complex, dynamic formulations can be translated into useful clinical instruments and methods. Finally, in a tribute to the Nebraska Symposium’s historic focus, their contribution culminates with a characterization of motivation as identical to the ongoing action of complex human cognitive processes operating to order and simplify a complex world. A second exemplar of complex modeling to achieve traditional goals is provided by Suzanne P. Lajoie. Dr. Lajoie uses theoretically grounded performance modeling in the development of computerbased “intelligent” tutoring systems designed to help learners master the complexities of real-world endeavors. Learning how experts go about problem solving and decision making through “cognitivetask analysis” is an important aspect in the process of developing an effective tutoring system. Dr. Lajoie highlights the importance of discerning experts’ relevant “dimensions of expertise” (e.g., self-monitoring), as expressed in a speciﬁc context, in developing effective models. She also emphasizes the importance of other variables, e.g., emotional, motivational, and social, and she describes strategies for determining what to model, whom or what should serve as the model, and how to model the content and/or process. She then translates these principles into design considerations for effective educational technology. The next contribution extends application of complex modeling from education to knowledge management. Mark A. Musen discusses past and current efforts to develop computer applications to support decision making and data representation in health care. Dr. Musen’s theory base is not psychology or neuroscience but artiﬁcial intelligence. His technology is the technology of computer engineering. Nevertheless, he envisions a future role for psychology in the development of artiﬁcially intelligent systems to manage our already enormous and rapidly expanding knowledge base. It is noteworthy in this regard that psychology has drawn from engi-

xiii Introduction neering as much as vice versa, from radar-inspired signal detection models of perception to band-ﬁlter models of attention to computer models of executive cognition. Herbert Simon, in his 1994 Nebraska Symposium on Motivation contribution, cited artiﬁcial intelligence as a promising model for human cognition. The impact of complex models for knowledge management may be in psychology’s future rather than its past. In the next contribution, Eduardo Salas, Kevin C. Stagl, C. Shawn Burke, and Gerald F. Goodwin scrutinize complex processes associated with small groups of people brought together for common purposes. They advance “the science of teams” by providing a detailed review of representative models of team performance in organizations and other naturalistic settings generated over the past quarter century. In their review, Drs. Salas, Stagl, Burke, and Goodwin ﬁnd the invocation of input-process-output (ipo) models, consonant with the “general systems” framework that has inﬂuenced many areas in the social sciences during the past several decades, to be a key commonality among these models. There is greater diversity among models with respect to emphasis on internal team processes versus greater attention to the inﬂuence of external, contextual factors. The authors conclude that both inﬂuences are important and that, consequently, more sophisticated modeling techniques are needed to successfully deal with the resulting dynamic complexity, particularly in naturalistic settings. Drs. Salas, Stagl, Burke, and Goodwin then turn to a description and elaboration of a new and unique multilevel integrative framework for understanding team functioning. This new model is distinctive in the importance that it attaches to individual team members’ cognition as an important moderating variable as well as group decision making, shared mental models, and external factors. Michael J. Mahoney’s contribution is a nuanced interlacing of several kinds of “models,” including verbal metaphor, narrative, photography, and poetry. Dr. Mahoney discuses various perspectives on “complexity” theory and its precursors in philosophy and science, including current theoretical frameworks such as dynamic systems theory, complexity studies, and chaos theory, placing them in the context of the history of ideas. He describes and elaborates on constructivism, an integrative framework and family of theories. Dr. Mahoney’s contribution includes two appendices. The ﬁrst provides

xiv modeling complex systems a synopsis of important aspects of human change from the perspective of constructivism. The second provides rich and provocative perspectives for incorporation in the practice of counseling, psychotherapy, coaching, and other educational pursuits.

Complexity, Systems, and the Nebraska Symposium on Motivation: A Brief History of Ideas The perspectives reﬂected in this volume are exemplars of an evolving set of conceptual frameworks that inﬂuenced thinking in many areas of science during the second half of the 20th century. These frameworks are most generally associated with general systems theory, and addressing complexity is one of their key common features. Having recently celebrated a half century of the Nebraska Symposium on Motivation, as this volume’s editors we saw a useful purpose in reviewing the more than 300 individual contributions that constitute the previous volumes, to identify ideas that anticipate or shape the approaches to complexity that we ﬁnd in contemporary work. We found a richness of such ideas, so many that only a few can be highlighted here. The remainder of our introduction to this volume is a review of ﬁve especially resonant contributions from volumes past: Heider (1960), Walker (1964), Leeper (1965), Newcomb (1953), and Barker (1960). We selected contributions that, in addition to showing the nascent ideas about complexity and systems theory discussed in this volume’s contributions, have clear relevance to practical application and translational research and especially to our own particular interests in physical medicine and psychiatric rehabilitation. One common characteristic of systems theories is an organizational scheme that orders speciﬁc mechanisms and processes according to their respective complexity. Two terms from classic learning theory, molar and molecular, serve to deﬁne the poles of these schemes. Processes and mechanisms are molar rather than molecular to the degree that they represent the integrated interaction of multiple components. Psychology itself reﬂects this type of ordering, ranging as it does from theories of neuronal activity to neuropsychology to the psychology of social cognition and interpersonal behavior. The historical Nebraska Symposium contributions reviewed here follow

xv Introduction a molecular to molar rather than a chronological sequence. Interestingly, contributions from common theoretical perspectives can differ with respect to the molecular-molar dimension. Heider (1960) and Barker (1960) integrate classic Gestalt principles with subsequent theories, but Heider addresses comparatively molecular expressions of key processes, while Barker is at the other end of the continuum, addressing “spontaneous” organizational processes at the level of a human community. Other contributions to be reviewed address system organizational processes at the cognitive level (Walker, 1964), the emotional level (Leeper, 1965), and at the social/interpersonal level (Newcomb, 1953). We have included extended excerpts from the original contributions, in an effort to preserve the style and tone of the original presentation. Also, we hope to provide enough of the language in sufﬁcient detail for readers to draw their own conclusions about the relation of the historic ideas to those of the present volume. However, these passages cannot fully convey the logic or the eloquence of the source materials. The reader is encouraged to consult the full chapters in areas of particular interest.

Fritz Heider (1960) Many Symposium contributions have included Gestalt psychology frameworks offering concepts and models that resonate with contemporary perspectives (e.g., self-organizing systems) that, in many cases, evolved from earlier, traditional Gestalt investigators. An exemplary illustration is Fritz Heider’s “The Gestalt theory of motivation” (1960). Heider’s work has been more inﬂuential on certain theorists than the relative paucity of scientiﬁc citations would indicate. George S. Klein, then editor of the journal Psychological Issues (itself representative of the rise of interest in ego functioning in the psychoanalytic literature the 1940s and 1950s), provides a brief preface to a set of selected papers published as Fritz Heider’s monograph On Perception, Event Structure, and the Psychological Environment (1959). In addition to describing Heider’s unique viewpoint, Klein’s comments highlight important aspects of the view of rehabilitation and the notion of participation as an important goal of rehabilitation (World Health Organization, 2001). The

xvi modeling complex systems ideas presented have direct relevance to current models of rehabilitation and crucial aspects of participation and quality of life, each intimately associated with the behavior settings available to the individual, the quality of his or her social networks, and other variables that will be discussed in the ﬁnal section of this introduction. According to Klein (1959): “Fritz Heider’s work . . . has had over the years a signiﬁcant, if relatively unobtrusive, impact on some of the most important theorists of our time, notably Kurt Koffka, Kurt Lewin, and Egon Brunswik. More recently, Heider’s inﬂuence has been detectable in perception theory, for example, in the work of James Gibson. Still, his writings cannot be called ‘popular’” (p. v). Klein continues: Heider’s emphasis on the “macrophysics” of things (in contrast to the reductionist emphasis on microphysics), the important distinction he develops between those parts of the environment which mediate (“medium”) and those which are mediated (“thing”), his analysis of how we may distinguish behavioral events attributable to the structure of the environment and those attributable to the structure of the perceptual system—all of these merit close study. . . . Throughout the papers the composition of a “unit”— whether spatial, temporal, or causal—is of central importance to Heider’s distinction between “thing” and “medium.” The deﬁning properties of a “unit,” therefore, come in for extended and penetrating analysis. . . . Heider has made [an attempt] to penetrate the essential nature of the concept of structure. The general macrostructures which he describes may apply to their subjective counterparts in ego organization. . . . A unique feature of Heider’s approach is his attempt to fathom environmental structure not from the response side— from the inside outward, as it were, as is common in psychological theories—but from the outside inward, that is, by specifying the architectural rules of the extrapersonal world of physical object and event units. The result, then, is an extraordinarily fresh confrontation of the external structures which are assumed but never speciﬁed in psychoanalytic notions of reality testing and adaptation. (p. vii)

xvii Introduction In his Symposium contribution Heider provides an overview of four “thought models or schemata” characterizing the Gestalt tradition at the time. Heider terms the initial model the classic Gestalt theory. The model is based on the work of investigators such as Wertheimer, Köhler, and Koffka. Describing this model, Heider highlights a number of aspects of this perspective that bring to mind such concepts as patterning and perceptual organizing processes that seem quite consistent with concepts used today. Heider begins by emphasizing the Gestalt concept of good ﬁgure, advanced by Wertheimer, who applied it to visual processing: This principle states that the perceived ﬁgure tends to be as good as the stimulus pattern will allow, or as Koffka says: “Psychological organization will always be as ‘good’ as the prevailing conditions allow. In this deﬁnition, ‘good’ is undeﬁned. It embraces such properties as regularity, symmetry, simplicity and others. . . .” For instance, slight irregularities in the shape of visual forms are usually not noticed. Kohler gives the following example: Faces of people usually appear to us symmetrical, in spite of the fact that they are rarely objectively symmetrical. We may notice this irregularity in another person’s face when we look at his mirror image; but ordinarily we don’t see it. Kohler has called attention to the fact that a tendency towards simplicity can also be observed in physical systems, as, for instance, Ernst Mach has pointed out. Kohler gives many examples in his book on Physical Gestalten [1924]. . . . Let us recapitulate: Wertheimer observed the tendency toward good form with percepts; Kohler then related this observation to a similar tendency found in physical systems. Of course the same tendency is then assumed to rule the process in the physiological brain ﬁeld. Since isomorphism is assumed, all this ﬁts very well together. The thought model is one of a complex process with many part events which interact in such a way that a certain endstate is reached, an end-state which is in some way distinguished, and which has characteristics the other possible states do not have; as long as this end-state is not reached something will happen. On the other hand, when it is reached, the process attains an equilibrium and nothing more will happen.

xviii modeling complex systems Furthermore, the end-state will come about regardless of what the beginning state of the system is: thus one can talk about a tendency, which implies direction, a reaching of the same endstate by different routes. (Heider, 1960, pp. 145–146) Heider follows his discussion of Köhler’s ideas with a perspective from Kurt Koffka: It is not surprising that Gestalt psychologists have applied this same thought model to behavior. For instance, Koffka does so in his book on The Growth of the Mind, which ﬁrst appeared in 1921. I should mention that Koffka uses the term “closure” for the distinguished end-state, a closed ﬁgure being a better ﬁgure than an open one. Tendency to closure is therefore only another name for tendency toward simplicity, or goodness of ﬁgure. This is what Koffka says (Koffka, 1925, p. 103): . . . The characteristics of closure . . . belong not merely to the phenomena themselves, but likewise to the behavior taken as a whole, including all reactions made to the environment. Instinctive activity then becomes an objective mode of behavior analogous to such phenomena as rhythm, melody, and ﬁgure. . . . He [Köhler] calls the state toward which the processes in the organism are directed a “standard state.” It has to be distinguished from the state of equilibrium (Kohler, 1938, p. 325), and he describes it as follows (Kohler, 1938, p. 303): The essential characteristic of regulation is an invariance of direction. Whatever initial conﬁguration may obtain in those systems when we begin to observe them—if we observe long enough their inner displacements or transformations will always be found to bring them nearer to a standard status. The word “standard” points here to the fact that the ﬁnal status is independent of the initial conﬁguration. . . . Essentially . . . the thought model of a system tending towards a standard state is applied to directed action, and this model had its origin in the principle of good ﬁgure. However, . . . when we try to ﬁnd out how it is carried out, we see that

xix Introduction two steps are necessary for the transition from the phenomena of the visual ﬁeld to action. First, we have to take into account not merely perceptual appearances but a space in which behavior occurs; and secondly, we have to consider the objective environment, and the way the organism effects changes in it. . . . Thus we have to substitute for the visual ﬁeld what Koffka called the behavioral ﬁeld and Lewin the life space. This behavioral ﬁeld is conceived of as having similarities with the visual ﬁeld. It also is a system containing a great number of part processes which interact, it exhibits forces and tensions, and tends to arrange itself in such a way that a distinguished end-state is reached. This distinguished end-state, in some way comparable to the simple ﬁgure, is the state of the person who has reached the goal in his life space. Now, this life space or behavioral ﬁeld is a concept which involves many difﬁculties and unsuspected depths and snares. . . . I can only say that in a ﬁrst approximation, which, however, is not entirely correct, one can conceive of it as representing the environment of the person as the person himself experiences it—and it is in some way related to the brain ﬁeld, to physical processes going on in the brain. The Gestalt psychologist would characterize this relation as one of isomorphism, i.e., of structural similarity. This is the ﬁrst step we have to accept when we apply the principle of good form to activity: namely, the step from perceptual to behavioral ﬁeld. The second step requires a more extensive consideration. So far we have only considered processes which are “inside” the organism in some way, which are “encapsulated,” as Brunswik says. How is it possible that they produce effects outside the organism, in his physical environment? We assume that this behavioral ﬁeld changes in the direction of a distinguished state, maybe a state of minimal tension, i.e., the state of the person being at the goal. But . . . we have to understand how the tension in the behavioral ﬁeld makes the person reach the goal in reality. (Heider, 1960, pp. 147–149) He later continues: In action, not only a part of the organism, but the whole organism is involved. . . . The idea of the feedback or circular process

xx modeling complex systems can be applied also in this case: as long as the person has not yet reached the goal, there is a tension in the behavioral ﬁeld; this tension is communicated to the executive system, which changes the relation between organism and objective environment in such a way that the goal is reached; via perception this objective state is communicated to the behavioral ﬁeld; and thus the tension in this ﬁeld is removed. I have used the term “feedback” to characterize this process. However, one has to keep in mind that this circular process is not a simple feedback process. What distinguishes the circular process of Gestalt theory from simple feedback is the interpolation of the behavioral ﬁeld with its tendency towards a distinguished state [emphasis added]. (p. 150) Heider goes on to describe two models advanced by Kurt Lewin—the person model and the environment model, together constituting what Heider terms Lewin’s spatialized psychology. He notes the move from the perceptual sphere to the behavioral realm in Koffka’s work and what Lewin terms the life space. Heider’s discussion of Lewin’s concepts is thought provoking, and the reader is encouraged to review those concepts in the source material. Heider then offers a relatively brief summary of his own recent theorizing, describing what he calls his balance theory, which he feels answers some questions left inadequately treated by the classic Gestalt theories he has summarized: This theory of balance deals mainly with conﬁgurations consisting of a number of entities between which exist certain relations. The entities can be persons—the own person or other persons—and other entities, as for instance, things, situations, or groups. The relations considered are mainly of two sorts: on the one hand attitudes of liking or disliking, and on the other hand unit relations of belonging. The main idea is that certain of these conﬁgurations are preferred, and that, if circumstances allow, they will be realized by the person either in such a mental reorganization as wishful thinking, or in an actual change through action. . . . In recent times a number of theories have been proposed which are similar to the one just outlined. I remind you, for instance, of Newcomb’s (1953) discussion of processes of com-

xxi Introduction munication [see below]. . . . These conceptions, symmetry, consonance, balance, and simplicity, are, of course, implied in that idea with which Gestalt theory started and which always was central to it, namely, the idea of a ‘good’ ﬁgure. We therefore have returned to the model we considered ﬁrst. This model implies a number of different entities with certain properties and standing in certain relations, which make up a constellation of factors tending toward a standard state. The properties of these conﬁgurations which determine their meaning and their fate are whole-qualities. Consonance or simplicity of the structure cannot be derived from the properties of the parts . . . If we study the p-o-x system1 which is composed of the own person (p), another person (o), and an impersonal entity (x), then we ﬁnd that the state of balance depends on the attitudes of p toward o or x. That is, the attitudes toward the parts of the conﬁguration, and the relation of these attitudes to each other enter as signiﬁcant factors, and determine the attitude toward the whole conﬁguration. . . . Thus, we are able to specify more exactly the conditions of goal selection, at least in some cases. The goal is not taken to be an unanalyzed entity which in some way acquired valence, but is derived from the properties of the structure [emphasis added]. . . . [The] difference between Lewinian theory and balance theory [is] in regard to the role structure plays. In Lewin’s environment model . . . structure is not intimately connected with the conditions of tendencies, nor with their effects. Structure helps us to derive the direction toward means from direction toward goal; but it does not help us to derive the direction to the goal. . . . . . . Thus we see that in these [Lewin’s] models the dynamic factors are not very closely linked with structure. Neither do the properties of the structure imply forces, nor do the forces affect the structure in a speciﬁable way. In the balance model the dynamic factors are intimately connected with the structure. The dynamic factors arise out of deﬁnable structural characteristics and the forces toward the standard state tend to change the structure in deﬁnite directions.

xxii modeling complex systems . . . [Unlike Lewin’s models], in the balance model structure in a state of equilibrium is deﬁnably different from one in a state of disequilibrium, and all the parts of the structure are relevant to this difference, not only the relation between two parts, person and goal. (Heider, 1960, pp. 167–170)

Edward L. Walker (1964) Edward Walker’s contribution to the 1964 Symposium, entitled “Psychological complexity as a basis for a theory of motivation and choice,” is generally congruent with Heider’s (1960) views, though the terminology employed by these scholars and the associated research traditions from which they come are quite different. Walker (1964) provides a concrete example of model building as well as an interesting perspective on the concept of complexity and mechanisms that contribute to the organization and self-regulation of behavior at the level of the organism. Also, as with Heider’s view, mechanisms of perception, cognition, learning, adaptation, and motivation are all seen as quite interdependent and quite closely related to environmental stimulus context. Walker states at the outset: This paper is an attempt to state what I believe to be the most basic questions of behavior theory, to elaborate the concept of psychological complexity as a potentially unifying concept, and to test its clarifying contributions with respect to some critical problems of behavior theory. The three basic questions of behavior theory are . . . 1. What is the mechanism that terminates an event? 2. What are the determinants of the next event? 3. What is the fate of an event after it is terminated? (Walker, 1964, pp. 48–49) He follows precedent in setting the temporal length of a psychological event at 0.5 seconds. Then he asks: When an event is terminated, what are the determinants of the next event? A great many subareas of psychology are devoted to an effort to discover and quantify the determinants of choice behavior. . . . “Habit,” “motive,” “subjective probability,” “util-

xxiii Introduction ity,” “set,” “attitude,” and “trait,” along with many other concepts in psychology, are reducible to names of intervening variables or theoretical constructs, each related to different sets of operations, but all, ultimately, determiners of choice behavior. It will be the argument of this paper that the concept of psychological complexity can be used to account for the termination of psychological events, and the choice of the next event over a wide range of traditional concepts of determiners of choice. Thus, the concept of psychological complexity can be useful in answering the ﬁrst two of the three basic questions. Psychological complexity can also be used to account for many of the phenomena associated with the trace of a past event, but this third basic question or problem is beyond the scope of the present paper. (pp. 51–52) Under the subheading “Psychological Complexity Theory,” Walker asserts: The major distinction that must be made is between “stimulus complexity” on the one hand and “psychological complexity” on the other. The ﬁrst is a characteristic of the external stimulus, more or less independent of the individual organism. Psychological complexity is a characteristic of the event itself, the organism’s response. Psychological complexity and neural process complexity will be assumed to be completely isomorphic. (Walker, 1964, pp. 52–53) In the context of a basic deﬁnition of terms, he turns to an elaboration of his concept of psychological complexity: Psychological complexity is a characteristic of the event itself and is thus a characteristic of the interaction of the organism with the distal stimulus when the event in question is initiated by a stimulus. Thus it is possible for two organisms to react with equal psychological complexity to stimuli with very different distal stimulus complexity values. The same organism may also react with different degrees of psychological complexity at different times to the same stimulus. (pp. 54–55) Walker offers a brief description of the (assumed) underlying nervous system basis of the experience of optimal complexity, under the heading “Neural Net or Neural Process Complexity”:

xxiv modeling complex systems Underlying any psychological event, is, of course, a pattern of neural events. Such events are spatially three dimensional and occur over a ﬁnite period of time, a fourth dimension. It is assumed that variation can occur in the relative complexities of two neural processes. We shall refer to the relative complexity of a four-dimensional neural process as the relative complexity of the relevant neural net. Furthermore, we shall assume a complete isomorphism between neural net complexity and psychological complexity (Walker, 1964, p. 55). He offers as relative characteristics of simple versus complex neural net in terms of the four dimensions noted above: “simple” nets consist of processes that are relatively “small,” “short,” “focal,” and “central” (i.e., origin within the central nervous system); “complex” nets are relatively “large,” “long,” “diffuse,” and “peripheral.” Having offered a neurologically oriented substrate, Walker turns to the concept of psychological complexity itself. Under the heading “Optimal Complexity,” he asserts: The key concept of the theory I am attempting to fabricate is the concept of optimal complexity. The simplest and most straight forward psychological deﬁnition of optimal complexity is the following: Optimal complexity is that degree of psychological complexity the organism will seek to maintain. If a psychological event is more complex than the optimum, the organism will behave in such a manner as to reduce the complexity of the event. If a psychological event is less complex than the optimum, then the organism will behave in such a manner as to increase the complexity of the event. Optimal complexity can be bracketed by other values of psychological complexity. In perception, an input level far above optimum produces “mental dazzle.” A lower limit is a level of complexity that is below the threshold of consciousness. In motor activity, psychological complexity far above optimum results in discoordinated tetany, and there is a lower value which constitutes the threshold for action. The optimum is a “normal” percept or a smoothly coordinated movement. . . . The sequence is inevitable and the fall in neural net complexity is an automatic result of observable and fairly well understood neurophysiological characteristics. Since neural net

xxv Introduction complexity and psychological complexity are assumed to be isomorphic, psychological complexity may be said to rise and fall as well. With a sufﬁciently complex stimulus, psychological complexity will rise to and exceed the threshold of consciousness, will rise to and exceed optimal complexity, will fall below optimal complexity, will then drop below the threshold of consciousness automatically, and will usually be followed by another psychological event. [The reader] will recall that the ﬁrst of the three basic questions of behavior theory was: What is the mechanism which terminates an event? The answer is: Whether the stimulus for an event is continued or not, a psychological event undergoes a sharp and automatic drop in complexity during a period of approximately one-half second after its initiation. (Walker, 1964, pp. 56–58) With respect to repeated activation of an event, he asserts: Repeated activation of a neural net will result in a progressive decrease in the psychological complexity of the event involved. (p. 59) With respect to the second major question (What are the determinants of the choice of the next event?), he continues: The principle of optimal complexity incorporates the dynamism that the organism will seek such a level. The termination of one event occasioned by the automatic reduction in its psychological complexity below the optimum level, literally forces choice of that event among available next events which will be nearest optimum. Therefore: Among available alternatives, an organism will choose as a next event that activity which is nearest optimal psychological complexity. It is assumed as a working hypothesis that many of the major determinants of choice behavior such as reinforcement, habit, motivation, curiosity and other collative2 variables, subjective probability and utility, and others ultimately can be reduced to a single concept—psychological complexity. (p. 60)

xxvi modeling complex systems Walker offers a brief recapitulation: The theory can be stated in an abbreviated form. During the course of a psychological event that has a duration of approximately 500 [msec], the psychological complexity of the event rises abruptly and falls more slowly. The automatic reduction in the psychological complexity of an event insures that it will drop promptly below the optimum to be replaced by that one of the available alternative events which is nearest optimum. The psychological complexity of alternative behaviors or events will be a function of four variables. They are: (1) the stimulus complexity of the initiation stimulus; (2) the time since this particular event has occurred previously; (3) the number of times that event had occurred before; (4) the arousal properties of the stimulus or event. (Walker, 1964, p. 60) Walker addresses the issue of what behavior(s) might be expected when no near-optimum event is present. In the case of all available alternatives below optimum, an individual might commonly respond in one or more ways (slightly adapting Walker’s text): 1. Search the environment or his own repertory for more complex events; 2. Find a more complex stimulus in the environment to which he had not attended previously, or he might fall to daydreaming; 3. React by locomoting, getting up and moving about; 4. Seek arousing stimuli; 5. Seek to differentiate previously unexplored potential complexity in his environment or in old thought sequences and problems. All of these devices would serve to increase the complexity level of the sequences of events which are occurring. All would serve to move the sequence nearer an optimal level of complexity. (Walker, 1964, p. 61) At the other extreme: Situations in which the psychological complexity levels are above optimum are usually situations in which the sensory inputs into the nervous system are providing more information than the organism can process. This may result when the exter-

xxvii Introduction nal environment is too complex, when the problem one is working [on] is beyond immediate solution, or when the motivational or emotional system is in a highly aroused state. (pp. 61–62) In such circumstances, he notes common reactions: 1. The organism may shift attention or narrow attention to a limited portion of the stimulus input; 2. If the overload is of external origin, the organism may locomote to a less complex circumstance; 3. If either are difﬁcult or impossible, the organism may attend repeatedly to the same stimulus in an effort, usually successful, to produce a reduction in the psychological complexity of the situation through repeated activation of the relevant event; 4. An associated result of repeated activation is to organize a very complex stimulus into a smaller number of “chunks.” (p. 62) Walker’s contribution concludes with a survey and critique of relevant research and theoretical distinctions to provide support for and elucidation of the conceptual framework adduced therein. He buttresses his concept by applying it to existing experimental data that are not easily explainable by any other existing theory. For example, he applies his theory of optimal psychological complexity to the often-observed (but less frequently reported) decremental variations in “conditioned responses” following “learning” experiments that have been taken by many to be unexplainable anomalies. After noting a few such “anomalous” observations: For the sake of the argument I am certain to get, let me take the position that the appropriate “learning” curve shape in running studies, conditioning studies, and selective learning studies, is one that rises and falls to zero or to a steady level below the maximum performance. The curve that rises to a steady maximum and remains there indeﬁnitely is likely to be rare. The reason that we see few “learning” curves of the postulated type is that most experimenters know in advance what a learning curve is supposed to look like. As a result of this knowledge, they stop training when the “asymptote” is reached, or, if they obtain a curve which does not ﬁt their conception of what one should look like, they ﬁnd a great many other ways to respond

xxviii modeling complex systems . . . other than to publish their sin against respectability. They throw away their data. They restructure the apparatus. They change the parameters of the study. They change the design. This process is known as the establishment of experimental(er) control. Sooner or later they manage a situation in which they obtain the “right” answer. I can attest that this process is carried out in good faith and under the assumption that in so doing, one is behaving like a sound, rigorous, and careful experimentalist. I can attest to this because I am one of the sinners. Thus psychological complexity theory handles the usual learning curve, extinction, and the drop in performance that often occurs under continued reinforcement. It predicts that most experimental situations will produce a drop in performance if training is continued. (Walker, 1964, pp. 85–86) In later comments concerning Walker’s presentation, his fellow presenter, Frank Logan, in addition to suggesting caveats to the former’s views, concludes with an important observation: There are several features of Walker’s approach with which I am in strong agreement. A language that avoids the artiﬁcial separation of stimuli and responses more nearly captures the uniﬁed, interdependent inseparability of psychological events. It is also becoming increasingly recognized that the fundamental behavioral operation of an organism is selection or choice. . . . Walker’s attempt to develop a system that can deal with behavior dynamically, i.e., continuously over time, is perhaps the critical feature necessary to achieve a general integration. And, by whatever means, visualizing such disparate concepts as habit, motivation, and decision-making in terms of a single construct certainly is one we should applaud. (Logan, 1964, p. 98)

Robert Ward Leeper (1965) Robert Leeper’s contribution, “Some needed developments in the motivational theory of emotions” (1965), is focused on urging greater attention to understanding emotions as motivational factors, rather than “lower,” simply “energizing” or “arousing” factors that are then guided by “higher-order” functions (e.g., perception and/

xxix Introduction or cognition). In fact, Leeper highlights the ultimately inseparability of processes of perception, motivation, and emotion: Still earlier, David Krech (1949, 1950a, 1950b, 1951), in his usual impassioned style, had reasoned that it is unrealistic to conceive of psychological phenomena in terms of separate processes of perception, motivation and learning. Instead, he urged, we ought merely to conceive of “Dynamic Systems.” These, he said, are so deﬁnitely organic unities that no single aspect of such a system can be changed without changing the other aspects as well—we have been dealing in myths in believing that we could vary some one of these aspects while keeping the other aspects constant. Though proposing a less drastic statement on this point, E. C. Tolman (1932, 1948) had been suggesting some perceptual factors in motivation in his view that motivation is partly a matter of reward expectations and punishment expectations. Kurt Lewin similarly had been discussing many problems of motivation in terms of factors in the organism’s “psychological environment.” . . . In my own previous writing, my original paper on a motivational theory of emotion (1948) was extended to some extent into the perceptual-motivational theory which has been elaborated in the present paper. . . . One odd fact about these various earlier discussions of a perceptual or conceptual interpretation of motivation is that their authors have made practically no references to the related ideas of the other papers. This is the more surprising in view of the fact that most of this group are more or less closely related to one another both personally and as regards their general theoretical outlooks and interests. It seems, therefore, as though each of these persons had to grope to the concept on his own, even though possibly helped in ways that he did not recognize by his predecessors or colleagues. I make this suggestion with somewhat more conﬁdence because I remember that, in my own case, when I ﬁrst read Krech’s papers on “Dynamic Systems,” they did not make much sense for me. . . . And, peculiarly, it took me a long time to recognize that Lewin’s ideas might be thought of as a perceptual theory of emotion. . . . Maybe this sort of thing will continue to be the case. If a perceptual theory of motivation is to become more common,

xxx modeling complex systems perhaps each psychologist will have to ﬁgure it out for himself. (Leeper, 1965, pp. 111–112) Leeper summarizes some major themes: The suggestion that comes from a number of sources, therefore, is, ﬁrst of all, that emotions are motives, and then, second, that emotional processes, along with all other motives, are perceptual or representational processes. The suggestion that comes is that emotions and other motives do not exist or operate in any less complex sense than this. . . . Even though perceptual habits are hard to change in some cases . . . it seems that all perceptual habits can be modiﬁed by learning and that sometimes such modiﬁcations can occur suddenly and dramatically. If emotional habits are perceptual habits, these same possibilities should exist for them. (pp. 113, 115)

Theodore Newcomb (1953) Theodore Newcomb’s contribution, to the very ﬁrst volume of the Symposium, was “Motivation in social behavior” (1953). Newcomb makes clear that he does not believe that a psychology of motivation in social situations should be fundamentally different or discrepant from a general psychology of motivation. Rather, it should be subsumed by a broader model of motivation that describes human motivation in any situation. However, he also notes that the breadth of such an overarching model would not lend itself to making predictions or heighten understanding of speciﬁc processes or variables within a particular subdomain of psychology in general (e.g., individual behavior in a learning situation; behavior in a social situation): A general theory, whether of motivation or of evolution of species, is never speciﬁc enough to predict within a speciﬁc area those details in which we are often most interested. Indeed, it is from the relatively limited theories that the relatively inclusive theories must in the long run emerge. (Newcomb, 1953, p. 139) The relevance of his contribution to the topic of self-organized systems is seen in his analysis of the dynamics or organization of communicative behavior in an interpersonal context.

xxxi Introduction Newcomb goes on to deﬁne terms and delimit his focus to communicative behavior between individuals: The properties of objects may be studied either objectively or phenomenally—preferably in both ways—but in any case, they are studied not as things-in-themselves, but as related to persons. Thus the characteristic way in which social psychologists study motivation is in terms of person-object relationships (the term “object” includes other persons, of course). Since, as we have all learned in recent years, motivational phenomena are intimately interlinked with perceptual phenomena, it is often necessary to distinguish two aspects of person-object relationships, which may be labeled the cathetic and the cognitive. Often, however, one does not need to make this distinction, while still bearing in mind that both aspects are involved, and in such instances the term “orientation” is a useful one. The term is similar to the concept of “attitude,” except that it connotes “existing directedness” and not merely a predisposition or a readiness. Orientations are known, of course, only as they are inferred from observable behavior. Insofar as such behavior involves reciprocal stimulation and response (or anticipations thereof) it is traditionally referred to as “interaction.” But one cannot observe interaction-in-general; one must observe discriminable units of behavior. I propose, therefore, to use as such an interaction unit the communicative act, deﬁned as any observable behavior by which information, consisting of discriminative stimuli, is transmitted from a human source to a human recipient. For present purposes, it is assumed that the discriminative stimuli have an object as referent. Thus in the simplest possible communicative act, one person (A) transmits information to another person (B) about something (X). Human social behavior is thus to be studied in terms of the conditions and consequences of varying communicative acts. And problems of motivation in social behavior are to be studied in terms of orientations toward the two kinds of objects necessarily involved in communicative acts—i.e., persons as recipients of transmitted information and objects (including persons) as referents of transmitted information. The relationship between orientations and communicative acts, as we shall see, is a

xxxii modeling complex systems circular one, so that it will be necessary to consider each of them, in turn, as varying with the other [emphasis added]. (Newcomb, 1953, pp. 140–141) Following a second section summarizing relevant ﬁndings concerning group membership, orientations, and communication, Newcomb moves to a section titled “Communicative Behavior as Varying with Orientation toward Persons and toward Objects.” In this section he explicates the systemic relations between communicative acts and the orientations of individuals in a communication setting: I can hardly imagine anything that would surprise you less than to hear that communicative acts are learned in ways that seem to have something to do with rewards and punishments. I shall stop, however, only to indicate in the most general kind of way what seems to be the nature of the learning conditions of communicative behavior. These conditions have to do with what I have already referred to as the individual’s necessity for co-orientation—i.e., relating himself simultaneously both to objects and to persons as actually or potentially related to those objects. . . . We may start with the assumption that orientations both toward persons as potential co-communicators and toward other objects have adaptive value; not to be oriented to them would mean to have no cognitive content regarding them and to have no “hypotheses” (in the Postman-Bruner sense) as to their potentialities for reward or punishment. The further assumption that co-orientation has adaptive value stems from what I believe to be the fact that neither kind of orientation occurs singly and independently of the other, in connection with communicative acts. First, the orientation of any communicator, A, toward B, a potential recipient of his communication, rarely, if ever, occurs in an objectless vacuum. . . . Secondly, and conversely, the orientation of the communicator, A, toward almost any conceivable X rarely, if ever, occurs in the total absence of an orientation toward B, the potential recipient of his communication. (“Autistic” verbalization, of the kind Piaget reports in young children, would, of course, represent an exception . . .). The very fact that B is a potential recipient requires some kind of orientation toward him. . . . Al-

xxxiii Introduction most invariably, moreover, there is included in this orientation toward B some assumption—however accurate or inaccurate— about B’s orientation toward the object of communication. From this elaboration of what is perhaps only too obvious, I want to deduce a single point—that there is a necessary interdependence between co-orientation (which itself involves an interdependence of orientations) and communicative acts. . . . Since, according to these assumptions, there are relationships of interdependence among several distinguishable orientations, it is convenient to regard them as together constituting a system. For some purposes the system is best treated as an objective one—i.e., a model employed by the observer. The elements in this system are, minimally, A, B and X (a source, a recipient and an object of communication); the interdependent orientations among them are A’s toward B and toward X, and B’s toward A and toward X. . . . The implications of this model are: (1) that while at any given moment the system may be conceived of as being “at rest,” it is characterized not by the absence but by the balance of forces; and (2) that a change in any part of the system may lead to changes in any of the others. I shall make one further set of assumptions about the system. . . . These assumptions are that (under the stated conditions) communication tends to result in increased similarity, or congruence, of A’s and B’s orientations toward X, and that, as a result of learning, communicative acts are instigated by the anticipation of increased similarity or congruence (or, alternatively, by the threat of decreased similarity or congruence). (Newcomb, 1953, pp. 147–149) After positing adaptive advantages of his concept of congruence, he goes on to articulate an important aspect of his “A-B-X” system, which he considers a “strain toward congruence” (p. 149). The systemic perspective of Newcomb’s contribution and his initial observations of the relation between explanations/models at the level of subareas of psychology concludes with a view to the future: I should like to suggest (with a good deal of tentativeness) that something along the lines of the framework of co-orientation which I have roughly sketched out may ﬁnd a place in

xxxiv modeling complex systems general motivation theory. Many, among the higher forms of animal life, at least, are capable of plural orientation, and the actual direction of behavior at any given moment often cannot be accounted for in terms of any single object-orientation, others being held experimentally or hypothetically constant. . . . I suspect that the study of social behavior can provide evidence, in ways other behavior cannot, of how behavior directedness varies with multiple orientations. If so, an adequate theory of motivated social behavior will have contributed something to a general theory. Last, and far from least, an adequate general theory would take fuller account than it does today of self-orientation. . . . Here, as in the case of other concepts of peculiar relevance to social motivation, it is my belief that more extrapolations from a general theory will not sufﬁce. Theorists from McDougall and Freud to Murphy and Rogers have properly accorded to the self a central place; though not always, in my judgment, have all of them seen that place in its full social context. Not only are self-orientations part and parcel of other-orientations, I would insist; they are inextricable from the eternal triangle of self, other persons, and the common environment. A general theory of motivation, when it is mature enough to include these interdependent orientations, will have borrowed from a theory of motivation in social behavior, as well as helping to establish it. (Newcomb, 1953, p. 159)

Roger Barker (1960) Roger Barker’s contribution, “Ecology and motivation” (1960), includes an account of alteration of individual state(s) as a function of external, higher-order patterns or change. Like Heider and Newcomb, Barker underscores the necessity of taking individual and environment into account as a unit in any thorough analysis of behavior, and, hence, his contribution advances themes consistent with the conceptual frameworks advanced at the 2004 Symposium. Barker’s conceptual framework is like that of Heider (1960) and Kurt Lewin’s concepts of ﬁeld and life space (Lewin, 1938). This provides an intellectual context in which to consider the importance of taking both in-

xxxv Introduction dividual and environment into account. At a practical level, this resonates with major themes in rehabilitation, for example, the World Health Organization’s recent emphasis on the construct participation as the ultimate aim of rehabilitation efforts (see World Health Organization, 2001). This construct is of importance because it underscores the importance of including assessment and modiﬁcation of an environment, in addition to clinical treatment, as a vital part of the rehabilitation process. Barker outlines features of his concept of psychological ecology, including the central concept of behavior settings, which provides an important window on our understanding of a range of psychological phenomena as a “system” and is, at times, a very useful unit of analysis for psychology. The relevance of Barker’s concepts for the issues addressed in the present volume is that, like Newcomb (1953), Barker describes a framework that explicitly relates systems concepts to adaptive processes at the social/community level. Barker begins by incorporating from the work of Egon Brunswik an emphasis on the critical importance of including in accounts of perception and behavior the environment in which an individual acts and perceives. In Barker’s words: Brunswik (1955) described psychological schools and theories in terms of their positions upon a macro-unit he considered to be the true vein of psychological ore. This vein extends from the environment to the environment; namely, from distal objects in the ecological environment, through proximal stimuli at the receptor surfaces of a person, through the person’s peripheral receptor mechanisms, through his central processes, and through his peripheral effector systems, to his proximal reactions, or means behavior; and it ﬁnally terminates in the focus of the total unit: the person’s achievement with respect to the nonpsychological world of things. The three major sectors of this unit are . . . (1) the ecological sector of objects and physical stimuli (preperceptual), (2) the organism or intrapersonal sector, and (3) the behavioral sector which occurs, again, in the ecological environment. (Barker, 1960, p. 1) Barker, along with Brunswik, regards the entire span of the E-E (environment-environment) unit as the fundamental unit of analysis with respect to psychology; it is “the basic psychological entity.”

xxxvi modeling complex systems He takes issue with some of Brunswik’s conclusions, advancing the hope that taking the entire E-E span into account can lead to more than a probabilistic framework for psychological explanations: I hold the hope that a more detailed, conceptual, and explanatory account of the whole course of events can be achieved, particularly at the junction point between the ecological and the intrapersonal sectors of the unit, and especially with respect to motivation. This, in fact, is the theme of my paper. (Barker, 1960, p. 3) The E-E unit, then, is to be taken as the ultimate unit of analysis; one obvious way to understand this unit is as a multisectored system. As can be seen in the material reproduced below, one theme that recurs in other examples of systems approaches is that of the close relation between the processes of change, perception, and motivation. A second major theme is the inﬂuence of changes or properties in one part of the system on the qualitative status of other parts of the system, a deﬁning aspect of all exemplars of models informed by general systems theory. Finally, the interaction of sectors of the E-E unit is considered with respect to emergent social aspects in Barker’s system: For a psychology deﬁned in terms of E-E units, the usual considerations of motivation are not adequate. These considerations almost always make personal motives the whole story of the energetics of behavior, and place them within the organism. But a unit is a unit; it is indivisible. When it is a psychological unit, the environment, the organism, and the behavior are all involved, and energetics must occur in all of the parts. Either the E-E unit is false, or motivation theory is too limited. (Barker, 1960, p. 4) Barker goes on to lay the groundwork for discussion of his theory of behavior settings by introducing some concepts that provide the context for his central theses. In a later section he provides further elucidation of the relation between the “entity” and the “environment” elements of his model: Ecology is concerned with relations between entity and environment. But before this statement has any useful meaning,

xxxvii Introduction entity and environment must be deﬁned. . . . Where does each entity end and its environment begin? . . . . . . To clarify this problem, it is necessary to revert to the levels of phenomena in science. . . . I have emphasized that the essential distinction between levels is this: The laws, the explanations, which have been devised to account for occurrences on one level are inadequate to explain occurrences on a different level, yet the levels are coupled systems. Another distinction that is crucial for the deﬁnition of environment is that between inside and outside. Every entity has a discriminable boundary; what is within the boundary constitutes the entity’s inside, and what is without constitutes its outside. . . . The environment of an entity is made up of those parts of the outside regions with which the entity is coupled by laws on a different level from those which govern the entity itself. . . . Here, for ecological problems, is the basis for delimiting an entity from its environment. The test is this: As we move from any discriminable thing to more remote, surrounding parts, a point is reached at which the governing laws, so far as we know them, become incommensurate, yet the linkage remains. This point marks the boundary of the entity and the beginning of the environment. (Barker, 1960, pp. 7–8) Barker asserts the desirability of taking the entire E-E continuum as the crucial unit of analysis, rather than abstracting only elements of it for psychological examination. He offers the study of psychological principles of learning as an example: The ﬁeld of learning is interesting in this connection. Learning is usually interpreted as the process, par excellence, by which the environment inﬂuences the organism and its behavior. This is the predominant way, almost the only way, a culture is presumed to shape the personality and behavior of the individuals born into it. The facts of learning demonstrate, however, as almost all learning theory recognizes, that even here the organism is the locus of driving forces without which learning does not occur. Indeed, within the context of learning it is, paradoxically, the behaving organism that endows the environment with behavior-controlling properties; the guiding and coercing powers of the environment have been shown to

xxxviii modeling complex systems depend upon what activities the organism has previously had with it, and these depend more upon the organism than upon the environment. Indeed, learning studies have demonstrated that almost every discriminable part of the ecological environment can be coupled with almost every kind of behavior. This is important information; it deﬁnes the range of an organism’s power to transform its connections with the ecological environment, and it implies that parts of the environment are almost equipotential. . . . It will be clear now where ecology enters the environment-environment unit, which Brunswik took as the realm of psychology. Psychological ecology deals with the relations between the nonpsychological sectors of this unit, governed by the laws of geometry, chemistry, economics, etc., and the intrapersonal and the behavior sectors, governed by psychological laws. (Barker, 1960, pp. 11–12) He then attempts to formulate an account of how these incommensurable system elements might be related (or, at least, an approach to a satisfactory understanding) by exploring earlier ideas of Fritz Heider’s. Barker makes a gradual transition to his concept of behavior settings. Because these ideas are readily available to the interested reader, a detailed presentation will not be offered here. However, one would highlight a particularly important element of his concepts concerning behavior settings: Barker highlights how behavior settings are regions in a community that offer certain opportunities and, along with these, require certain responsibilities. Furthermore, there is a relation between the peopling of these settings and both the number of responsibilities and the adequacy of performance that can be expected to occur. This relation between the demands of a given behavior setting and the impact on the behavior and life of the individuals populating these settings seems very compatible with more recent concepts. Implications for the analysis of complex settings and behavior are evident. Barker continues, in a section entitled “Theory of Behavior Settings”: Field studies in which I and my associates have been engaged, of the behavior of children in their natural habitats,

xxxix Introduction have brought us to the hypothesis that under certain precisely deﬁned and frequently occurring conditions, people stand in the relationship of media to behavior settings; and that under certain other less common conditions, people stand in the relationship of things to behavior settings, imposing certain absolute constraints on them. This hypothesis brings some order into data upon American-English differences in the behavior of children and adults, into data upon differences in the behavior of individuals in settings of different sizes, and into data concerning the behavioral consequences of physical disability. The wide ramiﬁcations of these simple ideas suggest that they may have a basic signiﬁcance for psychology, and particularly for the psychology of motivation. . . . It is ﬁrst necessary to describe behavior settings. When a mother writes, “There is a baseball game in progress on the playground across the street,” she does not refer to any individual’s behavior, but to the behavior of children en masse. The same is true of a newspaper item which reports, “The annual fete held in the St. Ambrose Church garden was a great success.” These are behavior settings. They are highly visible behavior phenomena; laymen mention them in conversation and in writing as frequently as they do individual persons. . . . Here are . . . [some] behavior settings: Streets and sidewalks Kane’s Grocery Clifford’s Drug Store Gwyn Café Pearl Café Midwest State Bank Of special relevance in the present connection, however, are the following characteristics of settings: 1. Behavior settings involve ongoing patterns of extraindividual behavior whose identity and functioning are independent of the participation of particular persons. 2. A behavior setting has a circumjacent soma of physical objects: of walls, doors, fences, chairs, dishes, typewriters, ad inﬁnitum, arranged in a characteristic spatial pattern, at a particular temporal and physical locus.

xl modeling complex systems 3. Behavior settings are homeostatic systems; they normally persist, often for years, at a relatively stable, characteristic level. . . . A behavior setting is a behavior entity, but its laws of operation are not the laws of individual psychology . . . Most of what we know about behavior settings is simple description, with any conceptualizations being not far removed from the surface appearance of settings. However, this is enough to make a beginning in tracing the connections along the Brunswikian unit which has its origins in this part of the ecological environment. For our purposes, the self-regulatory characteristic of behavior settings is crucial and must be considered further; it is this, indeed, which gives behavior settings, under certain conditions, the position of things which impose their own patterns on the people within them, who have the position of media. Behavior settings exhibit a stability-within-change, a persisting functional level which is due to a balance of many inﬂuences. Some of these issue from the larger community, some are intrinsic to the setting itself, and some originate within the individuals who populate the setting . . . . . . Forces operate in every setting. These multiple balanced forces assure that the level of a setting is more stable than most of its parts or conditions singly. One frequently occurring means of balancing the forces and maintaining the homeostatic level of a behavior setting is compensating for a deﬁciency in the number or docility of the parts of the medium by an increase in the amount of energy applied to each of them, and vice versa. When the media of a setting, the machinery, the tools, or the workmen, for example, are in short supply, those available have to work longer and/or “harder.” (Barker, 1960, pp. 15–21) In the last section of his contribution, Barker discusses “People: Media of Behavior Settings”: Six features of the relationship between people and behavior settings must now be mentioned. 1. People are part of the inside manifold of behavior settings.

xli Introduction 2. Of all the attributes of settings, people are the sine qua non. . . . 3. Each quasi-stationary level of a setting has its optimal population requirements. . . . 4. Of all the equipment and paraphernalia of a setting, people are among the most immediately malleable and adjustable. 5. Different behavior settings on the same level of functioning, and therefore with the same optimal population requirements, actually differ greatly in population. . . . 6. These ﬁve features of the relation between people and behavior settings emphasize the position of people as the media of behavior settings. This is true. However, there is one important exception. When the number of people in a setting, its population, falls below the minimal number required by its homeostatic level, the setting will be modiﬁed. (Barker, 1960, pp. 21–22) He concludes: Behavior settings with less than optimal people for their homeostatic levels are self-disciplining settings. The opportunities within them are matched by the obligations they contain. . . . We sometimes call them self-discipline. In reality they are controls built into the structure and the dynamics of the setting, into the ecological environment. . . . I would like to close with two remarks: (1) Brunswik’s environment-environment unit appears to be subject to more than empirical probabilistic laws, and (2) the ecological environment appears to be, especially, the seat of motivating inﬂuences. (pp. 48–49) Barker’s contribution—as is his body of scholarly work in general—is novel and interesting and would seem to have continuing applications today. In particular, the growing acknowledgment that community reintegration and quality of life are vitally important ends of rehabilitation efforts and that rehabilitation cannot really be considered a successful endeavor unless an individual is supported to the point of maximal participation in the life of the community, with the greatest degree of independence possible, leads inevitably to the recognition that there must be a satisfactory awareness of the

xlii modeling complex systems environment—the behavior settings and social/interpersonal environment to which an individual will be returning—before an optimal rehabilitation treatment plan can be developed and delivered. Barker’s (and his students’) techniques for identifying and cataloging community venues can serve as a guide in expanding rehabilitation practice to include such analyses. In this regard, in addition to Barker’s work and the models of Brunswik and Heider (already referenced), additional useful resources include Gibson (1979) and Wicker (1984).

Notes The editors would like to acknowledge and express a special thanks to Mr. Joe Brown for his superlative work as copy editor of this volume of the Nebraska Symposium on Motivation. Mr. Brown’s consistently resourceful and creative suggestions concerning the substance, organization, and presentation of chapter materials contributed immeasurably to the clarity and coherence of the ﬁnal product. As will be evident, the conceptual breadth of material and diversity of perspectives reﬂected in this volume are considerable. Mr. Brown’s timely and precise questions and observations, clear and patient counsel, and unﬂagging good humor throughout were notably catalytic in successfully bringing the 52nd edition of the symposium together in its present form. All volume editors should be so fortunate! 1. Compare Theodore Newcomb’s “A-B-X” model, discussed below. 2. Variables such as curiosity, novelty, and stimulus change, stimulus aspects sometimes thought to provoke increased engagement, interest, and/or increase arousal.

References Barker, R. (1960). Ecology and motivation. In M. R. Jones (Ed.), Nebraska symposium on motivation (Vol. 8, pp. 1–49). Lincoln: University of Nebraska Press. Brunswik, E. (1955). The conceptual framework of psychology. In International encyclopedia of uniﬁed science (Vol. 1, Pt. 2, pp. 656–750). Chicago: University of Chicago Press. Burke, C. J. (1966). Linear models for Pavlovian conditioning. In D. Levine (Ed.), Nebraska symposium on motivation (Vol. 14, pp. 49–66). Lincoln: University of Nebraska Press. Gibson, J. J. (1979). The ecological approach to visual perception. Boston: Houghton Mifﬂin.

xliii Introduction Heider, F. (1959). On perception, event structure, and the psychological environment. Psychological Issues, 1(3), 1–123. ———. (1960). The Gestalt theory of motivation. In M. R. Jones (Ed.), Nebraska symposium on motivation (Vol. 8, pp. 145–172). Lincoln: University of Nebraska Press. Klein, G. S. (1959). A note to the reader. Psychological Issues, 1(3), v–vii. Koch, S. (1956). Behavior as “intrinsically” regulated: Work notes towards a pre-theory of phenomena called “motivational.” In M. R. Jones (Ed.), Nebraska symposium on motivation (Vol. 4, pp. 42–87). Lincoln: University of Nebraska Press. Koffka, K. (1925). The growth of the mind. New York: Harcourt, Brace. (Original work published 1921) Köhler, W. (1924). Die physischen Gestalten in Ruhe und im stationären Zustand: Eine naturphilosophische Untersuchung von Wolfgang Köhler. Erlangen: Philosophische Akademie. Köhler, W. (1938). The place of value in a world of facts. New York: Liveright. Krech, D. (1949). Notes toward a psychological theory. Journal of Personality, 18, 66–87. Krech, D. (1950a). Dynamic systems as open neurological systems. Psychological Review, 57, 345–361. Krech, D. (1950b). Dynamic systems, psychological ﬁelds, and hypothetical constructs. Psychological Review, 57, 283–290. Krech, D. (1951). Cognition and motivation in psychological theory. In W. Dennis et al. (Eds.), Current trends in psychological theory (pp. 111–139). Pittsburgh: University of Pittsburgh Press. Leeper, R. W. (1948). A motivational theory of motivation to replace “emotion as disorganized response.” Psychological Review, 55, 5–21. Leeper, R. W. (1965). Some needed developments in the motivational theory of emotions. In D. Levine (Ed.), Nebraska symposium on motivation (Vol. 13, pp. 25–122). Lincoln: University of Nebraska Press. Lewin, K. (1938). The conceptual representation and the measurement of psychological forces (Contributions to Psychological Theory, Vol. 1, No. 4, Serial No. 4). Durham nc: Duke University Press. Logan, F. (1964). Comments on Edward L. Walker’s paper. In D. Levine (Ed.), Nebraska symposium on motivation (Vol. 12, pp. 96–98). Lincoln: University of Nebraska Press. Newcomb, T. (1953). Motivation in social behavior. In M. Jones (Ed.), Current theory and research in motivation (Nebraska Symposium on Motivation, Vol. 1, pp. 139–161). Lincoln: University of Nebraska Press. Nissen, H. W. (1954). The nature of the drive as innate determinant of behavioral organization. In M. R. Jones (Ed.), Nebraska symposium on motivation (Vol. 2, pp. 281–321). Lincoln: University of Nebraska Press. Simon, H. A. (1994). The bottleneck of attention: Connecting thought with motivation. In W. D. Spaulding (Ed.), Integrative view of motivation, cognition, and emotion (Nebraska Symposium on Motivation, Vol. 41, 1–21). Lincoln: University of Nebraska Press.

xliv modeling complex systems Tolman, E. C. (1932). Purposive behavior in animals and men. New York: Century. Tolman, E. C. (1948). Cognitive maps in rats and men. Psychological Review, 55, 189–208. Vinacke, W. E. (1962). Motivation as a complex problem. In M. R. Jones (Ed.), Nebraska symposium on motivation (Vol. 10, pp. 1–46). Lincoln: University of Nebraska Press. Walker, E. L. (1964). Psychological complexity as a basis for a theory of motivation and choice. In D. Levine (Ed.), Nebraska symposium on motivation (Vol. 12, pp. 47–95). Lincoln: University of Nebraska Press. Wicker, A. W. (1984). An introduction to ecological psychology. Cambridge: Cambridge University Press. World Health Organization. (2001). International classiﬁcation of functioning, disability and health. Geneva.

Composition and Uses of Formal Clinical Cognitive Science Richard W. J. Neufeld University of Western Ontario

“In every special doctrine of nature only so much science proper can be found as there is [applied] mathematics in it.” So wrote Immanuel Kant about the status of natural philosophy, as he saw it in 1786 (Kant, 1970, p. 470). This declamation seemingly reﬂected a certain frustration not unlike that possibly experienced almost two centuries earlier by Francis Bacon, who decried “the distemper of learning,” which occurs “when men study words and not matter” (Bacon, 1937, 200 (bk. 1)). These quotes are extracted from commentaries on activity done in the name of science by their authors’ contemporaries— activity possibly characterized by considerable redundancy of effort to resolve the day’s purportedly key issues, contentious debate devoid of clear resolution, a lack of precise place keeping regarding the status of important segments of “the body of knowledge,” and an absence of salient direction for potentially proﬁtable future thrusts of investigation. More contemporary, and emanating from within our discipline, Paul Meehl (1978) bemoaned the slow progress of “soft psychology” (which presumably included psychological clinical science). He indicted the dominant research paradigm (Kuhn, 1962), of which a prominent modus operandi was, and for that matter remains, Fisherian- and Pearsonian-based statistical methods. Theoretical predic-

2 modeling complex systems tions were derided as somewhat anemic, typically embracing simply the presence of nonzero relations among ad hoc measured variables; as nature is said to abhor an absence of association, so-called support rested on statistical power, whereby the ever-popular anova and related tables of results were, in effect, summary statements of such power (cf. Cohen, 1988, pp. 16, 17; Steiger & Fouladi, 1997). Meehl noted that “theories,” at least in ﬁelds of soft psychology, typically fell by the wayside, not because they were toppled by uncorroborated risky predictions, but because the discipline became inured to them and their proponents grew weary of marketing them. It might, nevertheless, be added that their proponents need not worry because it is well-known that old theoretical ideas frequently reemerge in different guises and with modernized labels (Staddon, 1991). (Most readers will have their favorite accounts of such instances from their particular ﬁelds of expertise.) I would contend that the foregoing lamentation retains considerable currency when it comes to contemporary psychological clinical science. This observation provides much in the way of motivation for the developments presented here. It is maintained that the implementation of formal theory in psychological clinical science can go far to redress current quagmires and accelerate progress. Emphasis of course is on the ﬁeld with which I am most familiar, clinical cognitive science. However, the merits of formal theory disclosed in that domain of investigation are, arguably, general, providing additional incentive for the particular sphere of application. To begin, a clear delineation of what does and what does not constitute formal theory is presented. Beneﬁts and selected (practical) liabilities are described. Following on that is an exemplary case in point from the domain of clinical cognitive science. Novel developments include proposals for monitoring cognitive aspects of an individual’s response to treatment over time and, likewise, dynamic assessment of treatment-program efﬁcacy with respect to cognitive functioning. Some surprising noteworthy spin-offs from excursions into formal modeling, spin-offs that impinge directly on prominent issues in this area of study, are expounded on. One such by-product emanates from a personal “labor of love,” entailing the appropriation of stochastic modeling to the study of cognitive efﬁciency as it relates to schizophrenia, to psychological stress, and to their combination. It constitutes a new form of construct validity, using

3 Formal Clinical Cognitive Science mathematical heuristics. Another bears on an issue that I have engaged more or less since its appearance in the literature, namely, the so-called psychometric-artifact problem in the study of differential cognitive deﬁcit. The problem is translated into simple but comprehensive measure-theoretic terms, revealing the shortcomings of typical efforts at redress. From there formal theory is shown to recast the entire issue in its own terms and in so doing render it as more or less obsolete.

Demarcations of Formal Theory Described here are the features of formal theoretical systems that for all practical purposes distinguish them from nonformal and quasiformal systems. Drawn on is the rigorous, still-cogent taxonomy of logical-deductive systems in science provided by R. B. Braithwaite (1968). The most prominent type of theoretical system in psychological clinical science, Braithwaite’s mixed deductive system, can profitably be used as a point of departure in explicating the nature and role of formal theoretical systems. Mixed deductive systems consist of formal and nonformal subsystems. A survey of contemporary issues of American Psychological Association clinical-science journals (e.g., the Journal of Abnormal Psychology, the Journal of Consulting and Clinical Psychology), and any number of psychiatry research journals, reveals the following predominant structure. Within the nonformal subsystem, predictions of empirical observations are dictated by syllogistic verbal reasoning, possibly accompanied by visual aids and most certainly by antedating empirical ﬁndings. There is, however, no particular governance by a set of rules constraining the format by which the verbal arguments are posed or the way in which one statement follows from another (e.g., mathematical derivation, computer language, or symbolic logic). The sequence of qualitative arguments culminates on transition to prediction(s) of results, or at least guiding question(s), credited with motivating the investigation at hand (further elaboration of informal and other systems, in selected domains of applied psychological science, is presented in Neufeld, 1989). Summarily speaking, the study is then executed and results obtained. It is at this point that the formal subsystem makes its entry. Mea-

4 modeling complex systems sures of dependent variables, guided, if not “prescribed” as such, by the nonformal theoretical apparatus (see, e.g., McFall & Townsend, 1998; Meehl, 1978), provide the data array submitted to statistical analysis, expedited by associated software. Statistical treatment is part and parcel of a formal subsystem because its constituent operations rest on closed-form formulas that, by and large, emerge from theorem-proof continuity (nowadays increasingly accompanied by numerical simulation). Note that so-called qualitative research methodology replaces this formal data-analytic arm of the logical-deductive system with nonformal, verbal operations. In this way, the entire logical-deductive enterprise is rendered informal. Conversely, formal theoretical approaches replace nonformal reasoning in the above mixed deductive design with a formal logical-deductive subsystem, thus creating an entirely formal logicaldeductive structure. Whereas the typical statistical apparatus (often elegant in its own right and with a virtual industry of software support) is substantively generic—its application transcending theoretical-content domains—formal theoretical models are content speciﬁc. For example, Doob (1953) has described one such type of theoretical formulation, a stochastic model, as a “mathematical abstraction of an empirical process, whose development is governed by probabilistic laws” (p. v). Whereas statistical computations constitute mathematical methods for assessing arrays of values rendered by empirical investigations, formal theoretical models are concerned with processes that make the data what they are (although, as will be demonstrated below, formal theories are far from mute about ways of analyzing the data whose genesis they address). In other words, in the explanatory chain of events, such models weigh in with quantitative constraints ab initio. The nature of authentically formal theory can be thrown into sharper relief by considering procedures that may appear to qualify as the former, pending closer scrutiny. Integrative and evaluative reviews, for example, potentially useful in their own right to be sure, are a case in point. Apart from a rare application of synthesizing quantitative models of possible agents of surveyed outcomes, verbal schemata are, as a rule, restatements of what is already known (e.g., McFall, Townsend, & Viken, 1995). Formal theory can also be distinguished from prominent higherlevel statistical methods, including “structural-equation causal mod-

5 Formal Clinical Cognitive Science eling” (sem; e.g., Kline, 1988; Tomarken & Baker, 2003; cf. MacCallum & Ashby, 1986), taxometric procedures (Meehl & Golden, 1982; Ruscio & Ruscio, 2004; Waller & Meehl, 1998), and hierarchical linear modeling (Bryk & Raudenbush, 1992; see Piasecki, Jorenby, Smith, Fiore, & Baker, 2003). With respect to sem, the subject of prediction comprises its parameters,1 for all intents and purposes meaning the path coefﬁcients of conjectured relations among the posited latent variables. (It is surmised that criterion values for contemporary indexes of overall statistical ﬁt, a sort of overarching prediction, have been met.) Note that, ordinarily, the hypothesized structure of interlatent-variable linkage and the relative magnitudes and/or signs of its associated path coefﬁcients themselves are posited nonformally. In a similar vein, conﬁrmatory factor analysis forecasts factor loadings and coefﬁcients expressing interfactor relations, including those involving higher-order factors. Concern is with an empirical covariance structure, but one whose features are deﬁned according to antedating empirical observations, or nonformal argumentation, rather than derivations spawned by speciﬁc axioms, deﬁnitions, and assumptions. The chief subject of prediction for taxometric methods, in turn, is certain conﬁgurations of data summaries (calculated via methods prominently known as Maxcov, Mambac, Maxslope, Maxeig, and L-mode) indicative of dichotomous versus continuous latent distributions of target variables (e.g., speciﬁc symptoms or syndromes). Selected parameters, such as base rates of a taxon (a discrete group with the symptom, genetic endowment, or other target characteristic), can be estimated. Once more, however, the data conﬁguration corresponding to the conjectured distribution is not in and of itself the product of formal-system reasoning. By these characteristics, taxometric methods and sem fall into the category of statistical analysis and, in that way, constitute the formal subsystem of a mixed deductive system (elaborated on in Neufeld, 1989). Sufﬁce it to say that similar observations attend typical applications of hierarchical linear modeling. Formal theoretical systems as well entail parameter estimation and testing for goodness of ﬁt between predictions and observations. Their parameters, however, are substantively signiﬁcant, as imbued by their explanatory roles in the theoretical model in which they participate (Braithwaite, 1968; exempliﬁed below). Moreover, their

6 modeling complex systems variation generates selective changes in empirical data sui generis to the localized problem (e.g., Link, 1982; Townsend & Wenger, 2004a, 2004b). This treatment contrasts an essentially generic interpretation of parameter values, or other model properties, across even vastly divergent content areas and a focus on highly similar aspects of data (e.g., empirical covariance structures). I will return momentarily to the nonformal system widely known as qualitative research methodology because of its growing popularity, especially in the social sciences. Qualitative research increasingly has been touted as a legitimate, even superior, alternative to systems of science implanted with formal methods. Nonformal strategies give the appearance of ﬂexibility in dealing with concepts and measures whose implementation may strain formal systems. The former, however, accommodate such challenges handily, not because they somehow transcend the explanatory merits of formal systems, but because they simply sidestep the constraints on precision of expression demanded by formal systems or subsystems (Staddon, 1984). As such, despite their purported checks and balances (e.g., Miles & Huberman, 1994), they are vulnerable to the onslaught of frailties in reasoning and inefﬁciencies accompanying dependence on essentially verbal contentions, defects that formal procedures are designed to confront (Hintzman, 1991; Kline, 1985). Indeed, the history of science makes it clear that systematic retreat from decidedly formal inferential methods inevitably detracts from progress (e.g., Braithwaite, 1968; Kline, 1985).

Assets of Formal Theory explanatory insights A major advantage bequeathed by formal developments is the unique insight afforded into the studied phenomena. Interplay of critical variables is disclosed by inspecting the structure of derived predictions and the steps to obtaining them. Examples in the hard sciences of course abound. Consider, for instance, a notable case in point, Newton’s universal law of gravity: m1m2/r2. The law deﬁnes the force of attraction between two particles, with masses m1 and m2, separated by distance r. The constant varies only with the units of

7 Formal Clinical Cognitive Science measurement. The actual physical mechanism of gravitational pull may be enigmatic, but the mathematical deﬁnition is not. Moreover, to be taken seriously, any proposed physical mechanism would have to comply with this mathematical statement. The function of each term, and of the law itself, in turn, is decipherable from the nature of the formal assertion and from the way in which the law serves in broader theoretical contexts. Similarly, something of the nature and function of Einstein’s limiting constant, the speed of light, can be appreciated contextually by considering its operation in a formal deductive system. Such exposition, for example, discloses that any meaningfulness of a larger value rests on a solution to the imaginary number (1)½. In psychological clinical science, elucidation of obtained data conﬁgurations can be bolstered by stipulating relations between observed and inferred variables in terms analogous to the above. As opposed to ad hoc interpretation, statistical interactions, for one, can be understood from formal depiction of the psychological processes under study and associated scales of measurement (e.g., Busemeyer, 1980). An illustrative instance entails the stochastic dynamic modeling (see, e.g., McFall et al., 1995) of the effects of stress (ambient noise) on cognitive efﬁciency (Weinstein, 1974). One such model expresses the expected (mean) latency for proofreading lines of prose (e.g., Glass & Singer, 1972). The expression is mr/(k 1) (Neufeld, 1996, 2002). In this formula, r conveys stress-related impairment of performance speed (r goes up with increasing stress, elevating the mean, reﬂecting diminished speed), k is an increasing function of task-relevant competence (associated here with practice), and m indicates task load, quantiﬁed as the mean number of processed stimulus elements (chunks of letters scanned per line of prose). The expected latency for a given level of stress, and amount of practice, for the summary task load m can, thus, be predicted as a function of substantively signiﬁcant parameters. The expression also discloses the fabric of interplay between stress and practice, making for their “interaction”: the effects of stress can be mitigated by practice, practice becomes more important as stress goes up, and the effects of this interplay on performance latency are “scaled” according to the prevailing task load m. In fact, the form of the stress-practice interaction is one of “superadditivity” of the interaction’s second-order difference: [(mean latency under higher stress and lower practice) (mean latency under

8 modeling complex systems lower stress and lower practice)] [(mean latency under higher stress higher practice) (mean latency under lower stress and higher practice)] > 0 (Townsend, 1984). Note that each of the above parameters is also entrenched in its own quantitative infrastructure, endowing it, accordingly, with further analytic and substantive meaning (Evans, Hastings, & Peacock, 2000; elaborated on below). Beyond their revelatory function, closed-form, analytic expressions can also serve the practical cause of computational economy. It is enticing to depend on numerical simulations as surrogates for analytic derivations, given contemporary computational technology. Analytic derivations, however, not only enhance understanding, according to their detailed structures, but can also reduce computational steps, ease computer memory demands, and increase speed when simulations are inevitable. Partly for this reason, theorists in longer-established sciences will opt to take analytic solutions to the limit, before invoking numerical algorithms.2

self-diagnostic, self-indicting properties Desiderata of useful theory are Popperian bold conjecture and falsiﬁability. In Bayesian terms, bold conjecture is tantamount to a predicted observation being negligible, apart from viability of the proposed theory. In like fashion, distinct falsiﬁability is identiﬁed with an essentially zero probability of the tendered theory being defensible on failure of its speciﬁed prediction. It is readily shown that, in the ﬁrst instance, the probability of the theory being tenable, given that the predicted evidence occurs, approaches 1.0 and, in the second instance, the probability of the evidence, given the theory, also approaches 1.0 (Neufeld, Carter, Boksman, Jetté, & Vollick, 2002). Predictions emanating from strong theory stand to be formidably precise. A weak meteorological forecast, for example, predicts rain in the Northwest during November. At the opposite extreme, a strong prediction stipulates the precise amount to fall within a given time period in a speciﬁc location (Meehl, 1978). Formal theory arguably makes for desirably strong prediction, including that located in psychological clinical science (Neufeld et al., 2002). This is why formal theories die while nonformal theories seemingly defy all laws of natural selection (Staddon, 1991).3

9 Formal Clinical Cognitive Science Formal theories provide speciﬁc predictions that can compete directly in terms of differential conformity to observed values. Predictions set in quantitative terms, in turn, expedite quantitative expressions of their relative superiority (e.g., tests for goodness of ﬁt and comparative goodness of ﬁt). In addition, however, models can prescribe their own qualitative signatures of validity. The latter entail the contour of data across gradations of independent variables and their combinations (e.g., superadditivity of means, as discussed above; Neufeld & Williamson, 1996). Such patterns can, moreover, be mutually exclusive, in that, even with very weak assumptions, competing models cannot produce each other’s conﬁgurations (e.g., Townsend & Wenger, 2004a; cf. “empirical equivalence,” e.g., Townsend, 1990). If formal models mandate their own diagnostics, they necessarily ease the problem of empirical measurement. In fact, formal deductive systems stand the selection of measures on its head. As with longer-established sciences (e.g., physics examples, as discussed above), theory is potentially strong enough to mandate certain empirical ramiﬁcations. Displaced, therefore, is the often ad hoc process of selecting off-the-shelf measures, conjecturally connected to dependent variables (McFall et al., 1995; Meehl, 1978). Owing to the precision of the formal deductive system, theory and measurement become intertwined. Assessment of the generated measures, for the usual statistical properties of fallible estimates (e.g., bias, maximumlikelihood status), to be sure remains applicable (e.g., Riefer & Batchelder, 1988; Townsend & Wenger, 2004a). Nevertheless, those properties apply to indexes substantively tied to the theory at hand. Individual differences in model expression can be accommodated within the context of model diagnostics, especially beﬁtting clinical science and assessment. Included is variation in the structure of cognitive transactions (e.g., serial processing, variations on parallel processing, and selected combinations of these structures; Townsend & Fiﬁc, 2004). Variation in model properties, notably parameter values, across individuals can also be proﬁtably accommodated through Bayesian techniques (Batchelder, 1998; Chechile, 1998; Neufeld et al., 2002). Not least among the endowments of formal theory is its inherent challenge to preferred interpretations of addressed phenomena. Explicitness of expression helps make salient current boundaries of

10 modeling complex systems knowledge—ﬂaws and limitations in extant formulations. In particular, empirical equivalence, on current measures, of nonetheless opposing positions is disclosed. However, so too are alternatives that can break the encountered experimental-mimicking logjams (Casti, 1989; Townsend, 1990).

absorption of theory-exogenous variance With formal logical-deductive systems, random empirical variance can be explained by the theory itself. Quantitative theory, in other words, encompasses apparent observational indeterminacy, instead of relegating it to “error” or “noise variance” (cf. anova linear model, e.g., Kirk, 1994; “local independence,” Bollen, 2002). An additional classiﬁcation of theoretical systems, cogent apropos this avowed asset, is that of Busemeyer and Townsend (1993; depicted in Table 1). Systems are categorized as dynamic versus static and as stochastic versus deterministic. First, static stochastic models (in the lower-left quadrant) predict frequencies of response values (e.g., categories, ratings, and so on), if not the times for producing them. An example is Tversky’s elimination-by-aspects model of preference and choice (Tversky, 1972a, 1972b; see Batsell, Polking, Cramer, & Miller, 2003). Comparative frequencies of choosing items from those available are considered to follow a multinomial distribution (e.g., Evans et al., 2000). Accordingly, probabilities of selection are the main focus. The probabilities are governed by parameters of the elimination-by-aspects model, which are subjective utilities of the item attributes (e.g., the likelihood of winning a lottery vs. its payoff; Rappaport, 1983; for applications in stress-and-coping research, cf. Kukde & Neufeld, 1994; Morrison, Neufeld, & Lefebvre, 1988). Note that an earmark of stochastic models is their prediction of probabilities, or distributional frequencies of events, versus predictions of an all-or-none nature. Turning to the upper-right quadrant, we see that deterministic dynamic theory is exempliﬁed most notably by nonlinear dynamic systems models, popularly known as chaos theory (e.g., Gregson & Pressing, 2000; Koçak, 1989). To illustrate, variables such as psychological stress and coping can be depicted as being in constant ﬂux, reciprocally inﬂuencing and being inﬂuenced by one another. The

11 Formal Clinical Cognitive Science Table 1. Schema of Theoretical Systems, with Selected Examples

Deterministic

Stochastic

Static

Dynamic

Verbal theory; Nondynamic ﬂow diagrams

Deterministic nonlinear dynamic systems (“chaos-theoretic systems”)

Some decision-and-choice models (e.g., selected subjective-utility models)

Dynamic extensions of decision-and-choice models; classic stochastic models; nonlinear dynamic-systems models, with a stochastic element

Note: Adapted from Busemeyer & Townsend (1993).

system is represented by differential equations, each indicating the momentary change in a variable. The changes themselves are deﬁned in these equations essentially by the variable’s own extant state and those of the other variables making up the network (Neufeld, 1999b). Stochastic dynamic theories are exempliﬁed by time-related extensions of decision-and-choice models. The elimination-by-aspects model (discussed above), for example, has been reformulated by Marley and his colleagues (e.g., Marley, 1989; Marley & Colonius, 1992) so as to add temporal predictions about selection responses to the original model’s predicted frequencies of selection from among the competing items. Busemeyer and Townsend (1993; also Roe, Busemeyer, & Townsend, 2001) have presented an empirically compelling model of decision and choice, providing for speed of choice, type of choice, and their interconnections. Variability of expressed preferences and their latencies are, once again, accommodated, in that the crux of prediction involves frequency distributions. Stochastic dynamic models represent possibly the most widely used version of formal theory in psychology. Generally speaking, they depict probability distributions of events as a function of time (see, e.g., Johnson, Kotz, & Balakrishnan, 1994, 1995; Luce, 1986; Townsend & Ashby, 1983). Part of the model fabric, then, is variability of event latencies across occasions of observation (e.g., experimental trials). The provision for variability is graphically illustrated by considering the model-prescribed variance of time taken to complete a relatively simple cognitive process.

12 modeling complex systems The process comprises k stages, each stage having a speciﬁed frequency distribution, as a function of time. Each stage is executed at the rate of v stages per unit of time (whereby the mean stagewise completion is 1/v time units). Such a process may correspond to the encoding of k stimulus features (e.g., curves, lines, and intersections of an alphanumeric item) into a cognitive format facilitating further processing, such as selected manipulations involving the item in short-term memory (e.g., Sternberg, 1975). By the present model, known as the Erlang distribution (Evans et al., 2000), the mean time across trials for process completion is k/v. The across-trial variance of completion, in turn, is k/v2. Note that this statement of dispersion is squarely embedded in the substantively signiﬁcant parameters of the theoretical model of process completion. Further provision for stochastic properties is available by allowing for individual differences in the values of the parameters k and v. This extension (a mixture model) “acknowledges” interparticipant variation in the above variance and related model expressions (e.g., Neufeld et al., 2002). Latency variance for a given value of k and v maps onto the familiar within-participant variance in (m)anova, and variation across participants in k and v maps onto between-participant variance (see, e.g., Parzen, 1962, p. 200; also “A Stochastic-Modeling Translation of the Statistical-Property Issue” below). Other contemporary exemplars of dynamic stochastic models include selected extensions of multidimensional scaling (see, e.g., Schiffman, Reynolds, & Young, 1981). These extensions quantitatively relate categorizations, or ratings of multidimensional stimuli, to their cognitive-process underpinnings, where the variable of time plays a central role (e.g., Carter & Neufeld, 1999, in press; Nosofsky, 1992; Takane & Sergent, 1983). Before leaving this subsection, it should be noted that nonlinear systems theory can, in principle, be augmented to incorporate a stochastic component. A speciﬁcally random element can be added to the otherwise deterministic differential equations describing intervariable inﬂuences across time (e.g., Brown & Holmes, 2001; Huber, Braun, & Krieg, 1999). This interlaced stochastic thread perturbs the momentary change in each variable and the consequent inﬂuence of those changes on the coupled network of variables. Interestingly, the temporal trajectories of such a “stochasticized” system can deviate dramatically from those of its strictly deterministic counterpart.

13 Formal Clinical Cognitive Science Sufﬁce it to say that, overall, unaccounted-for variance in observed values may never be eliminated (cf. Gilden, 2001). However, such “model-exogenous variance” stands to be substantially diminished with increasing comprehensiveness of theory construction.

aesthetic appeal P. A. M. Dirac, a Nobel laureate in physics, once declared: “It is more important for our equations to be beautiful than to have them ﬁt the experiment” (quoted in Freedman, 1993). Given the historical success of beautiful equations, their apparent failure was held to be a temporary aberration. Perhaps, in their presented form, the equations were an inadequate approximation of empirical events. Or, possibly, the experimental observations themselves were ﬂawed. Either of these shortcomings could be redressed. Beautiful equations were intrinsically self-vindicating, and the rest was detail. At the very least, if equations were to stand up to empirical challenge, they must spring forth from among the ranks of the beautiful, whereas equations that are not beautiful are inevitably doomed to experimental defeat (cf. Smolin, 2006). Formal theory thus potentiates the scientiﬁc merit of equational elegance. There is no inherent reason why psychological clinical science should be barred from this asset, which is, arguably, a proprietorship of formal developments. What, then, makes an equation aesthetically appealing? One criterion is succinctness. The simplicity of beautiful formal expressions belies their profundity, which becomes apparent when their copious and far-reaching implications are unveiled (Polkinghorne, 2003). One example, emanating from the ﬁeld of nonlinear dynamics, is Mandelbrot’s formal representation of fractals. Although the representation is parsimonious, a feature of its constituent variables is that they take on a multitude of intriguing trajectories. The proliferation of exotic patterns has been widely disseminated, thanks to modern computer-graphics technology (e.g., Gleick, 1988; Townsend, 1994). A more modest example is available from the Erlang distribution of cognitive-process completion (described above). Here, the probability of completing the stated process by a given time interval, t, is

14 modeling complex systems (k’ – 1)! – (k’, vt). (k’ – 1)!

(1)

Note that one can safely bypass the deﬁnition of each term in the equation and its equivalents (below) and still appreciate the pithiness of this expression. The somewhat involved operations, summarily contained above, include j 1 – vt x k’–1 e –x dx/(k’ – 1)! = j=k’ (vt)/j! e–vt. 00

00

(2)

These formulas, in turn, can be used to create visual aids that facilitate insight into the interplay of model parameters (for a treatise on theory-related intuitions afforded by visual imagery, see Clark & Paivio, 1989). Three-dimensional response surfaces convey variation in the present probability across alternate combinations of k, v, and the time window, t. Two examples are provided in Figure 1, for values of k ranging from 1 to 10 and values of v ranging from 0.01 to 3.0. The lower response surface is constructed for t = 4 and the upper one for t = 7. This ﬁgure presents but two instances of a virtually inﬁnite number of possible graphic portrayals of the formally stated process. This is why, if a picture is worth a thousand words, a formula is worth a thousand pictures.

liberating qualities Formal logical-deductive systems are constrained by observationally and quantitatively principled initial assumptions, deﬁnitions, and axioms.4 Building on this foundation, rigorous derivations render deductions about events of principal interest. In this way, precisely framed empirical implications can, in effect, be evaluated for their potential importance, in advance of actual empirical testing. Fruits of these labors, then, include, not only the resulting insights, but also a potential increase in the overall economy of inquiry. Important questions can be posed before launching out empirically. Will the forecasted yield of information justify the associated investment? Is that which may be found actually worth ﬁnding? Does the theory posit results that, if corroborated, could signiﬁcantly affect explanation, measurement, and, indeed, future predictions? The point to be made is that the theorist is emancipated to safely

15 Formal Clinical Cognitive Science

Figure 1. Probability of completing k stages in four time units (lower surface) and seven time units (upper surface) with each stage transacted at rate v (see Equation [2]; k ranges from 1 to 10, and v ranges from .01 to 3.0.

wade into the staked out substantive territory, with the above and related issues in hand, because her deliberations are constrained mathematically. These constraints represent an antidote to the hazards of armchair philosophy and psychology’s ensuing reaction against exploration free of constant empirical vigilance (cf. instrumentalist vs. realist distinctions in scientiﬁc theorizing; e.g., Casti, 1989). As for the empirical enterprise, formal theory can increase the latitude of investigation in clinical science. It can do so by incorporating abnormalities associated with psychopathology into hypothesis formation and testing. Such possibilities are exempliﬁed in the cognitive-science application presented below. In the enterprise of clinical science, certain hypotheses regarding pathognomonic disturbances may defy often-used methodological options, and formal theory can supply essential alternatives. To elaborate, one tack to take when testing hypotheses about cognitive debilities, for example, involves attempting to duplicate performance deviations in normals using experimental manipulations that are designed to mimic their etiology in patients. Formal theory can complement this strategy and, moreover, pick up where it leaves off. The approach is to modify quantitative mod-

16 modeling complex systems els of performance in ways implied by the hypothesized agent of deviation and then to evaluate the accuracy of the resulting predictions (elaborated on below). Such methodological endowments come into play especially where tendered sources of disturbance are beyond the reach of direct manipulations. Selected agents of dysfunction, although model hospitable, may be fundamentally alien to normal functioning or demand unethical extremes of experimental induction.

barriers: attitudes, aptitudes, and investigator role Ideally, formal theory at minimum should be a ready-to-hand arrow in an investigator’s hypothesis-development quiver. It is recognized, however, that there are certain practical barriers to this ideal. One such barrier quite simply involves prevailing attitudes toward formal theory. Mathematical models are regarded with suspicion in some circles unless their implications are accompanied by readily appreciated parallels. In clinical science, such parallels often take the form of behavioral neurological mechanisms that clearly align with the mathematical developments.5 Interestingly, this position is counter to that of longer-established disciplines, where accepted mathematical necessity predates identiﬁcation of a corresponding physical mechanism (e.g., Newton’s universal law of gravity, discussed above). One price paid for suspended belief in quantitative formulations is the forfeiture of potential uniﬁcation of coexisting explanatory systems and new angles on synthesizing current evidence (e.g., Herrnstein, 1979). Another practical impediment concerns formal theory’s demands for hands-on tasking by someone closely involved with the subject matter. This requirement contrasts with practices where assignments can be delegated to an associate or a research assistant (as might be done, e.g., in data entry or statistical analysis). Formal modeling, furthermore, being customized to the immediate theoretical problem, is not a matter of output from routinely applied software. This is not to say, of course, that advances in computational power and computer-algebra programs such as maple and mathematica do not substantially aid the theorist’s job.

17 Formal Clinical Cognitive Science On balance, it can be said that such hurdles to formal theorizing may be daunting at ﬁrst blush. Current practices, however, need not be abandoned while cultivating the formal alternative as an increasingly friendly ally. Eliminating the present hurdles can reap the substantial dividends on outlay in time and labor that have been enumerated above.

Application to Clinical Cognitive Science: The Case of Stimulus-Encoding Dysfunction in Schizophrenia This section leads off with a presentation of the philosophical blueprint guiding the present application. Its exposition culminates in the appropriation of group-level inferences about cognitive functioning to the individual participant. This appropriation is mediated by the use of Bayesian-based procedures. Their promised value launches an exposition of the parent Bayesian framework dominating the current development. Positioning of stochastic models of cognitive performance within this framework is pinpointed. Aided by the present instantiation, entailing selected stimulusencoding dysfunction in schizophrenia, potential payoffs of the approach are expounded. They include (a) precision in estimating performance-model parameters for the individual; (b) customization of the process model’s performance-latency distribution to the individual; (c) monitoring changes in cognitive functioning with the progression of treatment, at both the individual and the group levels; (d) importing the informational infrastructure endowed by the model’s cognitive-science roots; (e) complementing neuroanatomical regions of interest with times of interest, thus contributing to contemporary technology for measuring evoked neurophysiological responses; and (f) facilitating links of cognitive-performance indexes with clinically signiﬁcant quantitative formulations on stress and coping.

philosophical orientation Associations among cognitive science, clinical cognitive science, and applications to individuals for clinical assessment or other purposes

18 modeling complex systems

Figure 2. Transitions from basic cognitive science among normals to cognitive assessment among patients.

are depicted in Figure 2. Its arrows denote a smooth, if not seamless, transition among these investigative domains. In the ﬁrst instance, a model of cognitive performance addressed to the targeted territory of mentation is chosen. Deliberations include viability of the experimental paradigm and relevance of its response parameters. Addressed cognitive faculties, for example, may be those identiﬁed with visual, memory, and collateral processes (e.g., stimulus encoding and response mechanisms); the paradigm may take the form of a conventional visual- or memory-search task (Sternberg, 1975; Townsend & Ashby, 1983); and response parameters may include performance latency and/or accuracy.6 Several candidate models may present themselves. Selection from among contenders is based on congruence between the model’s features (e.g., its parameters or other properties) and concepts indigenous to the substantive research question (e.g., cognitive capacity, processing architecture [serial, parallel, or hybrid], efﬁciency of capacity allocation, and so on). Extant models may require elaboration or modiﬁcation, beﬁtting the currently driving issues. It may even be necessary to develop a model ab initio (see “Barriers: Attitudes, Aptitudes, and Investigator Role” above), depending on points of contact between the available armamentarium and the clinical-science queries that are pressing. The formal account of cognitive processing is then adjusted so as to accommodate performance deviations of disordered individuals. Alterations can be directed toward the architectural structure of the modeled processing system and/or toward model parameters.

19 Formal Clinical Cognitive Science Architectural structure refers to the arrangement of constituent processes involved in task transaction. To illustrate, in ascertaining and registering the existence of a presented alphanumeric item in a memorized set of items, component processes composed of encoding, memory-search, and response operations may be conducted serially, in parallel, or in some combination of the two. Structure refers as well to the arrangement of constituent stages (subprocesses) of the said processes (e.g., in the case of encoding subprocesses, parallel vs. serial input of the physical item features for template matching to those of the memorized item set; Townsend, 1984; Townsend & Ashby, 1983). Parameters refers to variables characterizing the processes, such as their size (operationalized as the number of subprocesses making up the process) and capacity (operationalized as the speed with which these subprocesses are transacted)—the parameters denoted above as k’ and v, respectively. For the sake of parsimony, the normal model is minimally perturbed, pending realization of the target conﬁguration. Necessitated adjustments purportedly indicate disorder-affected functions, and those remaining intact signify functions that are spared. In general, processing architecture tends be preserved; however, the same cannot be said for the values of model parameters (Neufeld et al., 2002; Neufeld, Vollick, et al., in press; Townsend, Fiﬁc, & Neufeld, in press). It is apparent from the foregoing that the models most closely aligned with the present purposes are those that are stochastic and dynamic (see “Absorption of Theory-Exogenous Variance” above). By certain extensions of their stochastic element, beyond that attending response parameters per se, these models become poised to accommodate unique aspects of individual performance. Speciﬁcally, mixture models are fashioned to incorporate unequal values of performance-model parameters across units of observation—in this case, members of diagnostic groups (Batchelder, 1998; Berger, 1985; cf. Busemeyer & Stout, 2002). The distribution of individual parameter values becomes, in effect, a “prior distribution,” in Bayesian statistical terms. Through the implementation of Bayes’s theorem (discussed below), performance-model parameter estimates, as well as performance-latency distributions, stand to be individualized with considerable precision. Analogous to the procedure followed in a medical-laboratory scenario, where a modest blood specimen is

20 modeling complex systems referred to the broader body of hematological knowledge, a modest cognitive-performance sample is now embellished with information conferred by the prior distribution of parameter values emanating from the participant’s group. Note that, with the introduction of Bayesian-prior distributions of parameter values, deviation-accommodating adjustments no longer take place directly at the level of the task-performance process model. In the case of mixture models, there is, instead, movement of the deviant group’s Bayesian-prior distribution of parameter values. In this way, effects of disorder are registered as features of the prior, and, inasmuch as what has been altered is a distribution, provision for individual differences in the expression of these effects is retained. The arrows in Figure 2 are unidirectional, conveying the thrust of the current argument. There is, nevertheless, reciprocity among the above investigative domains. Successful capturing of deviant performance among clinical samples speaks to model robustness; those that readily accommodate performance extremes are preferred to those that are strained or fail in this regard (Neufeld, 1982; Neufeld & Mothersill, 1980). This widening of application represents a unique opportunity for generalization testing (Busemeyer & Wang, 2000) and, thus, potentially reciprocates beneﬁts back to basic cognitive science. Nor are the layers of investigation mutually exclusive (see the leftmost column of Figure 2). Successful titration of stochastic models can lead to parallel manipulations at the connectionist level of analysis, whose own success or failure adjudges theory robustness (Carter & Neufeld, in press; Roe et al., 2001). Neurophysiological levels of investigation also stand to beneﬁt. As described above, performance-model parameters (e.g., k, v) can be customized to individuals by amalgamating each one’s performance sample with the parameter-distribution prior, ascertained for the associated diagnostic group. With the individual estimates in hand, parametrically homogeneous subgroups can be formed. The resultant speciﬁcation of cognitive-process-latency distributions (e.g., that of cognitively encoding a presented item into a task-facilitative format) can, in principle, be used to establish the likely epoch of occurrence of the process for the constructed group (e.g., Neufeld et al., 2002, elaborated on below). When it comes to measuring evoked neurophysiological responses (as in fmri and erp), estimated “time

21 Formal Clinical Cognitive Science windows of interest” may, thus, be prescribed, complementing neuroanatomical regions of interest, in calibrating the space-time coordinates of measurement.

group and individual differences in the encoding process: bayes’s theorem in the current context Recall that Bayes’s theorem states: Pr(A|B) = [Pr(A)Pr(B|A)]/Pr(B),

(3)

where A and B are events; Pr(A) is the prior probability of A, or the probability of A preceding consideration of B; Pr(B|A) is the conditional probability of B, given A; and Pr(B) is the unconditional probability of B. As in the present context, A is often a hypothesized event, and B refers to experimental observation. This statement appears simple enough, until its wide-reaching ramiﬁcations are unraveled. The parsimony of expression that, nevertheless, belies its profundity obviously qualiﬁes this theorem as one of statistics’s aesthetically exquisite equations. Perhaps a more familiar general expression of Bayes’s theorem is Pr(Ai|B) = [Pr(Ai)Pr(B|Ai)]/[iPr(Ai)Pr(B|Ai)],

(4)

where Ai is one of a set of hypothesized events, Ai. In its present implementation, Ai’ becomes a parameter value of a test-performance process model, and B becomes a participant’s performance sample comprising N empirical latencies {t1, t2, . . . , tN},7 or {*} for short. The latencies refer to stimulus encoding because cognitive translation of presented items into a format that facilitates collateral functions (e.g., search for a matching memory-held item) tends to be elongated in schizophrenia (Neufeld et al., 2002). The following formula adapts these equations to the present setting of application and, in so doing, provides a comprehensive backdrop for explicating the potential assets enumerated at the beginning of this section: w(|{*}) =

G

({Pr(g|{*})}[w(|{*}, g)]). g=1

(5)

22 modeling complex systems In this equation, [w( |{*}) is the probability, or probability density (in the case of discrete and continuous parameters, respectively), of the parameter value , given the set of empirical latencies {*}; Pr(g|{*}) is the probability that the participant belongs to diagnostic group g, given {*}; and w( |{*}, g) is the probability (density) of , given {*} as well as the prior distribution of for group g. As stated above, it is through variation in the prior distributions of that group-related performance deviations become registered in Equation (5). It is the presence of cognition-relevant parameters

in this equation, moreover, that takes Equations (3) and (4) beyond a theoretically barren actuarial status to substantive signiﬁcance. The precise role played by the group-speciﬁc priors in Equation (5) is pinpointed in its expansion (presented in Appendix A). Note that the equation is partitioned by elongated braces { } and elongated square brackets [ ] according to the respective terms bearing on the main developments (outlined below).

experimental paradigm and process of interest What then corresponds to in the present application? The answer requires consideration of the cognitive task on which the model parameters bear and whose performance supplies the latency sample {t1, t2, . . . , tN}. A variant of a memory-search paradigm (Sternberg, 1975) eliciting performance deviations among schizophrenic patients is described (Highgate-Maynard & Neufeld, 1986). The latter are representative of deviations occurring with other memory-search and related paradigms. Encoding of presented items is identiﬁed as the source of observed abnormalities in performance latency, which composes the subject of prediction. Pertinent developments in quantitative cognitive science are exerted to discern responsible parameters. The theoretical design of the encoding process is delineated. Its model parameters are deemed to vary randomly across participants, with their distribution being perturbed according to group membership. The several potential beneﬁts of this Bayesian layout, enumerated above, are then detailed. The study reported in Highgate-Maynard and Neufeld (1986) extended typical memory-search methodology by incorporating meth-

23 Formal Clinical Cognitive Science ods emanating from Paivio’s dual-coding theory (Paivio, 1971, 1986). Individuals with schizophrenia and controls indicated “as quickly and accurately as possible” whether or not the real-life size—“overall volume”—of a presented item (the probe item was either an object or an animal) was similar to that of a member of a previously memorized set of items (the memory set; “ﬁxed-set procedure”). Memory-set size, ranging from 1 through 4, varied randomly across trials. Presence versus absence of the probe’s size properties in a member of the memory set (positive vs. negative trials) was similarly scrambled. Half the participants in each group were presented with similarly sized drawings of probe items, whereas the other half were presented with names. Encoding demands were, accordingly, deemed to be greater in the latter instance because the initial verbal presentation of the probe would require referral to the imagery system for access to the demanded spatial-size properties (Paivio, 1986; for an examination of dual-coding and related predictions among schizophrenia participants, see George & Neufeld, 1984). Scanning memory-held items for the presence of the probe’s size properties was considered to be comparatively taxing. It required comparison of the probe and memory-held items’ overall volume, as set against subjective criteria of similarity in this property (Hockley & Murdock, 1987; Wright, 1977). Size attributes of items in the memory set were spaced such that the means of their normative-size ratings (Paivio, 1975) were at least 2 standard deviations of their Thurstonian discriminal-difference size dispersions apart (practically, at least 2 standard deviations of the distribution of difference scores between their normative-size ratings; Highgate-Maynard & Neufeld, 1986). Responses were designated as correct or incorrect on the basis of whether or not they conformed to the following criteria for positive and negative trials. Negative trials, whose correct response was a “no” button press, meant that the probe item’s size did not resemble that of a memory-set item. Here, the probe item’s mean real-life size rating was at least 1 standard deviation of the Thurstonian discriminal-difference dispersion from the normative mean rating of each item in the memory set. Positive trials, whose correct response was a “yes” button press, were those where the mean of the probe item, and that of a memory-set member, were, for all intents and purposes, identical. Practice trials, ensuring familiarity with task requirements and the nature of correct responding, were similar in number for each group.

24 modeling complex systems Further speciﬁcs, including those surrounding provision for potentially confounding clinical and demographic variables and ascertainment of the viability of the paradigm for each diagnostic group of participants (e.g., applicability of normed item properties to each one), are detailed in Highgate-Maynard and Neufeld (1986) and summarized in Neufeld et al. (2002). Considering the current emphasis on latency, namely, estimated encoding durations for correct responses, it should be emphasized that error rates were comparable across groups and did not contribute to latency in any confounding way. Finally, trimming from the examined latency data those times estimated for processes residual to the one of current focus—namely, encoding—is detailed in Neufeld et al. (2002; see also Neufeld, Vollick, & Highgate, 1993). Speciﬁc processes thought to be tapped by this task, then, include, ﬁrst, the cognitive translation of the probe item into a format abetting comparisons with the memory set or the memory set’s relevant properties; second, the comparisons proper; and third, response processes. Results from Highgate-Maynard and Neufeld’s (1986) study represent a case in point of delayed completion of the encoding process among schizophrenia participants, amid integrity of remaining processes. Individuals with schizophrenia taking part in this study included paranoid and nonparanoid subgroups. Comparison groups involved nonschizophrenic psychiatric controls as well as nonpatient controls. The present developments are expedited by focusing on the paranoid-schizophrenia participants and the nonpatient controls. The clinical signiﬁcance of this delineation has been described in Nicholson and Neufeld (1993); paranoid subgroups have also tended to be more elongated in probe-item encoding than nonparanoids. Convergent support for this form of deﬁcit, evinced on the present and related paradigms, has been presented in other sources (Neufeld, 1991; Neufeld et al., 2002; Neufeld, Vollick, et al., in press; Neufeld & Williamson, 1996). Convergent support for the modeled composition of the deﬁcit (elaborated on below) arises from investigations of divergent paradigms and forms of analysis. These include studies analyzing expressly the architectures and parameters of item encoding (Vollick, 1994; Vollick & Neufeld, 2004), studies combining memory-trace theory and multidimensional scaling in assessing deviant similarity judgments (Carter & Neufeld, 1999), and neuroconnectionist assays of temporal and substantive aspects of item categorization (Carter & Neufeld, in press). Finally, potential signiﬁcance

25 Formal Clinical Cognitive Science as a source of symptomatology, notably thought-form disorder (delusions and thematic hallucinations), has been discussed in other sources (Neufeld et al., 1993; Neufeld & Williamson, 1996).

parameter estimation at the diagnosticgroup level of analysis Equation (5) above is relatively complex (see the expansion in Appendix A), but it comprehensively allows for the possibility that we may not know the group of origin for the individual supplying the current cognitive-performance sample. Thus, we ﬁrst evaluate the probabilities of group membership, given the performance sample {Pr(g|{*})}, and then cipher the groupwise posterior probability (density) of , conditional on group membership [w( |{*}, g)]. This provision is brought into play (see below) when we wish to use performance samples taken over time to proﬁle an individual’s progress in terms of proximity to variously symptomatic groups (cf. Nicholson & Neufeld, 1993). It comes into play as well when attempting to evaluate dynamic aspects of treatment-regimen efﬁcacy in terms of estimated temporal changes in group density. Meanwhile, we may know perfectly well the group to which the present individual belongs. Our concern is, therefore, strictly with the expression [w( |{*}, g)]. We may isolate on the latter by setting Pr(g) = 1.0, where g is the relevant group. This constraint reduces the parent equation, Equation (5), to the target expression and spotlights its associated computations (consider remnants left by cancellation of terms, in the expanded version of Equation (5) presented in Appendix A, when Pr(g) is set equal to 1.0).

process-latency model The process-latency model has already been alluded to in the earlier discussion of the nature of stochastic dynamic models (see, e.g., “Absorption of Theory-Exogenous Variance”). It consists of the Erlang distribution, which is invoked to portray the stimulus-encoding process. The encoding process is highlighted because, of the several processes putatively involved in task transaction in toto, it is one that is diagnostic of group-performance differences. The rationale

26 modeling complex systems

Figure 3. Erlang distribution, with various pairs of its parameters.

for the adoption of the Erlang and other distributions used here and the qualifying considerations surrounding the present applications have been detailed in Neufeld et al. (2002). The model postulates k stages of encoding (e.g., constituent operations of covertly accessing size properties of the probe item). The stages proceed at the rate v per unit of time (seconds, in this instantiation). In the deterministic case, the process would invariably terminate after an interval of k/v. Ambient trial-to-trial inﬂuences naturally inﬁltrate processing time and perturb this pristine expression. These inﬂuences are summarized according to the probability density of process completion at time t: f(t) = (vt)k1/(k 1)! vevt,

(6)

which renders a mean latency of k/v and a variance across trials of k/v2.

27 Formal Clinical Cognitive Science The values of parameters k and v are not invariant, but posited to vary across group members. Thus, the present mixture model provides for individual differences according to the group’s Bayesianprior distribution of parameter values that randomly enter into the above process-latency model. Figure 3 displays the f(t) for four combinations of parameter values, for four hypothetical individuals.

parameter-value mixing distributions It is now apparent that two parameters, k and v, are involved in the process-latency model. Equation (5) may be readily extended to accommodate the present as representing this pair of parameter values (Neufeld et al., 2002). I consider the distribution of each, in turn. The parameter v is held to be gamma distributed, its gamma distribution having the parameters r and k. Its probability-density function, proportional to the relative frequency of its population occurrence, is (7) (rv)k1/(k)rerv. Note that is the gamma function, which is a continuous-variable analogue to the factorial; where k is an integer, (k) is (k 1)!. The parameter r tenably expresses stress-induced impairment in processing speed. In contrast, k arguably conveys the speed-augmenting effects of task-related competence or performance skill (Neufeld, 1994, 1996). As v formally conveys “processing capacity” (e.g., Townsend & Ashby, 1978; cf. Wenger & Townsend, 2000), its distribution is identiﬁed with the taskwise capacity pool, and r and k correspond to inﬂuences on this pool of stress and performer competence. Construct validity for the mixing-distribution parameter interpretations, particularly that of k, is examined under “Construct Validity of Distribution Parameters: Analytic Considerations” below. Introduced is a new format of construct validity endowed by the parameters’ formal moorings (cf. Haynes, 2001). Sufﬁcient for the purposes at hand is the observation that the distribution of v parsimoniously is regarded as common across diagnostic groups and conditions of encoding load (supporting analyses are enumerated in Neufeld & Williamson, 1996; Vollick & Neufeld, 2004). In the case of schizophrenia, therefore, processing capacity is

28 modeling complex systems

Figure 4. Gamma distribution of v.

spared; its distribution parameters remain unaltered. The estimates of r and k, obtained essentially through ﬁtting theoretical to empirical moments (e.g., Townsend, 1984), were 0.03735 and 2.5044, respectively. Note that, in order to circumvent certain interpretative hurdles apropos the present purposes, incorrect responses were excluded, as were positive trials (Neufeld & Gardner, 1990; Neufeld et al., 1993). The common distribution of v is presented in Figure 4. Unlike the distribution of v, that of k varies with diagnostic group and encoding-load condition. Speciﬁcally, this distribution is shifted upward in a similar fashion, with elevation in the encoding load, and with paranoid-schizophrenia diagnostic status. In this sense, these factors represent exogenous and endogenous sources of encoding-load increase, respectively. The distribution of k is considered to be Poisson (Evans et al., 2000). The single parameter of this distribution is designated m. Its mean—the modeled average value of k—is moved to the right by a constant amount for each source, as is its variance. This perturbation of the mixing distribution for k therefore signiﬁes an additive

29 Formal Clinical Cognitive Science increase in encoding subprocesses attributable to each factor of the diagnostic groups by encoding-load layout. The Poisson “intensity parameter,” m, is taken to indicate tendency toward encoding-subprocess recruitment on engagement of the probe item. The mean and variance of the Poisson distribution, then, are both m. The distribution’s base value for the nonpatient controls, under the low-encoding-load condition, is denoted m. The base value is increased by the amount h for each group under the higher-encoding-load, verbal-probe format and by the amount g with paranoidschizophrenia diagnostic status under both higher and lower encoding demands. Reference to the binomial distribution’s counterpart to m allows a more concrete understanding of this parameter. This counterpart, denoted np, approximates m as follows. The quantity n indicates the pool of encoding subprocesses potentially transacting the encoding process for any participant. The term p, in turn, portrays the probability of a latent subprocess being marshaled to the encoding operation. The value of n is m + h + g + x, where x is merely some very large number. That of p approximates m/n, where the base value of m, for controls under low-encoding demands, is m, to which is added h and/or g, depending on whether exogenous (encoding condition) and/or endogenous (paranoid schizophrenia) related increases in propensity toward greater subprocesses are present. The term n is tantamount to a dormant “subprocess pool,” and p may be viewed as a “disinhibition” term. Clearly, m increases with each term. The present interpretation designates p as the source of variation in m (Neufeld et al., 2002). In keeping with this stance, as n approaches the limiting value of inﬁnity and np approaches the limiting value of m, the Poisson and binomial distributions converge. It is the value m = np that governs the probability of a speciﬁc value of k. This probability is Pr(k; m) = mk/k!em.

(8)

The estimated value of m was .00001, that of h was 19.7390, and that of g was 70.0. The value of m suggests that the controls required hardly any processing to encode the rudimentary limned drawings of the probe items. Plots (smoothed) of the Pr(k; m) for the present values of m are plotted in Figure 5. The general structure of the mixture-distribution model is sum-

30 modeling complex systems

Figure 5. Poisson distribution of k (smoothed) for differing values of m.

marized in Figure 6. The gamma and Poisson mixing distributions of v and k, respectively, correspond to those displayed in Figures 4 and 5 above. The Erlang distribution of t, on which v and k bear in the individual case (exempliﬁed in Figure 3 above), is the mixture’s “base distribution.” With this structure in hand, estimation of the base distribution’s parameters for an individual can, in principle, be substantially sharpened; all that is needed is a modest sample of the participant’s own process latencies {*}.

individualized parameter estimation With appropriate application of [w( |{*}, g)] to the case of {v, k}, estimation of these parameters stands to become substantially more precise (computational formulas are presented in Neufeld et al., 2002). To illustrate the point, representative performance samples from each diagnostic group, under each condition of encoding load, are brought into play. In each instance, a sample was extracted from

31 Formal Clinical Cognitive Science

Figure 6. Design of mixture model.

one of the 10 participants in each of the four factorial combinations (2 diagnostic groups 2 encoding-load conditions). Each sample size was four, one for each memory-set size, under negative trials. Adjusting these latencies to highlight the encoding process has been described in Neufeld et al. (2002). The sets of trimmed values are presented near the top of Table 2. A graphic example of tailoring a parameter distribution to the individual is exempliﬁed for a tentative case. Speciﬁcally, consider the performance sample supplied by the control participant under the low-encoding condition {1.037, . . .}. Although unlikely according to Table 3, this sample still could have been produced by a member of the paranoid-schizophrenia group, under the low encoding load. If so, the customized distribution of k would have moved markedly to the left, relative to the prior distribution aligned with the paranoid schizophrenia/low encoding load combination. This relocation reﬂects the joint probability of comparatively lower latencies making up this encoding-process performance sample. The Bayesian posterior distribution of k (smoothed) is presented in Figure 7, alongside the prior distribution corresponding to m = 70.00001. Note as well the increased precision of estimation, indicated by the diminished width of the posterior distribution. In fact, the standard deviation of this posterior distribution of k is 1.8110, whereas that of the prior is 8.36. Similar observations attend the individualized estimation of v.

.32930(10-7) Pr{*}→

Control: High encoding loadc Low encoding loadd

b

Parameters of prior distributions: r = .03735, k = 2.5044; m = 89.73901. Parameters of prior distributions: r = .03735, k = 2.5044; m = 70.00001. c Parameters of prior distributions: r = .03735, k = 2.5044; m = 19.73901. d Parameters of prior distributions: r = .03735, k = 2.5044; m = .00001.

a

.17176(10-12) .999776(10-11)

High Encoding Load {3.476, 7.150, 9.00, 4.238}

Paranoid schizophrenia: High encoding loada Low encoding loadb

Classiﬁcation

.00083 .1486(10-12)

.00684 .00737

Low Encoding Load {2.177, 3.184, 2.411, 2.498}

Paranoid Schizophrenia

.00002 .80080(10-12)

.54107(10-11) .70547(10-9)

High Encoding Load {1.137, 3.204, 1.371, 2.207}

.75783(10-13) .65324(10-12)

.29397(10-40) .23614(10-32)

Low Encoding Load {1.037, 0.554, 0.191, 6.390}

Control

Representative Individual-Participant-Latency Sets

Table 2. Unconditional Joint Densities of Encoding-Latency Performance Samples under Alternate Bayesian Priors

.99987 Pr{g} → 0

Control: High encoding loadd Low encoding loade .027 .23972(10-11)

.022 .9511

Low Encoding Load {2.177, 3.184, 2.411, 2.498}

.99993 .200186(10-12)

.27083(10-7) .00007

High Encoding Load {1.137, 3.204, 1.371, 2.207}

.188326 .81167

.730536(10-29) .23473(10-20)

Low Encoding Load {1.037, .554, .191, 6.390}

Control

Base rates for group membership = .05 for paranoid schizophrenia/high encoding load, .20 for paranoid schizophrenia/low encoding load, .5 for nonpatient controls/high encoding load, and .25 for nonpatient controls/low encoding load. b Parameters of prior distributions: r = .03735, k = 2.5044; m = 89.73901. c Parameters of prior distributions: r = .03735, k = 2.5044; m = 70.00001. d Parameters of prior distributions: r = .03735, k = 2.5044; m = 19.73901. e Parameters of prior distributions: r = .03735, k = 2.5044; m = .00001.

a

.52528(10-6) .00012

High Encoding Load {3.476, 7.150, 9.00, 4.238}

Paranoid schizophrenia: High encoding loadb Low encoding loadc

Classiﬁcationa

Paranoid Schizophrenia

Representative Individual-Participant-Latency Sets

Table 3. Posterior Probability of Group Membership, Given Representative Individual Latency Sets

34 modeling complex systems

Figure 7. Posterior distribution of k, Pr(k|{1.037, .554, .191, 6.39}; m = 70.00001), and prior distribution of k, Pr(k; m = 70.00001).

individualized latency distributions These slenderized posterior parameter-mixing distributions can be used to create posterior latency distributions f(t|{*}; g). The customizing effects of a performance sample can be appreciated by contrasting f(t|{*}; g) with those of the alternate mixture distributions f(t|g), sans reference to the performance sample {*}. The selected performance sample is that for the paranoid-schizophrenia low-encoding condition {2.177, . . .}. Figures 8 and 9 show the posterior latency distributions constrained by {*}, and Figures 10 and 11 give their unconstrained mixture-distribution counterparts. Observe that, in Figure 10, the distribution for m = 0.00001 is, essentially, not visible because the stochastic mechanisms involved result in its hugging the y axis.

posterior probability of group membership I now consider the case where a performance sample has been obtained from an individual and is to be used to evaluate that per-

35 Formal Clinical Cognitive Science

Figure 8. Posterior latency distributions f(t|{2.177, 3.184, 2.411, 2.498}; g).

son’s “proximity” to variously symptomatic groups. This scenario may apply to treatment evaluation over time: Is the individual being edged closer to normal functioning, as assessed by methods rooted in contemporary cognitive science? Estimates are available in terms of the posterior probability of a representative of the target group—a group with less or no symptom severity—generating the sample, in contrast to a member of a symptomatic or more severely symptomatic group. In effect, what is being asked is: In Bayesian terms, what are the compatibilities of the obtained process-latency specimen with each group’s pair of prior parameter distributions? The expression of Equation (5) above enclosed in elongated braces { } is now brought forth. Allowing that the performance is sampled at selected points in time, it is assumed that the representativeness of each symptom group along the way is known. In other words, transitions in prevalence of the candidate groups are, ostensibly, monitored so that base rates for a testee’s group membership, Pr(g), g = 1, 2, . . . , G, are available. It is assumed as well that the mixing distributions for these

36 modeling complex systems

Figure 9. Posterior latency distributions f(t|{2.177, 3.184, 2.411, 2.498}; g).

groups, established at the outset, remain valid or, if they do not, that new priors have been formulated for the ensuing points of performance sampling. In the present application, the assumption of initial-prior stability is invoked. Finally, the present computations imply that the G classiﬁcations are mutually exclusive and exhaustive. The classiﬁcatory system is, therefore, regarded as comprehensive. This feature underwrites the intended description of the presenting individual, which, in fact, is a ﬁnite, discrete posterior-probability mixture. The adopted approach entails an “odds-ratio evaluation strategy”—an odds-ratio evaluation strategy for assessing the appropriateness for the current performance sample {*} of the G respective priors. It is deemed to facilitate the ﬂow of the present developments; other Bayesian strategies, with unique statistical merits, are, nevertheless, available (Karabatsos, 2006). In contrast to the scenario painted above, it is possible that diagnostic-group base rates are, indeed, not available for the designated time points. Instead of monitoring an individual’s progress by proﬁling his or her likelihoods of belonging to symptom groups of known base rates, the primary issue may entail the base rates themselves.

37 Formal Clinical Cognitive Science

Figure 10. Mixture-model latency distribution f(t|g).

Here, the focal problem may be one of evaluating the collective effectiveness of treatment by ascertaining movement in size of symptom-group membership. A question of this nature can, once more, be broached through the elongated-braced expression { } of Equation (5) above. This second case is taken up in its turn below. To expedite the exposition, the present classiﬁcations formed by factorially combining diagnostic status and encoding-load condition are appropriated. These combinations again provide the Bayesian mixing-distribution priors of k and v as well as the representative performance samples listed in Table 2 above. We conveniently allow for a progressive decrease in symptomatology with movement across groups formed from combinations of paranoid-schizophrenia status/high encoding demands, to paranoidschizophrenia status/low encoding demands, through nonpatient status/high encoding demands, to ﬁnally nonpatient status/low encoding demands. Apropos the encoding manipulation as a surrogate for symptom

38 modeling complex systems

Figure 11. Mixture-model latency distribution f(t|g).

severity, to be sure, responding to increased encoding demands with an increased number of encoding stages is hardly indicative of disorder. It nevertheless conveys an escalation in recruited subprocesses, as a stand-in for the elevated number that may well go along with increased symptomatology (Nicholson & Neufeld, 1993). In all events, this pretense for expedience is inconsequential to the ensuing line of argument and in no way detracts from its analytic points. We posit a scenario where encoding-latency samples have been obtained from four participants, as arrayed in the ﬁrst row of data in Table 2 above. The values have been obtained, say, at a time of interest speciﬁed by the treatment regimen. We wish to estimate probabilities of symptom-group membership for each participant, given the obtained performance data. These groupwise posterior probabilities are endowed by the prior parameter-mixing distributions aligned with the candidate groups. To proceed, it is necessary to know the values of the uncondi-

39 Formal Clinical Cognitive Science tional joint probability density of each latency sample with respect to each group ucd({*}|g). The ucd({*}|g) is the joint density of the obtained sample of latencies, given group g, all values of v and k considered. The numbers assumed by ucd({*}|g) will change with g because the parameter-mixing distributions are group speciﬁc. The values of ujd({*}|g) for the current performance samples are presented in Table 2 above. Base rates assigned to the respective groups Pr(g), proceeding from the left to the right of Table 2 (from paranoid schizophrenia/high encoding load to nonpatient controls/low encoding load), are 0.05, 0.20, 0.5, and 0.25. The posterior probabilities of group membership are entered in Table 3 above. Each individual is proﬁled in terms of distance from the respective groups, enumerated as the Bayesian posterior probability of group membership Pr(g|{*}). Not surprisingly, the highest values tend to be aligned with the groups from which the samples originated. The sole exception occurs for the sample from the paranoid schizophrenia/high encoding load combination. This result is attributable to both ucd({*}|g) and Pr(g) being highest in connection with the nonpatient controls/high encoding load amalgam. The point of the exercise, however, mandates that the clinical scientist/practitioner is blind as to the actual origins of the latency samples. Table entries can, then, provide pivotal information regarding participant status, as conveyed by the Bayesian expansion of the present mixture model. Allowing that all individuals were members of the same group at the outset, clearly the far-right column presents the most encouraging proﬁle and the second-from-the-left column the least so. As intimated above, such multivariate proﬁles can be charted across successive times of interest. The resulting individual conﬁgurations of treatment (non)response, at least as quantiﬁed by focal cognitive functioning, may, in turn, be statistically mediated to psychometric and other predictors.

maximum-likelihood estimation of symptom-group base rates In the ﬁrst case presented above, base rates for the variously symptomatic groups were known, and the desired information consisted of the alternate likelihoods of the individual’s membership, given his

40 modeling complex systems or her specimen of process latencies. In the present case, we entertain a sample of individuals who have been allocated to one or another group on the basis of presenting symptomatology at the selected time of interest. However, the present case differs from the ﬁrst case in that the base rates are now unknown, and it is this information that we seek. These densities of enclaves with differing symptomatology may be useful in assessing treatment efﬁcacy across the designated times of interest. It may be argued that the desired course of action is simply to directly examine the array of diagnostic assignments that have been made at the selected times. It may be prohibitive, however, to diagnostically interview (or assess by some other means, such as psychometric) a large enough sample to ensure reasonable ﬁdelity of estimates (see, e.g., Batchelder, 1998; Luke & Homan, 1998). If so, it seems desirable to bring to bear our available bank of information on cognitive functioning. In importing this information, we may settle, not only for a modest sample of individuals whose classiﬁcations are updated, but also for a modest performance sample from each. The required sampling may, feasibly, be tacked onto sessions hosting symptom assessment at the successive times of interest. The strategy for estimation is straightforward enough. It consists of constructing a likelihood function for the updated participant assignments and then maximizing the function with respect to Pr(g), g = 1, 2, . . . , G. The likelihood function is made up of the joint probability of the updated participant categorizations. This joint probability comes down to the product of the posterior probabilities of those categorizations, given the latency samples {*} linked to the categorized individuals. The procedure constitutes a somewhat unusual but valid combination of maximum-likelihood estimation of a meta-Bayesian parameter (see Appendix A). The likelihood function and related computational details of maximum-likelihood estimation of Pr(g), g = 1, 2, . . . , G, are presented in Appendix B. The instantiation of this application once again makes use of entries in Table 2 above. For reasons of mathematical tractability, and in the interests of illustrative neatness, only three of the four data sets listed in this table are considered. The set corresponding to the nonpatient controls/high encoding load is excluded, deleting the third row and third column of ujd({*}|g) values in the table. However, in

41 Formal Clinical Cognitive Science practice, such a restriction is not necessary, as described momentarily. The maximum-likelihood estimates of Pr(g) were obtained by solving for each value using closed-form solutions to two simultaneous equations (two, rather than three, as the base-rate probabilities sum to 1.0; see Appendix B). It was found that increasing G beyond 3 became infeasible, even with the assistance of a computer-algebra program. In larger-scale applications, however, where G > 3 and/ or the number of diagnostically updated participants is somewhat larger, one may resort to a numerical-search algorithm (e.g., matlab Optimization Toolbox), rather than pristine closed-form solutions, to estimate the respective values of Pr(g). For the present purposes, the three sets of performance samples in the provisionally reduced Table 2 now represent those putatively obtained at the period of updated individual classiﬁcation. It is, moreover, sufﬁcient to allow each individual to have been assigned to the group adjacent to the displayed performance sample. In other words, three individuals have been classiﬁed, one to each group. Furthermore, the classiﬁed individuals have putatively supplied the very process-latency samples atop the columns of their respective groups. With these arrangements in place, base-rate estimates of the three groups, proceeding from left to right, become 0.6748, 0.0852, and 0.2399. The most symptomatic category thus remains in the majority at this time of evaluation, and those who are moderately symptomatic are sparsest. Nevertheless, the least symptomatic category is reasonably well represented, with an estimated relative frequency of nearly 0.25. It is tempting to insert these base rates into the expression enclosed by { } in Equation (5) above and, in turn, to evaluate the posterior probabilities of membership in the respective groups, given the associated performance samples {*}—as was done in the ﬁrst case presented above. Note, however, that this exercise is nothing more than instructive, regarding the involvement of ujd({*}|g) and Pr(g) in arriving at the values of Pr(g|{*}). The latter, again proceeding from left to right in Table 2, are 0.119786, 0.119736, and 1.00. The ﬁrst amount resembles the second because the comparatively low ujd({*}|g) connected with the paranoid schizophrenia/high encoding load group for the individual so assigned is offset by that group’s

42 modeling complex systems having the highest value for Pr(g). The value of 1.00 for the nonpatient controls/low encoding load assignment evinces the fact that the ujd({*}|g) associated with this speciﬁc group exceeds the next highest value by a factor of .2766325061(1021).

qualifying assumptions The above estimates of Pr(g) assume that the parameter-mixing distributions initially prescribed for the symptom-based classiﬁcations extend to the time of estimation. This assumption is intrinsically bound up with stability of conditions under which the current sample of task performance is obtained. The assumption is also yoked to continuation of the initial classiﬁcatory criteria leading to participant allocation. These requirements come down to retention of their earlier values by parameters of the mixing distributions r, k, and m, continuing tenability of the distributions themselves, and transference of the previous meaning of g in the presently estimated Pr(g). Dislodgement of the established priors from their host symptom groups may occur because of a shift in the nature of task performance, owing, for instance, to practice effects. Potential redress may require the substitution of different task items, analogous to using parallel forms of psychometric tests. To guard steadfastness of the G categories, symptom criteria for designations need, by and large, to be consistent. It is additionally assumed that differential dropout across the G groups and other sources of distortion in their representativeness are prevented. Finally, the computations once again imply that the G classiﬁcations are mutually exclusive and exhaustive. In other words, transitions in symptomatology eventuate in one or the other of the G classes addressed at the outset. It is evident from these developments that tenability of the proposed analytic strategies makes for new and potentially useful avenues for formulating clinically signiﬁcant information. Available are inferences about the status of speciﬁc patients, as conveyed by cognitive-functioning aspects of symptomatology. Available as well are broader inferences about treatment efﬁcacy. Furthermore, assumptional constraints are relatively salient. Finally, the developments remain entrenched in a formal cognitive-process model.

43 Formal Clinical Cognitive Science

when events are processes: potential contributions to fmri Recent years have seen an explosion of advances in fmri technology (e.g., Frackowiak et al., 2004; Huettel, Song, & McCarthy, 2004; Reiman, Lane, Van Petten, & Bendettini, 2000). Clinical cognitive scientists have capitalized on these successes, mapping deviations in neuronal-activation patterns corresponding to symptom-signiﬁcant mentation (e.g., Boksman et al., 2005; Lanius et al., 2004). Note that, in documenting neuronal-activation concomitants, the ultimate subject of interest is not that of external-stimulus events per se. Rather, it is the cognitive processes that are instigated by these events and that parallel the activation proﬁles that are, thus, set in motion (cf. Friston, Harrison, & Penny, 2003; although not emphasized here, similar considerations attend other measures of neurophysiological functioning, such as magnetoencephalography, or “meg”; see, e.g., Fokas & Marinakis, 2006). Considerable dynamic modeling of the blood-oxygen-level-dependent (bold) response, the kernel of mri measurement, has been put forth (e.g., Aguirre, Zarahn, & D’Esposito, 1998; Boynton, Engel, Glover, & Heeger, 1996). Similarly impressive advances in the statistical treatment of fmri signals have taken place. Quantitative cognitive science is, meanwhile, poised to complement these developments. Implicated are stochastic dynamic models, as follows. Consider the stochastic dynamic trajectories presented in Figure 9 above. The areas under these curves are proportional to relative frequencies of process completion. They can be integrated into posterior cumulative probability functions F(t|{*}) whose values for the respective curves in Figure 9 are shown in Figure 12. The amounts displayed in Figure 12 may, in principle, be consulted to synchronize times of interest in bold-response measurement, on the one hand, with probability of target-process occurrence, on the other. In moving from right to left in Figure 12, for example, the probability of process expiry decreases; however, the time window for measurement does as well. The interplay of such considerations now has a formal backdrop against which to estimate times of interest, complementing regions of interest, in calibrating mri measurement. Moreover, imputation of cognitive functions now stands to be anchored in a verisimilitudinous performance model, avoiding the often-encoun-

44 modeling complex systems

Figure 12. Cumulative probability distributions corresponding to probability densities of Figure 12 above.

tered circular reference to the very neurocircuitry whose functional signiﬁcance is under study. The trajectories in these ﬁgures have been constructed for an individual performance sample/prior distribution combination. They could, however, be extended to accommodate subsets of individuals whose sets of performance samples are similar and to whom the same priors apply. The resulting amalgams should render subsets with relatively homogeneous parameter proﬁles (see Figure 7 above), along with associated latency trajectories. Aggregation across participants is often necessary to attenuate noise of mri signals or for selected statistical treatment. Amalgamation of parametrically homogeneous participants guards against an amalgam of systematic individual differences in the expression of cognitive functions under study.

multiple processes Typically, even the simplest cognitive task recruits multiple stochastic processes (Smith, 1995; Townsend, 1984). Separability of the respective processes then comes into play. The processes may occur

45 Formal Clinical Cognitive Science in parallel, meaning that they commence simultaneously. Their ﬁnishing times are, nevertheless, staggered, according to their respective stochastic distributions (Townsend & Nozawa, 1995; Wenger & Townsend, 2000). Let the probability-distribution function for one of a pair of processes be F1(t) and that for the second be F2(t). It is stated here without elaboration that the greatest likelihood of the second process surviving, with the ﬁrst having elapsed, is deﬁned by max([1 F2(t)] [1 F1(t)]) = max(F1(t) F2(t)). Measurement epochs addressed to the second process would be synchronized accordingly. The processes may, instead, proceed serially. Let the probabilitydensity function of the initial one be f1(t). Then the greatest likelihood that it has been accomplished, while the second has not yet elapsed, occurs at the modal value of f1(t), or max(f1(t)).

cognitive debility and stress susceptibility The adaptive signiﬁcance of cognitive functions may be appreciated by deciphering how processing operations interface with environmental exigencies. Consideration of cognitive psychopathology invites the following extension: examination of quantitatively dissected cognitive debilities in relation to likewise quantiﬁed environmental pressures on which they bear. As applied to stress negotiation, the above approach implies, among other analyses, delineating information-processing requirements for selecting advantageous, threat-minimizing options when stressor situations are multifaceted. This endeavor is potentiated through the application of mathematical combinatorics, and auxiliary procedures, to environmental prototypes. Products of the analysis include the speciﬁcation of (a) cognitive costs, in the form of information-processing demands; (b) stress-reducing beneﬁts, in the form of minimization of threat (e.g., physical danger/discomfort; social sanction); and (c) sources of vulnerability stemming from information-processing frailties. The synthesis should, ultimately, provide a more precise picture of compromised negotiation of environmental demands, harboring implications for prediction and intervention. A prominent form of coping with stress, alluded to above, is labeled decisional control (Averill, 1973; cf. Lees & Neufeld, 1999). This

46 modeling complex systems type of coping basically takes the form of situating oneself in a multifaceted stressing context so as to engage the situation’s most innocuous option. Dissection of decisional control, through the application of mathematical combinatorics and related methods to the essential structure of this form of coping, discloses it to be a prototypical expression of selection and choice (Morrison et al., 1988; Neufeld, 1999b). Unveiled, in turn, are its demands on cognition, including option-cue encoding, visual and memory search, and response processes. Unveiled alongside is the undermining potential of debilities in the required cognitive functions, including those comprising retarded completion of encoding processes (Neufeld, 1991, 1999a; Nicholson & Neufeld, 1992). A factor potentially exacerbating the above risk is the toll taken on encoding speed by stress itself (Neufeld & McCarty, 1994; Neufeld, Townsend, & Jetté, in press). Apropos the formal depiction of encoding, this effect translates into a general reduction in the parameter v of the base distribution of latencies (Figure 3 above). Recall that a stress-induced movement of the distribution of v to the left is identiﬁed with an increase in its mixing-distribution parameter r (Figure 4 above; Neufeld, 1994; Neufeld & Carter, 2000).

Construct Validity of Distribution Parameters: Analytic Considerations Construct validity is a form of psychometric validity entailing a corpus of evidence brought to bear on the espoused interpretation of a measure. Haynes and O’Brien (2000) have summarily stated that it “comprises the evidence and rationales indicating the degree to which data from an assessment instrument measures the targeted construct; includes all evidence bearing on the measure, and includes all types of validity” (p. 75). In line with this deﬁnition, construct validity for the purported psychological meaning of scores from a test, an inventory, or some other mode of assessment emphasizes the measurement tool’s portfolio of empirical correlates (Campbell & Fiske, 1959; Cronbach & Meehl, 1955; Embretson, 1983; for a mathematical critique of selected data-analytic strategies for adducing psychometric construct validity, see Neufeld & Gardner, 1990). An overriding principle governing evidence for a measure’s

47 Formal Clinical Cognitive Science construct validity is that the measure act in accordance with theory (Wiggins, 1973, p. 406). In the present treatment, “measures” consist of model parameters. Construct validity for their imputed meaning, in turn, emanates from their mathematical properties, which are part and parcel of the stochastic dynamic models in which they participate. The light thrown on the nature of parameters by the mathematical properties they possess arguably embodies a new form of construct validity (cf. Cronbach & Meehl, 1955). The parameters of the appropriated process-latency model included v and k. Their interpretations stemmed from the assumed structure of the encoding process, as translated into a corresponding distribution of latencies. Distributions of the above parameters, in turn, have their own parameters, r and k, in the case of a gammadistributed v, and m, in the case of a Poisson-distributed k. Because they pertain to the mixing distributions for v and k, the parameters r, k, and m are a step removed from the empirical touchstones of an individual’s latency data. Accordingly, called for is support for the constructs that they purportedly express, arguably from a commensurate level of analysis. Recall that the parameter m was aligned with “encoding load” and conveyed exogenous (task demands) and endogenous (paranoid schizophrenia) sources of increase. Its binomial-based dissection cast a certain light on the potential composition of this parameter (cf. “construct representation,” Embretson, 1983). The parameter r was identiﬁed with stress effects on processing speed. Its increase moves the stage-dispatching capacity parameter v to the left. Finally, the parameter k ostensibly corresponds to process-transaction competence, and its increase shifts the distribution of v to the right. The parameter k is endowed with additional mathematical properties that effectively illustrate the type of construct validity imbued by analytic derivations. The behavioral signiﬁcance of these properties hinges on the concept of maintenance of cognitive performance, over and against its breakdown, as follows. Within the present analytic context, integrity, versus collapsing, of performance requires that the value of k exceed a certain threshold, paralleling a critical level of performer skill. In contrast, increasing r and m prolongs performance times, but, in this case, as opposed to that of k, actual failure of the processing system is not precipitated when a ﬁnite range of values is violated.

48 modeling complex systems The current formal expression of performance breakdown takes the form of certain latency-distribution moments acquiring inﬁnite values: more succinctly, E(Tn) = , where n is the order of the computed moment. Behaviorally, the result signiﬁes that the population of trials occurring under the prevailing value of k includes a critical mass of extremely long, or essentially outstanding, completions (in effect, noncompletions). Consequently, the integral

f(t)t dt = E(T ) 00

n

n

0

does not converge (for further technical exposition, see the ﬁrst section of Appendix C). The ﬁrst four moments of a stochastic distribution (n = 1, 2, 3, 4) are somewhat familiar, comprising or being involved in its mean, variance, skewness, and kurtosis. In model evaluation, empirical moments beyond the second are seldom sampled because of instability. The present theoretical deliberations, however, operate in the realm of population properties, bypassing issues of sampling and measurement weaknesses. The order of the moment n can, therefore, be driven to whatever amount is prescribed by the governing theoretical analysis, which, in this case, concerns consequences of variation in the processing system’s constituent parameters. At the same time, such theoretical explorations retain a bona ﬁde bearing on the interpretation of the parameter values, as fallibly estimated from empirical data. Observe that this approach to discerning the nature of a theoretical system’s parameters has much precedent in ﬁelds ranging from mathematical ecology, to economics, to astrophysics. The above integral indicates how values of t and the order of the moment n act in concert to affect E(Tn). As n increases, its exponentiation of t correspondingly magniﬁes the consequences of elongated latencies for the computed moment, including its transition to inﬁnity (Appendix C). Apropos the model parameter k, strengthening process-performance skill should inﬂuence the distribution of t downward, resulting in E(Tn) being bounded to an even greater extent by ﬁnite values. That is to say, it should enable E(Tn) to withstand heightened values of n before transmogrifying to inﬁnity. In effect, k acts to diminish densities f(t) for higher latencies, including extremes, through its shifting the distribution of the process-completion rates, v, upward. Indeed, the threshold value of k

49 Formal Clinical Cognitive Science issuing in bounded moments is deﬁned precisely in terms of n (see both below and Appendix C). Insufﬁcient k implies that appreciable densities of v, f(v), are perilously close to v = 0, generating the critical magnitudes of f(t) at acutely high ranges. The above support for the construct validity of k as a capacityaffecting competence parameter is robust, as assayed from multiple analytic standpoints. It transcends process-latency base distributions with diverse characteristics; its interpretation is enhanced by pertinent Bayesian extensions; its effects are modiﬁed in expected ways by formally deﬁned efﬁciency of capacity application by the operative base distribution; its effects align with those of related parameters from similar mixing distributions; and it is evinced in familiar base-distribution statistical properties. Finally, its effects are distinguishable from those of parameters to which a different meaning is ascribed, notably, r, the avowed index of stress effects on processing capacity. These variations on analytic support are dealt with in succession. Their largely verbal descriptions are presented below, with technical justiﬁcations allocated to the several sections of Appendix C.

critical values of k : convergence from multiple base distributions Finite moments of mixture-model latencies require that k exceed the order of the moment, n, for diverse process-latency base distributions. The current selection of base distributions is prominent in the stochastic-modeling literature and expresses alternate structures of a processing system. The descriptions of these base distributions are available in various sources (e.g., Evans et al., 2000; Johnson et al., 1994, 1995; Ross, 1996); presentation here is restricted to theoretical analysis of the espoused substantive signiﬁcance of k. Note that the present treatment, although theoretical, is malleable with respect to empirical application. Almost any empirical process-latency distribution can be well approximated by creating a ﬁnite probability mixture of one or more base distributions (defensibly expressing the relative percentage of trials best characterized by the respective base distributions; e.g., Townsend & Fiﬁc, 2004). Importantly, results of the present analyses regarding moment ﬁniteness are preserved with such a mixture.

k stages fixed, each with constant rate v

k stages random, v constant across stages

Compound Poisson

Distinguishing Features

Erlang

Base Distribution

-M

(rv) k – 1rerv , (k)

r2m(2(k – 1) + m) (k – 1)2(k – 2)

mr ; (k – 1)

Var (T) =

E(T) =

and m is the parameter of the Poisson distribution of k. For example,

f(v) =

i

v , (v + )

T

f(v)[(em(M i ( – 1))-e |v]dv,

MT ( ) =

00

0

where MT ( ) =

(– 1) ndn MT ( )/d n| = 0 ,

rn (k + n – 1)!(k – n) (k – 1)!(k)

k>n

a

k = n1

00 (1 – e –m) = 00

r k (k + n – 1)! ln(r + t) (k)(k – 1)!

nth Moment

] t= 0

t= 00

= 00

Comments

continued

E(T n|k) = 00, hence E(T n) = 00, also by results for the general gamma distribution below

E(T n|k) = 00, by results for the Erlang distribution above

For the exponential distribution, k = 1.0

Table 4. Process-Latency Base Distributions and Their Moments with a Gamma-Distributed “Capacity” (Intensity) Parameter, v

-e-M

a

Rate for the ith stage is civ, where ci = (k– (i+1)); i = 1, 2, . . . , k

Single stage; rate is continuous functions of time

Pure death general gamma process with linear death rate

Weibull

k

Ckcin!r –n (k – n) , (k)

k

–1

k i

n

n

– n) () r n!(k (k)i

cj

ci

rn (n/ + 1)(k – n) , (k) where > 0

i=1

(– 1) i+1

j=i –

Cik = j=1 1–

k

( )

where ci is a scalar of v for the ith stage, ci > 0; the rate thus in effect for stage i is civ; and

i=1

k

k

k i

k

i+1

00 , as (k – n) = (0)

i=1

n+1

ln(r + it) () ir n (k)i ]

= (– 1) , 00 = 00

i=1

(– 1) i+1

00 , as (k – n) = (0)

t= 0

t= 00

Expression of moment infinity as (0) or ln(00) merely rests on integration with respect to t first or v first, respectively.

Stagewise rate transition

General gamma

52 modeling complex systems The above relation between k and n occurs for the Erlang distribution (Figure 3 above), with a ﬁxed-stage parameter k. A speciﬁc case is the single-stage exponential distribution, where k = 1.0. This relation includes as well the extension to k as a Poisson-distributed random parameter, otherwise known as the compound Poisson distribution (e.g., Feller, 1966; Ross, 1996). Together with the mixture on v, the latter distribution makes up the dual-parameter mixture model depicted in Figure 6 above. In the present context of base distributions whose parameter v is gamma distributed, the compound Poisson distribution is considered simply to be a base distribution whose Erlang stage parameter, k, is Poisson premixed. The rate parameter, v, is constant for each stage in both the Erlang and the compound Poisson distributions. The above relation between k and n, however, transcends base distributions with a static rate parameter. It extends to those with rate transitions across stages, known as the general gamma distribution (McGill & Gibbon, 1965). A notable version of the general gamma is a pure death system with a linear death rate, which is a parametric version of an independent parallel-processing model with unlimited capacity (see Townsend, 1990; Wenger & Townsend, 2000). This model has ﬁgured prominently in selected theorizing about the architecture of human information-processing mechanisms (see, e.g., Townsend & Ashby, 1983, pp. 80–95). Another well-known and frequently used distribution to which the present relation applies is one with a single stage of process completion but with a rate that changes continuously with time (Weibull distribution). The selection of base distributions, and their distinguishing properties, to which the relation between k and n applies is summarized in Table 4 (further details are brieﬂy presented in the second section of Appendix C, additional elaboration being available in Neufeld, 1994). Also listed are the moments for k > n and k = n. Apropos the latter, note that, if the nth-order moment is unavailable (inﬁnite), as in the case of n = k, so are moments of order exceeding n (for k as an integer, see, e.g., Harris, 1966, p. 102; where k is not an integer and k < n, the function (k n) appearing in the moment’s expression takes on a succession of negative, inﬁnite, and positive values as k n decreases from 0, whereby moments about the origin become meaningless. At the heart of the matter is the attestation to k as a process-

53 Formal Clinical Cognitive Science transacting competence parameter, according to invariance of its inﬁnity-avoiding values across base distributions of heterogeneous makeup. Robustness of this nature bears a certain resemblance to empirical convergent validity, which entails agreement among divergent methods of measuring the same trait (Haynes, 2001).

bayesian extension Additional light can be thrown on the proffered interpretation of k by dissecting its properties within a Bayesian context. Insight is afforded according to the give-and-take between k and both empirical performance samples and collateral parameters. A poignant exemplar involves the Erlangk,v distribution whose rate parameter, v, is gamma mixed. The current Bayesian extension once more incorporates a set of process latencies {t1, t2, . . . , tN} = {*}, and the nature of latency-distribution moments again occupies center stage. The ﬁxed-parameter status of k streamlines the present exploration without curtailing the generality of available inferences. Findings tenably apply to classes of individuals sufﬁciently homogeneous regarding k that this simpliﬁcation is a reasonable approximation of the quantitative goings-on. The required moment expression, so conditionalized on k, can be immediately extracted from derivation (A.10) of Neufeld et al. (2002). It appears as follows: N

n

E(T |{*}, k) =

[r + ti ]n(Nk + k – n)(k + n – 1)! i=1

(Nk + k)(k – 1)!

.

(9)

This expression affords a certain perspective on the posited interpretation of k from the standpoint of rudimentary classic psychometric theory surrounding performance-sample size or test length. The excursion into the psychometric analogy is followed once more by observations on the viability of the proposed interpretation of k. Isolating ﬁrst on the term (Nk + k n), an inﬁnite value of E(Tn|{*}; k) is avoided if Nk + k exceeds n. In this case, it is possible to have k = n without E(Tn|{*}; k) = if and only if Nk > n k, allowing, of course, that the present ti < , i = 1, 2, . . . , N. A value of k = n signiﬁes that a ﬁnite E(Tn|{*}; k) does not characterize all members of the class of individuals to whom the current value of k applies.

54 modeling complex systems The speciﬁc member at hand is, nevertheless, characterized by a latency moment bounded by ﬁnite values. The psychometric analogy seemingly throws light on this conclusion as follows. Within this analogy, the stage parameter, k, corresponds to the number of parallel items composing a psychometric test. Second, N corresponds to the number of independent test administrations. The collective item sample is, therefore, Nk. With a sufﬁciently large empirical sample of ﬁnite latencies, combined with the extant value of k, a ﬁnite value of E(Tn|{*}; k) is assured for the testee at hand. The argument can be carried further as follows. Consider that the Bayesian posterior moment E(Tn|{*}; k) can, in principle, be partitioned according to a probability mixture of two partial expectations, (E(Tn|{*}; k) = ) and (1 )(E(Tn|{*}; k) < ), where is the mixing parameter, 0 1.0. A sufﬁciently large performance sample of ﬁnite process-completion latencies, in concert with k, in effect renders equal to 0. Conversely, a value of Nk such that Nk + k = n fails to nullify the viability of the partial expectation E(Tn|{*}; k) = , 0 < 1.0, and an inﬁnite moment ensues for the present individual. Bringing the above formulation to bear on convergent support for k as a process-transaction competence parameter, the following inference from Equation (9) is apparent. A performance sample size that is inadequate to ensure stability of ﬁnite latency values for the person at hand can be offset by a sufﬁciently large value of k for the class of individuals to which the present performer belongs, speciﬁcally, k > n Nk. Inspection of Equation (9) reveals further interplay between k and N and between k and k, both informative as to the nature of k. Allowing Nk + k > n, consider the following ratio, extracted from Equation (9): 1 (Nk + k – n) . = (Nk + k) (Nk + k – 1)(Nk + k – 2) . . . (Nk + k – n)

E(Tn|{*}; k) evidently decreases as N and k increase, other things remaining constant. This result stems from the participant’s performance-based maximum-likelihood estimate of v, mle(v) being Nk . N

t i=1 N

i

Clearly, mle(v) increases with N, ti , along with k remaining coni=1

55 Formal Clinical Cognitive Science stant. Likewise, the estimated process-completion capacity of the current participant is raised in a parallel fashion with elevation in k (Equation (A.6) of Neufeld et al. 2002), as should be the case, considering the imputed meaning of k as a capacity-enhancing competence parameter. Finally, Equation (9) discloses that the interplay between k and k is comparatively complex. Computer computations indicate that changes in E(Tn|{*}; k) with elevation in k can be nonmonotone, decreasing and then increasing. The decrease is understandable in light of the inﬂuence of k on mle(v), considered above. Note, however, that, as k goes up, so does the process magnitude, with respect to the number of stages involved. Evidently, effects of the latter ultimately surpass those on mle(v), eventuating in a movement upward of the Bayesian posterior moment. Apropos the current thrust regarding k, such eventual elevation can, interestingly, be replaced with continuing decline if k is made increasingly large. This result again should be observed if k represents a competence parameter, whose inﬂuence can override that of a progressive increase in task load k on E(Tn|{*}; k).

interplay with capacity utilization Unlike the Erlang or general gamma distributions, but similar to the Weibull distribution (discussed above), the nonhomogeneous Poisson distribution speciﬁes a continuous change in processing rate with time. This dynamic rate is vt1, 0. If > 1, the rate continuously increases; if < 1, there is a continuous decline in rate, reﬂecting, say, a fatigue effect. The parameter is, therefore, designated a capacity-application efﬁciency parameter. Metaphorically, it signiﬁes the exploitation of capacity dealt to the processing system by the “capacity-resource pool” (gammar,k-distributed v). It is apparent that the Weibull distribution can be obtained from the nonhomogeneous Poisson by setting k = 1 and taking v to the power of . Releasing these constraints in the study of k, however, affords certain insights stemming from the relation of its effects to those of . As the rate parameter of the Erlang distribution is now replaced with vt1, f(t|v) = ((vt–1)t) k–1 –1

(k – 1)!

vt–1e–(vt )t.

56 modeling complex systems The coefﬁcient provides for

0 f (t|v)dt = 1.0. 00

Consequently,

E(T n ) = 0

00

=

0

00

(rv) k–1 –rv (vt) k–1vt–1e –vt tn re dt dv (k) (k – 1)!

rn/ (k – n/ )(k + n/ ) . (k – 1)!(k)

On inspecting the structure of this moment, it is apparent that the burden placed on k for retention of ﬁnite values is now alleviated, or, indeed, intensiﬁed (for > 1.0 and < 1.0, respectively), by a factor of 1/. Observe in passing that stress effects on moment magnitude are also modiﬁed according to in the term rn/. Similar to the escape of E(Tn) from ﬁnite values if k = n/, E(Tn) breaches ﬁnite bounds if = n/k. In this way, performer competence as conveyed by k and capacity-application efﬁciency as conveyed by reciprocate with respect to moment retention of ﬁnite values. A complementary angle on the relation of k to capacity utilization, , is afforded by a rigorous and robust index of “process-completion potential,” as articulated in contemporary cognitive science (Townsend & Ashby, 1983; Wenger & Townsend, 2000 with respect to the clinical-science implementations, see Neufeld, Townsend, & Jetté, in press). This index is aligned with E(Tn); it consists of 1n(F (t)), where F (t) is the process-latency distribution’s survivor function, or the complement of the cumulative probability-distribution function, 1 F(t) (see the ﬁrst section of Appendix C; also Figure 12 above and related text). The current interpretation of k implies effects on 1n(F (t)) that should closely resemble or mimic those of . The parameters k and should also play off one another in maintaining speciﬁc levels of 1n(F (t)). These associations are apparent in Figure 13, which plots the above index of process-completion potential as a function of k and for selected values of r, k, and t (formulas underlying Figure 13 are presented in the third section of Appendix C). The parameter k, then, behaves as it should as a capacity-supply competence index, as set against as a capacity-application efﬁciency index. It does so with respect to both E(Tn) and the allied expression of process-completion potential ln (F (t)).

57 Formal Clinical Cognitive Science

Figure 13. Values of process-completion potential 1n(F (t)) for the nonhomogeneous Poisson distribution, with rate parameter vt1 and v gammar,k distributed, for k varying from 1 to 3 and varying from 1 to 5, r = .03735, k = 18, t = 2.25.

convergence with allied mixingdistribution parameters The above developments surrounding construct validity for k specify that performer competence is a key contributor to capacity resources bearing on process transaction and that this shape parameter of gamma-distributed v is an index of such performer skill. There is nothing about this formulation that excludes other capacity mixing-distribution parameters from assuming a role like that of k. The shape parameter of related mixing distributions could, tenably, express performer competence, with properties like those of k. In other words, k need not have a corner on conveying performer ability; indeed, parameters that take on a similar role in other distributions and affect performance indicators in a similar fashion to k indicate that the derived analytic properties are not idiosyncratic to a gamma mixing distribution. The concept of capacity-infusing competence should transcend the shape parameter k of the gamma distribution as a particular quantitative instantiation. This version of analytic construct validity again is analogous to empirical convergent validity, where a trait evinces parallel values across divergent methods of measurement (e.g., Haynes, 2001).

58 modeling complex systems Table 5. Moments of Weibull and Erlang Process-Latency Base Distributions, with Capacity Parameters Either WeibullV, or Gammar,k Distributed. Base Distribution Mixing Distribution

Weibull

Gamma

Weibull; v Is Randomly Mixed

Weibull; v Is Randomly Mixed

Erlang; v Is Randomly Mixed

Vn (n/ + 1)(1 – n/B) Vn/ (n/ + 1), (1 – n/B)) Vn (k + n)(1 – n/ B) (k – 1)! rn (n/ + 1)(k – n) (k)

rn/ (n/ + 1)(k – n / ) (k)

rn (k + n – 1)!(k – n) (k – 1)!(k)

A notable example whose distributed variate, like gamma’s, ranges from 0 to , with intensity parameter v and shape parameter , is the Weibull distribution, described above. As the Weibull now serves as a mixing distribution, these parameters are designated V and B, respectively. The second section of Appendix C shows that, whether they are mixing or base distributions, the gamma and Weibull are nested in the generalized gamma distribution (additionally allowing k of the generalized gamma to assume the status of k as a continuous variable). An exhaustive analysis of the two distributions’ effects seems unnecessary. Sufﬁcient to make the point are deﬁnitions of moment ﬁniteness in relation to k and B. Selected process-latency base distributions now include the Erlang and the Weibull itself. In the latter case, the intensity (capacity) parameter v alone and, then, v exponentiated by , v, constitute the gamma- or Weibull-mixed random variate. The respective moments are presented in Table 5. Isolating on the far-right-hand gamma function, (x), in each case, maintaining moments within ﬁnite bounds evidently requires identical values of k and B from gamma and Weibull mixing distributions, respectively. Convergently, then, these equivalent thresholds of k and B add to the constellation of analytic properties endorsing the construct validity of k. Finally, similar observations attend the discrete-variable analogue of the mixture involving the Erlang base whose parameter v is gammar,k distributed. In contrast to the Erlang base distribution, whose functions are those of continuous time t, the Pascal distribu-

59 Formal Clinical Cognitive Science tion deﬁnes the probability of completing a process comprising k stages on the jth trial, j = k, k + 1, . . . . Its parameters are k and p, where p denotes the probability of successful stage completion on a trial (Patil & Joshi, 1968). The parameter p is, thus, the discrete-variable analogue of stage-completion capacity and, in that way, maps onto v of the Erlang distribution. The Pascal-gamma mixture deﬁnes p as eu, where u is gamma distributed, with parameters r and k. Clearly, the directions in which r and k affect the distribution of the base processes’ capacity parameter are now reversed, suggesting that moment ﬁniteness may depend on critical values of r. Accordingly, the mean and variance of process-completion trials, for example, escape ﬁnite bounds if r = 1, and the variance does so if r = 2.

summary statistics of prototypical performance samples The ﬁnal aspect of analytic construct validity considered here pertains to statistical summaries of latency distributions for representative individuals, as related to k for the class to which these participants belong (for discussions of empirical estimates of classcomposing homogeneity, see, e.g., Carter, Neufeld, & Benn, 1998; Neufeld & McCarty, 1994). The analysis affords another glimpse of k as indicative of the performance-ability level required to contain the statistical summaries of performance within ﬁnite bounds. Each formally prescribed hypothetical participant is representative, inasmuch as the amount appropriated by the capacity parameter v of his or her Erlang distribution generates the mean or expected value of the summary statistic for the represented class. The ﬁrst summary statistic is the mean process latency, which is k/v. With v being gammar,k distributed, the grand mean taken across all participants is kr/(k 1). The result shows that the value of v for an individual whose own mean latency equals that of the grand mean is, necessarily, (k 1)/r, which is the modal value of the gammar,k distribution. Similar observations apply to the individual whose standard deviation in latency equals the expected standard deviation, or k½r/(k 1). Finite values of these summary statistics demand that k > 1.0.

60 modeling complex systems In like fashion, v for an individual whose variance in process latencies is equal to that of the expected value—the average variance—is ((k 1)(k 2))<1/2>/r. The result follows because the latency variance, given v, is k/v2 and the expected or mean variance is kr2/ ((k 1)(k 2)). (Note that the theoretically prescribed value of v for the expected standard deviation and that for the expected variance are not identical, as E((SD(T|v))2) = E(Var(T|v)) (E(SD(T|v)))2.) Here, ﬁnite values of the summary statistic require that k > 2.

complementing empirical support The above exposition of the analytic side of construct validity should dovetail with empirical support in its varied forms (enumerated in Haynes & O’Brien, 2000). A tactic of choice, when it comes to formal-model parameters, involves selective sensitivity to experimental manipulations or to variation in pertinent organismic, participant variables. Concrete illustrations from the literature include those addressed to the understanding of parameters symbolizing determinants of retroactive interference with traces of memorized items (Chechile, 1987). They address, as well, parameters linked to the detection and retention of informational sources (Riefer, Knapp, Batchelder, Bamber, & Manifold, 2002). As for the present parameters, k has been released as a free parameter to express variations in cognitive (visual-scanning) performance across groups differing in stress proneness (Hamilton, 1980). Model predictions of empirical performance conﬁgurations failed to compete with those where r was, instead, freed to vary with stress susceptibility (Neufeld & Carter, 2000). In other instances (Neufeld, 2002), variation in k effectively expressed effects of practice, unlike a competing parameter considered to indicate the number of stimulus features to be processed during task trials (Weinstein, 1974).

in summary The method of illustration was used to motivate and exposit engagement with analytic aspects of construct validity as it related to interpretation of model parameters. Focus centered on support for

61 Formal Clinical Cognitive Science espoused meaning of a parameter bearing on the capacity-resource pool of a cognitive process (e.g., stimulus encoding). The capacityresource pool was operationalized as a stochastic distribution of a process-model parameter conveying speed of process transaction. Construct validity of the imputed substantive signiﬁcance of the capacity-affecting parameter, k, in turn, took the form of its mathematically derived analytic properties. The stochastic distribution of capacity values was gamma, with parameters r and k. Increasing values of these parameters shifted the capacity distribution downward and upward, respectively. The shape parameter, k, was tendered as an indicator of process-completion skill or competence. Evidence constellated for the posited interpretation of k included substantively meaningful properties transcending speciﬁc processlatency (base) distributions. Contrasting structures of base-distribution models were employed, tenably standing for diverse types of cognitive-task operations. Bayesian extensions lent additional support to educed inferences. Consequences of varying k interfaced with the status of other processing-system parameters in ways to be expected with the interpretation accorded k. Convergent support was also endowed by alternate-distribution parameters, whose effects on task performance should resemble those of k, given that they occupy a similar mathematically prescribed role. Next, performance of hypothetical participants, whose distributions of process latencies typiﬁed those of their embedding class, were inspected for workings of k. Performance properties (summary-statistic functions of k) of these idealized participants again conformed to the claimed substantive signiﬁcance of this parameter. Finally, the compatibility of analytic results with empirical support for construct validity was described.

Stochastic Modeling and the Psychometric-Artifact Issue in the Study of Differential Cognitive Deﬁcit Stochastic dynamic modeling of cognitive performance can speak to certain long-standing measure-theoretic predicaments accompanying the study of differential cognitive deﬁcit in schizophrenia and

62 modeling complex systems other disorders (Carter & Neufeld, 1999; Neufeld et al., 2002). At the forefront is the “psychometric-artifact problem,” which states that the apparent separation between pathological and control groups on a given measure conﬂates differences on the addressed cognitive faculty with the measure’s psychometric properties. A measure can, therefore, spuriously indicate greater deﬁcit on the faculty it putatively taps because it possesses greater psychometric precision than measures directed to other faculties (Chapman & Chapman, 1973, 2001). Advocated redress of the problem has entailed prematching measures’ psychometric properties using a standardization group composed of nonpatients who vary widely on the faculties being examined. Matching should, purportedly, incorporate variance in the measures’ observed scores and in their respective reliabilities. In addition, simplicity of factorial structure is recommended (Strauss, 2001; cf. Neufeld, 1984; Neufeld & Broga, 1981). The issue and its endorsed solution continue to elicit considerable adherence (e.g., Brenner, Wilt, Lysaker, Koyfman, & O’Donnell, 2003). The proposed solution of psychometric-property matching has also received substantial challenge, and alternatives have been articulated (Carbotte, 1978; Knight & Silverstein, 2001; Neufeld, 1984, 1989; Neufeld & Broga, 1977, 1981; Neufeld et al., 2002). The issue is dissected here by translating it into simple but arguably sufﬁcient measure-theoretic terms and then evaluating effects of typical psychometric-adjustment practices. Boiled down to essentials, the argument has, for some three decades, been more or less as follows. More psychometrically precise measures will have more statistical power for detecting control/ pathological-group separation and, thus, be differentially favored over psychometrically inferior measures in indicating statistically signiﬁcant deﬁcit or disproportionate deﬁcit (expressed as an interaction with corresponding groupwise simple main effects). Comparative effect-size estimates should, of course, conform to the above intermeasure differences in statistical properties, other values (notably, sample size) being equal. Note that, for two groups, statistical power for detecting a difference on a selected measure is an increasing function of the squared noncentrality parameter n(1/4d2), where d refers to Cohen’s (1988) d, or the standardized population-group difference in means (1 2)/ . In this expression, is the common within-group standard devia-

63 Formal Clinical Cognitive Science tion, and n is the common group sample size. The expression 1/4(1 2)2 conveys between-group effects and is, tenably, composed of some amalgam, ﬁrst, of those that are unique to group membership and constant among its members (corresponding to taxonic variables in Meehl & Golden, 1982), denoted here as , and, second, of those that represent between-group inequalities with respect to latent variables that also generate true-score variance across participants, within groups . The net contribution to the numerator of 1/4d2 from the latter source is denoted c, where the scalar c increases as sources of between-group separation on the measure overlap with those of within-group, interparticipant variation, 0 c 1.0. The expression 1/4(1 2)2 can, thus, be replaced with c + . Finally, duly replacing 2 with + e, where the term e represents measurement-error variance, makes 1/4d2 itself become (c + )/( + e).

(10)

Although not essential, for mathematical tractability and the purpose of illustration, it is assumed that none of c,, e, and are functions of one another. Turning to the issue of psychometric adjustment using standardization-group data, it is apparent from Equation (10) that the nub of between-group inequality, c and , is bypassed; within-standardization-group measure reliability is equal to /( + e), and observedscore variance is equal to + e. Interestingly, by Equation (10), where all between-group variance is conveyed by , > 0, and c = 0, 1/4d2 can, evidently, be increased with reduction in . That is to say, the measure becomes less group discriminating as its standardization-group psychometric precision goes up. In fact, the effects of reducing are unveiled by differentiating (10) with respect to , the result being (ce )/( + e)2. Clearly, increasing —and, along the way, increasing measure reliability, /( + e)—has adverse effects on power if ce < and favorable effects if ce > . Thus, the consequences for d2 of varying are complex, and, in fact, unknowable from standardization-group variance, because the latter is mute about the structure of group differences composed of c and . Of course, decreasing e invariably increases reliability as well as d2, as would be expected, and as is obvious from Equation (10). But c and remain unknown, leaving psychometric matching to grope blindly when it comes to amounts by which e

64 modeling complex systems should be altered (say, through increased accuracy of instrumentation or preaggregating measurements; see, e.g., Neufeld & Gardner, 1990). The upshot is that equalizing a pair of measures on and e has no problem whatsoever in leaving them grossly unequal in their group-discriminating power.8 Empirical examples instantiating the theoretical dissociation between control/pathological-group discriminability and classic psychometric properties have been amply documented elsewhere (Knight & Silverstein, 2001; Neufeld, 1984, 1989; Neufeld & Broga, 1981). These instantiations need not be recapitulated here. More productively, it is shown that Equation (10) can be superposed over and against a stochastic-modeling formulation. The conversion metamorphoses the equation and arguably leaves behind as vestigial the statistical-property issue.

a stochastic-modeling translation of the statistical-property issue The terms of Equation (10) can be mapped onto substantive stochastic-modeling expressions. Those of the dual-parameter mixture model are invoked in the present instance, but the resulting observations transcend the immediate case. Each translated term potentially becomes a model prediction of correspondingly partitioned data. Statistical power considerations now pertain to rejection of a nontrivial competing model (cf. Cohen, 1988, chap. 7, pp. 16–17). More generally, ushered in is the substantial methodology of model selection. Included are procedures for evaluating the empirical ﬁt, and the comparative ﬁt, of model predictions that go beyond conventional goodness-of-ﬁt statistical-signiﬁcance tests (e.g., Myung, Forster, & Browne, 2000; Wagenmakers & Waldorp, 2006). Along with accessing advances in model testing and reﬁnement, the translation of Equation (10) breathes content into its otherwise substantively desiccated terms. The term e in Equation (10) signiﬁes “psychometric measurement error,” taking the form of random variation in participants’ empirical values across observational trials. The current mixture model, in turn, quantiﬁes participantwise intertrial variance in latencies as the expected or mean conditional variance, given k and v:

65 Formal Clinical Cognitive Science E(Var(T|k ∩ v)m

1,2

,r,k

).

Here, m1,2 signiﬁes that the expected variance is taken across both control and pathological groups, which, for simplicity and in keeping with the design depicted in Figure 6 above, differ only with respect to m. The actual amount prescribed by the mixture model is (r2m1,2)/((k – 1)(k – 2)).

(11)

The model counterpart of is the variance in the conditional mean or expected latency, given k and v: Var(E(T|k ∩ v)m

1,2

,r,k

)=

m1,2r 2(k – 1 + m 1,2) . (k – 1)2(k – 2)

(12)

Both the expected conditional variance and the variance in the conditional expectations (discussed above) provide for within-group random variation in v and k among participants. Model analogues of c and cast k as Poissonm mixed and ﬁxed parameter, respectively. In the ﬁrst case, k is randomly distributed across participants within groups. The distributions themselves are separated according to the values of m1,2. For example, m1 may be m of Figure 6, and m2 may be m + g. The differences in m1 and m2 are, thus, afﬁliated with c of Equation (10) above. The scaled value of , c, therefore has a model representation as the following function of the between-group difference in means:

( (

(m r)

(m r)

) )

2

1 2 1/4 (k –1) – (k –1) .

(13a)

The analogous value for k as ﬁxed within groups is (k1r)

(k2r)

2

1/4 (k –1) (k –1) . –

(13b)

Equations (11)–(13a)/(13b) together embody twelve model predictions of empirical terms. The data of each diagnostic group, under each experimental condition (encoding load, of Figure 6), can be assembled into the expected conditional variance, variance in conditional expectations, and group mean. Parameter estimates, obtained, for example, through moment-ﬁtting procedures, are ﬁve in number according to the prevailing model design. It is apparent that each parameter is called on simultaneously to participate in empirical values whose sources range from within-participant variation to intergroup separation. Among other considerations, severity of model

66 modeling complex systems testing increases with the ratio of empirical observations to parameter-set size and the heterogeneity of conditions to which the model applies. Inferential validity regarding proposed dysfunction, then, ultimately rests on the rigor of the model diagnostics that are exerted. Predictions desirably include qualitative properties of empirical conﬁgurations. For example, the present model prescribes additive effects on means of diagnostic groups and experimental conditions, which is tantamount to a predicted second-order difference (see above) being equal to 0 (e.g., Neufeld et al., 1993; Neufeld & Williamson, 1996). Augmenting signiﬁcance tests for goodness of ﬁt involving predicted distribution moments (crafted from Equations (11)–(13)) can also be constructed (Neufeld, Vollick, et al., in press). Further tests are available from distribution properties, which, in the case of dynamic stochastic modeling, are functions of time t (García Pérez, 1994, 2000; Van Zandt, 2000). These properties ﬂow from the very design spawning the above distributional moments (e.g., density functions, illustrated in Figures 10 and 11 above). Moreover, Bayesian extensions stand to further proliferate model predictions. Posterior density and distribution functions, given an individual sample of latencies (e.g., Figures 8–9 and 12 above), once more are engineered by the stochastically modeled system giving rise to Bayesian-posterior distributional moments themselves (illustrated in Neufeld et al., 2002).

Concluding Comments Placing clinical cognitive science on a decidedly formal platform can unearth productive routes of study and reveal otherwise hidden substantive nuances. Brokering the clinical ramiﬁcations of contemporary cognitive science makes for a potentially seamless transition between these two domains of investigation. Overarching model designs can be constructed so as to mediate group-level ﬁndings to individual differences in their expression among constituent members. Developments self-disclose implications for evaluation of the cognitive side of treatment-program efﬁcacy, and progress of an individual’s own response to treatment, with respect to cognitive functioning. Also disclosed are assumptive underpinnings of the ad-

67 Formal Clinical Cognitive Science umbrated assessment technology, along with lingering ambiguities awaiting resolution. Formal clinical cognitive science uniquely convenes a quantitative infrastructure for assessing construct validity of meaning assigned to its parameters and, potentially, other properties. It also engages measurement and statistical challenges to inferences about cognitive dysfunction. The current offering at minimum represents a beachhead from which, it is to be hoped, a stronghold of formal clinical cognitive science and assessment methodology may be consolidated.

Appendix A: Bayesian Platform for the Current Developments Expansion of {Pr(g|{*})} and [w( |{*}, g)] of Equation (5) results in the following expression: G G ] { }[ ( 1/[ g=1 Pr(g) ujd({*}|g)] Pr(g) ujd({*}|g) w(|g) cjd({*}|)/ ujd({*}|g) ).

g=1

In this expression, Pr(g) is the current base rate of group g, corresponding to the relative frequency of members of g in the participant sample at large; ujd({*}|g) is the unconditional joint density of the obtained latency data set, given group g, meaning the combined probability density of {t1, t2, . . . , tN}, given group g, all values of

considered; w( |g) is the probability (density) of , given group g; and cjd({*}| ) is the conditional joint density of the obtained latency data set, given the parameter value . Note that the above expansion reveals two sets of Bayesian priors. One comprises the groupwise mixing distribution of . The other comprises the discrete, ﬁnite mixing probabilities of the groupwise priors being operative, corresponding to the base rates Pr(g). The latter is a “meta-prior” of this “compound Bayesian formulation.” Note as well that the densities of , including that of any speciﬁc value , are determined by the group at hand’s prior distribution of . Where becomes a set of parameters (e.g., v and k; see the text), each with its own independent group-related mixing distribution, one or another member of the set may be considered separately. Separate consideration occurs, for example, when constructing a posterior parameter-mixing distribution, following the acquisition of a

68 modeling complex systems performance sample {*}, as exempliﬁed in Figure 7 above. Computations entail aggregating over values of the set-aside parameter(s) according to its (their) group-prescribed prior probabilities (densities) while addressing speciﬁc values of the parameter that has been singled out (“marginals”), also according to its group-speciﬁed prior probabilities (densities). Computational details undergirding the present developments have been presented in Neufeld et al. (2002).

Appendix B: Estimation of Base Rates of Symptom Severities The joint probability of an array of J classiﬁcations with respect to G groups, by Equation (5) and observations in Appendix A, is J

J

G

Pr(g|{*}) j = Pr(g) ujd({*}|g) j / Pr(g) ujd({*}|g) j .

j=1 j=1 g=1

(B.1)

This product can be show to be a reduction of the joint multinomial probability of the J observed classiﬁcations under the model Pr(g|{*}), each probability being an expression of the multinomial likelihood of classiﬁcation j (see, e.g., García Pérez, 1994; Riefer & Batchelder, 1988). Maximum-likelihood estimates of the Pr(g), g = 1, 2, . . . , G 1 G–1 (recalling that Pr(G 1) = 1 – Pr(g) ), are obtained by differentiatg=1 ing (B.1) with respect to each term in question, setting each derivative to 0, and simultaneously solving the G 1 equations in the G 1 unknowns. A numerical-search algorithm will be required to replace the analytic solutions as computations become unwieldy with increasing J and/or G. Note that Equation (B.1) beguilingly invites a 2 test for empirical goodness of ﬁt of the prevailing model of Pr(g|{*}). Speciﬁcally, 2 ln(B.1), which here can be shown to equal 2 ln(B.1)/1.0, is tantamount to 2 ln(multinomial-likelihood ratio), an expression that stands to be distributed as 2 with its associated degrees of freedom (cf. Carter & Neufeld, 1999). In fact, this expression is dubiously distributed as 2df=J(G1) in the present case because the current data array suffers from the problem of “extreme sparseness of cell-wise observations” (see Delucchi, 1993; Tollenaar & Mooijaart, 2003).

69 Formal Clinical Cognitive Science

Appendix C: Finite Values of Distribution Moments values of distribution moments The nth moment of a distribution of latencies t, including those of the mixtures discussed in the text, will be inﬁnite if its integral does not converge:

0 f (t) tn dt = 00 . 00

Equivalently,

(– 1) nE(T n) = dm/d n MT ( )| = 0 = 00 .

Here, MT( ) is the moment-generating function

0 f (t) e– tdt, 00

whose nth derivative with respect to encounters a singularity at

= 0. E(Tn) can also be written as – n 0 (1 – F(t))tn–1dt = n 0 F (t)tn–1dt, 00

00

where F(t) is the cumulative probability-distribution function (see Figure 12 above and related text) and 1 F(t) = F (t) is the survivor function. This version of E(Tn) relates distribution moments, and their availability as ﬁnite values, to “process-completion potential.” The latter is deﬁned as – 0 f(t)/F(t)dt, (C.1) where f(t)/F (t) is the hazard function H(t), indicating the conditional rate of process completion at time t, given noncompletion by – t. Because F (t) = exp( – 0 (H(t)dt), (C.1) is readily seen to be simply 1n(F (t)) (see the discourses of Townsend & Ashby, 1983; Wenger & Townsend, 2000). This expression of process-completion potential, and its association with E(Tn), is called on in developments presented in the text under “Interplay with Capacity Utilization” and the third section of Appendix C below. t

t

moments of selected base distributions mixed on a gamma-distributed “capacity” (intensity) parameter Mixture-model latency distributions considered in this section are composed of alternate process-latency base distributions whose “ca-

70 modeling complex systems pacity” (distributional intensity or scale) parameters are randomly distributed according to a gamma distribution whose own intensity or scale parameter is denoted r and whose shape parameter is denoted k. Moments are computed for the case where k > n, n being the order of the moment, and where k = n. Three of the base distributions, the exponential, Erlang, and Weibull, are nested in the generalized gamma distribution, whose probability-density function is

(vt va)k1/((k))v exp((vt va)).

(C.2)

The probability-density function for the Erlang distribution is obtained from (C.2) by setting equal to 1.0 and a equal to 0. The density function for the Weibull distribution results when k = 1.0 and a = 0. The exponential distribution, in turn, is nested in both the Erlang and the Weibull distributions, as its density function is derived from (C.2) when k = = 1.0 and a = 0. Loosely speaking, each base distribution has a conjugate prior, in that both prior and base entail exponentials, making for mathematical tractability (see, e.g., Geddes & Scott, 1989). The probability-density function of mixture-model latencies f(t) for the Erlang base distribution, with its capacity (also rate) parameter v being distributed as gammar,k, is Then f(t)tndt = E(T n)

(

rktk–1 (k + k) . (k)(k – 1)! (r + t) k + k

rn (k + n – 1)!(k – n) (k)(k – 1)!

=

r k (k + n – 1)! ln(r + t) (k)(k – 1)!

]

< 00 ,

if k > n,

= 00

if k = n,

t= 00

t= 0

,

The next base distribution is the compound Poisson (e.g., Feller, 1966; Ross, 1996), of which the special case at hand is composed of an Erlang distribution whose stage (distributional-shape) parameter k is Poisson distributed with parameter m (see Figures 5 and 6, as well as Equation (8), above and the related text). The Erlang rate parameter of this base distribution, in turn, is gamma distributed, with parameters r and k. A concise method of obtaining the moments of this second-order mixture capitalizes on its moment-generating function

71 Formal Clinical Cognitive Science MT( ); the nth-order derivative of the latter, evaluated at = 0 and scaled by (1)n, yields the nth-order moment (for an especially lucid and still current exposition of the moment-generating function, see Kenny & Keeping, 1951). Proceeding accordingly, MT ( ) = 0 f (v)(MT ( )|v)dv 00

=

k= k e –m (MT ( )|k ∩ v)dv 0 m 0 f (v) k= k!

=

e –m (MTk ( ) – e –m )|v dv 0 m 0 f (v) k= k!

=

0 f (v) [(e m(M () – 1) – e –m|v]dv,

00

00

[

00

k= 00

]

k

1

00

T1

where MT ( ) is the moment-generating function for each of the i k-stage intercompletion times, or v/(v + ) for the Erlang distribution with rate parameter v. As the k intercompletion times are identically and independently distributed for a given value of v, MT ( )|(k ∩ v = M kT( )|v . Note that 1

Thus,

MT ( )|(k = 0 ) = 0.

MT ( ) =

(rv) k–1 –rv+m(v/(v+ )–1) re dv – e –m , (k)

0

00

E(T ) = (– 1) d MT ( )/d | = 0 . and The nth-order moment of this second-order compound Poisson distribution will be inﬁnite if k = n, as follows: n

n n

E(T n) =

k= 00

n

k

m k! e–m E(T n|k)|

k=n

k= 0

k= k = m e –m. 00 – e –m. 00| k= 0 k! 00

k= n

(by results for the Erlang base distribution, presented above) = eme –m. 00 – e –m. 00| = 00 (1 – e –m)| k= n

k= n

= 00| . k= n

By similar reasoning, E(Tn) = |k=n also for the general gamma base distribution (discussed below). Turning to the general gamma base distribution, unlike that of the Erlang, the rate parameter is stage speciﬁc. In the present implementation, the rate for stage i is civ, ci > 0, i = 1, 2, . . . , k. For this distribution,

72 modeling complex systems MT ( ) = 0 f(v)MT ( )|vdv 00

= 0

00

(rv) k–1 –rv k Cik civ re dv, i=1 c v + (k) i

where Cik is deﬁned in Table 4 of the text:

E(T n) = (– 1) nd n/d nMT ( )| = 0 k

= Cikci n! rn i=1

(k – n). (k)

For the special case, where ci = (k (i 1)), the density function f(t) can be computed, from developments supplied by Morrison (1979), as i k

i=1

(– 1)

+1

(ki ) r ik . k

(r + it) k + 1

From here, it is a simple matter to obtain E(T n) = 0 f(t) tn dt 00

=

(

()

k (k – n) (– 1) i + 1 k rn n! i (k)i n i=1

()

k ir kn! ln(r + it) (– 1) i + 1 k i (k)i n + 1 i=1

]

< 00 ,

if k > n

= 00

if k = n.

t= 00

t= 0

The probability-density function for the Weibull distribution is

v(vt)1 exp((vt)), whereby E(Tn|v) = (n/ + 1)/vn. E(Tn), in turn, is readily available as n (n/ + 1) (k – n) , 0 f(v)E(T n|v)dv = r 00

(k)

which again is severed from ﬁnite values if k = n.

formulas underlying figure

13

For the nonhomogeneous Poisson process, with dynamic rate vt1, v having been gammar,k distributed,

73 Formal Clinical Cognitive Science f(t) =

and

r k(k + k)t k– 1 , (k – 1 )!(k) (t + r ) k+ k

k– 1 – F(t) = j= 0

(j + k )rktj . j! (k)(t + r ) j + k

The hazard function H(T) = f(t)/F (t) enters into the index of process-completion potential as follows: 0 H(t)dt. Happily, this integral is available as 1n(F (t)), especially considering the possibility of complex f(t) and F (t), such as those above (see the ﬁrst section of Appendix C). t

Notes Sources of support for this research include operating grants from the Social Sciences and Humanities Research Council of Canada and the Workplace Safety and Insurance Board and operating grants and a New Emerging Teams grant from the Canadian Institutes of Health Research. Appreciation is expressed to the University College of the Fraser Valley, the Department of Psychology, University of British Columbia, Kenneth D. Craig, sabbaticant host, and to Ruth and Douglas Barr, private citizens, for providing facilities during the writing of this manuscript. 1. A parameter is a variable in a mathematical expression whose values can change without altering the structure of the expression in which it is embedded (cf. Borowski & Borwein, 1989, p. 435). 2. Consider, as a representative example, numerical vs. symbolic integration, a special case being numerical solutions to differential equations composing nonlinear dynamic (“chaos-theoretic”) systems, as contrasted with desired (but most often intractable) explicit solutions of the “initialvalue problem” (Koçak, 1989). 3. Formal statements foster place keeping regarding the current state of progress in the problem area. They make clear what has been formulated, whether new formal proposals differ, and, if so, in what respects. Blind alleys are ﬂagged and promising leads discerned. The place of precedent is usually more fuzzy in the case of verbal statements. Even if they seem to rediscover preexisting ideas, subsequent proposals almost always escape indictment for ignoring antecedents and, if taken to task, usually appeal to allegedly important nuances of difference. 4. These explicit starting points can always be challenged, precisely because they are explicit. Informal theories escape because they are not forced to lay bare their assumptive moorings before commencing (cf. Bandura, 1984; Staddon, 1984). 5. Neuroconnectionist modeling, entailing computer simulations of task-transacting activations among neuronal units or unit modules (Rum-

74 modeling complex systems melhart, McClelland, & pdp Research Group, 1987), arguably has received a warmer welcome (including in the ﬁeld of clinical cognitive science) because of its face validity as a “brain metaphor” (notwithstanding certain challenges to biological plausibility) and because of commercial software expediting the use of simulation algorithms virtually since their introduction (McClelland, Rummelhart, & pdp Research Group, 1986; cf. Carter & Neufeld, in press). 6. Added to considerations of response properties of interest are those of their distributional properties (e.g., moments of performance latencies, including means, and intertrial variances, illustrated above; distribution functions, as exempliﬁed in Equation (2) above; and similar properties). The concept of target levels of inference, or levels of analysis, and associated boundaries of inference are underscored. Model properties and response parameters entailing task-performance latency, e.g., are deemed to express dynamic properties of operations occurring at other strata, including the algorithmic (neuroconnectionist) and the implementational (neurophysiological; Busemeyer & Townsend, 1993, pp. 443–445; Marr, 1982). The interstrata associations are isomorphic in the same way that a computer-machine language and a computer-programming language are coextensions of one another. Targeting one or the other stratum then becomes a matter of the nature of the focal problem. 7. The performance sample is represented as a set of observations, as opposed to either an ordered set or a response vector. The observations are deemed to be independent, and algebraic operations to which they are subjected do not depend on their ordering. 8. Although compelling in other respects to be sure, methodology developing from item-response theory (irt) can be shown not to resolve the above dilemma, either in its own right or via irt-informed measure composition (see, e.g., Santor & Ramsay, 1998).

References Aguirre, G. K., Zarahn, E., & D’Esposito, M. (1998). The variability of human, bold hemodynamic responses. Neuroimage, 8, 360–369. Averill, J. R. (1973). Personal control over aversive stimuli and its relationship to stress. Psychological Bulletin, 80, 286–303. Bacon, F. (1937). The Advancement of Learning. In Richard Foster Jones (Ed.), Essays, Advancement of Learning, New Atlantis, and Other Pieces (pp. 169– 235). New York: Odyssey. (Original work published 1605) Bandura, A. (1984). Representing personal determinants in causal structures. Psychological Review, 91, 508–511. Batchelder, W. H. (1998). Multinomial processing tree models and psychological assessment. Psychological Assessment, 10, 331–344. Batsell, R. R., Polking, J. C., Cramer, R. D., & Miller, C. M. (2003). Useful

75 Formal Clinical Cognitive Science mathematical relationships embedded in Tversky’s elimination by aspects model. Journal of Mathematical Psychology, 47, 538–544. Berger, J. O. (1985). Statistical decision theory and Bayesian analysis (2nd ed.). New York: Springer. Boksman, K., Théberge, J., Williamson, P., Drost, D., Malla, A., Densmore, M., Takhar, J., Pavlovsky, W., Menon, R., & Neufeld, R. W. J. (2005). A 4.0 Tesla fmri study of brain connectivity during word ﬂuency in ﬁrst episode schizophrenia. Schizophrenia Research, 75, 247–263. Bollen, K. A. (2002). Latent variables in psychology and the social sciences. Annual Review of Psychology, 53, 605–634. Borowski, E. J., & Borwein, J. M. (1989). Dictionary of Mathematics. New York: HarperCollins. Boynton, G. M., Engel, S. A., Glover, G. H., & Heeger, D. H. (1996). Linear systems analysis of fmri in human v1. Journal of Neuroscience, 16, 4207– 4221. Braithwaite, R. B. (1968). Scientiﬁc explanation. London: Cambridge University Press. Brenner, C. A., Wilt, A., Lysaker, P. H., Koyfman, A., & O’Donnell, B. F. (2003). Psychometrically matched visual-processing tasks in schizophrenia spectrum disorders. Journal of Abnormal Psychology, 112, 28–37. Brown, E., & Holmes, P. (2001). Modeling a simple choice task: Stochastic dynamics of mutually inhibitory neural groups. Stochastics and Dynamics, 1, 1–33. Bryk, A., & Raudenbush, S. W. (1992). Hierarchical linear models for social and behavioral research: Applications and data analysis methods. Newbury Park ca: Sage. Busemeyer, J. R. (1980). Importance of measurement theory, error theory, and experimental design for testing the signiﬁcance of interactions. Psychological Bulletin, 88, 237–244. Busemeyer, J. R., & Stout, J. C. (2002). A contribution of cognitive decision models to clinical assessment: Decomposing performance on the Bechara gambling task. Psychological Assessment, 14, 253–262. Busemeyer, J. R., & Townsend, J. T. (1993). Decision ﬁeld theory: A dynamiccognitive approach to decision making in an uncertain environment. Psychological Review, 100, 432–459. Busemeyer, J. R., & Wang, Y. (2000). Model comparisons and model selections based on generalization test methodology. Journal of Mathematical Psychology, 44, 171–189. Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56, 81–105. Carbotte, R. M. (1978). Converging operations or matched control tasks? Journal of Psychiatric Research, 14, 313–316. Carter, J. R., & Neufeld, R. W. J. (1999). Cognitive processing of multidimensional stimuli in schizophrenia: Formal modeling of judgment speed and content. Journal of Abnormal Psychology, 108, 633–654.

76 modeling complex systems Carter, J. R., & Neufeld, R. W. J. (in press). Cognitive processing of facial affect: Neuro-connectionist modeling of deviations in schizophrenia. Journal of Abnormal Psychology. Carter, J. R., Neufeld, R. W. J., & Benn, K. D. (1998). Application of process models in assessment psychology: Potential assets and challenges. Psychological Assessment, 10, 279–298. Casti, J. L. (1989). Alternate realities: Mathematical models of nature and man. New York: Wiley, 1989. Chapman, L. J., & Chapman, J. P. (1973). Problems in the measurement of cognitive deﬁcit. Psychological Bulletin, 79, 380–385. Chapman, L. J., & Chapman, J. P. (2001). Commentary on two articles concerning generalized and speciﬁc cognitive deﬁcits. Journal of Abnormal Psychology, 110, 31–39. Chechile, R. A. (1987). Trace susceptibility theory. Journal of Experimental Psychology, 116, 203–222. Chechile, R. A. (1998). A new method for estimating model parameters for multinomial data. Journal of Mathematical Psychology, 42, 432–471. Clark, J. M., & Paivio, A. (1989). Observational and theoretical terms in psychology: A cognitive perspective on scientiﬁc language. American Psychologist, 44, 500–512. Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale nj: Erlbaum. Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52, 281–302. Delucchi, K. L. (1993). On the use and misuse of chi-square. In G. Keren & C. Lewis (Eds.), A handbook for data analysis in the behavioral sciences: Statistical issues (pp. 295–320). Hillsdale nj: Erlbaum. Doob, J. L. (1953). Stochastic processes. New York: Wiley. Embretson, W. S. (1983). Construct validity: Construct representation versus nomothetic span. Psychological Bulletin, 93, 179–197. Evans, M., Hastings, N., & Peacock, B. (2000). Statistical distributions (3rd ed.). New York: Wiley. Feller, W. (1966). An introduction to probability theory and its applications (Vol. 2). New York: Wiley. Fokas, A. S., & Marinakis, V. (2006). The mathematics of the imaging techniques of meg, ct, pet and spect. International Journal of Bifurcation and Chaos, 16, 1671–1687. Frackowiak, R. S. J., Friston, K. J., Frith, C. D., Dolan, R. J., Price, C. J., Zeki, S., Ashburner, J., & Penny, W. (Eds.). (2004). Human brain function (2nd ed.). San Diego: Academic. Freedman, D. Z. (1993, November). Some beautiful equations of mathematical physics. Dirac Lecture, International Centre for Theoretical Physics, Trieste. Friston, K., Harrison, L., & Penny, W. (2003). Dynamic causal modeling. Neuroimage, 19, 1273–1302. García Pérez, M. A. (1994). Parameter estimation and goodness-of-ﬁt test-

77 Formal Clinical Cognitive Science ing in multinomial models. British Journal of Mathematical and Statistical Psychology, 47, 247–282. García Pérez, M. A. (2000). Exact ﬁnite-sample signiﬁcance and conﬁdence regions for goodness-of-ﬁt statistics in one-way multinomials. British Journal of Mathematical and Statistical Psychology, 53, 193–207. Geddes, K. O., & Scott, T. I. (1989). Recipes for classes of deﬁnite integrals involving exponentials and logarithms. In E. Kaltofen & S. M. Watt (Eds.), Computers and mathematics (pp. 192–201). New York: Springer. George, L., & Neufeld, R. W. J. (1984). Imagery and verbal aspects of schizophrenic informational-performance. British Journal of Clinical Psychology, 23, 9–18. Gilden, D. L. (2001). Cognitive emission of 1/f noise. Psychological Review, 108, 33–56. Glass, D. C., & Singer, J. E. (1972). Urban stress. New York: Academic. Gleick, J. (1988). Chaos: Making a new science. New York: Penguin. Gregson, R. G., & Pressing, J. (2000). Dynamic modelling. In J. T. Cacioppo, L. G. Tassinary, & G. Berntson (Eds.), Handbook of psychophysiology (Vol. 42, pp. 924–948). New York: Cambridge University Press. Hamilton, V. (1980). An information processing analysis of environmental stress and life crises. In I. G. Sarason & C. D. Spielberger (Eds.), Stress and anxiety (Vol. 7, pp. 13–30). Washington dc: Hemisphere. Harris, B. (1966). Theory of probability. Reading ma: Addison-Wesley. Haynes, S. N. (2001). Clinical applications of analogue behavioral observation: Dimensions of psychometric evaluation. Psychological Assessment, 13, 73–85. Haynes, S. N., & O’Brien, W. O. (2000). Principles of behavioral assessment: A functional approach to psychological assessment. New York: Plenum/ Klewer. Herrnstein, R. J. (1979). Derivatives of matching. Psychological Review, 86, 486–495. Highgate-Maynard, S., & Neufeld, R. W. J. (1986). Schizophrenic memorysearch performance involving nonverbal stimulus properties. Journal of Abnormal Psychology, 95, 67–73. Hintzman, D. L. (1991). Why are formal models useful in psychology? In W. E. Hockley & S. Lewandowsky (Eds.), Relating theory and data: Essays on human memory in honor of Bennet B. Murdoch (pp. 39–56). Hillsdale nj: Erlbaum. Hockley, W. E., & Murdock, B. B., Jr. (1987). A decision model for accuracy and response latency in recognition memory. Psychological Review, 94, 341–358. Huber, M. T., Braun, H. A., & Krieg, J. C. (1999). Consequences of deterministic and random dynamics for the course of affective disorders. Biological Psychiatry, 46, 256–262. Huettel, S. A., Song, A. W., & McCarthy, G. (2004). Functional magnetic resonance imaging. Sunderland ma: Sinauer. Johnson, N. L., Kotz, S., & Balakrishnan, N. (1994). Continuous univariate distributions (2nd ed., Vol. 1). New York: Wiley.

78 modeling complex systems Johnson, N. L., Kotz, S., & Balakrishnan, N. (1995). Continuous univariate distributions (2nd ed., Vol. 2). New York: Wiley. Kant, I. (1970). Metaphysical foundations of natural science (J. Ellington, Trans.). Indianapolis: Bobbs-Merrill. Karabatsos, G. (2006). Bayesian nonparametric model selection and model testing. Journal of Mathematical Psychology, 50(2), 123–148. Kenny, J. F., & Keeping, E. S. (1951). Mathematics of statistics (2nd ed.). New York: Van Nostrand. Kirk, R. E. (1994). Experimental design: Procedures for the behavioral sciences (3rd ed). Monterey ca: Brooks/Cole. Kline, M. (1985). Mathematics and the search for knowledge. Oxford: Oxford University Press. Kline, R. B. (1988). Structural equation modeling. New York: Guilford. Knight, R. A., & Silverstein, S. M. (2001). A process-oriented approach for averting confounds resulting from general performance deﬁciencies in schizophrenia. Journal of Abnormal Psychology, 110, 15–30. Koçak, H. (1989). Differential and difference equations through computer experiments (2nd ed., pp. 39–56). New York: Springer; Hillsdale nj: Erlbaum. Kuhn, T. S. (1962). The structure of scientiﬁc revolutions. Chicago: University of Chicago Press. Kukde, M. P., & Neufeld, R. W. J. (1994). Facial electromyographic measures distinguish covert coping from stress response to stimulus threat. Personality and Individual Differences, 16, 211–228. Lanius, R. A., Williamson, P. C., Densmore, M., Boksman, K., Neufeld, R. W. J., Gati, J. S., & Menon, R. S. (2004). The nature of traumatic memories: A 4-t fmri functional connectivity analysis. American Journal of Psychiatry, 161, 36–44. Lees, M. C., & Neufeld, R. W. J. (1999). Decision-theoretic aspects of stress arousal and coping propensity. Journal of Personality and Social Psychology, 77, 185–208. Link, S. W. (1982). Correcting response measures for guessing and partial information. Psychological Bulletin, 92, 469–486. Luce, R. D. (1986). Response times: Their role in inferring elementary mental organization. New York: Oxford University Press. Luke, D. A., & Homan, S. M. (1998). Time and change: Using survival analysis in clinical assessment and treatment evaluation. Psychological Assessment, 10, 360–378. MacCallum, R., & Ashby, F. G. (1986). Relationships between linear systems theory and covariance structure modeling. Journal of Mathematical Psychology, 30, 1–27. Marley, A. A. J. (1989). A random utility family that includes many of the “classical” models and has closed form choice probabilities and choice reaction times. British Journal of Mathematical and Statistical Psychology, 42, 13–36. Marley, A. A. J., & Colonius, H. (1992). The “horse race” random utility model for choice probabilities and reaction times, and its competing risks interpretation. Journal of Mathematical Psychology, 36, 1–20.

79 Formal Clinical Cognitive Science Marr, D. (1982). Vision. San Francisco: Freeman. McClelland, J., Rummelhart, D. E., & pdp Research Group. (1986). Parallel distributed processing: Explorations in the microstructure of cognition: Vol. 1. Foundations. Cambridge ma: mit Press. McFall, R. M., & Townsend, J. T. (1998). Foundations of psychological assessment: Implications for cognitive assessment in clinical science. Psychological Assessment, 10, 316–330. McFall, R. M., Townsend, J. T., & Viken, R. J. (1995). Diathesis stress model or “just so” story? Behavioral and Brain Sciences, 18, 565–566. McGill, W. J., & Gibbon, J. (1965). The general-gamma distribution and reaction times. Journal of Mathematical Psychology, 2, 1–18. Meehl, P. E. (1978). Theoretical risks and tabular asterisks: Sir Karl, Sir Ronald, and the slow progress of soft psychology. Journal of Consulting and Clinical Psychology, 46, 806–843. Meehl, P. E., & Golden, R. R. (1982). Taxometric methods. In P. C. Kendall & J. N. Butcher (Eds.), Handbook of research methods in clinical psychology (pp. 127–181). New York: Wiley. Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis. California: Sage. Morrison, D. G. (1979). An individual differences pure extinction process. Journal of Mathematical Psychology, 19, 307–315. Morrison, M. S., Neufeld, R. W. J., & Lefebvre, L. A. (1988). The economy of probabilistic stress: Interplay of controlling activity and threat reduction. British Journal of Mathematical and Statistical Psychology, 41, 155–177. Myung, I. J., Forster, M. R., & Browne, M. W. (Eds.). (2000). Special issue on model selection. Journal of Mathematical Psychology, 44(1). Neufeld, R. W. J. (1982). On decisional processes instigated by threat: Some possible implications for stress-related deviance. In R. W. J. Neufeld (Ed.), Psychological stress and psychopathology (pp. 240–270). New York: McGrawHill. Neufeld, R. W. J. (1984). Re: The incorrect application of traditional test discriminating power formulations to diagnostic-group studies. Journal of Nervous and Mental Disease, 172, 373–374. Neufeld, R. W. J. (1989). Methodological aspects of laboratory studies of stress. In R. W. J. Neufeld (Ed.), Advances in the investigation of psychological stress (pp. 71–132). New York: Wiley. Neufeld, R. W. J. (1991). Memory in paranoid schizophrenia. In P. Magaro (Ed.), The cognitive bases of mental disorders (Annual review of psychopathology, Vol. 1, pp. 231–261). Newbury Park ca: Sage. Neufeld, R. W. J. (1994). Theoretical stress and stress-proneness effects on information processing in light of mathematical models of stochastic processes (Research Bulletin No. 720). London: University of Western Ontario, Department of Psychology. Neufeld, R. W. J. (1996). Stochastic models of information processing under stress (Research Bulletin No. 734). London: University of Western Ontario, Department of Psychology.

80 modeling complex systems Neufeld, R. W. J. (1999a). Dynamic differentials of stress and coping. Psychological Review, 106, 385–397. Neufeld, R. W. J. (1999b, July). Quantitative schemata of decisional stress-control. Paper presented at the 32nd annual meeting of the Society for Mathematical Psychology, Santa Cruz. Neufeld, R. W. J. (2002, June). Stochastic modelling of stress effects on proofreading. Paper presented at the annual Summer Inter-Disciplinary Conference on Cognitive Science, Squamish bc. Neufeld, R. W. J., & Broga, M. I. (1977). Fallacy of the reliability-discriminability principle in research on differential cognitive deﬁcit (Research Bulletin No. 360. London: University of Western Ontario, Department of Psychology. Neufeld, R. W. J., & Broga, M. I. (1981). Evaluation of information-sequential aspects of schizophrenic performance, 2: Methodological considerations. Journal of Nervous and Mental Disease, 169, 569–579. Neufeld, R. W. J., & Carter, J. R. (2000). Formally deciphering effects of stress on cognitive performance. Paper presented at the 33rd annual meeting of the Society for Mathematical Psychology, Queen’s University, Kingston on. Neufeld, R. W. J., Carter, J. R., Boksman, K., Jetté, J., & Vollick, D. (2002). Application of stochastic modelling to group and individual differences in cognitive functioning. Psychological Assessment, 14, 279–298. Neufeld, R. W. J., & Gardner, R. C. (1990). Data aggregation in evaluating psychological constructs: Multivariate and logical-deductive considerations. Journal of Mathematical Psychology, 34, 276–296. Neufeld, R. W. J., & McCarty, T. (1994). A formal analysis of stressor and stress-proneness effects on basic information processing. British Journal of Mathematical and Statistical Psychology, 47, 193–226. Neufeld, R. W. J., & Mothersill, K. (1980). Stress as an irritant of psychopathology. In I. Sarason & C. D. Spielberger (Eds.), Stress and anxiety (Vol. 8, pp. 31–56). New York: Hemisphere. Neufeld, R. W. J., Townsend, J. T., & Jetté, J. (in press). Quantitative response time technology for measuring cognitive-processing capacity in clinical studies. In R. W. J. Neufeld (Ed.), Advances in clinical cognitive science: Formal modeling and assessment of processes and symptoms. Washington dc: American Psychological Association. Neufeld, R. W. J., Vollick, D., Carter, J. R., Boksman, K., Levy, L., George, L., & Jetté, J. (in press). A mathematical process account of group and individual differences in memory-search facilitative stimulus encoding, with application to schizophrenia. In R. W. J. Neufeld (Ed.), Advances in clinical cognitive science: Formal modeling and assessment of processes and symptoms. Washington dc: American Psychological Association. Neufeld, R. W. J., Vollick, D., & Highgate, S. (1993). Stochastic modelling of stimulus encoding and memory search in paranoid schizophrenia: Clinical and theoretical implications. In R. L. Cromwell & R. C. Snyder (Eds.), Schizophrenia: Origins, processes, treatment, and outcome: The Second Kansas Series in Clinical Psychology (pp. 176–196). Oxford: Oxford University Press.

81 Formal Clinical Cognitive Science Neufeld, R. W. J., & Williamson, P. (1996). Neuropsychological correlates of positive symptoms: Delusions and hallucinations. In C. Pantelis, H. E. Nelson, & T. R. E. Barnes (Eds.), Schizophrenia: A neuropsychological perspective (pp. 205–235). London: Wiley. Nicholson, I. R., & Neufeld, R. W. J. (1992). A dynamic vulnerability perspective on stress and schizophrenia. American Journal of Orthopsychiatry, 62, 117–130. Nicholson, I. R., & Neufeld, R. W. J. (1993). Classiﬁcation of the schizophrenias according to symptomatology: A two-factor model. Journal of Abnormal Psychology, 102, 259–270. Nosofsky, R. M. (1992). Similarity scaling and cognitive process models. Annual Review of Psychology, 43, 25–33. Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart & Winston. Paivio, A. (1975). Perceptual comparisons through the mind’s eye. Memory and Cognition, 3, 635–647. Paivio, A. (1986). Mental representations: A dual coding approach. New York: Oxford University Press. Parzen, E. (1962). Stochastic processes. San Francisco: Holden-Day. Patil, G. G., & Joshi, S. W. (1968). A dictionary and bibliography of discrete distributions. New York: Hafner. Piasecki, T. M., Jorenby, D. E., Smith, S. S., Fiore, M. C., & Baker, T. B. (2003). Smoking withdrawal dynamics: 1. Abstinence distress in lapsers and abstainers. Journal of Abnormal Psychology, 112, 3–13. Polkinghorne, J. (2003). The friendship of science and religion (Plenary Lecture: Veritas Forum). London: University of Western Ontario. Rappaport, A. (1983). Mathematical models in the social and behavioral sciences. New York: Wiley. Reiman, E. M., Lane, R. D., Van Petten, C., & Bandettini, P. A. (2000). Positron emission tomography and functional magnetic resonance imaging. In J. T. Cacioppo, L. G. Tassinary, & G. G. Bernston (Eds.), Handbook of psychophysiology (2nd ed., pp. 85–118). New York: Cambridge University Press. Riefer, D. M., & Batchelder, W. H. (1988). Multinomial modeling and the measurement of cognitive processes. Psychological Review, 95, 318–339. Riefer, D. M., Knapp, B., Batchelder, W. H., Bamber, D., & Manifold, V. (2002). Psychological Assessment, 14, 184–201. Roe, R. M., Busemeyer, J. R., & Townsend, J. T. (2001). Multialternative decision ﬁeld theory: A dynamic connectionist model of decision making. Psychological Review, 108, 370–392. Ross, S. M. (1996). Stochastic processes (2nd ed.). New York: Wiley. Rumelhart, D., McClelland, J., & pdp Research Group. (1987). Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 2). Cambridge ma: mit Press. Ruscio, J., & Ruscio, A. M. (2004). A conceptual and methodological checklist for conducting a taxometric investigation. Behavior Therapy, 35, 403–447. Santor, D. A., & Ramsay, J. O. (1998). Progress in the technology of measure-

82 modeling complex systems ment: Applications of item response models. Psychological Assessment, 10, 345–359. Schiffman, S. S., Reynolds, M. L., & Young, F. W. (1981). Introduction to multidimensional scaling: Theory, methods and applications. New York: Academic. Smith, P. L. (1995). Psychophysically principled models of visual simple reaction time. Psychological Review, 102, 567–593. Smolin, L. (2006). The trouble with physics: The rise of string theory, the fall of a science, and what comes next. Boston: Houghton Mifﬂin. Staddon, J. E. R. (1984). Social learning theory and the dynamics of interaction. Psychological Review, 91, 502–507. Staddon, J. E. R. (1991). “The distemper of learning . . .”: A review of S. B. Klein and R. R. Mowrer (Eds.) Contemporary learning theories: Instrumental conditioning theory and the impact of biological constraints on learning. Contemporary Psychology, 36, 506–507. Steiger, J. H., & Fouladi, R. T. (1997). Noncentrality interval estimation and the evaluation of statistical models. In L. L. Harlow, S. A. Mulaik, & J. H. Steiger (Eds.), What if there were no signiﬁcance tests? (pp. 221–257). Mahwah nj: Erlbaum. Sternberg, S. (1975). Memory scanning: New ﬁndings and current controversies. Quarterly Journal of Experimental Psychology, 27, 1–32. Strauss, M. E. (2001). Methodology for identifying speciﬁc psychological deﬁcits: Introduction to the special section. Journal of Abnormal Psychology, 110, 4–5. Takane, Y., & Sergent, J. (1983). Multidimensional scaling models for reaction times and same-different judgments. Psychometrika, 48, 393–423. Tollenaar, N., & Mooijaart, A. (2003). Type I errors and power of the parametric bootstrap goodness-of-ﬁt test: Full and limited information. British Journal of Mathematical and Statistical Psychology, 56, 271–288. Tomarken, A. J., & Baker, A. J. (2003). Introduction to the special section on structural equation modeling. Journal of Abnormal Psychology, 112, 523– 525. Townsend, J. T. (1984). Uncovering mental processes with factorial experiments. Journal of Mathematical Psychology, 28, 363–400. Townsend, J. T. (1990). Serial vs. parallel processing: Sometimes they look like Tweedledum and Tweedledee but they can (and should) be distinguished. Psychological Science, 1, 46–54. Townsend, J. T. (1994). A visual approach to nonlinear dynamics: Just cartoons or a serious pedagogical device? [Review of A visual introduction to dynamical systems theory for psychology, by Frederick David Abraham]. American Journal of Psychology, 107, 117–123. Townsend, J. T., & Ashby, F. G. (1978). Methods of modeling capacity in simple processing systems. In J. Castellan & F. Restle (Eds.), Cognitive theory (Vol. 3, 200–239). Hillsdale nj: Erlbaum. Townsend, J. T., & Ashby, F. G. (1983). Stochastic modelling of elementary psychological processes. Cambridge: Cambridge University Press. Townsend, J. T., & Fiﬁc, M. (2004). Parallel versus serial processing and indi-

83 Formal Clinical Cognitive Science vidual differences in high-speed search in human memory. Perception and Psychophysics, 66(6), 953–962. Townsend, J. T., Fiﬁc, M., & Neufeld, R. W. J. (in press). Assessment of mental architecture in clinical/cognitive research. In T. A. Treat, A. Kraut, & T. Baker (Eds.), Psychological Clinical Science: Recent Advances in Theory and Practice. Hillsdale nj: Erlbaum. Townsend, J. T., & Nozawa, G. (1995). Spatio-temporal properties of elementary perception: An investigation of parallel, serial, and coactive theories. Journal of Mathematical Psychology, 39, 321–359. Townsend, J. T., & Wenger, M. J. (2004a). The serial-parallel dilemma: A case study in a linkage of theory and method. Psychonomic Bulletin and Review, 11(3), 391–418. Townsend, J. T., & Wenger, M. J. (2004b). A theory of interactive parallel processing: New capacity measures and predictions for a response time inequality series. Psychological Review, 111, 1003–1035. Tversky, A. (1972a). Choice by elimination. Journal of Mathematical Psychology, 9, 341–367. Tversky, A. (1972b). Elimination by aspects: A theory of choice. Psychological Review, 79, 289–299. Van Zandt, T. (2000). How to ﬁt a response time distribution. Psychonomic Bulletin and Review, 7, 424–465. Vollick, D. N. (1994). Stochastic models of encoding-latency means and variances in paranoid schizophrenia. Unpublished doctoral dissertation, University of Western Ontario, Faculty of Graduate Studies. Vollick, D. N., & Neufeld, R. W. J. (2004). Stochastic modelling of encodinglatency means and variances in paranoid schizophrenia. Manuscript in preparation. Wagenmakers, E., & Waldorp, L. (Eds.). (2006). Special issue on model selection: Theoretical developments and applications. Journal of Mathematical Psychology, 50(2). Waller, N. G., & Meehl, P. E. (1998). Multivariate taxometric procedures: Distinguishing types from continua. Newbury Park ca: Sage. Weinstein, N. D. (1974). Effect of noise on intellectual performance. Journal of Applied Psychology, 59, 548–554. Wenger, M. J., & Townsend, J. T. (2000). Basic response time tools for studying general processing capacity in attention, perception, and cognition. Journal of General Psychology, 127, 67–99. Wiggins, J. S. (1973). Personality and prediction: Principles of personality assessment. Reading ma: Addison-Wesley. Wright, B. D. (1977). Solving measurement problems with the Rasch model. Journal of Educational Measurement, 14, 97–116.

A Dynamics-Oriented Approach to Psychopathology Wolfgang Tschacher and Zeno Kupper University of Bern

Theoretical Framework: Optimality in Self-Organizing Systems Psychopathology addresses disorders of the psyche. Researchers in this ﬁeld explore maladaptive cognition, emotion, and action, thereby starting from a number of different perspectives; such perspectives are usually given by the scientiﬁc disciplines from which such exploration was initiated: • Psychology focuses on cognitive and behavioral models of psychopathology. • Biology focuses on genetics and on neuronal processes associated with psychological dysfunction. • Pharmacology studies chemical compounds that act on the neuronal substrate. • Sociology identiﬁes social and societal factors that are associated with the disorders in individuals. Thus, psychopathology is a genuinely interdisciplinary ﬁeld, and practitioners in this ﬁeld (traditionally, psychiatrists) commonly apply a range of quite different instruments and interventions when they, for instance, combine psychotherapy with neuroleptic medication in a milieu-oriented ward of a psychiatric hospital.

86 modeling complex systems The perspective for psychopathology that we wish to put forward in this chapter is in itself interdisciplinary—of its two ingredients, one is dynamic systems theory, the other cognitive science. In this introductory section, we will outline our theoretical guidelines and suggest how the two main ingredients of our approach—cognitive science and dynamic systems theory—may be combined. We will argue that cognitive acts can be viewed as coherent patterns emerging spontaneously in a complex cognitive system. Pattern formation via self-organization is the cornerstone of this approach to psychopathology. Motivational processes are, consequently, deﬁned as those forces or gradients that “drive” the complex cognitive system. It is with respect to such motivational gradients that cognition is more or less “optimal.” Cognitive science is a ﬁeld to which neuroscientists, psychologists and psychiatrists, philosophers of mind, workers in artiﬁcial intelligence (ai), and many others contribute. This ﬁeld has gained momentum in the last decades especially because of advances in brain research (e.g., the use of brain-imaging techniques and the profound implications of those techniques for psychiatry) and because of the perennial questions of consciousness that have been receiving increased attention recently (Carter, 2002). Cognitive science is, thus, a natural foundation from which psychopathological states of cognitive functioning may be regarded. Among the current approaches in cognitive science, the distinction between the computationalist perspective and the dynamic approach has been debated intensely. The computationalist perspective in cognitive science elaborates on the theory of information processing and continues the tradition of “classic ai,” which is based on the physical-symbol-systems premise (Newell & Simon, 1972). According to this premise, the mind is regarded as a physically implemented symbol system. Consequently, mental processes are identical to symbol processing, which can, in principle, be simulated using digital computers. The major part of today’s cognitive psychology, as well as the cognitive-behavioral therapies currently in use in psychiatry and clinical psychology, is derived from computationalist theory. The dynamic approach to cognitive science (e.g., Clark, 1997; Pfeifer & Scheier, 1999; Tschacher & Dauwalder, 1999, 2003) has challenged this idea of cognition as computation. Bridging assumptions of dynamic systems approaches to cog-

87 Approach to Psychopathology nition have been identiﬁed: “ofﬂine” reasoning is continuous with “online” motor-control strategies in the sense that abstract cognition may be decoupled from behavior in an actual environment but still work with the same principles (Mechsner, Kerzel, Knoblich, & Prinz, 2001). There is agreement in the dynamics community that cognitive patterns are not provided by programs, as in the computationalist perspective, but are emergent properties of the complex brain/mind system. Thus, the colloquial distinction between software and hardware makes sense only with respect to computers, not with respect to naturally occurring cognition. By these assumptions cognitive science and dynamic systems theory have joined forces to formulate a non-Cartesian theory of cognition. Emergent properties are modeled in the mathematical framework of self-organization theory (Haken, 1983, 1996; Kelso, 1995; Nicolis & Prigogine, 1977), the core part of the theory of complex systems. Self-organization theory (synergetics) deals with systems composed of several or many components. By means of their interaction, these components can produce new qualitative features on macroscopic scales; such features are often directly observable as patterns. In other words, the emergence of new qualities is studied. The main question is whether there are general principles that govern the behavior of complex systems when qualitative changes occur. These situations are probably of particular interest. And it has, in fact, been shown that a large number of systems are accessible to unifying mathematical and conceptual approaches. Synergetics starts from the observation that the behavior of many systems is strongly determined by environmental conditions. These conditions may be divided into constant conditions or constraints (e.g., those constants of a physical environment that conﬁne behavioral systems) and further environmental conditions that “energize” or “drive” the systems (e.g., free energy in physical systems, task affordances in psychological systems). In the mathematical approach these latter environmental conditions are taken care of by control parameters. In many cases control parameters have the form of externally applied gradients. The general strategy of synergetics is as follows: It sets out from a state of a system that is already known under a certain control-parameter value. When one or several control parameters are changed, the system may become unstable. Then it may tend to leave its state and develop a new structure or behavior.

88 modeling complex systems

Figure 1. Bistable Gestalt array (Rubin, 1921). Either two (black) faces or a (white) vase is recognized (for details, see the text).

Synergetics shows that the behavior of the system close to instability points is described and determined by order parameters. According to synergetics, the—in general few—order parameters determine the behavior of the many individual components. This implies an enormous information compression because it sufﬁces to describe the order parameters instead of all the components. On the other hand, the individual components react on the order parameters and, in this way, even generate the order parameters. Thus, the relation between order parameters and components is based on circular causality. Quite often order parameters show very simple behavior, for instance, bi or multistability; that is, a system can acquire different states under the same external conditions. An example of visual perception is shown in Figure 1, Rubin’s (1921) well-known vase-face Gestalt, a bistable Gestalt array. In this ﬁgure two order parameters compete with each other: perception of the pattern face(s) or perception of the pattern vase. Both order parameters organize the components (the black-and-white details of the display) in a speciﬁc fashion. Perception of these patterns is mutually exclusive—it is impossible to perceive both at the same time. What are viewed as noses in the domain of one pattern become merely background as soon as a vase is perceived. Under neutral circumstances, transitions between the patterns occur frequently; such Gestalt ﬂips are analogous to phase transitions in dynamic systems theory. Moreover, control parameters such as task affordances can

89 Approach to Psychopathology

1 "chaos" 2 "order"

"chaos"

3

"chaos"

"order"

4 "order"

Figure 2. Sequence of Gestalt-like written words (left column, cells 1–4) that give rise to the perception of either chaos or order. Right column, depiction of respective potential landscape (for details, see the text). After Tschacher (1997).

modulate the emergence of perceived patterns. If, for example, Figure 1 is presented in a series of similar displays together with the instruction to determine the gender of faces in the pictures, the vase pattern is hardly ever perceived. A further essential property of pattern formation via self-organization can be illustrated with the help of Figure 2. Words of a somewhat ambiguous quality are presented (see the left column of Figure 2); in cell 1 the word chaos is quite reliably identiﬁed. Here again, the percept chaos emerges as the order parameter, the qualitative pattern recognized instantaneously in the stimulus of cell 1. The stability of the percept chaos resulting from this stimulus is represented by the potential sink depicted in Figure 2 (see the right column next to cell 1). The state of the system, symbolized by the black circle, rests in the minimum of a potential landscape (the “attractor” of the system). The stimulus is, then, altered gradually in the sequence of cells 2, 3, and 4. Most viewers, however, still read chaos in cell 2 and cell 3 (the system remains in the respective attractor of the potential land-

90 modeling complex systems scape), a phase transition occurring only in cell 4. Here, the attracting capacity of the Gestalt chaos has given way to the Gestalt order, a phase transition has taken place, and the state of the system has been qualitatively changed. In this way the stability of emergent patterns can be explored: a Gestalt may, and often does, remain unchanged in spite of alterations of the actual stimulus. Self-organized patterns have attractor-like dynamic properties. Interestingly, when the stimulus sequence is reversed in experiments with several subjects, it is found that the timing of the phase transition may depend on sequence. When the state of the system is started from order (see cell 4; the circle dwells in the potential minimum on the right-hand side) and passes through cells 3 and 2, the phase transition occurs in cell 1 instead of cell 4. This phenomenon of delayed phase transitions depending on past sequences is called hysteresis. Hysteresis is usually viewed as a hallmark of nonlinearity in system dynamics. In the present context, hysteresis is an additional ﬁngerprint of attractors governing the perceptual system. In summary, we have shown that Gestalt perception may be analogous to pattern-formation processes described in the framework of dynamic systems theory. We claim that these are pivotal processes found throughout cognition and action. As we cannot go into more detail here, it must sufﬁce to point to some active ﬁelds of research where pattern formation is at the center of investigations. In the movement sciences, for instance, cognitive coordination and movement coordination are modeled along such lines (Kelso, 2003). In neuroscience, synchrony in transient aggregates of neurons in the cortex is seen as the hallmark of the self-organization of cell assemblies (Miltner, Braun, Arnold, Witte, & Taub, 1999; Singer & Gray, 1995; Varela, 1995). In clinical psychology and psychiatry, therapeutic strategies have been developed on the background of dynamic systems concepts, for example, in psychotherapy research (Grawe, 2004; Mahoney, 1991) and in theoretical psychiatry (Ciompi, 1997). Patterns of the mind emerge from the complex mind/brain system via self-organization processes—this general premise puts great emphasis on the connectivity among the components as well as on the driving of systems by environmental-control parameters. We will brieﬂy explore both topics in the rest of this section.

91 Approach to Psychopathology

connectivity There are two general strategies employed to investigate neurocognitive systems with respect to their interaction and self-organization, top-down and bottom-up. The top-down strategy begins with observable (“macroscopic”) phenomena, such as the perception of Gestalts and the ﬁngerprints of phase transitions introduced above. The bottom-up approach ﬁrst identiﬁes components of a system and then bases the pattern emerging from the system on a description of the components. In focusing on the connectivity of systems components, we would adopt a bottom-up approach. We have argued that the synergy (i.e., the working together) of many components leads to the emergence of macroscopic patterns. The comparable term coined by the Gestalt-psychological tradition is nonsummativity. The Gestaltist holistic position claimed that aggregates are not identical with the sum of their components (e.g., Köhler, 1920). With this core tenet of Gestalt theory, psychology departed from associationist psychology—whereas associationism stated that properties of aggregates can be derived from the properties of their components, Gestalt psychology insisted on aggregates possessing qualities that are absent at the level of components. Therefore, the historical roots of contemporary emergentism and of complexity theory lie in Gestalt theory. The crucial point about the bottom-up approach is that we must clarify the nature of the components. What, however, are the components of (neuro)cognition? Deﬁning an elementary cognitive act is notoriously a nontrivial task in phenomenological psychology, by which one can easily be lead into the shallow waters of introspectionism. Attempts to deﬁne the components of behavior in a cognitivebehavioral fashion, for example, as conditioned stimulus-response associations, have remained rather hypothetical. In a biopsychological framework, however, the task appears much easier—the activity of a single neuronal cell in the brain is a good candidate for a component because it is well deﬁned and, at least in principle, observable. Ensembles of neurons are formed in concordance with Hebb’s law: “Cells that ﬁre together wire together.” The research program of connectionism is, thus, based on the connectivity of single cells. Connectionist models are good examples of emergent complex systems (e.g., Haken, 1987).

92 modeling complex systems Several issues in psychopathology have recently been approached with a focus on connectivity. Schizophrenia, for instance, can be seen as a misconnection disorder characterized by altered interactions among brain regions. The cognitive-dysmetria hypothesis (Andreasen, Paradiso, & O’Leary, 1998) predicts that the circuitry between areas of the (especially frontal) cortex, the thalamus, and the cerebellum is unbalanced in schizophrenia spectrum disorder. A related approach views the ion channels of glutamatergic synapses as responsible for the “loosening of associations” that Bleuler (1911) identiﬁed as the core symptom of spectrum disorders. Here, the emphasis is again on changed neuronal connectivity, which may result in a host of cognitive impairments, for example, Gestalt perception changes (Phillips & Silverstein, 2003) and decreased “binding” capabilities (Garcia-Toro, Blanco, Gonzalez, & Salva, 2001).

control parameters In this section, we introduced the synergetic theory of pattern formation. In addition to the connectivity of many components in a complex system, we pointed out that the driving of the system by control parameters constitutes a necessary condition for self-organization to occur. Such energetic driving of cognitive systems is traditionally termed motivation. In the previous two perception examples (Figures 1 and 2 above), task affordances and attention/vigilance determined what was being perceived. In the absence of motivation, or in situations with competitive motivational parameters, some order parameters may never be generated, even if they appear obvious to unbiased observers. This has been demonstrated, for example, in experiments on inattentional blindness (Simons & Chabris, 1999) where observers were found to be functionally and consciously blind to certain items in a visual scene when their attention was focused on an alternative task, even if this task was performed using the same visual stimulus material. Most psychopathological states, in addition to the obvious affective disorders, are associated with speciﬁc motivational and affective deﬁcits or excesses. The prototypical courses of schizophrenia, for example, often have ﬂuctuating levels of negative symptoms, such

93 Approach to Psychopathology as ﬂat affect and emotional withdrawal (see the next section). Thus, it is to be expected that the motivational driving of the cognitive system in such patients may undergo marked changes during the course of the illness. We conclude the theoretical introduction to this chapter by mentioning the interaction of (cognitive) pattern formation and (motivational) control parameters. So far we have concentrated on the “control side” of the system-environment coupling, that is, how control parameters affect the generation of patterns. Yet this reﬂects only one side of the coin of interaction: it is self-evident that the patterns, in turn, exert an inﬂuence back on the control parameters. As we have claimed above and discussed elsewhere (Tschacher & Haken, in press), motivation interacts with cognition in a manner analogous to the driving of complex systems by gradients of free energy (also called exergy in ecology; Schneider & Kay, 1994). In such systems, emergent patterns are consistently found to present the most efﬁcient ways to reduce the gradients that caused their emergence in the ﬁrst place. We argued that, when applied to cognitive systems, this principle of efﬁciency or “optimality” in reducing environmental gradients was a key to understanding intentionality in cognition. In this theoretical section we have sketched a formal systems model for psychopathology. We think that this may turn out to be an encompassing “big picture,” from a natural science perspective, of how cognition works. Admittedly, however, this big picture must be ﬁlled out with empirical ﬁndings and scrutinized for its predictive value. In the remainder of this chapter, we will examine the evidence that has been accumulated and put forward ideas about how one might proceed in this line of research on psychopathology.

Empirical Research in Psychopathology: From Description to Testing Dynamic Hypotheses Understanding phenomena by observing change is a common and often successful path to scientiﬁc insight. In the following, explorative studies of the dynamics of psychopathology are presented. These studies used various approaches to the dynamics of psychopathology. The study questions and corresponding methodological

94 modeling complex systems approaches can be summarized under three levels of analysis. On a ﬁrst level of analysis, the study questions relate to “how things change in time.” The results on this level are descriptions of longitudinal developments, called trajectories in the language of dynamic systems theory. This type of analysis is mainly descriptive. A second level of analysis investigates “how things interact in time.” In other words, the interrelation of the systems’ variables is analyzed. Systematic interrelations result in stable patterns of a system, that is, attractors. Finally, a third level of analysis is related to the emergence of patterns—nonlinearity, degrees of order, and pattern formation. It is important to note that these different levels of analysis are complementary rather than mutually exclusive. Both descriptive and more reﬁned analytic approaches are necessary to illuminate self-organizing systems.

level of analysis 1 : trajectories and attractors What is commonly termed psychopathological impairment is in most, if not all, instances more than a trait of a person. Constant changes in symptoms have, instead, been described as a hallmark of mental disorders (Wittchen, Lieb, Pﬁster, & Schuster, 2000). On a descriptive level, the resulting longitudinal patterns have for a long time been important categories. As an example, “insidious” versus “acute” onset of disorders has been related to different prognoses of schizophrenic disorders (Wiersma, Nienhuis, Slooff, & Giel, 1998). By deﬁnition, “chronic” forms of mental disorders show considerable temporal stability of psychopathological impairments. Nevertheless, even in chronic mental disorders, several typical temporal patterns of psychopathology can be distinguished (Kupper, 1999). Systems theory provides mathematical models for the evolution of mental disorders (Globus & Arpaia, 1994; Kupper, 1999; Tschacher, 1997). From this perspective, temporal evolutions are termed trajectories. Except for the case of purely stochastic dynamics, trajectories can be traced back to underlying attractors that describe the inﬂuence of forces that act on the evolution of the process variables (Figure 3). Examples of prototypical dynamics are trajectories that tend to a stable constant state. The corresponding attractors are

95 Approach to Psychopathology Prototypical

Dynamics and

Attractors

Dynamics Convergent

Time

Cyclic

Time

Chaotic

Time

Attractors Point

0-dimensional

Limit Cycle

1-dimensional

Chaotic

>2-dimensional

Figure 3. Prototypical dynamics and attractors.

called points attractors. Another prototypical dynamics is characterized by oscillations: limit cycle attractors. Finally, unpredictable yet nonrandom ﬂuctuations are produced by certain nonlinear dynamics that have been discovered only recently (Rössler, 1976). In such a way psychopathological systems may be classiﬁed as exemplars from a “zoo” of prototypical attractors arising from various linear and nonlinear dynamics. Psychopathology in acute schizophrenia, for example, is often characterized by dramatic and repeated changes in symptoms. To date, however, research on the dimensions of psychopathology in schizophrenia has been mainly cross-sectional in nature (Peralta & Cuesta, 2001). Correspondingly, whereas the cross-sectional structure of schizophrenic symptoms has been studied extensively, little is known about the development of symptoms during acute episodes. In a study of symptom trajectories, 46 schizophrenia spectrum patients (18 females; mean age 24.7 years) were examined daily during an average treatment period of 104 days (Kupper & Tschacher, 2002). A novel time-series approach was used to identify initial phases of response and other descriptive features of the trajectories. Figure 4 shows the symptom processes of four exemplary pa-

96 modeling complex systems

Figure 4. Symptom trajectories of psychotic episodes.

tients. Each process consisted of three trajectories, termed psychoticity, excitement, and withdrawal, depicting the positive symptoms, anxiety and tension, and negative symptoms together with depression observed in patients. The time-series method used in this study allowed the identiﬁcation of “dynamic factors,” consisting of trends, levels, and durations of these trajectories that were assessed in two periods of treatment, the initial improvement phase and the ensuing stabilization phase. These factors summarized the temporal structure of symptom evolution throughout the psychotic episodes. Five such dynamic factors were found: (1) overall level of positive symptoms, (2) duration of nonspeciﬁc response, (3) slope of response in all symptom domains, (4) enduring negative symptoms, and (5) duration of response regarding psychoticity. Compared to patients with an acute schizophrenia-like psychotic disorder, schizophrenia and schizoaffective disorder patients ranked higher on factor 4 (enduring negative symptoms). They tended toward a lower level of positive symptoms and showed a less prominent response to treatment. The examination of a subsample of 19 patients with relapse indicated a

97 Approach to Psychopathology prolonged duration of initial treatment response regarding psychoticity. These results supported the validity of this approach for the description of symptom trajectories. In this study the symptom change throughout psychotic episodes showed distinct temporal features. Compared to a merely cross-sectional view, this approach provided a dynamic analysis of the dimensions of schizophrenic psychopathology. Therefore, it was possible to separate the stable and the variable components of psychopathology in the patients. The trajectories of psychopathology found suggested that, whereas levels of symptoms are, in fact, often independent cross-sectionally, their change within a given patient was often strongly interrelated. Moreover, these results suggested that the different aspects of psychopathology might be related to some single, common process. In schizophrenia meaningful temporal patterns in psychopathology are not restricted to acute phases of the disorder. In a study of rehabilitation (Kupper & Hoffmann, 2000), courses of psychosocial functioning during intensive vocational rehabilitation of schizophrenia patients were examined. Although considerable research has been undertaken on psychosocial treatment and rehabilitation of patients with chronic schizophrenia, few studies have examined individual courses by means of repeated and frequent observations. A more dynamic view of rehabilitation was thought to disclose patterns of response useful for both understanding and treating symptoms and disabilities associated with chronic schizophrenia. In this exploratory study time series of 35 schizophrenia outpatients participating in a vocational rehabilitation program were examined by a novel quantitative approach by which dynamic patterns were identiﬁed. Time-series regression was applied on weekly behavioral ratings of psychosocial functioning, provided by the nurses’ observation scale (nosie). The mean, trend, and variability of each trajectory were calculated. Cluster analysis revealed ﬁve groups of trajectories: (1) stable at a high level, (2) ﬂuctuating at a middle level, (3) middlelevel functioning, tending toward a slight descent, (4) steep descent of functioning, and (5) unstable at a low level of functioning. Examples from groups 1 and 4 are shown in Figure 5. The ﬁve groups of patients with different trajectories varied at intake with respect to psychopathology, cognitive dysfunction, and measures of self-concept, locus of control, and coping. At program termination pronounced differences were found among the groups

98 modeling complex systems

Figure 5. Psychosocial functioning in two vocational rehabilitation courses of schizophrenia patients.

in vocational reintegration. The different trajectories can, thus, be understood as typical pathways linking patient characteristics to rehabilitation outcome. This study exempliﬁed that a broader use of dynamic designs can substantially clarify the variety of reactions of patients to psychosocial interventions. From a systems-theory point of view, these trajectories arose from patients’ characteristics interacting with environmental constraints that were induced by the introduction of intensive rehabilitation with increased levels of stress. The task affordances related to the rehabilitation environment acted as gradients (i.e., environmental-control parameters), leading to a potential destabilization of the system. This resulted in a transfer either to a higher level of functioning or to a deterioration in functioning. Similar patterns are known from other forms of psychological intervention. As an example, it is observed that psychotherapy can initially amplify the differences among clients. It should be noted that, generally, these few examples showed that the descriptive level of “how things change” in time provides information that is complementary to traditional research approaches. Cross-sectional analysis cannot explain this rich source of variation. On the grounds of dynamic systems theory, we may suppose that the diverse courses described here have resulted from attractors, the driving forces of longitudinal evolutions of psychopathology. The time-series methods used in the studies discussed above allowed the description and clustering of such evolutions but gave no indications as to the attractors and patterns underlying them. For this type of exploration additional methods must be applied, methods that

99 Approach to Psychopathology provide a closer look at the mechanisms of change leading to observable dynamics.

level of analysis 2 : the interrelation of system variables On this level the temporal interrelations of components of a system are analyzed. The interrelations among different components are expected to reveal information about underlying mechanisms that themselves are not directly observable, neither in cross section nor in longitudinal monitoring. Thus, the goal of these analyses was to model mechanisms that are supposed to fuel the courses described in the previous section. This notion can be tested empirically on symptom courses by utilizing methods of dynamic systems research, such as time-series analysis. In a study of psychotic episodes (Tschacher & Kupper, 2002), the symptom courses of 84 schizophrenia spectrum patients (mean age: 24.4 years; mean previous admissions: 1.3; 64% males) of a community-based acute ward were examined to identify dynamic patterns of symptoms and to investigate the relations between these patterns and treatment outcome. The symptoms were monitored by regular daily staff ratings using a scale composed of three factors: psychoticity, excitement, and withdrawal. Examples of such data are shown in Figure 4 above. Patients showed moderate to high symptomatic improvement documented by effect-size measures ranging from .65 to .92. Each of the 84 symptom trajectories was analyzed by time-series methods using the vector autoregression (var) approach (Lütkepohl, 1993). var models the day-to-day interrelations between symptom factors on the basis of the data of a single course (for an example, see Figure 6). In Figure 6 the arrow starting from withdrawal at day t 1, pointing toward excitement at day t, expresses a positive association between these two variables with a weight of .44 (the T-value of 2.94 is signiﬁcant at the 1% level). This var parameter may be labeled withdrawal → excitement. In other words, in this patient high withdrawal one day entailed increased excitement on the following day. The horizontal arrow in Figure 6 (withdrawal → withdrawal) indicates positive autocorrelation of withdrawal; that is, increased withdrawal on any one day predicts increased withdrawal the following day.

100 modeling complex systems

Figure 6. Individual time-series model of a patient (see the text). * p < .05; ** p < .01.

Multiple and stepwise regression analyses were then performed on the basis of the parameters of var models of all 84 patients. Two var parameters were found to be associated signiﬁcantly with favorable outcome in this sample: withdrawal → reduction of psychoticity as well as excitement → increase of withdrawal. These ﬁndings were interpreted as indicating how patients cope with psychotic episodes. It may be assumed that this illuminated mechanisms that drove the symptom evolutions in this sample. These ﬁndings must, however, be regarded as preliminary assumptions, as hypotheses about the underlying attractors of the symptom dynamics. As a further step in a consecutive unpublished study, we combined this analysis of schizophrenia courses with a comprehensive assessment of neuropsychological functioning and psychopathology (Spaulding et al., 1999). In the Schizophrenia Process Study, symptom courses of 114 schizophrenia spectrum patients were observed using daily ratings of psychopathology (mean duration of observation: 88 days). A 10-item scale for daily symptom assessment was applied (Today’s Evaluation of Psychopathology, or tep). The rating scale was composed of three factors: positive symptoms, negative symptoms, and anxiety-depression. The courses were analyzed by time-series methods using var to model the day-to-day interrelations between symptom factors. Comprehensive neuropsychologi-

101 Approach to Psychopathology cal assessments were performed at the beginning and at the end of the observed courses. Associations between the day-to-day patterns of symptoms, neurocognition, and outcome were calculated. Preliminary results (Kupper, Tschacher, & Hoffmann, 2004) revealed speciﬁc relations between longitudinal symptom patterns and both outcome and neurocognition. The interrelations between positive and negative symptoms were associated with outcome. If positive symptoms entailed negative symptoms, outcome tended to be unfavorable. In cross-sectional studies different symptom domains as well as symptoms and neurocognition are often found to be independent. The results of this study, again, suggested that the various symptom domains as well as symptoms and neurocognition are, in fact, associated if longitudinal patterns of symptoms in individual patients are examined. In a study of crisis intervention (Tschacher & Jacobshagen, 2002), the remediation processes after psychosocial crises were modeled. We included patients with diagnoses such as adjustment disorder and an affective disorder who were often hospitalized with a status of suicidal behavior or ideation. The focus of this study was on “cognitive orientation” (“outward” events centered vs. “inward” centered). A sample of 38 inpatients who were assigned to treatment in a crisis-intervention unit was monitored in order to study the interrelations of process variables. The process data consisted of patients’ self-ratings of the variables mood, tension, and cognitive orientation, which were assessed three times a day throughout hospitalization (on average, over 22.6 days). Again, var models of the process data were computed to describe the prototypical dynamic patterns of the sample and explore process-outcome associations. Outcome of crisis-intervention treatment was evaluated by pre/postquestionnaires. On a ﬁrst level of analysis, the expected linear trends were found pointing to an improvement of mood, a reduction of tension, and an increase of outward cognitive orientation. Time-series modeling (level of analysis 2) showed that, on average, outward cognitive orientation entailed improved mood (see Figure 7). The time-series models predicted the treatment effect, notably the outcome domain reduction of social anxiety, yet were unrelated to the domain of symptom reduction. From the results it was concluded that crisis intervention should focus on having patients increasingly engage in outward cognitive orientation, that is, reduce the levels

102 modeling complex systems

Whole Sample (N=38) t-1 MOOD

t .74**

.34*

TENSION .97**

COGNITIVE ORIENTATION

.90***

Figure 7. Interrelations among system variables in crisis intervention. * p < .05; ** p < .01; *** p < .001.

of self-focused attention. This therapeutic approach should support stabilizing mood, reducing anxiety, and activating the resources of patients. Similarly, “driving forces” of change were extracted from psychotherapy courses (Tschacher, Baur, & Grawe, 2000). A sample of 91 courses of dyadic psychotherapy using different treatment modalities was analyzed in order to study session-by-session dynamics. The presented problems generally belonged to the spectrum of neurotic disorders, the largest diagnostic groups being adjustment disorders, anxiety and phobic disorders, eating disorders, and relationship problems. The process data consisted of therapists’ and patients’ session reports. Therapy outcome was evaluated by pre/ postquestionnaires and direct measures of change. After data reduction by principal-component analysis, linear time-series models of the resulting factors were computed to describe the prototypical dynamic patterns of the sample and of the modality subsamples (cognitive-behavioral, client-centered, schema-theoretical cognitive psychotherapy). It was found that the factor patient’s sense of selfefﬁcacy/morale governed the observed dynamics of the sample, whereas the therapeutic-bond factors did have less of an effect on the dynamics (Figure 8). The dynamic patterns of client-centered therapies differed from other modalities. The dynamics-outcome ﬁndings showed that direct measures of change were associated with a spe-

103 Approach to Psychopathology

Figure 8. Interrelations among process factors in psychotherapy. * p < .05; ** p < .01.

ciﬁc process pattern in which the patient’s sense of self-efﬁcacy was, in turn, supported by other process factors. This time-series approach was also utilized to uncover psychopathological changes and side effects related to different substances employed in psychopharmacology. Clinical observations and recent ﬁndings suggested different acceptance of morphine and heroin by intravenous drug users in opiate-maintenance programs. Such programs are conducted nationwide in Switzerland. In a psychopharmacological study (Tschacher, Haemmig, & Jacobshagen, 2003), we postulated that this different acceptance may be caused by differences in the perceived effects of these drugs, especially how desired and adverse effects of both drugs interacted. Thus, we assumed different mechanisms of pharmacological action underlying the manifest effects. We measured the desired and adverse effects of high doses of injected morphine and heroin in patients included in the program. The interrelations between both types of effects were determined to test the hypothesis of a differential mechanism of action. Thirty-three patients (5 females, 28 males; mean duration of previous street heroin use 10.7 years, mean age 30.1 years) were randomly allocated double-blind to the substance groups. The average daily dose per participant in the heroin condition (n = 17) was 491 mg, in the morphine condition (n = 16) 597 mg. The observation period lasted 3 weeks; an average of 70 injec-

104 modeling complex systems tions was received. After each injection of either substance, various aspects of drug effects were recorded systematically. Ratings were summarized into the factors euphoria and adverse effects. Time-series models were computed for each participant on the basis of the factor scores, using the var format described above. A highly signiﬁcant difference between the substances was found in the interrelation between euphoria and adverse effects. Adverse effects of heroin preceded higher euphoria, whereas adverse effects of morphine preceded subsequent lower euphoria. Additionally, the ﬁnding of a generally higher level of adverse effects in morphine was replicated. These results pointed to different mechanisms of action of the two opioids when the perceived drug effects are evaluated in a ﬁeld setting. This may explain the better acceptance of heroin in opiate-assisted treatment of intravenous drug patients. In addition to the results concerning the studied sample, we proposed the method as a promising pharmacological tool for the comparison of substance groups other than opioids. In general, this second level of analysis allowed for the identiﬁcation of interrelations among process variables of many different origins and in various time frames. The monitoring units may be ﬁxed (e.g., daily ratings), natural (e.g., at each injection of a substance), or a combination of both (e.g., at each psychotherapy session). The interrelations of systems variables can be studied under naturalistic conditions, giving this approach high heuristic value and ecological validity. We view such interrelations as core attributes of dynamic systems that give rise to the patterns and attractors described above. Some premises for the application of level of analysis 2 methods must be regarded, notably, that the approach assumes stationarity of the processes. For this reason we used several techniques in the aforementioned empirical studies to ensure stationarity, for example, detrending of courses prior to modeling. On a next level of analysis, transitions in order and pattern formation were analyzed. As described in the introductory section of this chapter, such transformations of systems leading to the formation of new patterns are a striking feature of self-organizing systems. The next section therefore presents studies of nonlinearity and pattern formation in psychopathology.

105 Approach to Psychopathology

level of analysis 3 : nonlinearity and pattern formation In this section we will address some of the more speciﬁc predictions of the dynamic systems perspective proposed in the ﬁrst section. If the notion of pattern formation by self-organizing dynamics is true, and if phase transitions between stable patterns occur, the following ﬁngerprint phenomena should be expected. First, as soon as self-organization structures a complex system, an increase of order (to be deﬁned below) should be observable. Second, one may determine whether systems evolve in a nonlinear manner by detecting signs of chaotic dynamics and, more generally, nonlinearity. Third, one may explore the stability attributes of self-organized patterns and how they are linked with psychopathological symptoms. Order and Disorder in Complex Systems We studied the hypothesis of pattern formation in a variety of data sets in clinical psychology as well as in acute and rehabilitative psychiatry. A systematic study was performed in the ﬁeld of psychotherapy (Tschacher, Scheier, & Grawe, 1998). In this study 28 courses of dyadic psychotherapy (10 behavioral, 3 client centered, 9 heuristic, 6 schema oriented; 40–90 weekly sessions) were investigated. The hypothesis of order increase by pattern formation rested on our conceptualization of the therapeutic alliance as a complex psychological system that is embedded in an environment of various driving parameters. Using the method of therapy-session reports, all individual sessions of each therapy were assessed by the therapist (a 14-item questionnaire with items such as: “I have the impression that the client came up with what really bothers him/her”) and likewise by the client (a 19-item questionnaire with items such as: “Today I felt comfortable with the therapist”). For each therapy course, data on 33 variables per weekly session were collected. The multiple time series generated by this monitoring procedure served as the foundation of order analysis. Landsberg order was used as described in Banerjee, Sibbald, and Maze (1990; see Shiner, Davison, & Landsberg, 1999). This measure estimates the degree of order in a multivariate data set by calculating the ratio of the actual entropy, normalized by the potential entropy, both computed on the basis of the variance-covariance matrix of the p t window (p =

106 modeling complex systems items of session reports, t = session number). One minus this ratio then provides the measure of order. We computed this measure for the 20 ﬁrst and the 20 last consecutive sessions using the 33 items recorded in therapy-session reports. Results clearly supported the hypothesis of a signiﬁcant increase of order during the course of psychotherapy in a comparison of ﬁrst versus last sessions of each therapy course. This ﬁnding was corroborated using an extended data set of a later study (Tschacher et al., 2000). In a second step of this study we addressed a question that follows naturally from this ﬁnding: Is there a connection between order and therapy outcome? The general impression achieved by the outcome study was unequivocal: Order of the therapy system as well as increase of order were related to positive outcome of therapy. We monitored 41 outcome measures, 20 of which were correlated with Landsberg order in a statistically signiﬁcant way. These associations pointed to order and order formation as being functional for good treatment results. Improvement of the self-image of patients and a decrease in social anxiety were linked with both order and increase of order. Psychopathology of the psychotherapy patients was found to be connected to increase of order. Order was also computed to investigate the course of schizophrenia in acute treatment (Kupper, Tschacher, & Hoffmann, 1997) and to study rehabilitation processes in psychiatry. In these data the assessment of order was based on daily symptom ratings using the instruments introduced above and on weekly evaluations of progress in a psychiatric vocational rehabilitation program (Kupper, 1999). In the study of acute treatment we analyzed psychopathology courses of 21 patients (mean age 24 years, 10 of 21 female) during a psychotic episode. Daily ratings of psychopathology were performed. As in the study reported in Tschacher et al. (1998), order and order formation were calculated using the measure Landsberg order. The results of this study showed that order and order formation correlated negatively with the overall level of psychopathology and with the ﬁnal level of psychopathology at discharge of patients. Order generally increased throughout the treatment of psychotic episodes. In the vocational rehabilitation data (Kupper, 1999), order and order formation in courses of 30 participants (mean age 27 years, 10 of 30 female) with schizophrenic disorders were explored. The same methodological approach to order formation as in the previous study was

107 Approach to Psychopathology Table 1. Summary of Order and Order-Formation Results from Studies of the Treatment and Rehabilitation of Schizophrenia Patients Vocational Rehabilitation

Acute Treatment

Initial phase: no correlation

Initial phase: more negative symptoms and depression (outcome) Middle phase: less confusion and tension (outcome) Final phase: no correlation

Correlates of order (r approximately 0.4–0.7)

Middle phase: positive outcome (level of integration)

Ordering effect present

No

Correlates of ordering

Positive outcome (level of integration)

Yes Initial-middle: less overall tension, fear, and ambivalence Initial-ﬁnal: less negative symptoms and depression

used. This program for vocational rehabilitation included vocational training, social skills training, family interventions, job placement, and support by job coaches. The assessments were weekly ratings of behavioral and emotional indicators of functioning, from both participants’ and supervisors’ perspectives. The primary outcome was level of vocational reintegration after the conclusion of the program. The results showed that order and order formation were again related to a positive outcome, here regarding vocational reintegration. No general increase of order was found. Overall, the majority of results in these data were consistent with what we labeled order effect (level of Landsberg order correlating to favorable outcome) and ordering effect (increase of Landsberg order correlating to favorable outcome; see Table 1). Moreover, meaningful differences between studies are evident from Table 1. Acute treatment seems to consist in a transition from a disordered acute state to a more stable state toward the end of treatment. Rehabilitation programs with their higher demands, however, appear to start from a stable entry point. The increased stress induced by these demands does not allow a general ordering effect but leads instead to largely varying developments among patients. Generally,

108 modeling complex systems it can be hypothesized that, in these studies order reﬂected the stability of the mental system of the patients. Therefore, the tacit undertone of the word disorder (meaning mental illness) seems justiﬁed, even when we deﬁne order in the strict mathematical sense. Fingerprints of Nonlinearity and Chaos One of the hypotheses that have been derived from dynamic systems theory is that seemingly random temporal behavior may be generated by relatively simple deterministic nonlinear systems. Chaos theory, a branch of systems theory, has pointed out how such systems may provide both stability (attractor-driven dynamics) and innovation (sensitivity to environmental changes). It was, therefore, suggested that nonlinear systems might provide explanations for the often unpredictable trajectories of psychopathological disorders, especially schizophrenia (Ciompi, 1997). The review of Paulus and Braff (2003) lists the recent dynamic approaches used to investigate the temporal architecture and the disease process underlying schizophrenia. In a study of daily ratings of psychopathology in 14 schizophrenia spectrum patients where observations were possible for extended periods of time (between 200 and 760 consecutive days), we modeled the attributes of the disease process (Tschacher, Scheier, & Hashimoto, 1997). To do this, we ﬁrst estimated the forecastability of the time series of each individual patient. Then we applied a series of bootstrap tests that implemented a hierarchy of potential process attributes, such as “random,” “linearly autocorrelated stochastic process,” and “linear stochastic process with identical power spectrum.” Finally, the actual forecastability of the observations was compared to the forecastabilities of the bootstrap time series. Whenever signiﬁcant differences occur, we may conclude that the dynamics of the observed time series was not sufﬁciently explained by the process attributes, thereby pointing to nonlinearly generated dynamics of the observed data. The analysis suggested that 8 of 14 schizophrenia patients presented nonlinear dynamics. These investigations were compatible with the hypothesis that schizophrenia may be characterized by chaotic evolutions. Yet direct proof of deterministic chaos in relatively short empirical time series is unattainable, therefore, one cannot assess the chaoticity of the schizophrenia courses we studied. Visual inspection of forecasting accuracies showed, however, that the decay of predictability typical of deterministic chaos was present in the larger

109 Approach to Psychopathology part of the sample. Together with the rejection of alternative hypotheses (especially linearly correlated stochasticity), this allows that chaotic processes may underlie psychopathology in schizophrenia. We think that the chaos hypothesis of schizophrenia and other diseases (e.g., depression; Pezard et al., 1996) must be treated with considerable caution. Although there is some evidence from a variety of data sources, both behavioral and neurophysiological, the prerequisites for the generally very sophisticated dynamic methods are seldom realized in psychopathology data. Existence of chaos may well remain undetected because of the constant ﬂuctuations found in any environment outside a physicist’s laboratory. In addition to chaos representing a quite ephemeral quality of (some) nonlinear systems, it is probably not their most important property. We already introduced pattern formation as a functionally valuable and, at the same time, ubiquitous phenomenon and will concentrate on ﬁngerprints of cognitive patterns and their stability in the following. Stability of Cognitive Patterns in Schizophrenia Perceptual and cognitive patterns are characteristically changed during psychotic experience. The tradition of Gestalt psychiatry (Conrad, 1958; Matussek, 1952) views the evolution of schizophrenia by a succession of stages: everyday cognitive patterns may be impaired prior to the outbreak of psychosis (derealization as Gestalt loss) and may then be resurrected in a bizarre way (delusions and hallucinations as malfunctional Gestalt re-formation). These phenomenological studies thus pointed to a variety of alterations in Gestalt perception and cognition indicating that both disorganization and delusional hyperorganization should be found. A number of empirical studies have, consequently, focused on the changes in (predominantly perceptual) organization in schizophrenia spectrum disorder (Uhlhaas & Silverstein, 2005). Cognitive psychology views Gestalt-like processes as essential during preattentional stages of parallel stimulus processing; in these early stages stimulus features are grouped and “bound together.” The large majority of these studies suggested that perceptual organization is dysfunctional in schizophrenia. These changes in perceptual organization may reﬂect a more encompassing impairment in the coordination and binding of spatial and temporal information and of information originating from different sensory modalities (Phil-

110 modeling complex systems lips & Silverstein, 2003). Interestingly, Gestalt dysfunction does not necessarily go hand in hand with patients showing decreased task performance. Place and Gilmore (1980) and others found that schizophrenia patients did not present a generalized deﬁcit when compared to controls; in fact they were, for example, more accurate than controls in counting tasks where grouping of stimuli did not support performance. Research has shown that alterations of perceptual organization differ between subgroups of patients and, especially, correlate with symptomatology. A number of clinical symptoms were linked with Gestalt perception. In their review Uhlhaas and Silverstein (2005) cited evidence for a correlation of Gestalt dysfunctions with cognitive disorganization as well as with both positive and negative symptoms. In a recent study of Gestalt perception (Tschacher, Dubouloz, Meier, & Junghan, in press), we investigated several apparent motion paradigms. We included 32 schizophrenia patients (66% dayclinic patients, 34% inpatients; 81% males; mean age 27.3 years) and 32 control subjects matched for age, sex, and education. Clinical symptomatology of the patients was measured by the Positive and Negative Symptom Scale (panss). Although well in need of treatment, most patients were not acutely symptomatic, receiving average panss scores below 3 (“low symptoms”). For all computations we used the factorization of the panss according to Lindenmayer, Grochowski, and Hyman (1995), which consists of the ﬁve factors positive, negative, excitement, cognitive (disorganization), and depression. Nevertheless, this schizophrenia group had reaction times markedly decreased compared to controls. The mean reaction time was increased by 1.7 standard deviations in patients, pointing to their generalized impairment of cognitive functioning. Circular Apparent Motion (cam) The frames displayed in Figure 9 were presented alternately for 500 ms each. Generally, Gestalts in the shape of movement illusions result. The impression is always either that of a wandering of the dots in a clockwise or counterclockwise circular direction, or that of a rapid oscillation of the two apparent motions. Thus, the stimulus presentation is inherently multistable (Kruse, Stadler, Pavlekovic, & Gheorghiu, 1992) because the same display can give rise to different illusions in the same person under

111 Approach to Psychopathology

Figure 9. Display shown to test circular apparent motion (cam). The left- and the right-hand panels are shown alternately for 500 ms each, producing the motion perception.

identical circumstances. The perceived durations of respective Gestalts were suggested as a measure of cognitive stability (Kruse & Stadler, 1995). In the present sample, multiple regression showed that durations of Gestalt perceptions were signiﬁcantly linked to the panss factors. Especially the negative symptom factor was associated with prolonged durations. Yet there was no difference between the patient and control groups. Stroboscopic Apparent Motion (sam) Corresponding to the previous task, the frames displayed in Figure 10 were presented alternately for 500 ms each. Two qualitatively different Gestalt illusions are generated, vertical apparent motion (vsam) or horizontal apparent motion (hsam) of the dots. In the case of vsam, dot A is perceived as moving down to dot D and B as moving up to C. In the case of hsam, the perception is of A and B moving to C and D, respectively. The Gestalt impression depends on the horizontal distance d. The smaller is d, the more likely hsam is seen. With d gradually reduced from a maximum value to 0 and then increased again, ﬂip events are enforced (initially, vsam transiting to hsam, then the reverse, hsam giving way to vsam). We monitored the timing of these ﬂip events by having subjects press a button when a ﬂip occurred. The asymmetry of the timing of the initial and the reversed ﬂip events (i.e., the hysteresis effect) is a hallmark of phase transitions between stable attractors, as was seen above. Haken (1996) discussed the size of the hysteresis effect as a prominent measure of Gestalt stability.

112 modeling complex systems

Figure 10. Display shown to test stroboscopic apparent motion (sam). The left- and the right-hand panels are presented alternately for 500 ms each.

Again, no group effect was observed in our data. In other words, Gestalt stability measured by the hysteresis effect did not generally deviate in patients. The timing of the initial Gestalt ﬂip (vsam to hsam) was, however, strongly linked with symptomatology. The ﬁve panss factors explained 48% of the variance in this variable in a whole model test. The negative and the cognitive (disorganization) factors both delayed the initial Gestalt ﬂips, whereas positive symptoms accelerated the occurrence of ﬂip events. Motion-Induced Blindness (mib) The following phenomenon was described recently by Bonneh, Cooperman, and Sagi (2001): Subjects are instructed to observe the middle of a display such as that depicted in Figure 11. The display contains three stationary dots colored bright yellow and a grid of dimly illuminated blue crosses. The grid is rotated slowly. The phenomenon that most observers perceive is “apparent blindness”—one or all of the yellow dots subjectively disappear from the visual ﬁeld for a period of time, sometimes for several seconds. The subjects are asked to press a key whenever, and for the duration that, apparent blindness occurs. It was our premise that this task is, in fact, a Gestalt phenomenon in the sense of ﬁgureground rivalry. The blue grid may acquire the function of ﬁgure after some time has elapsed, with the result that the yellows dots are put in the background. This exchange of ﬁgure and ground may entail the subjective perception of subjects becoming blind to the yellow dots. The mib phenomenon is similar to inattentional blindness as investigated by Simons and Chabris (1999). The variable of interest in this paradigm was the number of mib

113 Approach to Psychopathology

Figure 11. Screen display (schematic) for motion-induced blindness (mib).

phenomena perceived by each participant during a 3-min period. There was a tendency in the patients’ group toward fewer mib counts (29.3, vs. 42.1 in the control group), yet this difference did not reach the signiﬁcance level. In multiple regression analysis of the patients group, however, we found a strong association of number of mib phenomena with symptomatology. The ﬁve panss factors explained 49% of the variance in this variable. The positive and the excitement factors were both related to increased numbers of mib phenomena (Tschacher, Schuler, & Junghan, 2006). Intersensory Binding The ﬁnal task of the Gestalt test battery addressed temporal binding of different sensory modalities (here, auditory and visual) into a coherent perceived pattern. We used a program developed by Scheier, Lewkowicz, and Shimojo (2003). Two disks are presented moving from either sides of the screen toward the opposing sides, following the same trajectory (Figure 12). When the disks approach one another in the middle of the screen, they are perceived either as “bouncing” off each other or as “streaming” through one another at the moment they overlap visually. Similar

114 modeling complex systems

Figure 12. Screen display (schematic) for the intersensory binding task. Two disks wander from the sides of the screen toward one another; at the time of their “collision” and visual overlap, a click sound is given. Which causal relation is attributed to this scenario, bouncing or streaming?

displays have been investigated in studies of apparent causality by Michotte (1954). In addition to the optical stimuli, an acoustic-click stimulus is presented with varying temporal latency relative to the time of disk overlap. The observer is instructed to indicate which of the two possible events (bouncing or streaming) was perceived. We report here the ﬁndings regarding the probability of bouncing versus streaming perceptions (Tschacher & Kupper, 2006). There was a tendency toward patients having fewer bouncing perceptions. In the patient group, a strong relation was found between symptoms and preferred causal attributions. The positive factor and the cognitive (disorganization) factor of the panss were strongly linked with increased probability of bouncing perceptions. As a whole, the Gestalt test battery showed few general group differences between patients and matched controls. This may be attributed in part to the low symptom burden in the patient group. Within the patient group, however, the panss ratings suggested that speciﬁc associations exist between psychopathology and Gestalt perception. Psychopathology has opposing effects on perception, effects that can explain the weak signiﬁcance of group comparisons. More speciﬁcally, we found that positive symptoms were linked with Gestalt instability and faster production of Gestalt perceptions, negative symptoms and the cognitive disorganization factor with persistence of Gestalts. Hysteresis, however, was unaffected by psychopathology. Additionally, apparent causality depended on symptoms. Attri-

115 Approach to Psychopathology bution of causality (the “bounce” perception) was a marker of positive and disorganization symptoms.

Discussion A common ﬁnding in research on mental disorders is that the level of psychopathology usually varies in each individual case in the course of time. Such ﬂuctuations have frequently been perceived as a problem for the validity and reliability of quantitative research, even more so as some psychopathological phenomena are inherently unstable or show unpredictable and erratic dynamics. With the theoretical and empirical approach presented here, such dynamics poses not so much a scientiﬁc problem as an opportunity for a novel approach to modeling. In this, a long-standing call for longitudinal studies is met. The dynamics-oriented approach views change as primary, especially in complex open systems. Structure and stability are, consequently, seen as the secondary results of dynamic processes that lead to the emergence of attractors. In the second section of this chapter, we have demonstrated the validity and feasibility of this approach. We have quantiﬁed and described the trajectories and processes ubiquitously found in various ﬁelds of practice; the resulting attractors and patterns have been detailed; some of the speciﬁc hypotheses of the dynamic approach have been tested in the data sets. The main results were these: Studies show a variety of dynamic patterns in courses of psychopathology concerning psychiatric rehabilitation and acute treatment of psychoses. For a wide range of settings and problems we applied methods by which the trajectories could be translated into condensed models, especially var models, that are capable of capturing the essence of the change processes. We proposed that these models reﬂect the attractors underlying the observed dynamics. We concluded that these models were psychologically meaningful because the attractors were related to outcome and/or patient characteristics. The ﬁnal part of the second section addressed some of the core assumptions of dynamic systems theory applied to ﬁelds of psychology, particularly order formation, nonlinearity, and the stability of Gestalt-like attractors. On the basis of the empirical evidence we may claim that the search for ﬁngerprints of nonlinear dynamics has been successful.

116 modeling complex systems The data again suggested that the phenomena of nonlinearity, such as order and stability of patterns, are associated with markers of psychopathology and clinical outcome. From these applications of the dynamic approach can be drawn a series of implications for psychiatry and psychology, especially as regards the ﬁelds of therapy, diagnosis, and the theory of cognitive science.

therapy We expect that time-series methods can elucidate mechanisms of action in practice ﬁelds of psychiatry and psychology, such as rehabilitation and psychotherapy. The kind of longitudinal assessment that we proposed stands in contrast to the established cross-sectional approach; this has signiﬁcant clinical implications. The most important beneﬁt lies with the possible identiﬁcation of change mechanisms in therapeutic settings. As time-series methods enable the modeling of single systems, the effective mechanisms can be assessed directly in the respective ﬁelds of practice and in natural clinical situations with uncompromised ecological validity. The aggregation of many single systems, as demonstrated repeatedly in the previous section, allows the distinguishing of favorable from unfavorable patterns. New methods of online control and the adaptation of interventions in the context of quality control may, thus, come within reach as soon as favorable patterns are reliably identiﬁed. It is, nevertheless, necessary to be aware of the limitations of dynamic methods, limitations that are brought about by the demands of time-series methodology. These demands are the large number of sequential measurements necessary for modeling; during the necessary observation period the underlying causal mechanisms must remain unaltered (the premise of stationarity). Another caveat must also be kept in mind: var models give an account of sequential interrelations among the process variables that may likely, but need not necessarily, reﬂect causal mechanisms. Time-series models are so-called Granger-causal models—as in all correlational models, the relation could be affected by unknown third variables that have not been monitored. Therefore, dynamic approaches should go hand in hand with classic experimental therapy research wherever possible. When these limitations

117 Approach to Psychopathology are taken into account, the dynamic approach has proved fruitful in process-outcome research, with a vast potential for further studies of therapeutic applications.

diagnosis Categorical diagnosis and the dynamic view of psychopathology follow conﬂicting concepts of disorder. For example, the concept of dynamic diseases (cf. Bélair, Glass, an der Heiden, & Milton, 1995; Emrich & Hohenschutz, 1992) deviates from the common structural concept in that (some) diseases are viewed as emergent states and attractors. This is in line with what was found above. Yet the dynamics-oriented approach holds further implications for psychiatric diagnoses, implications that have as yet received little attention. Current dsm-style diagnostics is based on set theory; for example, to diagnose borderline personality disorder, ﬁve of nine criteria must be fulﬁlled. The deﬁnition of concepts in terms of a list of criteria, however, is a computationalist notion that cognitive science, especially dynamic cognitive science, has largely abandoned. Categorization along the lines of prototype theory (Rosch & Lloyd, 1978) has clear advantages and, in addition, would be more compatible with the approach favored in this chapter. We propose that psychiatric diagnosis must be reformed at its very basis by considering what is known about psychological categorization.

theory Our ﬁnal remarks, however, go beyond psychopathology: How can we address cognition in general on the basis of the dynamic view? Arguments for a novel view of cognition, one that treats cognition as an emergent property of the mind, have accumulated in the recent past (e.g., Tschacher & Dauwalder, 2003). As we said in the introductory section, this understanding of cognition deviates from the computationalist approach of cognitive psychology, where cognition is an abstract semantic structure running on neuronal hardware, much as a computer program would run on silicon chips. Dynamics-oriented theory is more encompassing in several respects. First, this theory is

118 modeling complex systems in accordance with transdisciplinary systems theories in the natural sciences that study and predict self-organization phenomena. This unitary approach has advantages with respect to the dissemination of both methods and phenomenological ﬁndings across disciplines. Second, in the dynamic framework, information processing and cognition is never separable from emotion and motivation. The exploration of cognition by abstracting from its very driving forces is increasingly viewed as unfeasible in psychology and neuroscience. This integrated view of cognition in context (embodied cognition) is a principal attribute of our model. Third, dynamic systems theory deals with complexity. We have demonstrated that, in cognitive science, two separate sources of complexity must be addressed. On the one hand, cognitive systems themselves are always complex, as only complex systems can, via a self-organizational process, produce adaptive pattern formation. On the other hand, cognitive systems always deal with a complex world, a world that enforces and affords complexity reduction. Continuous simpliﬁcation is, therefore, of the utmost survival value. The main demand on cognition is just this—simplifying the world by the extraction of signiﬁcant patterns on the basis of motivation.

References Andreasen, N. C., Paradiso, S., & O’Leary, D. S. (1998). “Cognitive dysmetria” as an integrative theory of schizophrenia: A dysfunction in corticalsubcortical-cerebellar circuitry? Schizophrenia Bulletin, 24, 203–218. Banerjee, S., Sibbald, P. R., & Maze, J. (1990). Quantifying the dynamics of order and organization in biological systems. Journal of Theoretical Biology, 143, 91–111. Bélair, J., Glass, L., an der Heiden, U., & Milton, J. (Eds.). (1995). Dynamical disease. Woodbury ny: American Institute of Physics Press. Bleuler, E. (1911). Dementia Praecox oder die Gruppe der Schizophrenien. Leipzig: Deuticke. Bonneh, Y. S., Cooperman, A., & Sagi, D. (2001). Motion-induced blindness in normal observers. Nature, 411, 798–801. Carter, R. (2002). Consciousness. London: Weidenfeld & Nicolson. Ciompi, L. (1997). Die emotionalen Grundlagen des Denkens: Entwurf einer fraktalen Affektlogik. Göttingen: Vandenhoeck & Ruprecht. Clark, A. (1997). Being there: Putting brain, body, and world together again. Cambridge ma: mit Press. Conrad, K. (1958). Die beginnende Schizophrenie. Stuttgart: Thieme.

119 Approach to Psychopathology Emrich, H., & Hohenschutz, C. (1992). Psychiatric disorders: Are they “dynamical diseases”? In W. Tschacher, G. Schiepek, & E. J. Brunner (Eds.), Self-organization and clinical psychology (pp. 204–212). Berlin: Springer. Garcia-Toro, M., Blanco, C., Gonzalez, A., & Salva, J. (2001). Psychopathology and the binding problem. Medical Hypotheses, 57(6), 718–723. Globus, G. G., & Arpaia, J. P. (1994). Psychiatry and the new dynamics. Biological Psychiatry, 35, 352–364. Grawe, K. (2004). Psychological therapy. Seattle: Hogrefe & Huber. Haken, H. (1983). Synergetics: An introduction. Berlin: Springer. Haken, H. (1987). Synergetic computers for pattern recognition and associative memory. In H. Haken (Ed.), Computational systems—natural and artiﬁcial: Proceedings of the International Symposium on Synergetics at Schloss Elmau, Bavaria, May 4–9, 1987 (pp. 2–23). Berlin: Springer. Haken, H. (1996). Principles of brain functioning: A synergetic approach to brain activity, behavior, and cognition. Berlin: Springer. Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge ma: mit Press. Kelso, J. A. S. (2003). Cognitive coordination dynamics. In W. Tschacher & J.-P. Dauwalder (Eds.), The dynamical systems approach to cognition (pp. 45–70). Singapore: World Scientiﬁc. Köhler, W. (1920). Die physischen Gestalten in Ruhe und in stationärem Zustand. Braunschweig: Vieweg. Kruse, P., & Stadler, M. (Eds.). (1995). Ambiguity in mind and nature. Berlin: Springer. Kruse, P., Stadler, M., Pavlekovic, B., & Gheorghiu, V. (1992). Instability and cognitive order formation: Self-organization principles, psychological experiments, and psychotherapeutic interventions. In W. Tschacher, G. Schiepek, & E. J. Brunner (Eds.), Self-organization and clinical psychology (pp. 102–117). Berlin: Springer. Kupper, Z. (1999). Dynamische Modelle für chronische psychische Störungen. Lengerich: Pabst Science. Kupper, Z., & Hoffmann, H. (2000). Course patterns of psychosocial functioning in schizophrenia patients attending a vocational rehabilitation program. Schizophrenia Bulletin, 26(3), 681–698. Kupper, Z., & Tschacher, W. (2002). Symptom trajectories in psychotic episodes. Comprehensive Psychiatry, 43(4), 311–318. Kupper, Z., Tschacher, W., & Hoffmann, H. (1997, March). Correlates of order and ordering in the time course of schizophrenia. Poster presented at the Joint Conference on Complex Systems in Psychology/6. Herbstakademie, Gstaad. Kupper, Z., Tschacher, W., & Hoffmann, H. (2004). Longitudinal symptom patterns in psychotic episodes are related to neurocognition and outcome. Schizophrenia Research, 67(1, Suppl.), 272. Lindenmayer, J.-P., Grochowski, S., & Hyman, R. B. (1995). Five factor model of schizophrenia: Replication across samples. Schizophrenia Research, 14, 229–234.

120 modeling complex systems Lütkepohl, H. (1993). Introduction to multiple time series analyses. Berlin: Springer. Mahoney, M. J. (1991). Human change processes: The scientiﬁc foundations of psychotherapy. Chicago: Basic. Matussek, P. (1952). Untersuchungen über die Wahrnehmung: 1. Mitteilung. Archive für Psychiatrie und Zeitschrift für Neurologie, 180, 279–319. Mechsner, F., Kerzel, D., Knoblich, G., & Prinz, W. (2001). Perceptual basis of bimanual coordination. Nature, 414, 69–73. Michotte, A. (1954). Gesammelte Werke. Bern: Huber. Miltner, W., Braun, C., Arnold, M., Witte, H., & Taub, E. (1999). Coherence of gamma-band eeg activity as a basis for associative learning. Nature, 397, 434–436. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs nj: Prentice-Hall. Nicolis, G., & Prigogine, I. (1977). Self-organization in non-equilibrium systems. New York: Wiley. Paulus, M. P., & Braff, D. L. (2003). Chaos and schizophrenia: Does the method ﬁt the madness? Biological Psychiatry, 53, 3–11. Peralta, V., & Cuesta, M. J. (2001). How many and which are the psychopathological dimensions in schizophrenia? Issues inﬂuencing their ascertainment. Schizophrenia Research, 49(3), 269–285. Pezard, L., Nandrino, J.-L., Renault, B., Massioui, F. E., Allilaire, J.-F., Varela, F. J., & Martinerie, J. (1996). Depression as a dynamical disease. Biological Psychiatry, 39, 991–999. Pfeifer, R., & Scheier, C. (1999). Understanding intelligence. Cambridge ma: mit Press. Phillips, W. A., & Silverstein, S. M. (2003). Convergence of biological and psychological perspectives on cognitive coordination in schizophrenia. Behavioral and Brain Sciences, 26, 65–138. Place, E. J., & Gilmore, G. C. (1980). Perceptual organization in schizophrenia. Journal of Abnormal Psychology, 89, 409–418. Rosch, E., & Lloyd, B. B. (Eds.). (1978). Cognition and categorization. Hillsdale nj: Erlbaum. Rössler, O. E. (1976). An equation for continuous chaos. Physics Letters, 57a, 397–398. Rubin, E. (1921). Visuell wahrgenommene Figuren. Copenhagen: Gyldendalske. Scheier, C., Lewkowicz, D. J., & Shimojo, S. (2003). Sound induces perceptual reorganization of an ambiguous motion display in human infants. Developmental Science, 6, 233–244. Schneider, E. D., & Kay, J. J. (1994). Life as a manifestation of the second law of thermodynamics. Mathematical and Computer Modelling, 19, 25–48. Shiner, J. S., Davison, M., & Landsberg, P. T. (1999). On measures for order and its relation to complexity. In W. Tschacher & J.-P. Dauwalder (Eds.), Dynamics, synergetics, autonomous agents: Nonlinear systems approaches to cognitive psychology and cognitive science (pp. 49–63). Singapore: World Scientiﬁc.

121 Approach to Psychopathology Simons, D. J., & Chabris, C. F. (1999). Gorillas in our midst: Sustained inattentional blindness for dynamic events. Perception, 28, 1059–1074. Singer, W., & Gray, C. M. (1995). Visual feature integration and the temporal correlation hypothesis. Annual Review of Neuroscience, 18, 555–586. Spaulding, W. D., Fleming, S. K., Reed, D., Sullivan, M., Storzbach, D., & Lam, M. A. F. (1999). Cognitive functioning in schizophrenia: Implications for psychiatric rehabilitation. Schizophrenia Bulletin, 25, 275–289. Tschacher, W. (1997). Prozessgestalten—die Anwendung der Selbstorganisationstheorie und der Theorie dynamischer Systeme auf Probleme der Psychologie. Göttingen: Hogrefe. Tschacher, W., Baur, N., & Grawe, K. (2000). Temporal interaction of process variables in psychotherapy. Psychotherapy Research, 10, 296–309. Tschacher, W., & Dauwalder, J.-P. (Eds.). (1999). Dynamics, synergetics, autonomous agents: Nonlinear systems approaches to cognitive psychology and cognitive science. Singapore: World Scientiﬁc. Tschacher, W., & Dauwalder, J.-P. (Eds.). (2003). The dynamical systems approach to cognition. Singapore: World Scientiﬁc. Tschacher, W., Dubouloz, P., Meier, R., & Junghan, U. (in press). Altered perception of apparent motion in schizophrenia spectrum disorder. Psychiatry Research. Tschacher, W., Haemmig, R., & Jacobshagen, N. (2003). Time series modeling of heroin and morphine drug action. Psychopharmacology, 165, 188–193. Tschacher, W., & Haken, H. (in press). Intentionality in non-equilibrium systems? The functional aspects of self-organized pattern formation. New Ideas in Psychology. Tschacher, W., & Jacobshagen, N. (2002). Analysis of crisis intervention processes. Crisis, 23, 59–67. Tschacher, W., & Kupper, Z. (2002). Time series models of symptoms in schizophrenia. Psychiatry Research, 113, 127–137. Tschacher, W., & Kupper, Z. (2006). Perception of causality in schizophrenia spectrum disorder. Schizophrenia Bulletin, 32, S106–S112. Tschacher, W., Scheier, C., & Grawe, K. (1998). Order and pattern formation in psychotherapy. Nonlinear Dynamics, Psychology and Life Sciences, 2, 195–215. Tschacher, W., Scheier, C., & Hashimoto, Y. (1997). Dynamical analysis of schizophrenia courses. Biological Psychiatry, 41, 428–437. Tschacher, W., Schuler, D., & Junghan, U. (2006). Reduced perception of the motion-induced blindness illusion in schizophrenia. Schizophrenia Research, 81, 261–267. Uhlhaas, P. J., & Silverstein, S. M. (2005). Perceptual organization in schizophrenia spectrum disorders: Empirical research and theoretical implications. Psychological Bulletin, 131, 618–632. Varela, F. J. (1995). Resonant cell assemblies: A new approach to cognitive function and neuronal synchrony. Biological Research, 28, 81–95. Wiersma, D., Nienhuis, F. J., Slooff, C. J., & Giel, R. (1998). Natural course

122 modeling complex systems of schizophrenic disorders: A 15-year follow-up of a Dutch incidence cohort. Schizophrenia Bulletin, 24, 75–85. Wittchen, H. U., Lieb, R., Pﬁster, H., & Schuster, P. (2000). The waxing and waning of mental disorders: Evaluating the stability of syndromes of mental disorders in the population. Comprehensive Psychiatry, 41, 122–132.

Developing ComputerBased Learning Environments Based on Complex Performance Models Susanne P. Lajoie McGill University

A universal theme in my research, regardless of the population or the subject material studied, is the importance of modeling performance. The word model can be deﬁned in many ways. My use of the term is close to that of the online Merriam-Webster dictionary, which refers to a model as being something set or held before one for guidance or imitation. Interestingly enough, from a psychological perspective, modeling has been deﬁned in a similar manner. Bandura (1977) deﬁned it in terms of how learning can be facilitated by observing others perform the task in question. The beneﬁt of modeling would be that learners do not always have to learn from their own mistakes but can sometimes learn by observing others do a task efﬁciently and then incorporating the ideas thereby obtained. The relevance to learning of such modeling is that learners practice deliberately rather than indiscriminately (Ericsson, 2006). Even though there is some truth to the notion that practice makes perfect, we also know that practicing the piano for 10 years does not make one a concert pianist. Expertise can be developed when practice is sufﬁciently deliberate with feedback regarding performance. There are still three key questions regarding modeling. The ﬁrst one is what to model, the second who or what does the modeling,

124 modeling complex systems and the third how to model effectively. Some answers to these questions are presented, followed by examples from my own research that will help contextualize the arguments made in this chapter.

What to Model In learning and performance contexts it is often the cognitive processes involved in problem solving that need to be modeled. Studies of expertise and proﬁciency have helped in the identiﬁcation of such processes in speciﬁc contexts and have also led to the discovery of commonalities in such components across expertise theories (Alexander, 2003; Chi, Glaser, & Farr, 1988; Ericsson, 2002; Glaser, Lesgold, & Lajoie, 1987; Lajoie, 2003). The groundwork for examining expert/ novice differences was summarized in Chi et al. (1988). Many of the principles of expertise discovered then, and the methodologies for studying expert/novice differences, are used today. The assumption is that identifying dimensions of expertise can lead to improvements in instruction that will, ultimately, result in helping learners become more competent. Instructional improvements are based on modeling these dimensions of expertise. For instance, self-monitoring and metacognition are key elements of expertise; however, we need to identify what experts monitor in a speciﬁc context before we can model these processes for novices. The domain knowledge, the structure of the knowledge, and the strategies that lead to effective problem solutions can all be modeled for the learner, but it is not simply observing things that leads to learning but, rather, using such knowledge and interacting with it with speciﬁc feedback. Studies of expertise have been criticized for being too coldly cognitive (Alexander, 2003). Although the mind, body, spirit connection—where affect and emotions are considered to go hand in hand with the mind (James, 1890)—has long been recognized, we are still struggling as a community to see how these elements interact. Lepper (1988) reintroduced the connection when he called for the intersection of cognition and motivation theories, comparing cognitive psychology to the Tin Man in The Wizard of Oz (Baum, 1969), who wanted a heart, and motivation theory to the Scarecrow, who wanted a brain. It seems that the cognitive scientists are often considered “heartless” in the sense that they focus on the mind and how people

125 Developing Learning Environments learn. However, learning theories are increasingly more inclusive in that cognition, motivation, and the social context in which learning takes place are considered as interconnected (Anderson, Greeno, Reder, & Simon, 2000; Cordova & Lepper, 1996; Mayer, 1997). Mahoney (in this volume) describes lifelong learning from the perspective of embodied cognition, where higher learning and cognition are linked to emotion. Examining learning processes as they occur within “situations” or meaningful contexts goes hand in hand with personal interests when considering what is meaningful. Learning situations in the real world are often social in nature, and team members are part of the situation. Salas, Stagl, Burke, and Goodwin (in this volume) speak to this issue with reference to developing expert teams in natural settings. Their chapter addresses the notion of distributed expertise, which is important in real-world settings. The intersections of learning, motivation, and social situations are becoming elements of research design (Cobb, Confrey, diSessa, Lehrer, & Schauble, 2003). Alexander has edited a special issue on the topic of expertise in the journal Educational Researcher (Vol. 32, No. 8) that points to how researchers study expertise today, pointing toward a shift to a more inclusive theoretical model that parallels the discussion in the editors’ introduction to this volume of the conﬂuence of social, cognitive, and motivational theories in interesting combinations that lead us to a better understanding of human performance (see esp. Alexander, 2003). New models must consider the interleaving of such principles in speciﬁc contexts.

modeling competence across the curriculum The National Academy of Science report by Pellegrino, Chudowsky, and Glaser (2001) provides insight with regard to what to model across the K–12 curriculum. Pellegrino et al. report on a large body of scientiﬁc knowledge about the processes of thinking and learning and the development of competence in speciﬁc curriculum areas. Examining competency or proﬁciency within a speciﬁc context is the ﬁrst step in elaborating a model of thinking that can help the less competent become more proﬁcient in a speciﬁc domain. Pellegrino et al. support a curriculum-instruction assessment triad based on

126 modeling complex systems principles about learning and knowing that can assist learners along a learning trajectory within a speciﬁc ﬁeld. They report that effective learning environments are learner centered and, hence, that instruction must provide scaffolding for solving meaningful problems and supporting learning and understanding. Again, we cannot scaffold learning unless we understand the learning process; only then can we model these processes for novice performers. The same principles would apply in complex learning situations in the real world (see Lajoie, 2003). The implications for teaching and the design of instruction and assessment (Pellegrino, 2002) are clearly based on identifying what students know and what they need to know and having them reﬂect on the processes. Studies of expertise have provided the foundation for developing models of what students need to know with respect to complex performance across domains. Such studies supported the design of methodologies used today to support our understanding of verbal data (see Chi, 1997; Ericsson & Simon, 1993), which are central to documenting problem-solving activity. Cognitive task analysis (cta) is still an appropriate means of discovering what experts know with regard to their problem-solving plans, actions, and mental models (Lesgold, Lajoie, Logan, & Eggan, 1990). In fact, cta is an excellent mechanism for documenting an effective problem space that reﬂects a more inclusive analysis of how individuals who differ in their performance skills go about solving problems. Two examples from complex decision-making activities in the real world, avionics troubleshooting and clinical diagnostic reasoning, reveal that experts can differ in their solution strategies and steps but be guided by common underlying mental models to the correct solution. Analyses of avionics troubleshooters revealed no ideal solution paths and demonstrated that experts solve similar problems differently (Lesgold et al., 1990). However, similarities were found in the underlying mental models that guided their problem solving. Similarly, a cta of the clinical decision making of surgical nurses revealed that expert nurses reached similar decisions, albeit by different routes, demonstrating that there are many paths to solving problems (Lajoie, Azevedo, & Fleiszer, 1998). Differences were observed in hypothesis generation, planning of medical interventions, actions performed, results of evidence gathering, interpretation of the results, heuristics, and overall solution paths.

127 Developing Learning Environments Documenting such problem spaces gives clear guidance for how we might tutor individuals in their problem-solving decisionmaking strategies. Assessment must consider learning at a higher level of abstraction and a wider effective problem space in which assessment of learners takes place (Lajoie, 2003). Developing computer-based learning environments (cbles) and intelligent tutoring systems on the basis of cta can validate performance models by demonstrating that learning does occur in the predicted manner and that providing feedback based on such performance models can lead to greater proﬁciency in trainees. Such an approach was employed in the avionics troubleshooting example mentioned above. The cta led to the development of a cble, “Sherlock,” that provided a practice environment allowing ﬁrst-term airmen to become proﬁcient troubleshooters (Lesgold, Lajoie, Bunzo, & Eggan, 1992). The tutor teaches troubleshooting procedures for dealing with problems associated with an f-15 manual avionics test station. A study was conducted evaluating Sherlock’s effectiveness using 32 trainees from two separate air force bases (Nichols, Pokorny, Jones, Gott, & Alley, 1992). Pre and posttutor assessment was done using verbal troubleshooting techniques as well as a paper-andpencil test. Two groups of subjects per air force base were tested: (1) subjects receiving 20 hours of instruction on Sherlock and (2) a control group receiving on-the-job training over the same period of time. Statistical analyses indicated that there were no differences between the treatment and the control groups on the pretest (means = 56.9 and 53.4, respectively). However, on the verbal posttest as well as on the paper-and-pencil test, the treatment group (mean = 79.0) performed signiﬁcantly better than the control group (mean = 58.9) and equivalently to experienced technicians having several years of on-the-job experience (mean = 82.2). That is, the average gain score for the group using Sherlock was equivalent to almost 4 years of experience. All this is to say that identifying robust performance models in complex domains can lead to effective instruction for those less proﬁcient in problem-solving performance (Shute, Lajoie, & Gluck, 2000). Hence, knowing what to model can lead to better performance. Similarly, sicun, a computer tutor for nurses working in a surgical intensive-care unit (sicu), was designed on the basis of the cta described by Lajoie et al. (1998). The analysis led to the design of

128 modeling complex systems a system that encourages self-monitoring of decision-making processes. Nurses who worked with sicun were required to post their goals before conducting patient assessments and then specify outcomes of their assessments. For example, if their goal was to check the patient’s circulatory system for adequate blood supply to the heart, they might check pulse rate as well as skin (for swelling, edema, coloration, capillary reﬁll, and temperature). Prior to their moving on to a new goal or body system, the tutor would prompt them about the results of their assessment. Hence, plans, goals, actions, and outcomes were all built into the system, with decision trees designed to encourage self-monitoring and comparison to expert problem solvers at various phases of problem solving. A full-scale evaluation of this tutor was not conducted, but the subjects tested became more planful in their patient assessments. The discussion presented above has described how decisions can be made about what to model to learners. The issue of deciding who or what does the modeling is another matter entirely, one that can lead to choices about how to successfully model.

Who or What Does the Modeling? Discussions of cognitive performance models, how complete they are, and who and what should do the modeling abound (Derry & Lajoie, 1993; Lajoie, 2000). What and whom you choose to model complicate these questions. The increasingly integrated models of learning, social processes, and motivation will further challenge us in answering these questions. A simple answer to the “who or what models?” question is that both human beings and computers can model, tutor, or coach individuals and groups of learners. However, the issue of computers as modelers is more contentious than that of human tutors. Members of the artiﬁcial intelligence (ai) and education community continue to struggle with the question of performance models. This struggle seems to parallel changing conceptions of learning theories as well as the difﬁculties associated with methodologies used to instantiate such theories. Learning theories today are described in the context of knowledge construction through realistic tasks and situations, whereas in the past they were described

129 Developing Learning Environments in the context of contained, well-deﬁned laboratory tasks that had solutions. Well-deﬁned tasks are easier to model than some of the ill-deﬁned real-world tasks being modeled today. In moving toward more complex problem-solving tasks, even the most seasoned ai experts have moved away from the modeling camp, Elliott Soloway (1990), for example, suggesting that extracting expert models in more ill-deﬁned problems takes too much time and involves less certainty that all the possible solutions patterns can be captured with computer models. Instead of living with incomplete computer models, Soloway moved away from computer models to human models involving teachers and peers serving as guides on the side. It is somewhat paradoxical that we do not question whether human coaches, teachers, and tutors have all the possible models in their heads, but, when we build computer tutors, we are not content without robust testable models. Musen (in this volume) describes how intelligent databases and expert systems can serve as vehicles for expertise but that domain ontologies that reﬂect problem solving in a domain are becoming increasingly more important in assisting learners. Rather than debating whether human or computer coaches are better at modeling, it might be better to consider the two in tandem than to replace one with the other. Computers can serve as cognitive tools that can amplify, extend, and enhance learning (Jonassen, 1996; Jonassen & Reeves, 1996; Pea, 1985; Salomon, Perkins, & Globerson, 1991). They can do so by providing a mechanism for learners to generate and test hypotheses in complex problem-solving situations (de Jong & van Joolingen, 1998). Furthermore, such tools can be designed to provide information through multiple modalities, thus ensuring opportunities for multiple representations of knowledge (Kozma, Russell, Jones, Marx, & Davis, 1996; Mayer & Moreno, 2002). In addition, given the identiﬁcation and use of complex student models, cbles can be designed to provide scaffolding that is in the form of diagnostic feedback embedded in the learning situation to be offered when assistance is needed (Lajoie, Faremo, & Wiseman, 2001; Lajoie & Lesgold, 1992; Mislevy, Steinberg, Breyer, Almond, & Johnson, 2002). If such sophisticated models are not available, computers can still be used as cognitive tools and used in conjunction with human coaches and peers to scaffold the process (Palincsar & Herrenkohl, 2002).

130 modeling complex systems

How to Model? Just as theories of expertise have guided my decisions about what to model, other frameworks have led me in deciding modeling performance itself. The cognitive-apprenticeship (Collins, Brown, & Newman, 1989) and communities-of-learning (Brown, 1994; Brown, Ellery, & Campione, 1998; Palinscar & Herrenkohl, 2002) frameworks lead the way in terms of linking cognitive science and pedagogy. The cognitive-apprenticeship model (Brown, Collins, & Duguid, 1989; Collins, Brown, & Newman, 1989; Gott, 1989; Lave & Wenger, 1991; Williams, 1992) provides a template for connecting abstract and real-world knowledge by creating forms of pedagogy that are based on a model of learners, the task, and the situation in which learning occurs. When considering a traditional apprenticeship setting, such as apprenticing to become a chef, we think of novices learning from a head chef by observing and modeling physical skills. For example, an apprentice chef observes and learns from the head chef and usually starts with easier tasks, cleaning and preparing vegetables, rather than planning menus and meals for heads of state. The novice, in this example, is supported by the expert chef, who provides coaching in, for example, the correct cutting procedures, and whose active intervention gradually fades into the background when the novice demonstrates that he or she can perform the task independently. A cognitive apprenticeship is more difﬁcult to construct since cognitive rather than physical processes are being modeled for the learner. The literature on expertise reveals that experts are not always able to articulate their problem-solving strategies or their thinking processes since those things are highly routinized or automated (Anderson, 1983). Hence, as described above, methods for externalizing these decision-making strategies must be used in order to build a model of complex performance prior to modeling for novice learners. There are four aspects of the pedagogical model, namely, content, methods, sequence, and sociology. I have described methods for documenting the content knowledge earlier in this chapter. The pedagogical methods employed by this framework involve modeling expertise and coaching learners with feedback in a manner that supports differences in knowledge acquisition. Coaches scaffold knowledge by providing the right amount of support to learners while they are performing a task. Not everyone needs the

131 Developing Learning Environments same amount of help or scaffolding. Once students demonstrate that they can do something on their own, the scaffolding is gradually removed since the goal is for students to become independent learners and problem solvers. The methods include opportunities for articulation, reﬂection, and exploration. Learners articulate what they know either through performances or through communication of some sort. Reﬂection can consist of self-assessments, and it can include comparisons of student processes with more proﬁcient models of performance. Exploration and active problem solving in meaningful contexts are key features of the model. The sequence of instruction takes the complexity of the content material into account, ensuring that students are challenged with meaningful tasks but not overwhelmed by information that they are not ready for. Hence, students participate in the global or overall task by performing smaller components of the overall task prior to engaging in the speciﬁc local skills from the start. In this way they are still participating in producing a ﬁnal product and developing a conceptual model of how the overall task is performed rather than working in isolation. The sociology of such real-world tasks usually occurs in small groups or teams of people working together. Once again, head chefs have a team of sous chefs who help in the overall preparation of a banquet. Similarly, with tasks requiring cognitive skills, the work can be distributed among a team of individuals working toward a common goal. Learning in this way must take the sociology of the group into consideration. Learning theories and pedagogical models alike are considering issues of shared cognition more carefully. We see similar considerations with training situations in the real world (Cannon-Bowers & Salas, 2001). In both in- and out-of-school learning situations, communities of learning (col; Brown, 1994; Brown et al., 1998) and communities of practice (cop; Barab & Duffy, 2000; Wenger, 1999), respectively, are being described, created, and empirically studied. The col model engages students in meaningful research activities, each student in the community playing a role as the community progresses toward the goal of learning. The cop model is employed in real-world situations, allowing students of such professions as medicine and law to work together to solve cases. The social sharing of information in this approach is essential to the ﬁnal outcome of the task. Similarities exist between these models and the cognitive-ap-

132 modeling complex systems prenticeship framework discussed above. Both approaches include in their descriptions reference to the sociology of the environment or the notion of community. Both suggest ways of making schooled learning more meaningful as a way of easing the transition between school and practice. Both use models of scaffolding that are based on the assumption that multiple zones of proximal development (Vygotsky, 1978) exist within the group and, hence, that feedback must adapt to those developmental zones. Both approaches use the models of expertise within their design of effective instruction. The col approach is unique in the types of participant structures that it provides as examples of modeling, turn taking, and role-playing to serve as new ways of sharing expertise and constructing learning within a community. So far I have described ways of discovering what to model, who or what can do the modeling, and how to model cognitive processes. The remainder of this chapter will contextualize these ideas with examples from my own research.

What to Model: Differences in Diagnostic Reasoning My research-and-development efforts aim at integrating instruction and assessment in contexts that reﬂect real-world situations. In the area of high school science there has been a call for instruction that provides more meaningful, more challenging, and richer learning experiences, experiences that foster the development of reasoning abilities that can apply to real-world practice (Johnson & Lawson, 1998; Metz, 1995, 1997). In this regard I developed “BioWorld,” a cble that serves as a platform for evoking evidence of knowledge and capturing complex performance as high school students develop scientiﬁc-reasoning skills in the context of problem solving about patient cases (Lajoie, Greer, Munsie, Wilkie, Guerrera, & Aleong, 1995; Lajoie, Lavigne, Guerrera, & Munsie, 2001). BioWorld serves to instruct, model proﬁciency, and assess knowledge. Students learn about diseases associated with the body systems they are studying. Patient cases are presented in the context of hospital simulations, and students work collaboratively to collect evidence to conﬁrm or refute hypotheses. Patient cases were developed with the help of experts, and a cta established robust models of such ill-structured problems.

133 Developing Learning Environments

Figure 1: The current BioWorld interface: case history screen.

In determining what to model, we asked medical experts to solve the patient cases designed for BioWorld, in an attempt to get a robust expert model of problem solving for each case. By identifying what experts know, it is possible to develop assessment genres for examining transitions in learning in novice performers. Expert problem-solving traces were monitored and collected dynamically, as were verbal protocols. These data determined the unique proﬁles of experts in terms of plans, strategies, and actions within BioWorld demonstrating different clusters of competencies that served to establish benchmarks of performance (Lajoie et al., 1995). Figure 1 presents an annotated screen shot of the BioWorld interface. The patient case is presented and evidence from this case can be collected and stored in the evidence table. Hypotheses can be selected and changed throughout the problem-solving activity, as can one’s beliefs about said hypotheses. For instance, you might select cirrhosis as a hypothesis, but your conﬁdence that it is a correct hypothesis might be weak. Results have shown that learners’ conﬁdence levels increase as they acquire more knowledge (Lajoie, La-

134 modeling complex systems

Figure 2. Ordering diagnostic tests.

vigne, et al., 2001). Interestingly enough, similar ﬁndings are found in clinical psychology, where treatment outcomes are related to increases in clients’ self-efﬁcacy (Tschacher & Kupper, in this volume). Conﬁdence and knowledge building are intertwined. BioWorld also presents an online library where factual information is provided, as is a patient chart, allowing learners to practice ordering diagnostic tests. Consultation is also provided online. As experts solve patients’ cases, they select relevant evidence from the patients’ medical histories, collect pertinent information in an online medical library, perform diagnostic tests pertinent to the hypothesis (see Figure 2), use systematic plans and actions throughout problem solving, and make ﬁnal arguments based on evidence (see Figure 3). Records of expert competency clusters are established and used to monitor novice students electronically in terms of the proportion of overlap of their actions with each predetermined cluster. Instructional feedback was based on the dynamic assessment of actions throughout problem solving. Students also learn to self-assess by comparing their reasoning with that of an expert (as shown in Figure 3). With practice, students became more systematic in their

135 Developing Learning Environments

Figure 3. Expert evidence as a reﬂection device for self-assessing ﬁnal argument.

reasoning and more strategic about evidence collection and hypothesis testing, indicating that instruction based on dynamic assessment can lead to transitions in learning (for evidence of these claims, see Lajoie, Lavigne, et al., 2001). Decisions regarding what to model led to design decisions regarding the types of cognitive tools that BioWorld provides to learners. In essence the evidence palette provides a mechanism for learners to monitor what they thought was important to the solution process. They can reﬂect on their own thinking and also compare their solutions with an expert’s path at the completion of the problem. Rather than modeling a sequence of problem-solving processes, BioWorld highlights relevant information in the decision-making process. The consult button was designed to model expertise in the form of feedback. The success of BioWorld with high school students has led us to consider ways to reinvent this cble for medical students. In order to better understand developing expertise, Faremo (2004) examines how medical students and residents solve medical cases, compared

136 modeling complex systems to experts. Her analysis of the cognitive processes involved in diagnostic reasoning represents the knowledge levels and qualitative differences in argumentation and reasoning patterns in this medical cohort. Through verbal-protocol analyses she was able to capture a picture of emerging expertise as individuals reason through cases. Her coding schemes are based on the plans, evidence identiﬁcation, and factual and procedural knowledge speciﬁc to a task. As an example, Figure 4 presents a representation of how a medical student’s problem-solving trace compared to that of a resident and expert in diagnosing a patient with celiacs. The patient case took the form of a narrative that presents a 27-year-old male with a 2-year history of intermittent diarrhea, weight loss, and anxiety. Celiacs is a chronic intestinal malabsorption disorder caused by intolerance to gluten. It presents in a variety of ways, and some direct clues (diarrhea, abdominal discomfort, and distention) may be present. Iron deﬁciency is found in affected adults, and there can be (1) low albumin, calcium, potassium, and sodium and (2) elevated alkaline phosphotase and prothrombin time. The disorder is diagnosed by means of symptoms, lab studies, and X-rays (antigliadin antibodies are the gold standard test for detection). Figure 4 demonstrates the relation between the plans or goals experts had, the types of data they collected and reviewed, and the hypotheses they generated in the context of reasoning about this case. It also shows that the students identiﬁed some of the same information as relevant, but did not engage in the same systematic planning and evidence evaluation, as experts. This type of in-depth analysis of diagnostic reasoning across cohorts can eventually lead to robust models of problem solving that can extend the use of BioWorld to higher-education classrooms and complete the picture of emerging expertise in this domain.

Who or What to Model: Human and Computer Coaches in Medical School Medical tutorials present a naturalistic mechanism for studying apprenticeship. We studied human tutors in medicine to develop a model of the medical tutorial situation for studying clinical problem-solving tasks (Lajoie, Wiseman, & Faremo, 2000). As with any apprenticeship there are strengths and weaknesses. The strength of

Figure 4. Expert model with student overlay for the case of a celiac.

138 modeling complex systems the setting is that realistic patient cases are seen and handled. The weakness of the natural apprenticeship is that what you see is what you get. In other words, there is no standardization of curriculum or guarantee that all students will see the types of patient cases that will build on their existing knowledge. We were interested in documenting the types of diagnostic feedback that human tutors gave to students as well as describing the learning context. In doing so our attempt was to develop a cognitive-apprenticeship model for diagnostic feedback that might be built into a cble. What we discovered was that a col approach was common in these tutoring sessions, whereby small groups of individuals shared the problem and distributed their learning through shared representations, articulation, and reﬂection, with a human tutor scaffolding as needed. Two lessons were learned from this setting. The ﬁrst lesson was that contextualized and diagnostic feedback could be studied in these settings and that elements of scaffolding and gradual removal, articulation and reﬂection, were, in fact, all part of the tutorial process. Hence, an argument could be made that the pedagogical methods of the cognitive-apprenticeship model were used and could be documented. The second lesson was that this environment had to consider the sociology of the group and the roles of each member if it was to be an authentic tutorial.

How to Model: The Social Context We are in the process of using the above ﬁndings in the construction of better tools for medical tutorials. As we move the BioWorld cble into a medical setting, we have designed the “CaseBuilder” (see Lajoie, Faremo, & Wiseman, 2001) for instructors’ use in building their curriculum of cases in areas that they want their students to have more knowledge about. The elements of the CaseBuilder will provide the instructor with tools to build in links to library and test information. The library becomes more complex with medical populations, and more complex studies of the link between knowledge representation and decision making in the context of information-access systems are needed (Nakamura & Lajoie, 2003). More important, not only do the instructors build the case, but they can also add feedback at key locations where they anticipate instruc-

139 Developing Learning Environments tional moments for their students. They can also run the case with students to see whether they need to revise it for better instruction and assessment purposes. Incorporating the social context into performance modeling is more difﬁcult. Medical students learn from experienced practitioners and teachers (experts) and other students and residents (intermediates) during their training. Students attend lectures, participate in problem-based groups, observe others, and participate in discussions about hospital patients. We are currently looking at ways to build a more distributed model of learning for this population in the context of problem-based learning situations. Lu and Lajoie (2003) describe three types of cognitive tools that will be considered in this situation: visualization tools, to make knowledge representations more visible to the teams involved; knowledge-building tools, to help people add on to existing knowledge in speciﬁc ways; and management tools, to facilitate the structuring and tracking of information between small groups. On a clinical teaching ward, there are experts in many different content areas collaborating hierarchically to work on a given case. These experts also continuously interact within a hierarchy of novice to intermediate learners (medical students, interns, and residents). We are still researching ways of designing tools that will facilitate these exchanges in a somewhat controlled setting, allowing individuals to share knowledge collaboratively.

Conclusion By asking and answering a set of questions, I have attempted to address the intersection between performance modeling and the design of cbles. These questions are: What do we chose to model? Who or what should we model? How do we model complex decision making? Theory-based arguments were provided in response to each question. The literature on expertise and competency was discussed in the context of the empirical ﬁndings from this literature as well as the methodological tools that were offered from that literature. In essence, selecting what to model must be speciﬁc to the domain and problem tasks in question. Better-informed models of competency can provide transparent trajectories for learning in such domains.

140 modeling complex systems Decisions regarding who or what should do the modeling are based on guiding learning paradigms. For instance, social constructivists may prefer that modeling occur in groups, with more expert peers assisting the learning process. Cognitive constructivists may agree that models provided in a cble can extend learning effectively when individuals interactively engage in the problem-solving activity. Not all modeling is done through explicit coaching, whether the tutor be human or computer. The design of cbles with cognitive tools can support learning by constraining problem solving through support of the overall learning process. In answering the how-to-model question, the cognitive-apprenticeship, col, and cop approaches were described, approaches that support speciﬁc pedagogical methods in order to promote learning through effective models. Subsequent to these discussions, examples from science and medicine were presented to demonstrate how such approaches guide the design of instruction and assessment in complex learning situations. There are still unanswered questions. As modeling moves beyond the individual learner and includes a small group of learners, the complexity of assessment is increased. Although it is widely recognized that real-world learning often takes place in small groups, it is difﬁcult to develop both individual and group models of performance when deciding appropriate levels of feedback to move the problem-solving process forward. New ways of supporting differences will need to be developed, and new ways of sharing and documenting the group process will need to parallel authentic practice.

Note Research reported in this chapter was made possible through funding provided by the following granting agencies: the Canadian Social Sciences and Humanities Research Council; the Quebec Ministry of Industry, Commerce, Science and Technology; Valorisation Recherche Quebec; Ofﬁce of Educational Research and Improvement. Many graduate students (former and current) have contributed to the work that is reported here. Special thanks to Arshad Ahmad, Roger Azevedo, Gloria Berdugo, Janet Blatter, Andrew Chiarella, Lucy Cumyn, Sonia Faremo, Genevieve Gauthier, Claudia Guerrera, Nancy Lavigne, Susan Lu, Carlos Nakamura, Thomas Patrick, and Jeffrey Wiseman.

141 Developing Learning Environments

References Alexander, P. A. (2003). The development of expertise: The journey from acclimation to proﬁciency. Educational Researcher, 32(8), 10–14. Anderson, J. R. (1983). The architecture of cognition. Cambridge ma: Harvard University Press. Anderson, J. R., Greeno, J. G., Reder, L., & Simon, H. A. (2000). Perspectives on learning, thinking and activity. Educational Researcher, 29(4), 11–13. Bandura, A. (1977). Social learning theory. Englewood Cliffs nj: PrenticeHall. Barab, S. A., & Duffy, T. M. (2000). From practice ﬁelds to communities of practice. In D. H. Jonassen & S. M. Land (Eds.), Theoretical foundations of learning environments (pp. 25–55). Mahwah nj: Erlbaum. Baum, F. L. (1969). The wizard of Oz. Chicago: Children’s Press. (Original work published 1900) Brown, A. L. (1994). The advancement of learning. Educational Researcher, 2(8), 4–12. Brown, A. L., Ellery, S., & Campione, J. C. (1998). Creating zones of proximal development electronically. In J. G. Greeno & S. Goldman (Eds.), Thinking practices: A symposium in mathematics and science education (pp. 341–367). Hillsdale nj: Erlbaum. Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the culture of learning. Educational Researcher, 18(1), 32–42. Cannon-Bowers, J. A., & Salas, E. (2001). Reﬂections on shared cognition. Journal of Organizational Behavior, 22(2), 195–202. Chi, M. T. H. (1997). Quantifying qualitative analyses of verbal data: A practical guide. Journal for the Learning Sciences, 6(3), 271–316. Chi, M. T. H., Glaser, R., & Farr, M. J. (1988). Introduction to M. T. H. Chi, R. Glaser, & M. J. Farr (Eds.), The nature of expertise (pp. xv–xxvi). Hillsdale nj: Erlbaum. Cobb, P., Confrey, J., diSessa, A., Lehrer, R., & Schauble, L. (2003). Design experiments in educational research. Educational Researcher, 32(1), 9–13. Collins, A., Brown, J. S., & Newman, S. E. (1989). Cognitive apprenticeship: Teaching the craft of reading, writing, and mathematics. In L. B. Resnick (Ed.), Knowing, learning, and instruction: Essays in honor of Robert Glaser (pp. 453–494). Hillsdale nj: Erlbaum. Cordova, D. I., & Lepper, M. R. (1996). Intrinsic motivation and the process of learning: Beneﬁcial effects of contextualization, personalization, and choice. Journal of Educational Psychology, 88(4), 715–730. de Jong, T., & van Joolingen, W. R. (1998). Scientiﬁc discovery learning with computer simulations of conceptual domains. Review of Educational Research, 68(2), 179–201. Derry, S. J., & Lajoie, S. P. (1993). A middle camp for (un)intelligent computing. In S. P. Lajoie & S. J. Derry (Eds.), Computers as cognitive tools (pp.1– 11). Hillsdale nj: Erlbaum. Ericsson, K. A. (2002). Attaining excellence through deliberate practice: In-

142 modeling complex systems sights from the study of expert performance. In M. Ferrari (Ed.), The pursuit of excellence in education (pp. 21–55). Hillsdale nj: Erlbaum. Ericsson, K. A. (2006). The inﬂuence of experience and deliberate practice on the development of superior expert performance. In K. A. Ericsson, N. Charness, P. J. Feltovich, & Robert R. Hoffman (Eds.), The Cambridge handbook of expertise and expert performance (pp. 223–241). Cambridge: Cambridge University Press. Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data (Rev. ed.). Cambridge ma: mit Press. Faremo, S. (2004). Examining medical problem solving in a computer-based learning environment. Unpublished doctoral dissertation, McGill University, Montreal. Glaser, R., Lesgold, A., & Lajoie, S. (1987). Toward a cognitive theory for the measurement of achievement. In R. R. Ronning, J. Glover, J. C. Conoley, & J. C. Witt (Eds.), The inﬂuence of cognitive psychology on testing (Vol. 3, pp. 41–85). Hillsdale nj: Erlbaum. Gott, S. P. (1989). Apprenticeship instruction for real world cognitive tasks. Review of Research in Education, 15, 97–169. James, W. (1890). The principles of psychology. New York: Henry Holt. Johnson, M. A., & Lawson, A. E. (1998). What are the relative effects of reasoning ability and prior knowledge on biology achievement in expository and inquiry classes? Journal of Research in Science Teaching, 35(1), 89–103. Jonassen, D. H. (1996). Computers in the classroom: Mindtools for critical thinking. Columbus oh: Prentice-Hall. Jonassen, D. H., & Reeves, T. C. (1996). Learning with technology: Using computers as cognitive tools. In D. H. Jonassen (Ed.), Handbook of research for educational communications and technology (pp. 693–719). New York: Macmillan. Kozma, R. B., Russell, J., Jones, T., Marx, N., & Davis, J. (1996). The use of multiple, linked representations to facilitate science understanding. In S. Vosniadou, E. De Corte, R. Glaser, & H. Mandl (Eds.), International perspectives on the design of technology-supported learning environments (pp. 41–60). Hillsdale nj: Erlbaum. Lajoie, S. P. (2000). Breaking camp to ﬁnd new summits. In S. P. Lajoie (Ed.), Computers as cognitive tools: Vol. 2. No more walls (pp. xv–xxxii). Mahwah nj: Erlbaum. Lajoie, S. P. (2003). Transitions and trajectories for studies of expertise. Educational Researcher, 32(8), 21–25. Lajoie, S. P., Azevedo, R., & Fleiszer, D. (1998). Cognitive tools for assessment and learning in a high information ﬂow environment. Journal of Educational Computing Research, 18(3), 205–235. Lajoie, S. P., Faremo, S., & Wiseman, J. (2001). Tutoring strategies for effective instruction in internal medicine. International Journal of Artiﬁcial Intelligence and Education, 12(3), 293–309. Lajoie, S. P., Greer, J. E., Munsie, S. D., Wilkie, T. V., Guerrera, C., & Aleong, P. (1995). Establishing an argumentation environment to foster scientiﬁc

143 Developing Learning Environments reasoning with Bio-World. In D. Jonassen & G. McCalla (Eds.), Proceedings of the International Conference on Computers in Education (pp. 89–96). Charlottesville va: aace. Lajoie, S. P., Lavigne, N. C., Guerrera, C., & Munsie, S. D. (2001). Constructing knowledge in the context of BioWorld. Instructional Science, 29(2), 155–186. Lajoie, S. P., & Lesgold, A. (1992). Apprenticeship training in the workplace: A computer-coached practice environment as a new form of apprenticeship. In M. Farr & J. Psotka (Eds.), Intelligent instruction by computer: Theory and practice (pp. 15–36). New York: Taylor & Francis. Lajoie, S. P., Wiseman, J., & Faremo, S. (2000, June). Tutoring strategies for effective instruction in internal medicine. Paper presented at the International Conference on Intelligent Tutoring Systems, Montreal. Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. New York: Cambridge University Press. Lepper, M. R. (1988). Motivational considerations in the study of instruction. Cognition and Instruction, 5(4), 289–309. Lesgold, A., Lajoie, S. P., Bunzo, M., & Eggan, G. (1992). sherlock: A coached practice environment for an electronics troubleshooting job. In J. H. Larkin & R. W. Chabay (Eds.), Computer assisted instruction and intelligent tutoring systems: Shared goals and complementary approaches (pp. 201–238). Hillsdale nj: Erlbaum. Lesgold, A., Lajoie, S. P., Logan, D., & Eggan, G. M. (1990). Cognitive task analysis approaches to testing. In N. Frederiksen, R. Glaser, A. Lesgold, & M. Shafto (Eds.), Diagnostic monitoring of skill and knowledge acquisition (pp. 325–350). Hillsdale nj: Erlbaum. Lu, J., & Lajoie, S. P. (2003). Fostering medical problem solving in a collaborative Web-based learning environment. In Y. S. Chee, N. Law, K. Lee, & D. Suthers (Eds.), Proceedings of the International Conference on Computers in Education (pp. 337–340). Charlottesville va: aace. Mayer, R. E. (1997). Learners as information processors: Legacies and limitations of educational psychology’s second metaphor. Educational Psychologist, 31(3/4), 151–161. Mayer, R. E., & Moreno, R. (2002). Aids to computer-based multimedia learning. Learning and Instruction, 12(1), 107–119. Metz, K. E. (1995). Reassessment of developmental constraints on children’s science instruction. Review of Educational Research, 65(2), 93–127. Metz, K. E. (1997). On the complex relation between cognitive developmental research and children’s science curricula. Review of Educational Research, 67(1), 151–163. Mislevy, R. J., Steinberg, L. S., Breyer, F. J., Almond, R. G., & Johnson, L. (2002). Making sense of data from complex assessment. Applied Measurement in Education, 15(4), 363–389. Nakamura, C., & Lajoie, S. P. (2003, June). Informational and instructional content in digital learning environments. Paper presented at the Educational Media Conference, Honolulu hi.

144 modeling complex systems Nichols, P., Pokorny, R., Jones, G., Gott, S. P., & Alley, W. E. (1992). Evaluation of an avionics troubleshooting tutoring system (Tech. Rep.). Brooks City tx: Brooks Air Force Base, Human Resources Directorate, Armstrong Laboratory. Manuscript in preparation. Palincsar, A. S., & Herrenkohl, L. (2002). Designing collaborative learning contexts. Theory into Practice, 41(1), 26–32. Pea, R. D. (1985). Beyond ampliﬁcation: Using the computer to reorganize mental functioning. Educational Psychologist, 20(4), 167–182. Pellegrino, J. (2002, February). Understanding how students learn and inferring what they know: Implications for the design of curriculum, instruction and assessment. Paper presented at the annual National Science Foundation K– 12 Math, Science, and Curriculum Implementation Projects Conference, Reston va. Pellegrino, J., Chudowsky, N., & Glaser, R. (Eds.). (2001). Knowing what students know: The Science and Design of Educational Assessment. Washington dc: National Academy Press. Salomon, G., Perkins, D. N., & Globerson, T. (1991). Partners in cognition: Extending human intelligence with intelligent technologies. Educational Researcher, 20(3), 2–9. Shute, V. J., Lajoie, S. P., & Gluck, K. (2000). Individual and group approaches to training. In S. Tobias & H. O’Neill (Eds.), Handbook on training (pp. 171–207). Washington dc: American Psychological Association. Soloway, E. (1990, July). Interactive learning environments. Paper presented at the nato Advanced Studies Institute, Calgary. Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes. Cambridge ma: Harvard University Press. Wenger, E. (1999). Communities of practice: Learning, meaning, and identity. New York: Cambridge University Press. Williams, S. M. (1992). Putting case-based instruction into context: Examples from legal and medical education. Journal for the Learning Sciences, 2(4), 367–427.

Technology for Building Intelligent Systems: From Psychology to Engineering Mark A. Musen Stanford University

The Vision of Artiﬁcial Intelligence People in industrialized societies are surrounded by computer systems that demonstrate intelligent behaviors (Steﬁk, 1995; Winston & Prendergast, 1986). They help us prepare our tax returns. They help us navigate our cars. They evaluate our applications for credit. We take it for granted that human knowledge can somehow be embedded in computer systems and that computer systems can help people solve knowledge-intensive tasks—or can solve such tasks completely autonomously. For better or for worse, we live in “the age of the smart machine” (Zuboff, 1988). The formulation, communication, and application of knowledge are fundamental aspects of being human. All people share a basic motivation to share knowledge within their social groups and to codify knowledge so that others can take advantage of it. From ancient peoples who ﬁrst tracked the stars in the night sky to contemporary industrial workers who wish to clarify best practices for their organizations, there is an undeniable drive to elucidate and to share information. It is no wonder that computers and their associated technologies—the most powerful communication vehicles that have yet to be invented—provide the infrastructure for much of our attempt to manage knowledge in modern society (Liebowitz, 1999).

146 modeling complex systems Because computers have been mass-produced for only a few decades, it is not surprising that the ﬁeld of artiﬁcial intelligence (ai) is relatively young. What is surprising is that, nearly as soon as computers became available for general research, investigators began to study how computers might be programmed to do intelligent things. The 1950s saw the birth of quite sophisticated programs for playing checkers, for proving mathematical theorems, and for processing natural language. By 1956 there was sufﬁcient interest in the challenge of getting computers to demonstrate intelligent behavior that John McCarthy and several colleagues convened a two-month meeting of workers in computer science, psychology, philosophy, and economics at Dartmouth College to study a new ﬁeld that McCarthy dubbed artiﬁcial intelligence. In the original proposal for the Dartmouth conference, McCarthy, Minsky, Rochester, & Shannon (1955) declared: “The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to ﬁnd how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.” This grand vision for ai has been achieved, to date, only in science ﬁction. Enormous progress has been made on components of this montage of capabilities, but no computer system can be viewed as generally artiﬁcially intelligent. Some programs can analyze or generate natural language. Others can form abstractions from primitive data. Some can solve problems typically reserved for humans. For a host of reasons, no one has created ai in the composite sense. The computer programs that help us prepare our taxes and that offer driving directions are extremely useful, nevertheless, and represent a class of computer programs that have been enormously successful in our society. We refer to such programs as intelligent systems. This chapter traces some of the history of intelligent systems, outlining the obstacles that developers have overcome, and suggesting methods by which such systems can be built and maintained. The emphasis is on the representation of human knowledge in computers and on the primitive building blocks that make such representations possible. The story begins with a close alignment between developers of intelligent systems and scholars of cognitive psychology. It ends, however, with a perspective that construes the devel-

147 Building Intelligent Systems opment of intelligent systems more as software engineering than as applied psychology. By not attempting to replicate human cognition but instead attempting to take advantage of what is known about the construction of good software systems, developers now can build computer programs that are extremely useful, reliable, maintainable, and—many would say—intelligent. Our goals, however, must be more focused than those of the participants in the 1956 Dartmouth conference. We must recognize that the computer systems that will be most useful to society are not necessarily those that might replicate human cognition but rather those that allow us to store, communicate, and apply human knowledge in speciﬁc contexts when it is needed. The elucidation of the methods that are needed to enable system builders to create such software artifacts can be just as exciting as the elucidation of human cognition itself. I ﬁnd it particularly helpful to look at the evolution of thought regarding knowledge-based systems in a historical context. The broad picture of how developers of intelligent computer programs have viewed the role of psychology in their craft over the years tells a very interesting story. This is a book chapter, however, not a book, and, therefore, I have to be selective in what I say. The reader should be aware that the need to streamline the presentation requires that I simplify (and sometimes oversimplify) parts of the story.

The Rise of Rule-Based Systems When most computer scientists think of intelligent computer programs, they usually think of computer programs that reason using large sets of rules. Enthusiasm for rule-based systems started in the 1960s, when Ed Feigenbaum, Carl Djerassi, and Joshua Lederberg teamed up at Stanford University to create dendral, a program that could interpret mass-spectroscopy data from the chemical laboratory (Lederberg, Sutherland, Buchanan, Feigenbaum, & Lindsay, 1980). By the 1970s, Ted Shortliffe and Bruce Buchanan had built mycin, a program that could diagnose causes of infectious diseases and recommend treatment as well as physicians could (Buchanan & Shortliffe, 1984). By the end of the 1970s, there were scores of wellknown expert systems that used rule-based programming and the explicit representation of domain-speciﬁc knowledge to solve a variety

148 modeling complex systems of difﬁcult tasks in application areas that ranged from medicine to geology to computer-hardware conﬁguration. The notion of rulebased programming became so popular that the terms rule-based system and expert system began to be used nearly interchangeably; if a computer behaved intelligently, there was an assumption that the program was using rules. (In this chapter, I avoid the term expert system, which is now becoming dated. I use the terms intelligent system and knowledge-based system as synonyms that refer to any computer system that encodes domain-speciﬁc knowledge for the purpose of driving problem solving, regardless of whether it uses rules.) Rule-based systems encode domain knowledge in an intuitive if-then formalism. In the mycin system, for example, there were hundreds of rules such as the following: Rule 124 if: 1) The site of the culture is throat, and 2) The identity of the organism is Streptococcus then: There is strongly suggestive evidence that the subtype of the organism is not Group D Although the rules were actually written in a dialect of the lisp programming language, the computer could translate them into understandable English. The mycin system used the rules to reason about descriptions of patients entered into the program (Figure 1). A user of mycin presumed that the patient to be described had either bacteremia (bacteria in the blood) or meningitis. At the time of the user’s interaction with mycin, the patient ostensibly was ill, but it was not clear what the source of the infection was. The program would analyze the case description, asking questions of the user to gather all the relevant data, and then determine (1) the organism or organisms most likely causing the patient’s infection and (2) the antibiotics that should be administered for optimal treatment (Buchanan & Shortliffe, 1984). A program known as an inference engine interpreted the rules in mycin and allowed one rule to invoke other rules in order to reach the program’s ﬁnal conclusions. For example, when evaluating Rule 124 (given above), the mycin inference engine would examine the ﬁrst clause of the rule and ask: “Is the site of the culture the throat?”

149 Building Intelligent Systems

Figure 1. A trace of a user’s dialog with the mycin system. The program posed questions to the user, allowing the entry of information regarding a patient to be treated for presumptive bacteremia or meningitis. Each question seeks information necessary to determine the truth value of one of the program’s many production rules.

If the site from which the culture under consideration was taken was not the throat, then the rule would fail, and further evaluation of Rule 124 would not take place. If, however, the site of the culture was, indeed, the throat, then the mycin inference engine would consider the second clause of the rule and ask: “Is the identity of the organism under consideration Streptococcus?” If the organism was not streptococcus, then the rule would fail. If the organism was, indeed, streptococcus, however, then the rule would succeed, and mycin would conclude that the subtype of the streptococcus is not Group D. What if the identity of the organism under consideration were not known one way or the other? The mycin inference engine would examine the rule base to see whether there were any rules that had not yet ﬁred that would allow the system to conclude whether the identity of the organism was streptococcus. If there were such rules, mycin would suspend its evaluation of Rule 124 and go on to evaluate those other rules; once it was known whether the identify of the organism was streptococcus, the mycin inference engine would return to Rule 124 to determine whether the second clause was true

150 modeling complex systems or false. If the second clause evaluated true, then the system would conclude that the subtype is not Group D. mycin’s inference engine thus used the hundreds of rules in the knowledge base to evaluate the case at hand and to generate recommendations for the user. Starting with an overarching goal rule (basically: “if you know what are the likely organisms causing the infection then recommend treatment”), the inference engine would invoke rule after rule recursively to conclude everything it could about the case at hand so that the program could then determine the appropriate course of action. This style of computer programming is known as a production system. Each production rule in the system has a left-hand side (a condition) and a right-hand side (a conclusion). An inference engine such as the one in mycin starts with the goal rule and determines whether the condition predicating the rule is true. If so, then the rule “produces” its conclusion; if not, then the inference engine considers the truth value of other rules that can “produce” as their conclusion the condition of the rule currently under consideration. Examination of these rules may cause yet other rules to be invoked. The procedure stops when the truth value of the goal rule becomes known. The wonderful thing about the mycin inference engine was that it was a simple procedure. The inference engine contained no knowledge of infectious diseases or their treatment. It was designed only to invoke rules to follow chains of inference; all the domain-speciﬁc knowledge was encoded strictly in the rules. Thus, it was a relatively straightforward matter to apply mycin-style inference engines to an entirely different rule base—allowing the system to evaluate cases and make recommendations in new application areas far aﬁeld from medicine. The idea of rule-based systems caught on like wildﬁre. It was extremely appealing to view knowledge bases as collections of rules. The rules were viewed as modular “chunks” of knowledge that developers could drop into an expanding knowledge base as new capabilities were deﬁned. Rules were viewed as independent of one another and not subject to interaction effects as a knowledge base grew. Most important, workers in ai at the time believed that mycin-style rules had verisimilitude with components of actual human problem solving.

151 Building Intelligent Systems

Figure 2. A cryptarithmetic problem. Newell and Simon (1972) asked subjects to replace each letter consistently with a digit in order to create a mathematically correct formula.

Rule-Based Systems as Cognitive Models The surge of interest in rule-based systems in the 1970s followed the explosion of interest in cognitive psychology that took root in the 1960s. Although not all ai researchers believed that the goal of automating intelligence would, ultimately, require a deeper understanding of psychology, many computer scientists were greatly inspired by investigation to developer comprehensive models of human information processing. Perhaps most inﬂuential to workers on ai was the research that Newell and Simon (1972) had done to study human problem solving. Newell and Simon studied how people solve cryptarithmetic puzzles (Figure 2), prove theorems in logic, and select moves in chess. They presented subjects with well-deﬁned problems to solve and carefully recorded and analyzed statements made by their subjects as they “thought out loud” about the problems that they were given. They used the term protocol analysis to deﬁne the manner in which they carefully recorded the stimuli to which their subjects seemed to attend, the goals to which the subjects seemed to be responding, and the actions that the subjects considered when confronted with a task to address (Ericsson & Simon, 1993). (Developers of intelligent systems have borrowed heavily from this same methodology when attempting to elicit the knowledge relevant for building electronic knowledge bases.) Newell and Simon’s (1972) empirical analysis of protocols from scores of laboratory experiments encouraged them to adopt an information-processing perspective on cognition, viewing human problem solving as a process in symbolic reasoning. They described problem solving as the serial processing of symbols transferred from long-term memory to short-term memory in response to external

152 modeling complex systems and internal stimuli. They saw in their data a common problem-solving procedure that looked as if subjects were performing rule-based reasoning. They intentionally were guarded when they claimed to “confess to a strong premonition that the actual organization of human programs closely resembles the production system organization” (p. 803), but there was an obvious attractiveness to this conjecture. Human problem solving suddenly could be understood in terms of production systems (Young, 2001)—as the simple application of rule-based reasoning, using mechanisms that had been well studied by computer scientists and philosophers for decades. For the ai community, this work had a rather unguarded implication: human knowledge must surely be encoded as production rules, and the builders of intelligent systems can construct their electronic knowledge bases by ﬁnding out what those rules are. By the end of the 1970s, it was commonplace for developers of intelligent systems to view the problem of building electronic knowledge bases as one of eliciting rules from experts. More important, there was widespread belief that the rules in the knowledge base had correlates in representations in the head of the expert who served as the informant. Building an intelligent system was seen as a problem in the “transfer of expertise,” where the goal was to get the rules out of the expert’s brain and into the computer. Pioneers in the ﬁeld, such as Feigenbaum (1984), frequently used a metaphor in which system builders were said to “mine” the knowledge of human experts, capturing “golden nuggets” of expertise that could be added to a growing knowledge base, one rule at a time. The prevailing view was that the rules in the computer system were precisely the same rules that the experts used to solve real-world problems. By deﬁnition, every knowledge-based system became a cognitive model. Granted, there were some vocal naysayers at the time. Dreyfus (1981), for example, pointed to the qualitative transformation of problem-solving abilities that seems to take place as novices become experts and argued that the tacit nature of much professional knowledge makes it impossible for computers ever to reason like human experts. The ai community paid some attention to these concerns, but the apparent success of mycin and a host of other rule-based systems seemed too compelling to ignore. There was face validity to the claim that human expertise could be packaged within a com-

153 Building Intelligent Systems puter, and Newell and Simon’s suggestion that simple production systems could form the basis of human cognition seemed to be playing out—at least in silico.

The ai Boom and the ai Winter mycin never was asked to diagnose a real patient. Rule-based systems for playing chess routinely—and quickly—lost to human grand masters. But the boom was beginning by the early 1980s. Companies such as Digital Equipment Corporation with great fanfare installed rule-based systems to help conﬁgure the complex backplanes of their minicomputers (McDermott, 1982). Schlumberger touted the beneﬁts of employing rule-based systems to assist in drilling for oil. A system called “Prospector” was credited with the discovery of a new deposit of molybdenum ore (Duda & Shortliffe, 1983). American industry became obsessed with the idea of mining human expertise and putting it in a box, building machines to perform the complex cognitive problem solving previously believed to be the realm of only highly trained people (Winston & Prendergast, 1986). The U.S. military, which had funded much of the work in ai, had visions of autonomous robots ﬁghting battles in which there were no human casualties. New professional organizations, such as the American Association for Artiﬁcial Intelligence and the International Association of Knowledge Engineers, sprung up, as did new companies that built special-purpose computers ﬁnely tuned to run special-purpose programming languages such as lisp. There were new companies selling programs to help developers edit and debug rules. These companies offered inference engines that grew in their capabilities (and in their complexity) to perform increasingly sophisticated reasoning. Indeed, the inference engines often became so complicated that yet other companies were founded to advise customers what new hardware and software to buy and how to use their purchases most effectively. A new “knowledge industry” had been born, and the excitement was like nothing that computer scientists has seen previously. When the International Joint Conference on Artiﬁcial Intelligence convened in Los Angeles in 1985, the city had not seen such crowds since the 1984 summer Olympic games. Companies with names

154 modeling complex systems such as Intellicorp, Teknowledge, and Symbolics were busily busing hundreds of conventioneers—including bright-eyed graduate students who had no money at all to spend—to exclusive locations in Beverly Hills, offering abundant free food and a chance to see the companies’ products in action. Scruffy-looking researchers who had agreed to give tutorials on basic aspects of rule-based systems were stunned, not only by the numbers of attendees cramming into their lecture halls, but also by the unprecedented per capita honoraria that they received, allowing at least one to brag about buying a new sports car as a result! The word on the street was that human intelligence could be mined and put into machines. By 1988, ai had become a $1 billion industry in the United States. Within a few years, however, the expert-systems industry would be largely gone. The expectation that knowledge-engineering groups would be springing up in every cranny of the corporate world was wildly optimistic. It proved much more difﬁcult to encode and to maintain electronic knowledge bases than most ai practitioners had ever imagined. As we shall see in the next section, building intelligent systems is an enterprise that is extraordinarily more complicated than simply mining nuggets of knowledge. Rule-based systems, nevertheless, still remain essential commodities and are woven into the fabric of modern society. Such systems continue to review our applications for commercial credit, to determine the appropriateness of requested medical procedures, to audit our income tax returns, to interpret our electrocardiograms and other medical tests, to assist in industrial manufacturing, even to control the environment of the buildings in which we work. We have learned, however, that the knowledge needed to drive such systems does not exist as sets of production rules waiting to be plucked from the heads of human experts and dumped into computer programs.

The Breakdown of the “Transfer” Metaphor When the early developers of knowledge-based systems spoke of the transfer of expertise from humans to machines, they saw the construction of intelligent systems as a problem in relocating knowledge. They spoke of the parallels between transferring knowledge—

155 Building Intelligent Systems generally in the form of production rules—from human experts to computer programs and transferring knowledge from computer programs to naive users. Knowledge was a commodity, a substance, and the goal was to get it from one place to another. The view of knowledge-based development as the transfer of expertise hindered the now-obvious recognition that building electronic knowledge bases is a creative and inventive activity. When developers interact with application specialists to build knowledgebased systems, the developers form mental models of how the experts solve problems; the experts, of course, have mental models of their own that attempt to capture their professional problem-solving behavior. In the course of building the system, both the developers and the experts continually revise their respective mental models. Although the developers and the experts may have very different mental models at the outset of their collaboration, the models eventually converge. This convergence is possible (1) because the development process forces all parties to commit their mental models to a ﬁxed, publicly examinable form—typically the emerging knowledge base—and (2) because the frequent consideration of examples and test cases forces the system builders to assess, corroborate, and revise their respective models (Regoczei & Plantinga, 1987). By the 1990s developers of knowledge-based systems began to realize that the nuggets of knowledge were not always there for the mining. Builders of intelligent systems began to accept that the application experts, whose professional acumen was to be encoded as a knowledge base, often have no preexisting mental model of how they do their work. Often, the experts cannot begin to verbalize how they actually go about solving problems. Dreyfus (1981) and others had been warning about these difﬁculties all along, but suddenly the warnings were beginning to be take root. The problem of building knowledge-based systems was no longer seen as one of transferring expertise; much of the expertise often had to be invented in the ﬁrst place. At the core of the matter is the automatization of professional behavior. Human cognitive skills appear to be acquired in at least three generally distinct stages of learning (Fitts, 1964; Johnson, 1983). Although different authors have used different terms to describe the three phases, there is concurrence regarding the qualitative changes that occur in the way in which people seem to retrieve information during problem solving. Initially, there is the cognitive stage, during

156 modeling complex systems which an individual identiﬁes the actions that are appropriate in particular circumstances, either as a result of direct instruction or from observation of other people. In this stage, the learner often verbally rehearses information needed for execution of the skill. Next comes the associative phase of learning, in which the relations noted during the cognitive stage are practiced and verbal mediation begins to disappear. With repetition and feedback, the person begins to apply the actions accurately in a ﬂuent and efﬁcient manner. Then, in the ﬁnal autonomous stage, the learner compiles the relations from repeated practice to the point where he or she can perform them without conscious awareness. Suddenly, the person performs the actions appropriately, proﬁciently, and effortlessly—without thinking. The knowledge has become tacit (Fodor, 1968). There is substantial evidence that, as humans become experienced in an application area and repeatedly apply their know-how to speciﬁc tasks, their knowledge becomes compiled and, thus, inaccessible to their consciousness. Experts lose awareness of what they know. The knowledge that experts acquired as novices may be retrievable in a declarative form, yet the skills that these professionals actually practice are procedural in nature (Anderson, 1987). The inability of experts to verbalize these compiled associations is well accepted (Lyons, 1986; Nisbett & Wilson, 1977). The consequence is that the special knowledge that we would most like to incorporate into our intelligent systems is often that knowledge about which experts are least able to talk. Johnson (1983) has identiﬁed this phenomenon as the paradox of expertise. The problem for builders of intelligent systems is that, when professionals are asked to report on their compiled expertise, they often volunteer plausible answers that may well be incorrect. In experimental situations, subjects have been shown to be frequently (1) unaware of the existence of a stimulus or cue inﬂuencing a response, (2) unaware that a response has been affected by a stimulus, and (3) unaware that a cognitive response has even occurred. Instead, subjects give verbal reports of their cognition based on prior causal theories from their nontacit memory (Nisbett & Wilson, 1977). Furthermore, because Western culture mistakenly teaches us that accurate introspection somehow should be possible (Lyons, 1986), people freely explain and rationalize their compiled behaviors without recognizing that these explanations are frequently incorrect.

157 Building Intelligent Systems Developers of knowledge-based systems are, thus, not mining preexisting nuggets of knowledge. They are creating de novo theories of professional problem-solving behavior and representing those theories in terms of electronic knowledge bases. The developers serve the important function of detecting gaps in the articulated knowledge of their informants and helping them ﬁll in those gaps by deﬁning plausible sequences of actions that can achieve the necessary goals. The intelligent systems that result from this work may not achieve the same level of nuanced performance associated with the procedures actually used by domain experts. The underlying knowledge bases can, nevertheless, be observed, extended, and easily disseminated to other people in need of advice. It is simply incorrect to view an electronic knowledge base as an embodiment of actual professional knowledge; knowledge bases instead represent only models of surface-level behaviors—models that attempt to approximate, but that do not reproduce, the actual problem-solving steps used by humans (Clancey, 1989).

Further Assaults on Knowledge Bases as Cognitive Models At the time of the Dartmouth conference in 1956, there had been a conviction that the secret to creating ai was rooted in a deeper understanding of human psychology. By the time the ai community had declared that “winter” had settled in at the end of the 1980s, however, the relation between psychology and knowledge-based systems seemed increasingly disconnected. A major salvo came from the situated-action community (Suchman, 1987; Clancey, 1997). Psychologists who were studying human problem solving not only reafﬁrmed that much professional behavior involves knowledge that is completely tacit but also noted that the stimuli that drive problem solving in the ﬁrst place are often environmental and difﬁcult to isolate. Intelligence increasingly was seen as a holistic interaction involving the problem solver with the environment. As such, intelligence could never be decontextualized. Knowledge could never be “extracted” and formalized since it could never be decoupled from the situations in which it was applied. The notion of situated action is more a perspective on behavior than it is

158 modeling complex systems a formal psychological theory (Kushmerick, 1996), but it had a jarring effect on the way people viewed the role of representations in cognitive science. The situated-action perspective cast aside Newell and Simon’s hypothesis that human problem solving is fundamentally grounded in the processing of symbols (Newell, 1980). Intelligent behavior was seen instead as the consequence of myriad speciﬁc, often-unknowable stimuli in the environment leading to myriad speciﬁc actions. There were no preexisting abstract representations in the brain to deﬁne general plans of behavior; there were only particular behavioral responses to particular situations. Such responses might require the dynamic creation of a mental representation to guide the agent’s behavior, but there were no prefabricated representations for the abstract problem space (e.g., no production rules) to drive cognition. Although the emergence of the situated-action perspective caused a minor uproar in the cognitive-science community, its effects on the developers of knowledge-based systems were, ultimately, more muted (Menzies & Clancey, 1998). The primary consequence of the “situated” perspective was that virtually no developers of knowledge-based systems claimed that they were creating cognitive models anymore. The emphasis shifted almost completely away from psychology toward the construction of useful pieces of software. This move toward a software-engineering perspective was reinforced by other events in the ai community. Perhaps none was more important than the appearance of computer programs that could, ﬁnally, beat grand masters in games of chess. Workers at CarnegieMellon University built Deep Thought (reborn at ibm as Deep Blue) and achieved outstanding game-playing performance. They did so, not by building a computer system that mimicked the cognitive behaviors of human experts, but by capitalizing on things that only computers can do: blazingly fast computation, the execution of multiple computations in parallel on distributed processors, and the ability to use brute force to look ahead many moves to see the consequences of potential actions (Newborn, 2002). Deep Blue ultimately beat Gary Kasparov at his own game, but not because the program was cleverer than Kasparov or because it modeled the thought processes of a better chess player. Deep Blue demonstrated convincingly that knowledge-based systems would be successful when they could

159 Building Intelligent Systems take advantage of the unique abilities of the hardware and software of which they were made. The emphasis was placed squarely on better techniques for systems engineering, rather than on better cognitive modeling.

Knowledge-Based Systems Are Software During the mycin era, most developers of intelligent systems argued strongly that knowledge engineering was not the same thing as software engineering, that knowledge engineers were fundamentally different from software engineers. They maintained vehemently that the building of intelligent computer systems was different from the development of conventional computer programs. Creating intelligent systems involved the representation of large amounts of content knowledge and almost no programming of traditional computer code. There was nothing to ﬂowchart, nothing to compile. The claim was that knowledge engineers needed a skill set that was different from the one required by computer programmers. Some investigators, particularly in Europe, thought otherwise. The kads project (Schreiber et al., 2000), for example, began in the early 1980s to provide a comprehensive methodology for building knowledge-based systems. The kads consortium argued that the construction of knowledge bases and the construction of conventional software share an important element: the need to manage complexity. Psychology was important in the development of intelligent systems—not because of the need to model the cognitive processes of human informants, but because of the need for the system builders themselves to be able to keep track of a huge network of knowledge representations and the relations among them. Just as workers in traditional software engineering were grappling with methods to help computer programmers manage complexity and generate solutions creatively (Brooks, 1987), the kads project strove to create methods allowing knowledge engineers to keep track of elicited knowledge and to identify gaps in the emerging problem-solving solution. kads emphasized the role of predeﬁned, stereotypic patterns of problem solving that could help developers elicit knowledge. The approach also stressed ways of categorizing knowledge into different “layers” of speciﬁcity to help structure the emerging knowledge-based

160 modeling complex systems

Figure 3. Stages in the development of intelligent systems. From a conceptual model of system performance, developers create a design model that makes a commitment to an abstract pattern for how the system might be created using software. The design model then drives the implementation, which results in a physical software system that users can run.

system. At its core, kads was a software-engineering methodology for building intelligent systems. For many development groups, kads—and its subsequent reﬁnement, commonkads—transformed the process of knowledge engineering from an informal enterprise of mining nuggets to a stepwise, methodical practice for building complex software systems that manifested intelligent behavior. An essential contribution of the kads project was the view that, like conventional software, intelligent systems need to be developed in stages (Figure 3). Developers should never simply sit down and try to build a large, complex system all at once; instead, they need to approach the design process in a principled manner. The kads project viewed system building as the construction of a set of models and argued that those models should be made explicit and examinable. First developers should build a conceptual model that deﬁnes what task the system is to perform in a functional manner: What are the inputs? What are the outputs? How does the nature of the inputs constrain the nature of the outputs? Can the task be decomposed into subtasks, both to simplify the computation and to make the system more cognitively tractable? In the case of mycin, for example, the conceptual model would identify patient data as the input and a set of antibiotics to prescribe as the output. The overall task of determining what presumptive therapy to administer might be decomposed

161 Building Intelligent Systems into two subtasks: (1) identify what classes of organisms most likely were causing the infection and (2) construct a small set of antibiotics that, together, were likely to be effective against the most likely organisms. The conceptual model is just that; it is conceptual. There is no discussion of how the computer might actually perform the tasks required by the ﬁnished system. Next comes the creation of a design model. The design model is just that; it is a design. It is not a working piece of computer code but, rather, an outline of how software components might be brought together to implement a working system. A design model for mycin as it was originally implemented might describe how production rules could be used to program the organism-identiﬁcation task, how a constraint-satisfaction program could be used to determine the minimal set of antibiotics to prescribe, and so on. Just as developers of methodologies for software engineering were arguing that design must precede implementation, the authors of kads and commonkads insisted that builders of intelligent systems needed to articulate a coherent design before they actually began to encode the production rules and write the program components that would form the basis of the executable system (Schreiber et al., 2000). Construction of a design model is facilitated if developers have access to a set of primitives that can serve as building blocks for constructing a design. Those primitives could be the components of expert-system-building shells or abstract algorithms for performing tasks such as constraint satisfaction. I will have much more to say about possible building blocks for the design model in the next section. From the design model would come an implementation. The implementation would be the software artifact with which end users would interact and that actually would provide them with advice. Because any creative task is simpliﬁed when it can be decomposed into more tractable elements, ideally there are software components that can serve as building blocks for developing the implementation. When creating conventional software, object-oriented programming often plays an important role in helping developers decompose the implementation task into the construction of cognitively tractable software modules. The kads project initially argued for a “waterfall” model of software development (Sommerville, 2004), with information ﬂowing

162 modeling complex systems from the conceptual model to the design model to the implementation. A primary problem with the approach is that developers rarely get things right the ﬁrst time; they must revisit their conceptual models and design models as they gain experience with a particular implementation. Unfortunately, well-meaning developers can choose to tinker with the implementation and ignore the design model. In time, the design model may become increasingly irrelevant for explaining the programming choices made in the implementation. Similarly, the conceptual model can rapidly become out-of-date unless system maintainers are particularly conscientious about documenting all their changes. Thus, the behavior of the knowledgebased system can come to represent something different from the speciﬁcation documented in the original conceptual model. System maintainers can guard against such drifts when they use building blocks that have well-understood meanings and behaviors to create their design models and implement their systems. If each of the building blocks is ﬁxed and invariant in its semantics, then developers can have conﬁdence in the consequences of replacing one building block with another. Any drift in behavior can be cleanly ascribed to a change in the components and can be documented fairly easily (Krueger, 1992). As system builders have thought about the development process shown in Figure 3 above, a number of questions surfaced: What are the optimal building blocks for creating design models and, ultimately, for implementing intelligent systems? How can a set of building blocks help developers construct their models in a way that ensures comprehensive and useful designs? How can a set of building blocks ensure that the software that implements an intelligent system is maximally reliable, maintainable, and traceable in its behavior? As the knowledge-engineering community evolved in the 1990s, the focus for research shifted radically to elucidating components both for design models and for implemented systems. The emphasis was still on nuggets, so to speak, but suddenly the community was much more concerned about software components and elements of design than it was about nuggets of expert knowledge. Psychology still was important to the builders of knowledge-based systems, but their attention was turning to the cognitive processes of the system developers—with the goal of creating modeling and implementation primitives that would help software engineers manage

163 Building Intelligent Systems the tremendous complexity inherent in the construction of intelligent systems.

Components for Modeling Knowledge Models are abstractions. And an abstraction—the word’s most literal meaning tracing back to its Latin roots—takes away from some reality, leaving just the essential elements. In any modeling activity, there are two key questions: (1) What are the kinds of essential elements to abstract? (2) How should modelers talk about those elements? There are, thus, the notions of deciding what are the right building blocks for creating models and of deciding what is the right language (or languages) for writing down descriptions of those building blocks. In the knowledge-based-systems community, there is ongoing discussion regarding the details of how knowledge is best modeled (e.g., what the kinds of essential elements are), but there is also surprising consensus. Partly inﬂuenced by the kads project, partly inﬂuenced by observations from workers such as Chandrasekaran (1986) and Clancey (1985), developers consider two broad classes of elements when modeling knowledge. These classes of knowledge describe (1) propositions that we know about the world—concepts and the relationships among concepts—and (2) the procedures by which we can use those propositions to solve problems. To use the current buzzwords, it becomes convenient to think about knowledge in terms of ontologies and problem-solving methods. Let us consider these two modeling elements in turn.

ontologies: defining what exists For hundreds of years, ontology was an abstract science, the branch of metaphysics that concerned the study of existence. Philosophers since the time of Aristotle have tried to deﬁne and categorize what exists in the world, carving up his world into those objects that were immaterial (spirits) and those that were material, those that were inanimate and those that were animate, those that were humans and those that were beasts. Aristotle worked to categorize objects and to deﬁne inherent properties of those objects that could differenti-

164 modeling complex systems ate each class into subcategories. The study of ontology remained a relatively obscure branch of metaphysics until the last decade of the 20th century, when computer scientists co-opted the word, put an article in front of it, and began to speak of an ontology as a description of the concepts and relationships among concepts that existed in some world being modeled. Suddenly, the elucidation of ontologies was, not only the work of philosophers, but also that of software engineers trying to model the concepts and relationships needed to deﬁne some reality for the purposes of building a system (Guarino, 1998). There are now conferences and workshops devoted exclusively to the subject of building ontologies. There are specialty journals and an explosion of books on the subject. The study of ontology has swelled because formal data structures that can be used to represent some aspect of the world and to provide an enumeration of essential categories have become more important commercially than any philosopher could ever have imagined. Yahoo! initially had enormous success because it provided easy entrée to the Internet by allowing users to browse a detailed ontology of the kinds of Web pages that people had created. Soon, other commercial Web sites such as Amazon.com offered a detailed ontology of books and music that allowed Web surfers to examine an enormous inventory of online products. Ontology became a big business as the Internet revolution required computer programs to be able to represent detailed descriptions of hierarchically organized information—from Web sites, to commodities, to e-commerce transactions (Staab & Studer, 2004). This notion of ontology was, of course, hardly something that the Internet community could claim as its own. The very idea stemmed from centuries of work in metaphysics. In the modern era the formal classiﬁcation of entities—from Linnaeus and the speciation of living organisms, to Roget and the thesaurus of the English language, to Dewey and the classiﬁcation of literature—had had profound effects on human thought. Indeed, much of human cognition centers on the creation and application of categorizations and taxonomies (Bowker & Star, 1999). The rediscovery of the notion of ontology in the late 20th century, however, provided a principled way for computer scientists to begin to think about the world that their programs were attempting to model and allowed them to put their models into a well-established context (Guarino, 1998).

165 Building Intelligent Systems

Figure 4. An ontology of oncology protocols. The ontology models the entities and relations needed to specify how medical care should be given over time to patients who have cancer. In the ontology, clinical-trial protocols are a kind of clinical guideline. Clinical trials have a number of attributes, including the authors who developed the protocol, the eligibility criteria that determine the set of patients to which the protocol applies, and the end points that determine when treatment should terminate.

When building knowledge-based systems, developers now concentrate on fashioning a formal model of the entities in the domain and the relationships among those entities. For example, Figure 4 shows an ontology of a medical application area—standard protocols for treating patients who have cancer that specify how chemotherapy and other treatments should be administered to patients, either as part of a clinical trial or in accordance with a clinical management guideline that attempts to deﬁne a set of best practices. The ontology describes what may exist—what are the possible components of a protocol (e.g., eligibility criteria that determine whether a clinical trial actually applies to a given patient, algorithms that specify the sequence of treatments to offer, tests to perform that may predicate decisions in the clinical algorithm) and what are the relationships among those components (e.g., that

166 modeling complex systems steps in the clinical algorithm are related temporally). The ontology does not describe any particular protocol (i.e., a speciﬁc plan of care for a speciﬁc medical problem), instead laying out the classes of entities that exist in protocols in general. The protocol ontology can then form the basis for acquiring the content knowledge for speciﬁc protocols—providing a structure that allows developers to build multiple knowledge bases such that each knowledge base encodes the description of a particular type of guideline or clinical trial (Musen, 1998). In our laboratory, we have developed a computer-based workbench that helps system builders use ontologies to create detailed knowledge bases. This workbench is known as “Protégé” (Musen, Fergerson, Grosso, Noy, Crubézy, & Gennari, 2000; Gennari et al., 2003). The screen shot in Figure 4 shows how an ontology appears within the Protégé tool. The tree browser on the left-hand side of the screen shows the entities in the ontology. The form on the right-hand side of the screen shows the attributes of whatever class is highlighted on the left. Ontologies are typically represented within computer systems as hierarchies of objects. At the top of the hierarchy may be an object with a very general description (e.g., thing). There will be subclasses of this object that represent more specialized concepts. The subclasses in turn will have subclasses. Each object in the hierarchy has attributes—some of which are particular to that object, others of which may be inherited from objects higher up in the hierarchy. In Figure 4, for example, the concept guideline has three distinct subclasses: management guideline, consultation guideline, and trial protocol. In the ﬁgure, attributes of trial protocol include items such as authors and eligibility criteria. Some of the attributes of trial protocol are inherited from more general concepts, such as guideline. For any ontology, the Protégé system can generate a user interface with which developers can enter the content knowledge structured by that ontology (Figure 5). For each generic entity in the ontology (e.g., trial protocol), Protégé provides an interface with which users can enter instances of that entity (e.g., a particular clinical trial for treating breast cancer). The interface also allows developers to specify relationships among the instances (e.g., the sequence of steps that determine how chemotherapy and other inter-

167 Building Intelligent Systems

Figure 5. A knowledge-entry tool for the speciﬁcation of clinical trials. Protégé generated the user interface for this tool directly from the ontology shown in Figure 4 above. The ontology deﬁnes the classes of entries that a user can make (in this case, the particular blanks to be ﬁlled in and the choices in the palette available for drawing diagrams); users ﬁll in the blanks and draw diagrams to specify the content knowledge that constitutes particular knowledge bases. In this example, the Protégé-generated tool allows developers to create knowledge bases that deﬁne clinical trials. Here, the user is entering knowledge about a clinical trial for treating patients who have breast cancer. The particular protocol compares the effects of conventional chemotherapy and radiotherapy with those of high-dose chemotherapy followed by bonemarrow transplantation.

ventions should be administered over time). Tools such as Protégé divide the process of creating an electronic knowledge base into two phases: (1) description of a general ontology of the application area and (2) instantiation of that ontology to deﬁne particular knowledge bases. We and many other groups routinely use Protégé (and tools like Protégé) in this two-step process of knowledge-base deﬁnition.

168 modeling complex systems

problem-solving methods: defining what to do Ontologies specify what exists in the world (e.g., a description of a clinical trial), but they typically take no position on what should be done with that information (e.g., using a clinical trial description to recommend treatment for a particular patient). The second major class of modeling components for knowledge-based systems—problem-solving methods—speciﬁes what to do but typically takes no position on what exists (Eriksson, Shahar, Tu, Puerta, & Musen, 1995; McDermott, 1988). A problem-solving method is a general procedure for solving some well-deﬁned task (e.g., classiﬁcation, planning, constraint satisfaction). Any time developers choose to build an intelligent system, they must model both the ontology of the application area and the problem-solving method with which the system will address the task at hand. The distinction between problem-solving methods and ontologies mirrors the distinction in psychology between knowing how and knowing that. A problem-solving method deﬁnes what a system should do with the propositional knowledge in an ontology (Crubézy & Musen, 2004). The notion of problem-solving methods has been around since the early days of ai. Newell and Simon (1963) identiﬁed a number of “weak” problem-solving methods, such as means-ends analysis and generate and test, that were very general purpose in nature and could, therefore, be applied to a wide range of problems. The methods were considered weak because they could not take advantage of explicit domain-speciﬁc knowledge to drive all their inferences; the methods required only a single task-speciﬁc evaluation function that could inform the problem solver whether a selected means would lead to some desired ends, whether some generated solution would test satisfactorily, and so on. All knowledge about the desired outcome—and, thus, about whether the problem solver might have achieved its goal—was built into the evaluation function. In the early 1980s, when the hype regarding knowledge-based systems was reaching its peak, some workers in ai were taking a fresh look at the idea of a problem-solving method. Chandrasekaran (1986) and his colleagues at Ohio State University were identifying broad patterns in the problem-solving behavior of knowledge-based systems that seemed reusable from one system to another. The Ohio

169 Building Intelligent Systems

Figure 6. Heuristic classiﬁcation in mycin. Problem solving in the mycin system can be construed in terms of three major inference patterns: (1) feature abstraction, (2) heuristic match, and (3) solution reﬁnement. After Clancey (1985).

State group described recurrent problem-solving approaches that, unlike weak methods, could take advantage of domain knowledge to address tasks in more sophisticated ways than are possible with a single evaluation function. For example, Chandrasekaran described a method that he called establish-reﬁne, which performed diagnosis by examining a hierarchy of possible diagnoses (the most general diagnoses were at the top of the hierarchy) and performed a top-down analysis of the hierarchy to determine the most speciﬁc diagnosis that applied to the situation at hand. Each time a diagnosis was established and then reﬁned into a more speciﬁc diagnosis, additional domain knowledge was considered by the problem solver. Soon thereafter, Clancey had fully dissected the problem-solving activities of the mycin system. Clancey (1985) showed that the behavior of mycin was not the emergent result of hundreds of modular rules interacting seemingly at random. Rather, Clancey maintained, the developers of mycin had deliberately—although perhaps unintentionally—built into mycin a coherent set of inference patterns that collectively he called heuristic classiﬁcation (Figure 6). For example, the use of data about the case at hand to infer generalizations about the patient to be treated involved a collection of inferences known as feature abstraction. In Figure 6, for example, mycin is shown to use its domain knowledge to reach the conclusion that the patient under consideration may be a “compromised host” either because the

170 modeling complex systems white-blood-cell count is low or because the patient is an alcoholic. The system is, thus, using input data about the case to determine resulting abstractions about the case. Other inferences allow mycin to perform solution reﬁnement, using domain knowledge to distinguish among a group of organisms that may be causing the infection. In this case, mycin uses knowledge of microbiology to identify the most speciﬁc organism that may be causing the infection, adopting an approach much like establish-reﬁne. Special inferences, known as heuristics, allow mycin to associate an element of one abstraction hierarchy (e.g., that the patient is a compromised host) with an element of another abstraction hierarchy (e.g., that the organism causing the infection may be of the class Gram-negative rod). In the case of mycin, heuristics links the abstractions related to the condition of the patient to abstractions related to the possible causes of the infection. Clancey (1985) demonstrated a new coherency in the way in which mycin solved problems. The program was not just triggering production rules; it was following a well-deﬁned pattern of inference that involved (1) feature abstraction, (2) heuristic match, and (3) solution reﬁnement. Even if the developers of mycin had talked about their work as the elicitation and engineering of hundreds of production rules that related directly to the rules expressed by infectious-disease experts, those rules merged into a singular, consistent problem-solving method when an observer could obtain sufﬁcient distance. More important, Clancey identiﬁed dozens of other contemporaneous knowledge-based systems that also seemed to be using heuristic classiﬁcation as their principal problem-solving method. For example, Elaine Rich had built a system called “Grundy” that served as a kind of automated librarian, making recommendations to potential borrowers of books to read; Grundy would interview the borrower, perform feature abstraction to conclude some generalizations about the client, then use heuristic match to identify classes of books in which the borrower might be interested (e.g., mystery novels, science ﬁction, biography), and then use additional knowledge to be able to recommend speciﬁc books. Workers at sri International had built a system called “Prospector” that would identify geographic areas in which to drill for minerals; Prospector would elicit features about a given geographic area, use feature abstraction to determine appropriate generalizations, and match those generalizations heuristically to classes of minerals that might be present (Duda

171 Building Intelligent Systems & Shortliffe, 1983). The systems that Clancey (1985) studied were designed for different application domains, but all of them used a shared problem-solving paradigm, one based on feature abstraction, heuristic match, and solution reﬁnement. In subsequent years, developers have identiﬁed and written computer programs to implement several reusable problem-solving methods that can be applied to a variety of tasks (McDermott, 1988). These contributions include methods such as heuristic classiﬁcation and establish-reﬁne, where the problem is one of classiﬁcation, in that the solution set is preenumerated and the problem solver selects one or more solutions from that set. Other problem-solving methods, on the other hand, require the problem solver to construct a solution on the ﬂy. These constructive methods encode algorithms for planning, constraint satisfaction, simulation, and so on—problems where it is impossible (or impractical) to preenumerate the solution set in advance. Problem-solving methods themselves can be viewed as having three components: (1) a deﬁnition of the algorithm that includes the problem-solving steps that the method carries out (often in some formal language) and the relationships that must hold between the method’s input data and the method’s output data once the algorithm has ﬁnished running (e.g., in the case of heuristic classiﬁcation, the input data must be mapped to some classiﬁcation), (2) a description of the format of the method’s input data and output data (often referred to as a method ontology; Gennari, Tu, Rothenﬂuh, & Musen, 1994), and (3) software that actually implements the method as a computer program that can execute on some machine. Problem-solving methods can be implemented in virtually any computer language. What is essential is that the method must be programmed in a manner that allows the data on which the methods operate to be provided in terms of an appropriate domain ontology.

Putting the Pieces Together Because each problem-solving method has a method ontology that deﬁnes the problem solver’s inputs and outputs, developers can relate application-speciﬁc concepts deﬁned by a domain ontology (e.g., a patient with an infectious disease, laboratory-test results) to the cor-

172 modeling complex systems responding application-neutral requirements deﬁned for the method (e.g., a case to be classiﬁed, input data). Developers can relate the entries of an ontology of microorganisms to the abstract solution set evaluated by the heuristic-classiﬁcation method; they can relate deﬁnitions (such as that of compromised host) to the inferential knowledge required for feature abstraction. Thus, the domain ontology deﬁnes the propositional knowledge of the application area; the problemsolving method deﬁnes a generic strategy for obtaining a solution for a certain class of problems; the manner in which the developer relates elements of the domain ontology to the data requirements of the problem-solving method determines how the method actually will be used to solve particular cases in the application domain (Crubézy & Musen, 2004). Imagine that a developer were to re-create the mycin system using these types of building blocks (Figure 7). The domain ontology would include concepts such as patient, organism, and antibiotic. Attributes of these concepts would specify the signs and symptoms that patients might demonstrate; the laboratory tests that physicians might request; the relationships between signs, symptoms, and test results; possible classes of infecting organisms; and so on. The reengineered mycin system would use the heuristic-classiﬁcation problem-solving method to address the task of identifying the most likely infecting organisms (as in Figure 6 above). A separate method would be needed to compute the most parsimonious set of antibiotics that can provide protection against all the potential microorganisms named by the organism-identiﬁcation task. This latter method needs to ensure (1) that the physician will administer at least one antibiotic that will treat each of the possible pathogens, (2) that the patient receives the fewest number of antibiotics possible, and (3) that the patient does not receive any antibiotic to which he or she is allergic or that may interact with any other drug that he or she is taking. Thus, the mycin antibiotic-selection task might be best solved using a constraint-satisfaction problem-solving method. The system builder would need to relate the data requirements of both the heuristic-classiﬁcation method and the constraint-satisfaction method to the classes of data described in the domain ontology. In practice developers specify this wiring of domain ontologies to problem-solving methods by declaring explicit mappings between these two kinds of building blocks (Crubézy & Musen, 2004; Gennari et al., 1994).

173 Building Intelligent Systems

Figure 7. Reengineering the mycin system with modern components. Propositional knowledge would be represented as a domain ontology. The problem-solving knowledge required for the organism-identiﬁcation task would be encoded using heuristic classiﬁcation. The system would require an additional constraint-satisfaction problem-solving method to address the task of constructing a minimal set of antibiotics that can cover for all likely pathogens.

The knowledge-based-systems community has gained considerable experience over the past decade building intelligent systems using explicit domain ontologies, knowledge bases that add detailed content speciﬁcations to those ontologies, and reusable problem-solving methods (Musen, 2004). These classes of building blocks have enabled a wide range of developers to create extremely complex systems and, most important, to maintain those systems over time as needs evolve and situations change. The evolution in thought has taken place rather quietly. There has been no great hype about any of this. There has been no overbearing excitement, as there was when rule-based systems debuted in the early 1980s. Although the vast majority of small-scale knowledge-based systems continue to be built using the rule-based approach, knowledge-engineering groups are increasingly turning to the use of ontologies and problem-solving methods to construct systems in a way that makes it clear (1) what knowledge about the world actually has been encoded in the systems (what is the ontology) and (2) how the systems go about solving the tasks with which they are confronted (what is the problem-solving method). Ontologies and problemsolving methods exist at two different levels—both as conceptual entities (abstract models of concepts and of generic problem-solving approaches) and as software artifacts (implemented object systems

174 modeling complex systems that store concept descriptions and implemented problem solvers). When thinking about the design stages depicted in Figure 3 above, it is clear that—abstractly—developers can use ontologies and problem-solving methods as building blocks for purposes of conceptual modeling and—concretely—developers can use these same building blocks to implement actual computer programs.

Building Intelligent Systems Fifty years after the Dartmouth conference, few psychologists come to scientiﬁc meetings that concern the development of knowledgebased systems. Once, the replication of human cognition in silico was viewed as essential for the creation of systems that displayed ai. In the past, workers sought to build cognitive models—sometimes replete with obviously erroneous human behaviors such as the Tversky and Kahneman (1974) biases—in an effort to duplicate the perceived thought processes of real-world experts. They chose approaches that were consonant with the way in which humans seemed comfortable addressing problems and even avoided algorithms that might be more normative or more computationally efﬁcient. It once seemed natural to believe that, by mimicking the perceived functioning of the most advanced problem-solving system that we know of (the human brain), we could create ai of the highest order. Builders of early expert systems, however, did not anticipate that the human brain that they sought to imitate was so poor at articulating its own reasoning (Johnson, 1983). They did not anticipate either the ultimate complexity of the rule-based systems that they would create or the enormous difﬁculty of maintaining those rule bases over time (Bachant & McDermott, 1984). Deeper understanding of the psychology of intelligence and of the underlying neuroscience unfortunately did not lead developers to create better computational artifacts. The human genome is able to guide the creation of a central nervous system that contains more than 100 billion neurons in an utterly graceful and remarkably reproducible manner; that same central nervous system, however, fails miserably whenever it tries to program a rule base of more than a few hundred entries or to keep track of how all the rules might interact. We have a long way to go

175 Building Intelligent Systems before the human brain will be able to understand its own cognition or to manage in an unaided fashion the complexity of computer systems that attempt to simulate human reasoning. Despite our best intentions, cognitive models do not lead to practical computational artifacts that are either highly scalable or easily maintainable. During the 1980s there was a surge of excitement that diverted the attention of some workers in ai away from modeling psychological phenomena to the enticing prospect of modeling basic neurological structures. Artiﬁcial neural networks came on the scene with tremendous fanfare and the promise that they would overcome the limitations of many of the cognitive models of intelligence (Hinton, 1991). These “connectionist” architectures consist of layers of nodes (sometimes called neurons) linked together in a manner that is inspired by neural circuitry in the cerebral cortex but that is not representative of any speciﬁc neural pathway. Particular values of input data trigger speciﬁc nodes in the bottommost layer, with propagation of signals upward through the network. This activity ultimately causes the nodes in the topmost layer to signal an appropriate output—a combination of nodes that can be inferred to represent the classiﬁcation of the input signal into some category. Artiﬁcial neural networks have proved to be extremely successful in a large number of classiﬁcation tasks. They are incapable of performing constructive problem solving, however. The networks can only transform input signals at the bottom layer to output signals at the top layer, and, thus, the solution space (represented by the top layer of nodes) must, by deﬁnition, be preenumerated. Although artiﬁcial neural networks continue to be used to address multivariate classiﬁcation problems, they do not solve the general problem of engineering computer systems that demonstrate intelligent behavior. The major lesson of the past ﬁve decades is that the most pragmatic route to the creation of intelligent systems is, not through psychology or through neuroscience, but rather through engineering. This lesson has been learned the hard way, by building a large number of systems—many of them successful, many of them not. Our collective experience has resulted in our ability both to engineer intelligent systems from reusable software components and to maintain and enhance those systems in a maximally efﬁcient manner. Although there were no traditional engineers invited to the Dartmouth conference in 1956, it is the ability to deﬁne a formal engineering discipline for the cre-

176 modeling complex systems ation of intelligent systems that may, ultimately, have had the greatest effect on the ﬁeld. The construction of intelligent systems is a problem in constructing artifacts. The importance of the engineer—who cares about the design, implementation, and testing of novel manufactured entities—has eclipsed that of the psychologist.

The Protégé System My own laboratory has adopted an engineering approach to the construction of intelligent systems, emphasizing the development of tools and methods to support the engineering of these systems from domain ontologies and abstract problem-solving methods. Much of our work is manifest in the suite of tools and the supporting development methodology that we call “Protégé.” Protégé provides a workbench that assists in the construction of ontologies (see Figure 4 above), in the use of ontologies to guide the entry of content knowledge (see Figure 5 above), and in a variety of other tasks related to the construction of knowledge-based systems. Protégé is in many ways like the computer-assisted software-engineering (case) tools that developers use routinely in the construction of conventional computer programs (Sommerville, 2004). The goal of Protégé, however, is to assist in the engineering of intelligent systems. The development of intelligent systems using Protégé proceeds in accordance with a sequence of well-deﬁned steps: • Creation of a domain ontology that captures the structure of the application area, deﬁning the major classes and relations that need to be modeled (see Figure 4 above); • Use of the domain ontology to generate a knowledge-acquisition system with which users can enter the detailed content knowledge that deﬁnes a knowledge base for the particular application area (see Figure 5 above); • Selection of a problem-solving method that can automate the task at hand; • Creation of mappings between the domain ontology and the method ontology that deﬁne how the abstract data and knowledge requirements of the selected problem-solving method are to be met using the particular concepts in the domain ontology.

177 Building Intelligent Systems Protégé assists in the creation, selection, adaptation, and use of domain ontologies and problem-solving methods and aids developers in linking the two types of building blocks together. From the perspective of Protégé, a domain ontology must deﬁne a structure for an application area in a manner that system builders and system users can agree on. There is no need for either the users or the participants to have innate beliefs about the particular ontology in advance—the ontology is merely a social convention that speciﬁes the concepts in the application domain and the attributes of those concepts. Similarly, the choice of a particular problem-solving method to solve a given task is often pragmatic. There is no need for the problem-solving method in any way to approximate the approach that a person might take in solving the given task. The building blocks that Protégé users manipulate to create intelligent systems—like the intelligent systems themselves—are simply artifacts. Protégé is used regularly by thousands of developers worldwide. It is built in a modular fashion that facilitates the development of “plug-ins” that can extend the functionality of the basic system. The growing community of Protégé users (more than 10,000 developers at the time of this writing) have implemented a wide range of plug ins—many of which address functions that our original development group could never have anticipated. Like the Internet itself, where software engineers have augmented the original Web-browser metaphor with countless extensions that go well beyond the original vision for the World Wide Web, Protégé has provided a framework for building knowledge-based systems where much of the ownership belongs to the community of developers at large, rather than to the original programming team. Building intelligent systems is inherently a creative process, and the ways in which users choose to support their creativity by augmenting the basic Protégé system seem to be almost limitless (Musen et al., 2000). Protégé enables developers to work with domain ontologies and problem-solving methods as reusable building blocks for system development without the need to buy into any kind of psychological theory. For readers of this volume, it may, indeed, come as a disappointment that psychology does not play an obvious, central role when developers use Protégé to build intelligent systems. Where psychology does take center stage, however, is in studying the way in which developers might interact with systems such as Protégé and in

178 modeling complex systems evaluating how alternative design choices in the workbench might affect the ways in which system builders design ontologies and knowledge bases and map their ontologies to selected problem solvers. The modern study of human-computer interaction (Norman & Draper, 1986) and computer-programming behaviors (Weinberg, 1985) has its origins in cognitive psychology; there is an essential role that psychologists can play in analyzing the human problem solving required to build and debug intelligent systems from reusable building blocks. Developers of intelligent systems must manage complexity, visualize intricate relationships, and assess system performance in ways that lend themselves to formal psychological study. We are already at the stage where system builders may incorporate into their work ontologies that contain hundreds of thousands of concepts and problemsolving methods of enormous computational complexity. For our development methods to continue to scale up, there is an urgent need to provide more guidance to system builders, who may need to examine and assemble these intricate building blocks. We are beginning to see the emergence of new methods to help in the visualization of complex knowledge bases and ontologies (Storey, Noy, Musen, Best, Fergerson, & Ernst, 2002), but there is clearly much more work to be done.

Moving into the Next Generation of Development Tools At the time of the Dartmouth Conference in the summer of 1956, there was no Internet, no World Wide Web, no computing grid. Computers were housed in specially built rooms at academic institutions, in government laboratories, and in large corporations. Most interaction with computers over the next two decades would take place via punched cards. The ﬁrst commercially successful personal computers were still 25 years away. Now we live in a world were computers are commodities. They are ubiquitous and interact with one another seemingly instantaneously over wide networks. We take it for granted that we can launch a World Wide Web browser and read information stored on computers almost anywhere in the world. Although desktop publishing and even hypertext existed long before the Web, the advent of the Internet and the Web have changed qualitatively the nature of information access in our society.

179 Building Intelligent Systems The Internet and the Web are about to change qualitatively the nature of intelligent systems. Just as users can translate print documents into hypertext markup language (html) and post them as Web pages anywhere on the Internet, developers now can “publish” ontologies and knowledge bases on the Web and can make them available to any computer anywhere in the world. Problem-solving methods such as heuristic classiﬁcation can be made available as Web services, and, suddenly, intelligent systems can be fashioned out of components distributed throughout cyberspace (Benjamins, 2003). The next generation of intelligent systems will comprise the same kinds of components that have become popular as building blocks for creating stand-alone systems, but now those building blocks will be situated throughout the Internet. The result will be what investigators have come to call the Semantic Web, a World Wide Web designed for access, not by humans reading html pages, but instead by computers that will surf the Web in search of appropriate ontologies, knowledge bases, and problem-solving methods (Berners-Lee, Hendler, & Lassila, 2001). The vision is that intelligent agents will be able to locate on the Internet the right building blocks to solve realworld tasks, to access those building blocks, and to use them in their problem solving. Although, at present, creating the Semantic Web remains a research goal, the scientiﬁc community has become captivated by the possibility of extending the scope of intelligent systems throughout cyberspace and by the potential of intelligent systems that can draw on vast, virtual libraries of resources that have no physical limitations and to which a global community of developers can contribute (De Roure & Hendler, 2004). There are no psychological correlates for intelligent problem solvers that are based on building blocks distributed across the Internet. There is no model for human cognition that matches the behavior of an intelligent agent that determines dynamically how to optimize the processing of huge amounts of data in parallel and that takes advantage of arbitrary memory stores and processing units that may happen to be available anywhere in the world. Like Deep Blue, intelligent systems that will use the Semantic Web and the computing grid will be successful primarily because of their ability to exploit particular computational technology, not because they will model known cognitive processes. Nevertheless, this new generation of intelligent systems will still use the same kinds of ontologies,

180 modeling complex systems knowledge bases, and problem-solving methods that have become important in the engineering of current intelligent systems, such as those now developed using Protégé. Expanding the scope of automated intelligence from programs that run on stand-alone computers to applications that will draw on the resources of the Semantic Web will require a new class of development tools and engineering methods. The principal issue in building intelligent systems is one of managing complex models and their interrelations. As those models individually increase in scope and collectively become distributed in cyberspace, developers will face new challenges in conceptualizing ontologies and navigating their mappings to the various Web services that will provide problemsolving support. The need to visualize the dynamic ﬂow of information across networks and to debug systems when their components are widely distributed will stimulate the creation of a new generation of development tools and techniques. The availability of the entire Internet as a frontier for the creation of new kinds of intelligent applications is extraordinarily inspiring. The Internet alone is not enough to guarantee our success, however. The Semantic Web will become a reality only with the advent of appropriate engineering principles, of new tools that support those principles, and of new methods that can make the use of those tools cognitively tractable.

Conclusion Early developers of knowledge-based systems viewed their work as a problem in applied psychology. Their belief was that, the better one’s model of an actual cognitive process in humans, the better would be the computational artifact that would attempt to solve the same task in the world. Rule-based systems offered appealing computational architectures because much of human problem solving could be modeled in terms of production systems. In practice, however, it was difﬁcult for rule-based systems to scale up to the requirements of many real-world tasks. As rule-based systems were reaching the peak of their popularity, psychologists were strongly questioning the claim that such systems could even be viewed as appropriate cognitive models in the ﬁrst place. It is most helpful to view the construction of intelligent systems

181 Building Intelligent Systems as a problem in engineering, rather than in psychology. Intelligent systems can be created from reusable building blocks that include (1) domain ontologies, (2) knowledge bases that provide content knowledge structured by those ontologies, (3) problem-solving methods that automate stereotypical tasks such as heuristic classiﬁcation and constraint satisfaction, and (4) mappings that relate the concepts in domain ontologies to the knowledge and data requirements of problem-solving methods. Currently, Protégé is the most widely used workbench for building intelligent systems from reusable components and demonstrates the utility of this approach. The advent of the Semantic Web will entail the creation of new kinds of intelligent systems from the same kinds of components that we use now—but the components will be distributed in cyberspace. There is an exciting challenge to build a new generation of usable development tools that will address the enormous complexity that will result when systems can be assembled dynamically from computing resources that are scattered across the Internet. Psychologists will have an invaluable role to play in helping design and evaluate new methods and tools for piecing together the widely distributed building blocks.

Note The Protégé system is supported as a national biotechnology resource by the National Library of Medicine under grant lm007885. Development of Protégé has been supported by contracts from the National Cancer Institute and the Defense Advanced Research Projects Agency and by grants from the National Library of Medicine and the National Science Foundation. The system may be freely downloaded from our Web site under an open-source license (see http://protege.Stanford.edu).

References Anderson, J. A. (1987). Skill acquisition: Compilation of weak-method problem solutions. Psychological Review, 94, 192–210. Bachant, J., & McDermott, J. (1984). r1 revisited: Four years in the trenches. ai Magazine, 5(3), 21–32. Benjamins, V. R. (2003). Web services solve problems, and problem-solving methods provide services. ieee Intelligent Systems, 18(1), 76-77.

182 modeling complex systems Berners-Lee, T., Hendler, J. A., & Lassila, O. (2001). The Semantic Web. Scientiﬁc American, 279(5), 34–43. Bowker, G. C., & Star, S. L. (1999). Sorting things out: Classiﬁcation and its consequences. Cambridge ma: mit Press. Brooks, F. P. (1987). No silver bullet: Essence and accidents of software engineering. Computer, 20, 10–19. Buchanan, B. G., & Shortliffe, E. H. (1984). Rule-based expert systems: The mycin experiments of the Stanford Heuristic Programming Project. Reading ma: Addison-Wesley. Chandrasekaran, B. (1986). Generic tasks in knowledge-based reasoning: High-level building blocks for expert system design. ieee Expert, 1(1), 23–30. Clancey, W. J. (1985). Heuristic classiﬁcation. Artiﬁcial Intelligence, 27, 289– 350. Clancey, W. J. (1989). Viewing knowledge bases as qualitative models. ieee Expert, 4(2), 18–23. Clancey, W. J. (1997). Situated cognition: On human knowledge and computer representations. New York: Cambridge University Press. Crubézy, M., & Musen, M. A. (2004). Ontologies in support of problem solving. In S. Staab & R. Studer (Eds.), Handbook on ontologies (pp. 321–341). Berlin: Springer. De Roure, D., & Hendler, J. A. (2004). E-science: The grid and the Semantic Web. ieee Intelligent Systems, 19(1), 65–71. Dreyfus, H. L. (1981). From micro-worlds to knowledge representation: ai at an impasse. In J. Haugland (Ed.), Mind design (pp. 161–204). Cambridge ma: mit Press. Duda, R. O., & Shortliffe, E. H. (1983). Expert systems research. Science, 220(4594), 261–268. Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data (Rev. ed.). Cambridge ma: Bradford. Eriksson, H., Shahar, Y., Tu, S. W., Puerta, A. R., & Musen, M. A. (1995). Task modeling with reusable problem-solving methods. Artiﬁcial Intelligence, 79, 293–326. Feigenbaum, E. A. (1984). Knowledge engineering: The applied side of artiﬁcial intelligence. Annals of the New York Academy of Sciences, 246, 91–107. Fitts, P. M. (1964). Perceptual-motor skill learning. In A. Melton (Ed.), Categories of human learning. New York: Academic. Fodor, J. A. (1968). The appeal of tacit knowledge in psychological explanation. Journal of Philosophy, 65, 627–640. Gennari, J. H., Musen, M. A., Fergerson, R. W., Grosso, W. E., Crubézy, M., Eriksson, H., Noy, N. F., & Tu, S. W. (2003). The evolution of Protégé: An environment for knowledge-based systems development. International Journal of Human-Computer Studies, 58(1), 89–123. Gennari, J. H., Tu, S. W., Rothenﬂuh, T. E., & Musen, M. A. (1994). Mapping domains to methods in support of reuse. International Journal of HumanComputer Studies, 41, 399–424.

183 Building Intelligent Systems Guarino, N. (1998). Formal ontology and information systems. In N. Guarino (Ed.), Formal ontology in information systems: Proceedings of fois ’98 (pp. 3–15). Amsterdam: ios. Hinton, G. (1991). Connectionist symbol processing [special issue]. Artiﬁcial Intelligence, 46, 1–258. Johnson, P. E. (1983). What kind of system should an expert be? Journal of Medicine and Philosophy, 8, 77–97. Krueger, C. W. (1992). Software reuse. acm Computing Surveys, 24(2), 131– 183. Kushmerick, N. (1996). Cognitivism and situated action: Two views on intelligent agency. Computers and Artiﬁcial Intelligence, 15(5), 393–417. Lederberg, J., Sutherland, J. L., Buchanan, B. G., Feigenbaum, E. A., & Lindsay, R. K. (1980). Applications of artiﬁcial intelligence for organic chemistry: The dendral project. New York: McGraw-Hill. Liebowitz, J. (1999). Knowledge management handbook. Boca Raton ﬂ: crc. Lyons, W. (1986). The disappearance of introspection. Cambridge ma: mit Press. McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (1955). A proposal for the Dartmouth Summer Research Project on artiﬁcial intelligence. Retrieved from http://www-formal.stanford.edu/jmc/history/ dartmouth/dartmouth.html. McDermott, J. (1982). r1: A rule-based conﬁgurer of computer systems. Artiﬁcial Intelligence, 19, 39–88. McDermott, J. (1988). Preliminary steps toward a taxonomy of problemsolving methods. In S. Marcus (Ed.), Automating knowledge acquisition for expert systems (pp. 225–256). Boston: Kluwer Academic. Menzies, T., & Clancey, W. J. (1998). The challenge of situated cognition for symbolic knowledge-based systems. International Journal of HumanComputer Studies, 49(6), 767–769. Musen, M. A. (1998). Domain ontologies in software engineering: Use of Protégé with the eon architecture. Methods of Information in Medicine, 37, 540–550. Musen, M. A. (2004). Ontology-oriented design and programming. In J. Cuena, Y. Demazeau, A. Garcia, & J. Treur (Eds.), Knowledge engineering and agent technology (pp. 3–16). Amsterdam: ios. Musen, M. A., Fergerson, R. W., Grosso, W. E., Noy, N. F., Crubézy, M., & Gennari, J. H. (2000). Component-based support for building knowledge-acquisition systems. In Proceedings of the Conference on Intelligent Information processing (iip 2000) of the International Federation for Information Processing Sixteenth World Computer Congress (wcc 2000), Beijing, China, August, 2000 (pp. 18–22). Laxenburg: International Federation for Information Processing. Newborn, M. (2002). Deep blue: An artiﬁcial intelligence milestone. New York: Springer. Newell, A. (1980). Physical symbol systems. Cognitive Science, 4(2), 135–183. Newell, A., & Simon, H. A. (1963). gps: A program that simulates human

184 modeling complex systems thought. In E. A. Feigenbaum & J. Feldman (Eds.), Computers and thought (pp. 279–293). New York: McGraw-Hill. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs nj: Prentice-Hall. Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84, 231–259. Norman, D. A., & Draper, S. W. (Eds.). (1986). User-centered system design. Englewood Cliffs nj: Erlbaum. Regoczei, S., & Plantinga, E. P. O. (1987). Creating the domain of discourse: Ontology and inventory. International Journal of Man-Machine Studies, 27, 235–250. Schreiber, G., Akkermans, H., Anjewierden, A., de Hoog, R., Shadbolt, N., van de Velde, W., & Wielinga, B. (2000). Knowledge engineering and management: The Common kads methodology. Cambridge ma: mit Press. Sommerville, I. (2004). Software engineering (7th ed.). Boston: Addison Wesley. Staab, S., & Studer, R. (Eds.). (2004). Handbook on ontologies. Berlin: Springer. Steﬁk, M. (1995). Introduction to knowledge systems. San Francisco: MorganKaufmann. Storey, M.-A., Noy, N. F., Musen, M. A., Best, C., Fergerson, R. W., & Ernst, N. (2002). Jambalaya: An interactive environment for exploring ontologies. In Proceedings of the Seventh International Conference on Intelligent User Interfaces (p. 239). New York: acm. Suchman, L. A. (1987). Plans and situated actions: The problem of human-machine communication. Cambridge: Cambridge University Press. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124–1131. Weinberg, G. M. (1985). The psychology of computer programming. New York: Van Nostrand Reinhold. Winston, P. H., & Prendergast, K. A. (Eds.). (1986). The ai business: Commercial uses of artiﬁcial intelligence. Cambridge ma: mit Press. Young, R. M. (2001). Production systems in cognitive psychology. In N. J. Smelser & P. B. Baltes (Eds.), International encyclopedia of the social and behavioral sciences. New York: Elsevier. Zuboff, S. (1988). In the age of the smart machine: The future of work and power. New York: Basic.

Fostering Team Effectiveness in Organizations: Toward an Integrative Theoretical Framework Eduardo Salas, Kevin C. Stagl, C. Shawn Burke, and Gerald F. Goodwin University of Central Florida; University of Central Florida; University of Central Florida; U.S. Army Research Institute Organizations seeking to foster teamwork in ﬂuid environments are increasingly turning to teams as a preferred performance arrangement. The ongoing proliferation of teams is due in part to the rise of global market opportunities. Multinational corporations facing unprecedented domestic and foreign competition are utilizing teams as a means to organize production and service around a 24-houra-day process. At one time in history the sun never set on the British Empire; today, however, the sun never sets on Intel as software engineers collaborate around the globe in order to meet organizational ﬁnancial forecasts. The spread of teams is also due in part to the ongoing technological revolution. Once restricted to collocation, technology now makes it possible and proﬁtable for team members to be distributed in space-time. As organizations become more comfortable with, and dependent on, technology, we believe that opportunities for team-based collaboration will abound.

186 modeling complex systems Unfortunately, however, history has repeatedly shown that team performance is an elusive, dynamic, and complex phenomenon (Salas, Stagl, & Burke, 2004). Indeed, many recent tragic incidents (e.g., the uss Vincennes, the uss Stark, Occidental’s Piper Alpha platform, the space shuttle Columbia) have been attributed to breakdowns in teamwork. These failures are due in part to the often ﬂuid and sometimes chaotic environment in which teams operate. Thus, developing a thorough understanding of how teams interact in a synchronized fashion to achieve shared goals is critical to meeting short-term tactical objectives and to cultivating long-term strategic success. The proliferation of teams and the concomitant focus on orchestrating team effectiveness so that team members, teams, multiteam systems, and organizations can reach their full potential is supported by a growing body of theoretical and empirical research (see Belbin, 2000; Bell & Kozlowski, 2002; Beyerlein, Johnson, & Beyerlein, 2003; Cooper & Robertson, 2004; Edmondson, 2003; Hackman, 1990; Ilgen, Major, Hollenbeck, & Sego, 1993). In fact, during the past century a plethora of research initiatives have been undertaken with the goal of more fully illuminating the complexities of teamwork and, thereby, advancing the science of teams. Therefore, it is no surprise that, in a recent literature review, we identiﬁed more than 800 articles and chapters that present empirical evidence addressing some aspect of team effectiveness. We suspect that this body of empirical evidence will only continue to expand exponentially in step with the ongoing proliferation of teams. In addition to the growing number of empirical investigations noted above, there is a concomitant increase in team-effectiveness theory building. Those concerned with developing a deeper understanding of teamwork in order to promote team synchronicity have proposed a wide variety of theoretically grounded models and frameworks that address both team development and team effectiveness. Later in this chapter we extract a representative sample and expound on how each of these initiatives has taken a different path to framing team effectiveness, including midpoint effects (Gersick, 1988), episodic team processes (Marks, Mathieu, & Zaccaro, 2001), temporal entrainment (Ancona & Chong, 1999), and team maturation (Morgan, Salas, & Glickman, 1994). Although dozens of intriguing conceptualizations of teamwork have surfaced in the past 25 years, there has not been a compre-

187 Team Effectiveness in Organizations hensive review of team-effectiveness models and frameworks conducted since the early 1990s (see Campion, Medsker, & Higgs, 1993; Tannenbaum, Beard, & Salas, 1992). Since that time, organizations have changed in many ways, and much has been learned about teams (see Campbell & Kuncel, 2001; Edmondson, 1999; Hackman, 2002; Ilgen, 1999; Kozlowski & Bell, 2003; West, Tjosvold, & Smith, 2003). This state of affairs suggests that there are a growing number of unincorporated team-effectiveness research ﬁndings, models, and frameworks in the current body of relevant literature. If there are, indeed, a number of recent research initiatives undertaken to frame teamwork that were not already addressed by prior comprehensive reviews, then it is likely that they include constructs, construct operationalizations, and proposed construct relations that incrementally add to our understanding of the complex picture of team effectiveness. Furthermore, there are several problems associated with having such wide diversity in the deﬁnition and measurement of team constructs. Likewise, there are clear positives associated with maintaining a common language when engaging in team research (Salas & Cannon-Bowers, 2000). Also, accumulating and integrating what is known to date are challenges confronting both team researchers and industrial organizational psychology (Campbell, 1990b). Given the above arguments, we believe that undertaking a comprehensive review and integration of team-effectiveness models and frameworks will be beneﬁcial to team researchers by revealing common ground on the substantive team constructs of interest. Therefore, in this chapter we present a multilevel, multidisciplinary synthesis of team-effectiveness models and frameworks advanced during the last 25 years. This endeavor was undertaken to collect and integrate current comprehensive models and frameworks of team effectiveness in order to present a more coherent picture of the factors constituting and impinging on teamwork. Speciﬁcally, in this chapter we begin by deﬁning the terms team, teamwork, team performance, and team effectiveness. Once these preliminary issues are addressed, we describe a set of criteria utilized to guide our literature review. Our review of the team literature resulted in the identiﬁcation and subsequent integration of 138 initiatives that modeled or framed aspects of team performance or effectiveness. A representative sample of 11 models and frameworks that

188 modeling complex systems typify cutting-edge team advancements are reviewed in detail. On the basis of this research initiative, we advance an integrative, albeit preliminary, multilevel framework of team effectiveness. The advanced framework was designed as a simple yet rich heuristic that reﬂects the current state of the art within the team domain. Furthermore, we expect that our framework will serve as a departure point for team researchers concerned with addressing the salient features and conditions that constitute and affect teamwork in speciﬁc types of teams. After describing the novelties of our approach to framing team effectiveness, we discuss a few practical applications (i.e., measurement, training, stafﬁng) and how they can be more fully appreciated when viewed through the lens afforded by a theoretically grounded framework. This chapter concludes with a call for the generation and, ultimately, widespread adoption of a set of short-, mid-, and long-term goals that can serve to guide future team research. We believe that team research has reached its teenage years and that tackling the substantive issues that remain to be addressed by team researchers will help curtail the haphazardness that has characterized the selection of constructs investigated up until this point in time. Much work remains to be done, and a collective, coordinated effort among all team researchers is needed to accomplish it.

Clariﬁcation of Key Constructs The current diffusion of existing conceptualizations of team constructs mandates that, before advancing further, we take a moment to clarify and elaborate on some of the qualities that characterize teams, teamwork, team performance, and team effectiveness. Thus, in this section we attempt to foster a shared understanding between the authors and the reader of the conceptualizations of these related but unique phenomena that are applied throughout this chapter. The descriptions that follow also offer insight into the persons, times, and settings that our team-effectiveness framework generalizes to and across (Cook & Campbell, 1979). For example, at a broad level our proposed framework is tailored to teams characterized as having members who hold shared goals and act interdependently to achieve them.

189 Team Effectiveness in Organizations

teams The term team is deﬁned in the online Merriam-Webster dictionary as a number of persons associated together in work or activity. In addition to this conceptualization, a plethora of other perspectives have been advanced during the past 25 years to deﬁne team (Guzzo & Dickson, 1996). For our purposes we deﬁne team as follows: it is a complex entity consisting of (1) two or more individuals (2) who interact socially and (3) adaptively, (4) have shared or common goals, and (5) hold meaningful task interdependencies; it (6) is hierarchically structured and (7) has a limited life span; in it (8) expertise and roles are distributed; and it is (9) embedded within an organizational/environmental context that inﬂuences and is inﬂuenced by ongoing processes and performance outcomes (Salas, Dickinson, Converse, & Tannenbaum, 1992). This deﬁnition exempliﬁes action teams that perform some of the most critical work in the global economy, including emergencymanagement teams, medical teams, command and control military teams, cockpit crews, control-tower teams, oil-rig crews, submarine teams, sports teams, forensic-science teams, gene-sequencing teams, archaeological teams, and space-exploration teams. It also typiﬁes many types of teams appearing in more traditional business settings, such as ﬁnancial-analyst teams, research-and-development teams, construction teams, marketing teams, and fast food teams. All these team exemplars often perform in ﬂuid environments typiﬁed by (1) rapidly evolving and ambiguous situations, (2) no “optimal” answers, (3) information overload, (4) intense time pressure, and (5) severe consequences in cases of error (Orasanu & Salas, 1993). In fact, modern operational environments are characterized by a historically unparalleled accelerating rate of change that mandates team ﬂexibility, adaptability, and resilience. We suspect that operational environments will increase in ﬂuidity in step with emerging technologies.

teamwork Teamwork is deﬁned in the online Merriam-Webster dictionary as work done by several associates with each doing a part but all subordinating personal prominence to the efﬁciency of the whole.

190 modeling complex systems Although this deﬁnition is elegant, a more detailed one has been advanced in Salas, Sims, and Klein (2004): “Teamwork is a set of ﬂexible behaviors, cognitions, and attitudes that interact to achieve desired mutual goals and adaptation to the changing internal and external environments. Teamwork consists of the knowledge, skills, and attitudes (ksas) that are displayed in support of one’s teammates, objectives, and mission. Essentially, teamwork is a set of interrelated thoughts, actions, and feelings that combine to facilitate coordinated, adaptive performance and the completion of taskwork objectives” (pp. 497–498). This deﬁnition illustrates the core of what we mean when we refer to teamwork throughout the remainder of this chapter. In the previous paragraph, we deﬁned teamwork as adaptively enacted processes undertaken to achieve shared goals. Furthermore, we noted that a set of knowledge, skills, attitudes, and other characteristics (ksaos) underpins and enables the enactment of the steps constituting a given process. However, the question remains as to which ksas are central to teamwork. In other words, what constitutes teamwork? The answer to this question, not surprisingly, is highly contingent on the type of team under consideration. As noted in the previous section, this chapter emphasizes a particular type of team (e.g., one that is hierarchically structured, with tight interdependencies). Therefore, the ksas presented in the remainder of this section are most appropriate for teams with characteristics similar to those mentioned above. However, this caveat is not meant to imply that some of the ksas advanced below do not generalize to all types of teams, as it would be difﬁcult to imagine any interdependent collective that did not on occasion require communication or coordination. There have been many systematic initiatives undertaken to illuminate the ksas that constitute teamwork (see Cannon-Bowers, Tannenbaum, Salas, & Volpe, 1995; Dickinson & McIntyre, 1997; Fleishman & Zaccaro, 1992; Marks et al., 2001; Marks, Zacarro, & Mathieu, 2000; McIntyre & Salas, 1995; Salas, Sims, & Burke, 2005; Smith-Jentsch, Zeisig, Acton, & McPherson, 1998; Stevens & Campion, 1994). For example, McIntyre and Salas (1995) delineated teamwork skills and performance norms from their programmatic work with navy tactical teams (e.g., guided-missile teams). The four teamwork skills advanced by McIntyre and Salas are (1) mutual performance moni-

191 Team Effectiveness in Organizations toring, (2) feedback, (3) closed-loop communication, and (4) backup behavior. Also identiﬁed were two performance norms: (1) a team’s self-awareness and (2) the fostering of within-team interdependence. Another approach to understanding teamwork was taken in Cannon-Bowers et al. (1995). These researchers collected and synthesized prior research addressing teamwork. This effort resulted in the advancement of the ksa competencies proposed to underlie teamwork in organizations. Speciﬁcally, Cannon-Bowers et al. suggest that teams require several knowledge competencies, including (1) cue-strategy associations; (2) teammate characteristics; (3) role responsibilities; (4) shared task models; (5) team mission, objectives, norms, and resources; (6) relation to the larger organization; (7) task sequencing; (8) team role-interaction patterns; (9) procedures for task accomplishment; (10) accurate task models; (11) accurate problem models; (12) boundary-spanning role; and (13) teamwork skills. Furthermore, they suggest that teams require dozens of skills, skills that can be clustered into eight higher-order dimensions, including (1) adaptability, (2) shared situational awareness, (3) performance monitoring and feedback, (4) leadership/team management, (5) interpersonal relations, (6) coordination, (7) communication, and (8) decision making. Finally, they propose that teams require a variety of attitudinal competencies, including (1) team orientation, (2) collective efﬁcacy, (3) shared vision, (4) team cohesion, (5) interpersonal relations, (6) mutual trust, (7) collective orientation, and (8) a belief in the importance of teamwork. While McIntyre and Salas (1995) and Cannon-Bowers et al. (1995) made great strides in advancing our collective understanding of the core content constituting teamwork, the more recent research of Dickinson and McIntyre (1997), Marks et al. (2001), and Salas et al. (2005) has made a signiﬁcant contribution to our collective understanding of the dynamic interdependencies existing among the components of teamwork. These three particular initiatives are critical enough to warrant closer inspection, and, thus, a thorough description is provided in a later section reviewing extant theoretical models and frameworks of team effectiveness. It will satisfy for now to state that teamwork ksas operate, not in isolation, but dynamically, simultaneously, and recursively as they unfold over time to emerge as team performance.

192 modeling complex systems

team performance Performance is deﬁned in the online Merriam-Webster dictionary as either (a) the execution of an action or (b) something accomplished. What is noteworthy in this deﬁnition is the inclusion of both cause (i.e., action) and effect (i.e., accomplishment). This mingling of antecedent and consequence is rejected by those researchers concerned with modeling human performance. In fact, Campbell (1990a, p. 704) states: “Performance is not the consequence(s) or result(s) of action; it is the action itself.” Following this line of thinking, performance has been deﬁned within the domain of industrial organizational psychology as goal-directed behavioral or cognitive action (Campbell, 1990a; Campbell, Dunnette, Lawler, & Weick, 1970; Dunnette, 1963). While Campbell, Dunnette, and their colleagues’ assertions provide a basis for understanding human performance at a conceptual level, in practice the timing of performance measurement often leads to confusion. This is because operationalizations of performance are contingent on the timing of measurement. Thus, prior research has often indiscriminately conceptualized performance as either a series of behavioral or cognitive actions that unfold over time or the outcomes resulting from those actions (Kozlowski & Bell, 2003). Speciﬁcally, when performance is examined longitudinally, it is often conceptualized as a series of actions. However, cross-sectional research is typiﬁed by the measurement of performance outcomes or the result(s) of expended energy (e.g., number of products produced). Our point is that, regardless of whether longitudinal or cross-sectional research is conducted, team performance remains either a behavioral or a cognitive action and outputs of these actions are best conceptualized as performance outcomes. In both longitudinal and cross-sectional team research, a record can be compiled of actions and outcomes at multiple levels in the conceptual space. This speaks to the issue of what constitutes team performance. We assert that team performance consists of the display by one or more team members of taskwork competencies (e.g., running a computer program), teamwork competencies (e.g., backup behavior), and integrated team-level action (i.e., coordination, adaptation). This line of thinking leads to the conclusion that team performance is a multilevel phenomenon, a point echoed elsewhere in recent scientiﬁc literature (Kozlowski & Klein, 2000).

193 Team Effectiveness in Organizations Essentially, team performance can be conceptualized as a bottom-up emergent process unfolding from individuals to teams. In other words, team performance begins with team members and emerges upward to teams. However, team performance is also simultaneously a top-down process. This line of reasoning suggests that higher-level context has a direct or moderating effect on team performance. However, this does not preclude the foundation of team effectiveness residing in the cognition, behavior, and affect of individual team members. In fact, team members’ thoughts, actions, and feelings via social interaction, dyadic role exchanges, and ampliﬁcation have emergent properties that manifest at higher levels (i.e., bottom-up emergence). The assertion that team performance is both a top-down and a bottom-up multilevel phenomenon that emerges over time is consistent with the guiding tenets of general systems theory (von Bertalanffy, 1956) and its variants, which collectively acknowledge that teams are hierarchically nested, interdependent, open systems (Kozlowski & Klein, 2000).

team effectiveness Effectiveness is deﬁned in the online Merriam-Webster dictionary as producing a decided, decisive, or desired effect. Thus, effectiveness is, not the outcome produced from team performance, but rather the result of a judgmental process whereby an output is compared to a subjective or an objective standard. Similarly, effectiveness has been deﬁned within industrial organizational psychology as the evaluation of the results of performance (Campbell, 1990a). While this conceptualization of team effectiveness seems straightforward, Cohen (1994) disagrees: “Most group effectiveness researchers think about effectiveness as a multidimensional construct, but do not agree as to the criteria of work group effectiveness” (p. 71). For example, team effectiveness has been operationalized as (1) archival records of productivity, (2) self-report measures of employee satisfaction, and (3) manager judgments of effectiveness (Campion et al., 1993). Subsequent research reported in Campion, Papper, and Medsker (1996) operationalized team effectiveness as (1) immediate managers’ judgments at two points in time, (2) senior managers’ judgments, (3) peer managers’ judgments, (4) subordi-

194 modeling complex systems nates’ judgments, (5) archival records of satisfaction, and (6) archival records of performance appraisals. A wide variety of other operationalizations of team effectiveness have also been advanced. For example, Cohen (1994) adopted a tripatriate perspective asserting that the variables contributing to team effectiveness can be clustered into three broad categories, including (1) team performance, (2) team members’ attitudes about quality of work life, and (3) withdrawal behaviors. According to Cohen (1994), each of these three global factors encompasses a number of effectiveness-related variables. Speciﬁcally, the performance factor includes (1) controlling costs, (2) increasing productivity, and (3) increasing quality. The team members’ attitudes factor incorporates (1) job satisfaction, (2) team satisfaction, (3) satisfaction with social relationships, (4) satisfaction with growth opportunities, and (5) organizational commitment. The withdrawal behaviors include both (1) absenteeism and (2) turnover. Hackman (1987) also asserts that group effectiveness can be conceptualized in terms of three components. The ﬁrst component consists of a judgment by those stakeholders who review the work of teams in terms of whether it meets their standards for quality and quantity. The second is whether the needs of group members are satisﬁed by their team participation. The third is whether group interaction has served to maintain or strengthen the group’s ability to work together at some future date. Finally, Sundstrom, DeMeuse, and Futrell (1990) suggest that team effectiveness is composed of (1) managers’ and customers’ judgments about the acceptability of performance and (2) team viability, where team viability is deﬁned as commitment on the part of team members to continuing to work together. We suspect that some of the ambiguity underlying the nature of team effectiveness lies in an incomplete speciﬁcation of effectiveness across levels and time. To date, most team research has operationalized and measured effectiveness at both the individual (i.e., team members’ self-reports of satisfaction/commitment) and the team (i.e., ratings of team coordination) levels. In other words: “Team effectiveness is different from individual effectiveness” (Yetton & Bottger, 1982, cited in Tannenbaum et al., 1992, p. 117). Furthermore, individual effectiveness and team effectiveness are distinct from, but interdependent with, multiteam-system and organizational effectiveness. Certainly, a comprehensive discussion of team effective-

195 Team Effectiveness in Organizations ness must extend beyond the individual and team levels of analysis typically addressed in research in order to incorporate effectiveness factors at the multiteam system and organizational levels. This is because the criteria relied on by stakeholders to form effectiveness judgments are contingent on whether the system under consideration is the team member, the team, the multiteam system, or the organization. In addition, team effectiveness judgments are intertwined with temporal factors such as team maturation. As a team and its members mature, a given criterion, once heavily relied on by senior stakeholders in forming their judgments about the effectiveness of a particular aspect of the team, may no longer be salient. Moreover, the type of criteria adopted may change as teams pass through the early stages of formative development en route to achieving process gains. For example, during the norming stages of team development, judgments of effectiveness may be directed at the accuracy of team-member mental models. However, as the team matures, the emphasis may shift from individual mental models to shared and compatible mental models that serve as a cognitive reservoir for fueling coordinated adaptation to novel challenges. As noted above, the criteria adopted by stakeholders to form their effectiveness judgments change over time. Speciﬁcally, we noted that temporal dynamics such as team maturation inﬂuence what criteria are relevant and how those criteria are weighted. The criteria utilized by stakeholders to form their effectiveness judgments may also vary over time as a function of the rater’s level in the organization. In this example, the stakeholder’s organizational level is meant to convey expertise or systems thinking (see Senge, 1990). Certainly, a strong argument can be made that, as incumbents navigate a given organizational hierarchy and develop a history with the target unit under consideration, there is a concomitant evolution in their perspective on what and how much of a given set of qualities constitute effectiveness. The above-noted line of reasoning is supported by empiricalresearch evidence collected from the naturalistic decision-making paradigm suggesting that expert decision makers (e.g., ﬁreﬁghters, chess masters, police ofﬁcers) emphasize different criteria than novices do when forming judgments about stimuli (see Klein, 1993). Thus, it may be plausible to conceive of an interaction between the

196 modeling complex systems target unit being rated and the level of the stakeholder (i.e., rater) when predicting both (1) the types of effectiveness criteria attended to and (2) how chosen criteria are weighted to form global effectiveness judgments. If these arguments are valid, they have implications for the design of performance-appraisal systems and for rater training. Thus, these assertions warrant closer attention in future theory building, research, and practice. Essentially, effectiveness is a value judgment that is inﬂuenced by a number of factors. In this subsection, we noted how a particular criterion or set of criteria can change as a function of (1) the target unit’s level in the conceptual space, (2) the target unit’s level of maturation, and (3) the idiosyncratic characteristics of raters such as systems thinking. Although these three factors are undoubtedly important in understanding stakeholder judgments of team effectiveness, a number of other pertinent issues also exist. For example, a host of individual (e.g., latent cognitive resources, available cognitive resources, biases), team (e.g., performance phase), situational (e.g., time pressure, novelty), and organizational (e.g., culture, reward systems) factors also inﬂuence judgments of effectiveness. In sum, the substantive similarities and differences of the content constituting and the process underlying effectiveness judgments across system levels and time remain to be addressed in future research endeavors.

Extant Team-Effectiveness Models and Frameworks In the preceding section we broadly deﬁned some of the more salient characteristics of teams, teamwork, team performance, and team effectiveness for the purpose of framing our assumptions regarding the nature of these interrelated phenomena. Furthermore, we contextualized our thinking about these phenomena by describing the complex and ﬂuid nature within which teamwork often occurs. Thus, the foregoing section served to set the stage for our thinking about teams as presented throughout the remainder of this chapter. In this section we describe the results that were generated from a review of the team literature. The literature review, guided by our above-noted assumptions, resulted in the identiﬁcation of 138 teamperformance and -effectiveness models and frameworks that met

197 Team Effectiveness in Organizations Table 1. Team Models and Frameworks Aldag & Fuller (1993) Alper, Tjosvold, & Law (2000) Ancona & Caldwell (1992) Annett & Cunningham (2000) Argote & McGrath (1993) Arrow, McGrath, & Berdahl (2000) Avolio, Kahai, Dumdum, & Sivasubramaniam (2001) Baldwin & Bedell (1997) Balkundi & Harrison (2004) Barrick, Stewart, Neubert, & Mount (1998) Barry & Stewart (1997)

Cooper, Shiﬂett, Korotkin, & Fleishman (1984) Coovert & Dorsey (2000) Cuevas, Fiore, Salas, & Bowers (2004) Cummings (1978) Deeter-Schmelz, Kennedy, & Ramsey (2002) de Jong, Bouhuys, & Barnhoorn (1999) Denison, Hart, & Kahn (1996) DeSanctis & Poole (1994) Devine & Clayton (1999) Dickinson & McIntyre (1997)

Gibson, Zellmer-Bruhn, & Schwab (2003) Gist, Locke, & Taylor (1987) Gittell (2000) Gladstein (1984) Goodman, Ravlin, & Schminke (1987) Gupta, Umanath, & Dirsmith (1999) Guzzo & Shea (1992) Hackman (1983) Hackman (1987) Hackman (1990)

Doolen, Hacker, & Hackman & Oldham (1980) Van Aken (2003) Beck (2002) Drexler, Sibbet, & Forrester (1988) Hall (2001) Bolman & Deal (1992) Driskell, Radtke, & Salas (2003) Haward et al. (2003) Brandes & Weise (1999) Druskat & Kayes (1999) Heinemann & Zeiss (2002) Brodbeck & Greitemeyer (2000) Durham & Knight (1997) Helmreich & Foushee (1993) Bunderson & Sutcliffe (2002) Edmondson, Roberto, & Hertel, Konradt, & Watkins (2003) Orlikowski (2004) Burke, Stagl, Salas, Pierce, & Entin & Serfaty (1999) Higgins & Routhieaux (1999) Kendall (2006) Campion, Medsker, & Erez, LePine, & Elms (2002) Higgs & Rowland (1992) Higgs (1993) Campion, Papper, & Fleishman & Zaccaro (1992) Hinsz, Tindale, & Vollrath Medsker (1996) (1997) Choi (2002) Flemming & Monda-Amaya (2001) Hoegl, Praveen, & Gemuenden (2003) Cohen (1994) Flood, Hannan, Smith, Turner, West, Hogan, Raza, & & Dawson (2000) Driskell (1988) Cohen & Bailey (1997) Furst, Reeves, Rosen, & Hollenbeck et al. (1995) Blackburn (2004) Cohen, Ledford, & Gersick (1988) Hollingshead & McGrath (1995) Spreitzer (1996) Janz, Colquitt, & Noe (1997) Millitello, Kyne, Klein, Shea & Guzzo (1987) Getchell, & Thordsen (1999) Jung, Sosik, & Baik (2002) Mischel & Northcraft (1997) Sheard & Kakabadse (2001) Kahai, Sosik, Avolio (1997) Morgan, Dailey, & Kulisch Sheehan & Martin (2003) (1976) Karau & Kelly (1992) Morgan, Salas, & Smith-Jentsch, Zeisig, Acton, Glickman (1994) & McPherson (1998) Kelley (2001) Navarro (1994) Sonnentag, Frese, Brodbeck, & Heinbokel (1997)

198 modeling complex systems Table 1. (cont.) Kirkman & Shapiro (1997) Kirkman, Tesluk, & Rosen (2004) Kline (2001) Kline & MacLeod (1996) Kolodny & Kiggundu (1980) Kozlowski, Gully, Nason, & Smith (1999) Kristof, Brown, Sims, & Smith (1995) Kuo (2004) Leon, List, & Magor (2004) Losada (1999) MacMillan, Entin, Entin, & Serfaty (1994) Marks, Mathieu, & Zaccaro (2001) Mathieu, Goodwin, Heffner, Salas, & Cannon-Bowers (2000) McGrath (1984) McGrath (1991) McGrath (1997) McGrath, Arrow, & Berdal (2000b) McGrew, Bilotta, & Deeney (1999)

Neal & Hesketh (2002)

Spreitzer, Cohen, & Ledford (1999) Neck, Connerley, & Manz (1997) Spreitzer, Noble, Mishra, & Cooke (1999) Neuman & Wright (1999) Stewart & Barrick (2000) Nieva, Fleishman, & Stoker & Remdisch (1997) Reick (1978) Oetzel (1999) Strasser & Falconer (1997) Pagell & LePine (2002) Sundstrom & Altman (1989) Pearce & Ravlin (1987) Poulton & West (1993) Poulton & West (1999) Priest, Stagl, Klein, & Salas (2006) Rangarajan, Chonko, Jones, & Roberts (2004) Rentsch & Hall (1994) Robertson, Maynard, Huang, & McDevitt (2002) Rouse & Rouse (2004) Ruel (2000) Salas, Dickinson, Converse, & Tannenbaum (1992) Salas, Sims, & Burke (2005)

Sundstrom, DeMeuse, & Futrell (1990) Susman & Ray (1999) Tambe (1996) Tannenbaum, Beard, & Salas (1992) Tata & Prasad (2004) Tjosvold, Wong, Nibler, & Pounder (2002) Tompkins (1997)

Tubbs (1994) Walker & Vines (2000) Werner & Lester (2001)

West, Borrill, & Unsworth (1998) Schippers, Den Hartog, Koopman, Yeatts & Hyten (1998) & Wienk (2003)

our criteria for inclusion (see Table 1). Speciﬁcally, inclusion was dictated by whether a particular initiative addressed team performance or effectiveness and described the relations between three or more constructs or construct categories. The latter of these two criteria was designed to screen out the thousands of research initiatives that addressed only a few of the myriad constructs currently believed to constitute the complex nomological network of teamwork. Although our literature review, which included searches of both electronic databases and our own extensive collection of team literature, resulted in an achieved population of 138 models and frameworks, only 11 of the identiﬁed theories are reviewed here. A strong case can, undoubtedly, be made that the remaining 127 sources are

199 Team Effectiveness in Organizations signiﬁcant in their own right. Unfortunately, space constraints preclude a thorough examination of all identiﬁed models and frameworks. Space limitations also restrict the detail with which any given theory can be addressed. Thus, the inclusion of the 11 theories expounded on in this section is not intended as a covert endorsement of a particular line of research. However, with this caveat in mind it should also be noted that we believe that the sample of models and frameworks selected for review in this chapter is quite representative of the larger population of theoretical research prevalent within the team domain. In fact, the 11 models and frameworks reviewed below were deliberately sampled because they illustrate some of the key advancements contributing to our collective understanding of teams over the past 25 years. The particular initiatives selected for review in this section were vital in shaping our thinking about the key constructs constituting and contributing to team effectiveness. Thus, this section provides critical background information for the reader interested in gaining a deeper appreciation of why a given model or framework was strongly inﬂuential in the development of the conceptually integrative team-effectiveness framework presented later in this chapter. Prior to launching into a full discussion of the salient features of the 11 models and frameworks, we begin by describing the criteria utilized to select these particular research efforts from the larger population of initiatives identiﬁed via our literature review.

criteria for model/framework selection Prior to discussing the particularities of the 11 models and frameworks, we will brieﬂy describe the criteria that guided the process whereby a given initiative was selected for review. Three criteria guided our decision to expand on a given model or framework: (1) model form, (2) model dynamism, and (3) model focus. Speciﬁcally, we feel that the 11 models and frameworks reviewed in this section are representative of cutting-edge advancements in the teams domain, advancements that we capture in our three criteria. After brieﬂy discussing these three criteria, we chronologically review each of the 11 models and frameworks.

200 modeling complex systems Model Form Perhaps the most prevalent theme within the team-effectiveness literature reviewed was the widespread use of ipo (inputprocess-output) models as a guiding framework (McGrath, 1964). The epistemological foundation for ipo models originates within general systems theory and its many derivatives (see von Bertalanffy, 1956). Thus, the use of ipo models to frame team effectiveness is consistent with derivatives of general systems theory, including both sociotechnical systems theory (Emery & Trist, 1960) and open systems theory (Katz & Kahn, 1978). Although ipo models are predominant in the teams domain, it should be noted that there is ongoing debate about the adequacy of this approach (Hackman, 2002; Ilgen, Hollenbeck, Johnson, & Jundt, in press; Sundstrom et al., 1990). Generally speaking, ipo models highlight the importance of throughputs as mediators or moderators of the relations between input factors (e.g., team and individual characteristics) and outputs (e.g., team satisfaction, performance). Thus, the use of an ipo model to frame team effectiveness is particularly important for capturing the dynamic interactions and emergent states that constitute teamwork and illuminating the nomological network of these processes. Describing effectiveness through an ipo lens not only emphasizes the importance of the interactions between inputs, processes, and resulting outputs but also recognizes that team outputs can be fed back into input variables and that teamwork does not occur in a vacuum (Tannenbaum et al., 1992). The use of ipo models reﬂects the current state of the art within the teams domain, as witnessed by the fact that the preponderance of models and frameworks identiﬁed by our literature search adopted this format. Thus, describing teamwork through an ipo lens has advanced our collective understanding of the factors that constitute and promote team effectiveness. However, there have been other advances with regard to team effectiveness that are even more intriguing, and these innovations will be reviewed in the following subsections describing model dynamism and model focus. Model Dynamism A second critical difference between the models and frameworks of team effectiveness identiﬁed in our literature review is that they tend to diverge on the degree to which they explicitly consider how team processes may differ across the dynamic task cycle of teams. While initial attempts by researchers to describe

201 Team Effectiveness in Organizations teamwork focused primarily on team inputs as a mechanism for manipulating team outcomes (Goldstein, 1993; Guzzo & Dickson, 1996), more recent models and frameworks of team effectiveness have served to illuminate the “black box” of team process. For example, Campion and his colleagues’ research (Campion et al., 1993; Campion et al., 1996), although inclusive of many of the key constructs believed to relate to team performance, is relatively static in nature. Conversely, the research advanced by Dickinson and McIntyre (1997) adopts a more dynamic position. The movement to model ﬂuidity reﬂects a growing recognition within the teams community that collective task performance requires adaptive moment-to-moment interteam and intrateam interaction (Dickinson & McIntyre, 1997; Fleishman & Zaccaro, 1992; Marks et al., 2001; Shiﬂett, Eisner, Price, & Schemmer, 1985). Modeling the dynamic nature of interdependent action blurs the distinction between traditionally held notions of predictors and criteria and, therefore, mandates the building of increasingly sophisticated theories in order to capture the inherent complexities. In addition to raising a number of intriguing directions for future endeavors, research such as that presented by Dickinson and McIntyre (1997) and Marks et al. (2001) has implications for human-capital-management systems. For example, if particular processes take precedence over other processes at speciﬁc phases of teamwork, then performancemeasurement, -feedback, and -training systems should be appropriately tailored to reﬂect these contingencies. Model Focus A third primary difference between the research efforts identiﬁed in our literature review concerns the overall focus of the work. The focus factor pertains to whether a particular model or framework deﬁnes performance as being driven primarily by internal team factors or by external context. A similar approach was utilized by Ancona and Chong (1999) to map existing group theory. These authors assert that the key to understanding team behavior lies in a shift in focus from internal factors, such as midpoints and stages of team development, to the temporal role of team context. These assertions are theoretically supported by a branch of mathematics known as coupled oscillations (i.e., entrainment). The stream of thought concerning entrainment is grounded in the premise that team members’ behavior becomes entrained to

202 modeling complex systems

Figure 1. Conceptual model of team performance (Nieva, Fleishman, & Reick, 1978).

naturally occurring contextual cycles. For example, entrainment is witnessed in the recurring ﬂurry of goal-directed activity that takes place within a top management team as the ﬁscal quarter comes to a close. As the management team’s cycles become entrained to the quarterly business cycle, the environment serves as a pacer, rhythm setter, creator of opportunity, source of interrupts, and context for the meaning of time. Efforts to understand teamwork via entrainment suggest that performance is largely determined by contextual factors in the team’s external environment. However, many of the models and frameworks delineated in the material covered in our literature search conceptualize performance as a phenomenon that is actively driven by members and events within the team. The differences between these perspectives have important implications for framing team effectiveness. For example, interventions meant to enhance teamwork have traditionally targeted variables inside the team for change. Given the compelling argument supporting entrainment, it seems that traditional approaches emphasizing internal factors should be augmented by an understanding of key external factors. A simultaneous internal and external focus could illuminate insights into critical processes such as the timely delivery of scarce resources. In fact, some initiatives have already been undertaken to examine the joint inﬂuence of both internal and external forces (e.g., Kozlowski, Gully, Nason, & Smith, 1999; McGrath, 1991).

203 Team Effectiveness in Organizations

Figure 2. Model of group effectiveness (Gladstein, 1984).

models and frameworks of team effectiveness One of the earliest conceptualizations of teamwork was advanced by Nieva, Fleishman, and Reick (1978) (see Figure 1). Nieva et al. propose that team performance is composed of both individual task performance and team-level performance functions. This model also speciﬁes four categories of team-performance antecedent variables: (1) team environment (e.g., social context, standard operating procedures), (2) member resources (e.g., individual skills, abilities, personality characteristics), (3) team characteristics (e.g., communication, training), and (4) task characteristics (e.g., structure, complexity). Gladstein (1984) presents another relevant model that depicts the relations between group inputs, processes, and outputs (see Figure 2). Particularly noteworthy is the fact that Gladstein’s model has empirical support from a large sample (Goodman, 1986). This model incorporates individual-level input factors, including group compo-

204 modeling complex systems

Figure 3. Normative model of team effectiveness (Hackman, 1987).

sition variables (e.g., skills, heterogeneity) and group structure (e.g., formal leadership, work norms). In addition, it includes organizational-level input factors such as resources available (e.g., training, consulting) and organizational-structure variables (e.g., rewards, supervisory control). The relations between individual- and organizational-level input factors and team effectiveness are proposed to be mediated by group processes. The model also indicates that group task complexity, uncertainty, and interdependence moderate the relations between group processes and outcomes such as satisfaction. Hackman (1987) also advanced a model of team effectiveness (see Figure 3). This model illustrates how a broad range of variables can inﬂuence teamwork. There are a number of propositions that ﬂow from this model, including that input factors (e.g., organizational context, group design) are related to team processes, which, in turn, are related to team effectiveness (e.g., customer satisfaction with product, member satisfaction, capability of team members to work together over time). More speciﬁcally, the model highlights the importance of fostering an organizational context that supports and reinforces teamwork via rewards, education, and availability of information. Group design is a second input factor illustrated that is proposed to relate to team processes. According to Hackman, group

205 Team Effectiveness in Organizations

Figure 4. Punctuated-equilibrium model of team performance (Gersick, 1988).

design consists of such things as (a) task structure, (b) group composition, and (c) appropriate group norms regarding teamwork. Hackman’s (1987) model also speciﬁes process criteria of effectiveness that can serve to guide the diagnosis of team weaknesses. These process criteria include (a) level of effort, (b) amount of knowledge and skill, and (c) appropriateness of task-performance strategies. Finally, the model proposes two variables that may moderate the depicted relations. First, the relations between team inputs and team processes are moderated by the ability of the group to minimize process losses (i.e., gain group synergy). Second, the relations between team processes and team effectiveness are moderated by the material resources available to the team. Thus, no matter how well team members interact with one another in terms of effort, skill, and performance strategies, if there are not enough material resources, the task may not be completed. A fourth model of team effectiveness reviewed was advanced by Gersick (1988) and is known as the punctuated-equilibrium model (pem) (see Figure 4). The pem is empirically grounded in the ﬁndings of research conducted with a sample of eight diverse teams. The evidence accumulated from Gersick’s research suggests that teams determine an initial method of performance during their ﬁrst meeting and adhere to this method until the midpoint of the target objective is reached. Gersick argues that, at the midpoint, team members become aware of the time left to completion and switch their strategy accordingly. Although fairly simple in its depiction, this research makes an important contribution by showing that teamwork is dynamic and that teams adapt their performance strategies in accordance with temporal contingencies. In Tannenbaum et al. (1992), which reported one of the ﬁrst integrative efforts to frame team effectiveness (see Figure 5), the researchers reviewed many of the then-current ipo models of team effectiveness and developed an integrative framework that built on prior initiatives. This framework is more complex than the pre-

206 modeling complex systems

Figure 5. Team-effectiveness framework (Tannenbaum, Beard, & Salas, 1992).

viously described Gladstein (1984) model but includes some of the same variables. Speciﬁcally, it illustrates four distinct types of input variables, including (1) task characteristics, (2) work characteristics, (3) individual characteristics, and (4) team characteristics. It suggests that these input factors affect each other and also serve to affect both team members and team processes (e.g., backup behavior, coordination, adaptability) that occur over time. Not surprisingly, both the individual team member and team processes are proposed to affect team-performance outcomes (e.g., quality, quantity, time, errors). The framework also depicts system feedback, resulting from team performance and performance outcomes, cycling back as subsequent system input. Furthermore, it recognizes that training or teambuilding interventions may moderate the relations between inputs and processes as well as those between processes and performance outcomes. A ﬁnal addition of this framework over previous models is the recognition that organizational and situational characteristics affect team effectiveness, not just at the input stage, but throughout the entire ipo process. A sixth model of team effectiveness reviewed herein was advanced in Campion et al. (1993) (see Figure 6). Synthesizing all ﬁve team-effectiveness models already illustrated, Campion et al. devel-

207 Team Effectiveness in Organizations

Figure 6. Synthesized model of group effectiveness (Campion, Medsker, & Higgs, 1993).

oped a metamodel of team effectiveness. This hybrid model incorporates only those constructs proposed to directly affect team effectiveness, thus ignoring key mediators and moderators of the relations between team inputs and outputs. Campion et al.’s model parsimoniously frames the complexities of team effectiveness. Although it may oversimplify the dynamic, recursive nature of team performance, this programmatic theoretical formulation has received empirical support from two diverse samples of teams (Campion et al., 1993; Campion et al., 1996). Campion et al.’s (1993) metamodel describes ﬁve categories of variables that are proposed to affect team effectiveness: job design, interdependence, composition, context, and process. Speciﬁcally, job

Figure 7. Team-evolution and -maturation model (Morgan, Salas, & Glickman, 1994). * Gersick (1988). ** Tuckman (1965).

209 Team Effectiveness in Organizations design subsumes self-management, participation, task variety, task signiﬁcance, and task identity. Interdependence is proposed to encompass task interdependence, goal interdependence, and interdependent feedback/rewards. Composition is proposed to subsume heterogeneity, ﬂexibility, relative size, and preference for group work. Context is proposed to cover training, managerial support, and communication/cooperation between groups. The process category includes potency, social support, workload sharing, and communication/cooperation within groups. Building on the work of Gersick (1988) and Tuckman (1964), Morgan et al. (1994) developed a model that illustrates the stages that teams progress through before, during, and after task performance (see Figure 7). The model proposes that task-orientated teams progress through a series of developmental stages at varying rates (see Tuckman, 1965). The speciﬁc stage at which a given team begins and how quickly the team progresses through the proposed stages depend on such characteristics as (a) members’ experience as a team, (b) individual expertise, (c) task characteristics, and (d) environmental context. Morgan et al.’s model also proposes that, as a team progresses through these stages, there are two tracks of skills that must be mastered before it can perform effectively: taskwork and teamwork. Taskwork represents the “task-orientated skills that the members must understand and acquire for task performance” (Salas et al., 1992, p. 10). Conversely, teamwork skills reﬂect the behavioral interactions, cognitions, and attitudinal responses that must be mastered before a team can work together effectively. The advantages of the Morgan et al. model over previous theoretically grounded models are twofold. First, it illustrates that there are two sets of skills that must be mastered in order for a team to be effective, both task and teamwork. Subsequent research has provided empirical support for this assertion (Glickman et al., 1987). Second, by illustrating behaviors that occur within each developmental stage, the model serves to frame how teamwork develops during task performance and training. More recently, Dickinson and McIntyre (1997) proposed a model that describes the interrelations between essential teamwork processes such as communication, team orientation, team leadership, monitoring, feedback, backup behavior, and coordination (see Figure 8). In this model communication acts as the glue linking together all

210 modeling complex systems

Figure 8. Model of teamwork (Dickinson & McIntyre, 1997).

other teamwork processes. Another aspect of the model is the integration of both team skill competencies (e.g., team leadership) and team attitude competencies (e.g., team orientation). Together, team leadership and team orientation are proposed to facilitate a team member’s capability to monitor his or her teammates’ performance. Furthermore, the model proposes that performance monitoring drives both the content of feedback and timely backup behaviors. It also suggests that, when all the aforementioned teamwork competencies are occurring in unison, they synergistically serve as a platform for team coordination. In turn, the feedback resulting from team coordination serves as input back into team processes. Although Dickinson and McIntyre’s (1997) model incorporates many of the skill competencies underlying teamwork, it fails to model many of the critical antecedents and outcomes of team process. The ninth initiative reviewed in this section, undertaken to illuminate the complexities of team effectiveness, is the research of Marks et al. (2001) (see Figure 9). This systematic research has markedly advanced our understanding of the content constituting team process and the ﬂuid nature of teamwork as processes unfold over time. Marks et al. advance a temporally based framework of team effectiveness that extends recent notions of team process by categorizing throughputs into recurring phases. Speciﬁcally, this episodic model of team process consists of a series of recursive ipo loops proposed to occur sequentially and simultaneously during both a transition stage and an action stage of performance. Distinct competencies characterize the action (e.g., mission analysis, goal speciﬁcation)

211 Team Effectiveness in Organizations

Figure 9. The rhythm of team task accomplishment (Marks, Mathieu, & Zaccaro, 2001).

and the transition (e.g., systems monitoring, coordination) stages, suggesting that certain ksas take precedence depending on the timing of performance. Interpersonal processes are proposed to occur during both stages. As previously noted, Marks et al.’s (2001) effort to model teamwork is theoretically as well as practically intriguing. The implications for those charged with observing, assessing, and rewarding team performance include the need to develop a new appreciation of the timing of their measurements as well as new human-capitalmanagement systems that account for which processes are predominate at a given time. There is “nothing as practical as a good theory” (Lewin, 1951, p. 169), and this maxim is nowhere more applicable than when utilized to describe meaningful research that offers a more robust speciﬁcation of the dynamic nature of teamwork. Another recently advanced theoretical initiative reviewed was the “Big Five” model, proposed by Salas et al. (2005). This model was developed in an effort to highlight the “essence of teamwork” by illustrating the relations between the processes that they argue constitute the core of interdependent interaction (see Figure 10). Speciﬁcally, this attempt to model teamwork highlights the centrality of ﬁve core teamwork processes, including (1) team leader-

212 modeling complex systems

Figure 10. The Big Five teamwork model (Salas, Sims, & Burke, 2005).

ship, (2) team orientation, (3) mutual performance monitoring, (4) backup behavior, and (5) adaptability. Furthermore, the Big Five model also illustrates the importance of three ancillary team products and processes, including (1) shared mental models, (2) closedloop communication, and (3) mutual trust. Taken together, these eight constructs are dynamically related to one another and collectively form teamwork. At a broad level the Big Five model of teamwork proposed in Salas et al. (2005) adopts a similar approach to that of the above-described theory advanced by Dickinson and McIntyre (1997). However, the Big Five model proposes a set of constructs and construct interrelations that differ somewhat from Dickinson and McIntyre’s attempt to describe the dynamic nature of teamwork. For example,

Figure 11. ito model of team adaptation (Burke, Stagl, Salas, Pierce, & Kendall, under review).

214 modeling complex systems while Dickinson and McIntyre’s theory primarily highlights the centrality of skill competencies, that proposed by Salas et al. moves beyond modeling teamwork behavior to incorporate affective competencies (i.e., mutual trust), cognitions (i.e., shared mental models), and emergent team phenomena (i.e., adaptability). The 11th and ﬁnal programmatic effort undertaken to investigate team effectiveness reviewed here was advanced by Burke, Stagl, Salas, Pierce, and Kendall (2006). These researchers proposed a model of team adaptation within an ipo framework (see Figure 11). The model advanced by Burke et al. emphasizes the centrality of an adaptive process that unfolds over time to emerge as team adaptation. Speciﬁcally, this applied research initiative deﬁnes team adaptation as an emergent phenomenon that coalesces over time from the unfolding of an adaptive process whereby one or more team members utilize their resources to functionally change current behaviors, cognitions, or attitudes to meet expected or unexpected demands. Essentially, team members draw from their individual and shared resources to detect, frame, and act on a set of cues that signal the need for functional change. As this adaptive process is carried out, feedback is generated that subsequently serves to revise shared cognition and adaptive input factors. Thus, the adaptive process is recursive by nature. The team-adaptation model illustrated in Figure 11 is the only initiative reviewed in this chapter that has radically departed from the traditional predominant focus on either task or team performance. Certainly, this state of affairs is indicative of an area ripe for exploration, and, therefore, it is our opinion that additional theory building should be undertaken to develop meaningful models of (1) task, (2) team, (3) contextual, and (4) adaptive performance. Perhaps one fruitful avenue for future exploration involves framing these unique but interrelated aspects of performance in a nomological network of lawful relations. For example, research could be carried out to develop a deeper understanding of the similarities and differences between team and contextual performance, which seem to share somewhat similar content and emphasis. We believe that the similarities and differences between contextual performance (see Borman & Motowidlo, 1993) and team performance warrant closer attention in future research endeavors. It may be particularly interesting to examine whether indexes of contextual performance

215 Team Effectiveness in Organizations capture a practically meaningful amount of unique variance in substantive criteria beyond that already explained by indexes of teammember performance.

models and frameworks summary The 11 team-effectiveness models and frameworks described above have commonalties as well as differences. These points of similarity and distinction provided us with clues as to potential content to include in our integrative framework of team effectiveness. For example, with the notable exception of Dickinson and McIntyre (1997) and Salas et al. (2005), whose models focus strictly on team processes, one commonality shared by all the above-noted efforts to understand team effectiveness is that they adopt an ipo model (see McGrath, 1964). Thus, each of these initiatives acknowledges that inputs, processes (throughputs), and outputs need to be examined in order to gain a holistic understanding of team effectiveness. The prominent differences between the ipo models reviewed lie (a) in the speciﬁc primary variables or constructs highlighted and (b) in the moderators and mediators included (e.g., situational/organizational characteristics). Another difference lies in the degree to which the models explicitly consider how the relations proposed differ across the dynamic task cycle and/or life span of teams. For example, of the efforts reviewed above, Hackman’s (1987) model, Campion et al.’s (1993) metamodel, and Gladstein (1984) are the least dynamic, followed by Nieva et al. (1978), Tannenbaum et al. (1992), Gersick (1988), and Morgan et al. (1994). Although the last four adopt a less static lens than the ﬁrst three, they do not present as detailed a picture of the ﬂuidity of teamwork as that modeled by Dickinson and McIntyre (1997), Salas et al. (2005), Burke et al. (2006), or Marks et al. (2001).

Toward an Integrative Multilevel Theoretical Framework The 11 models and frameworks discussed in the previous section were utilized to guide an inquiry into the properties that seem to be essential to any truly integrative theory of team effectiveness. Fur-

Figure 12. Integrative framework of team effectiveness.

217 Team Effectiveness in Organizations thermore, we drew heavily on what we considered intriguing constructs and ideas from the broader domain of team literature. The outcome of this inquiry is a preliminary integrative framework of team effectiveness (see Figure 12). In the remainder of this section we review some of the critical components of the framework while emphasizing how it expands prior programmatic efforts to understand team effectiveness. The multilevel integrative framework depicted in Figure 12 offers researchers and practitioners a simple but meaningful heuristic that illustrates some of the most important aspects of team performance. On a broad level, the framework highlights the role of team inputs (i.e., individual characteristics, team characteristics, task characteristics, work structure) in promoting teamwork and team performance. In addition, the framework illustrates the moderating role of individual-level cognition (i.e., expectations about roles and requirements) on the relations between team inputs and throughputs. Essentially, team inputs are actively interpreted by team members, who, via this ongoing process, form stable yet malleable expectations regarding the nature of their obligations. Team members with accurate expectations are more likely to know which team processes to engage in and when particular activities should occur. As the processes constituting teamwork are dynamically, simultaneously, and episodically enacted over time, they lead to shared cognition. Shared cognition (e.g., shared mental models, team-situation awareness, psychological safety) accrues as teamwork occurs and, in a recursive fashion, inﬂuences subsequent teamwork activities. The framework also suggests that team performance results in individual- and team-performance outcomes. Performance outcomes produce system feedback that, over time, can serve to change both the organizational inputs and the teams inputs available to subsequent performance episodes. The organizational environment is characterized by a cue stream that is also interpreted by team members, thereby contributing to both individual and shared cognition. Furthermore, team leadership inﬂuences, and is inﬂuenced by, the accuracy of individual and shared cognition. In the remainder of this section, the various aspects of the framework are detailed in greater speciﬁcity. The integrative framework of team effectiveness illustrated in Figure 12 incorporates four categories of input factors, including individual characteristics, team characteristics, task characteristics,

218 modeling complex systems and work structure. Each of these four broad categories is an umbrella for multiple speciﬁc constructs. For example, the category of individual characteristics covers a wide range of phenomena, such as task ksas, motivation, team orientation, mental models, and personality. In turn, each of these constructs is related to team processes and team-performance outcomes to varying degrees. For example, the results of empirical research have repeatedly supported the importance of a team orientation (e.g., Driskell & Salas, 1992; Eby & Dobbins, 1997; Goodwin, O’Shea, Driskell, Salas, & Ardison, 2004). Likewise, research results suggest that personality variables such as extraversion, openness to experience, and adjustment are essential for coordinated performance (Driskell, Hogan, & Salas, 1987; LePine, 2003). The category of team characteristics also includes myriad constructs that may either have been formally established (e.g., power structure, performance arrangements) or have emerged upward from the unfolding of coordinated dyadic role exchanges of team members over time (e.g., team-level openness to experience, teamlevel team orientation). For example, empirical work has supported the relation between team cohesion and team performance (Dailey, 1980; Mullen & Cooper, 1994). Given the paucity of attention afforded team characteristics to date, additional research should be undertaken to investigate other critical factors such as team climate and culture. The task-characteristics category includes task organization, task type, and task complexity. Both theoretical research and empirical research suggest that task characteristics are, indeed, critical to fostering team effectiveness (Herold, 1978; McGrath, 1984; Steiner, 1972). Interestingly, the task-characteristics category reminds us that not all teams are alike and that task variations leading to different levels of interdependence can have a profound effect on how teams interact. These differences will be important to future research initiatives undertaken to develop team type–speciﬁc models and frameworks of team effectiveness. One implication of our emphasis on work structure is that social-network analysis (see Krackhardt & Brass, 1994) can be applied to analyze communication patterns and that the ﬁndings of this process can be utilized to model, predict, and manage team-member and team performance. The fourth category of input variables, work structure, encom-

219 Team Effectiveness in Organizations passes work assignment, team norms, and communication structure. Work structure is critical to team performance because, as open systems, teams consist of both a formal work structure and a unique but interdependent social structure. Together, the socio and technical systems serve to shape both what cues team members attend to and how they react to those cues. Furthermore, work characteristics such as communication structure dictate who has access to what information and when, which is essential for negotiating what one wants to accomplish. Also, team norms have a considerable inﬂuence on what behaviors are deemed appropriate (Hackman, 1990). It should be noted, however, that the particular constructs exempliﬁed in the above descriptions were largely adopted from the programmatic theoretical framework advanced by Tannenbaum et al. (1992). This is important because, while there exists a general consensus about the nature of the broad categories of input variables, the speciﬁc constructs proposed to be encapsulated within these categories vary from research program to research program. For an equally impressive yet somewhat different set of team-input variables, see Campion and colleagues’ (Campion et al., 1993; Campion et al., 1996) metamodel of team effectiveness as reproduced in Figure 6 above. Our framework draws heavily on many of the recent advancements appearing in the body of team literature and perhaps most of all from cutting-edge research in cognitive psychology. For instance, while we frame many of the same inputs (i.e., individual characteristics, team characteristics, task characteristics, work structure) that are traditionally advanced within the team-effectiveness domain, our framework suggests a novel purpose for these components. The hybrid framework depicted in Figure 12 reﬂects our belief that the myriad relations between input variables and teamwork processes are moderated by team-member cognition. Thus, team-member expectations, established and changed by team inputs, ongoing teamwork, team leadership, and the organizational context, moderate the relations between inputs and processes. Essentially, team members with accurate, ﬂexible expectations about their roles and requirements will be better positioned to engage in appropriate team-member and team processes at the optimal point in time. Shared cognition also plays a prominent role in the hybrid framework. Shared cognition such as shared and compatible mental mod-

220 modeling complex systems els and shared situation awareness serves to form templates that are drawn on by team members during teamwork. Shared templates or shared frames of reference imply processing objectives that constitute the social reality that team members share (Hinsz, Tindale, & Vollrath, 1997). Speciﬁcally, this set of instantiated cognitive structures enhances teamwork processes because it provides team members with the insight required to understand (1) what they should and should not be doing, thinking, and feeling, (2) how to go about accomplishing stated objectives, and (3) when to enact ksaos in support of processes in order to meet tactical and strategic objectives. Also, as teamwork occurs, the organizational environment changes, and/or team leaders intervene, shared mental models and other forms of shared cognition are revised. In the absence of this cognitive reservoir, teamwork processes will still ensue; unfortunately, however, process losses will likely abound. In addition to having a distinctive cognitive ﬂavor, our framework also incorporates advancements made in modeling the dynamic nature of teamwork (Dickinson & McIntyre, 1997; Fleishman & Zaccaro, 1992; Marks et al., 2001; Shiﬂett et al., 1985; Salas et al., 2005). Speciﬁcally, our hybrid framework reﬂects the cyclic, episodic nature of activity in high-performance teams by illustrating a sample of core teamwork processes in a cylinder. The point we are trying to communicate by framing teamwork processes with a cylinder is simple: that both team members’ and team-level competencies must be displayed in a dynamic, simultaneous, and episodic fashion in order to foster individual and team effectiveness. In fact, several revolutions may be needed to meet a single objective. Furthermore, as multiple objectives are interlaced or additional interrelated but unique projects undertaken, the teamwork processes displayed in a single revolution may not be homogenously directed at the accomplishment of a single goal. Our integrative framework of team effectiveness also acknowledges the centrality of team leadership throughout the life span of the team (see Marks et al., 2000; Stagl, Salas, & Burke, in press). While the construct of team leadership is a complex phenomenon with a variety of conceptualizations (e.g., functions, roles, traits, competencies, implicit perceptions, negotiated dyadic role exchanges) most would agree that team leaders and the leadership processes that they enact are essential to promoting team performance, adaptation,

221 Team Effectiveness in Organizations and effectiveness. In fact, recent meta-analytic evidence highlights the importance of team-leadership behaviors in achieving team outcomes (Burke et al., in press). It is through the leadership process that team leaders act to synchronize task and developmental cycles. Moreover, they institute the conditions that teams and their members draw on before, during, and after task episodes (Hackman, 2002). Consequently, team leaders act as major drivers and maintainers of team development, performance, and effectiveness over time. Another unique aspect of our integrative hybrid framework is the fact that it illustrates, albeit at a broad level, the interdependencies between team members’ competencies and team competencies. In our framework, both individual-level (e.g., interpersonal) and team-level (e.g., adaptive, coordination) processes are highlighted. Furthermore, the processes displayed in our framework are meant as a small sample of the number of competencies that must be ﬂuidly interlaced as team performance unfolds over time. Essentially, teamwork/team performance is composed of the dynamic display of (1) team-member taskwork ksas, (2) team-member teamwork ksas, and (3) team ksas. As the competencies constituting these three sets unfold over time, they serve to yield team performance. While prior programmatic efforts to frame teamwork have included both individual-level team members’ competencies and team-level competencies, the distinction between these ksa sets, as well as their symbiotic interdependencies, has seldom been explicitly considered. For further clariﬁcation of this issue, the inquisitive reader is directed to Campbell and Kuncel (2001), who provide a more complete discussion of this topic. As noted above, our framework illustrates both individual- and team-level processes; therefore, it is intuitive that it should also include both individual- and team-level performance outcomes. The individual- and team-level performance outcomes depicted in the model are meant to include both traditional output indexes such as the number of goods produced as well as emergent states (Marks et al., 2001) or psychosocial traits (Cohen & Bailey, 1997) that result from a team’s experiences while navigating its operational challenges. These emergent states (e.g., cohesion, collective efﬁcacy) are simultaneously proximal products of interdependent interaction and serve as subsequent inputs to new performance episodes. Incorporating multiple performance outcomes mandates the need for multiple ef-

222 modeling complex systems fectiveness evaluations. Thus, our framework acknowledges that the particular level of effectiveness achieved by a given team member or team will vary depending on the dimension(s) under consideration. Furthermore, the performance-outcome dimensions that are most relevant to the evaluation process will change as a function of time. In essence this suggests that teams and their members can be described in terms of effectiveness patterns. The degree to which variables constituting these patterns can compensate for one another when contributing to a global index of effectiveness that can subsequently be utilized by organizations for practical decision-making purposes remains an area ripe for exploration.

summary The multilevel integrative framework described in this section offers researchers and practitioners a useful, yet simple, heuristic that can be called on to quickly understand the most important aspects of interdependent performance. The integrative framework illustrated in Figure 12 above is intended to highlight often-ignored issues in team-effectiveness research. Speciﬁcally, our hybrid framework advances the science of teamwork by (1) adopting a cognitive approach to framing the dynamic and multilevel nature of team effectiveness from an ipo perspective, (2) illustrating some the interdependencies between team members’ competencies and team-level competencies, and (3) describing key antecedents to team performance and effectiveness. On further conceptual reﬁnement, we expect that our hybrid integrative framework will be useful in (1) deﬁning a nomological net of teamwork, (2) formulating empirically testable propositions, and (3) designing interventions to facilitate human-capital management. With regard to this latter point, in the next section we review the practical implications of our hybrid framework for performance measurement, training, and stafﬁng teams.

Practical Implications During the last 25 years, hundreds of research studies have been guided by the team-effectiveness theories presented in this chapter.

223 Team Effectiveness in Organizations In some cases, the accumulated ﬁndings from this research have been shaped into principles and guidelines (e.g., Salas, Burke, & CannonBowers, 2000; Salas, Burke, & Stagl, 2004). The delineated principles and guidelines offer essential insight to those concerned with designing, developing, and delivering human-capital-management interventions. Thus, for those individuals charged with the application of measurement tools, instructional strategies, and task-based simulations, it seems that, as we have seen, there is “nothing as practical as a good theory.” Keeping with this line of thinking, the use of theoretically grounded team-effectiveness models for measuring performance, conducting developmental interventions, and stafﬁng teams will be addressed in this section.

measuring team performance Team-effectiveness models and frameworks are critical to building appropriate team-measurement systems because they serve to illustrate both the complexities of engaging in teamwork and the many factors surrounding team process. As discussed above, teameffectiveness models and frameworks illuminate what factors are potentially important in a given situation. Furthermore, depending on whether the team-effectiveness model adopted incorporates the notion of time, it can also provide guidance as to when particular factors are important. Finally, because most initiatives to describe team effectiveness have adopted an ipo lens, the resulting models and frameworks specify how and why a factor is important. The answers to the above questions provide the informed practitioner with clues on what factors to measure, how to measure them, and when to measure them and, thus, serve to increase the utility of a given intervention. The importance of team-effectiveness models and frameworks to this process cannot be understated, as is reﬂected in the statement that performance measurement can be viewed as a purchase of information that a researcher makes in order to help guide his or her decision making (Brannick & Prince, 1997). This purchase of information is better informed when guided by a theoretical rationale such as those provided by the team-effectiveness models and frameworks described in this chapter. Furthermore, developing measures of teamwork must be guided by a number of

224 modeling complex systems factors, including the purpose of measurement, the nature of incorporated stimuli, target ksa competencies, measurement timing, and anticipated costs (McIntyre & Salas, 1995). While measures can be designed to assess any part of the ipo system, capturing diagnostic information mandates measuring processes as raw materials are transformed into ﬁnished products and services as well as team-performance outcomes.

training team members In addition to providing a platform for performance measurement, team-effectiveness models and frameworks also offer insight into the types of training that can be utilized to develop high-performance teams and team members. This is because team-effectiveness models and frameworks illustrate critical ksa competencies or individual and team-level processes and their nomological net. In turn, these ksa competencies serve as the core content when designing, developing, and delivering instructional strategies to train teams and team members. While there are slight variations in the labels and deﬁnitions produced by those initiatives undertaken to examine teamwork, most of these differences can be reconciled in a set of core universal team processes. These processes are illustrated in most exhaustive attempts to frame teamwork (see Cannon-Bowers et al., 1995; Marks et al., 2001) and were described earlier in this chapter. By illuminating the speciﬁc competencies underlying teamwork, the comprehensive research initiatives listed above provide guidance on the development of instructional strategies. Essentially, models and frameworks of team effectiveness help practitioners answer the question of how to turn a team of experts into an expert team. In addition to the insight provided by models and frameworks of team effectiveness, researchers have also offered general guidance that is essential for developing instructional strategies. For example, in order for team members to develop the adaptive expertise required for effectiveness in a chaotic context, “training systems must shift in orientation from off-site, single episode, individual-level skills delivery to multi-episode, on line, multilevel systems. Training must be shifted to the work environment, focused on the development of adaptive individual and team skills” (Kozlowski, 1998, p. 116). Care

225 Team Effectiveness in Organizations must also be taken when designing interventions to help ensure that both horizontal transfer and vertical transfer ensue (Kozlowski, Brown, Weissbein, Cannon-Bowers, & Salas, 2000). Together, theories of team effectiveness and principles delineated from their application can be a powerful source of information for fostering team performance and effectiveness.

staffing teams There is a growing awareness of the importance of managing human capital at multiple levels (see Campbell & Kuncel, 2001; Huselid, 1995; Ployhart & Schneider, 2005; Schneider, Smith, & Sipe, 2000; Stevens & Campion, 1999). Once dominated by a largely individualistic emphasis, private- and public-sector organizations now recognize that the maxim “the people make the place” (Schneider, 1987, p. 450) holds true at all levels in the conceptual space. Essentially, all teams must import energy in the form of human capital in order to avoid entropy and enjoy the prosperity that interdependent interaction can produce (see Katz & Kahn, 1978). Thus, the traditional concerns of person-job ﬁt and person-environment ﬁt must be balanced with equal attention directed at person-team ﬁt (Hollenbeck et al., 2002). Although, a few endeavors have already been undertaken to delineate guidelines for stafﬁng teams (see Driskell et al., 1987; Jackson & Ruderman, 1996; Klimoski & Jones, 1995), additional research is required in order to more fully reﬁne current processes as well as to develop new stafﬁng strategies that simultaneously enhance both individual and team effectiveness. Following this line of thinking, researchers have begun to explore the importance of individual differences for both team members and team performance. Particularly noteworthy are those investigations that offer compelling theoretical rationales for how individual-level characteristics synergize and emerge as team-level phenomena (see Barrick, Stewart, Neubert, & Mount, 1998; LePine, 2003; Neuman & Wright, 1999). Specifying the mechanisms whereby individual-level characteristics emerge upward to collective phenomena within various team types characterized by varying levels of interdependencies is an issue warranting closer attention.

226 modeling complex systems

Toward an Integrated Science of Teams In this chapter we synthesized the current state of the literature with regard to the ongoing proliferation of team-effectiveness models and frameworks. This endeavor resulted in the identiﬁcation of 138 team-effectiveness models and frameworks that have been proposed since the early 1980s. Speciﬁcally, our review identiﬁed 15 initiatives published in the 1980s, 68 models and frameworks published in the 1990s, and 51 efforts that had already been advanced by early 2004. As anticipated, the results of our literature review suggest that there was a 257% increase in the number of theoretically grounded models and frameworks proposed in the 1990s as compared to the prior decade. Furthermore, team researchers are on pace to eclipse all previous historical periods combined in just the ﬁrst decade of the 21st century. The above-noted ﬁndings are indicative of the proverbial double-edged sword. If you typically view the glass as half full, you can ﬁnd solace in knowing that the quest to understand the complex, dynamic, and elusive nature of teamwork continues and is, indeed, quickening. Those viewing the glass as half empty, however, have undoubtedly by now surmised that our review also suggests that, while there are some commonalties among the models (Ruel, 2000), there remains a signiﬁcant amount of intereffort variation. The accelerating rate of production, coupled with the concomitant disparities among the initiatives advanced during the past 25 years, foreshadows a looming crisis in the teams domain, an impending conundrum paralleling the general lack of theoretical integration plaguing the behavioral sciences today (see Campbell, 1990b). Why are there such a wide range of perspectives on supposedly the same underlying team phenomena and the corresponding intereffort variation, as noted above? We believe the answer is simple: team researchers and, in fact, whole disciplines fail to communicate. Why is this so? The answer is certainly complex. However, aside from environmental and interpersonal issues, the answer is relatively straightforward: team researchers are speaking different languages, different in terms of underlying paradigms, assumptions, values, goals, and missions. Similar problems often undermine mergers, acquisitions, and multinational teamwork. So what can be done when social psychologists are speaking

227 Team Effectiveness in Organizations Latin while cognitive psychologists are versed in the queen’s English? We believe the answer is twofold. Foremost, the current state of affairs characterizing team research ﬁnds the area largely devoid of collectively set, overarching short-, mid-, and long-term goals. Agreed-on tactical and strategic objectives must be articulated if the domain of team research intends to continue maturing in an integrative, productive fashion. After all, differentiation ultimately results in chaos without integration. A second reason for the rampant disjointedness is the lack of an integrative structure that, if present, could serve to house what is currently known, guide future endeavors, and be seamlessly updated by distributed stakeholders to reﬂect accumulating ﬁndings. For example, O*Net has radically reshaped the outmoded Dictionary of Occupational Titles. Perhaps it is time for T*Net to emerge. A failure to systematically encode and integrate is far from the only problem contributing to the frailty of research investigating teams. Unfortunately, team research is also, on occasion, methodologically weak. For example, far too often, in their haste to advance the ﬁeld, team researchers settle for single-source expediency at the expense of scientiﬁc rigor. Speciﬁcally, increased use of triangulation methodology through the use of multiple measures is needed. Transparent Likert scales, subject to a host of contrived response protocols, continue to dominate the landscape of team members’ attitude assessment when seemingly superior measurement techniques such as Thurstone scaling and the conditional-reasoning approach are readily available. Is the extra effort and cost of developing a conditional-reasoning instrument not offset by the more accurate picture of a respondent’s standing on an underlying attribute that would be provided? Keeping with the above line of thinking, a similar issue warranting considerable attention concerns the measurement of team cognition (Salas & Fiore, 2004). Perhaps there is no single greater concern, as it is widely accepted that shared cognition is the foundation of the coordinated dyadic role exchanges that underlie much of team performance. However, for all the fanfare, there is little known about measuring cognition in teams (Cooke, Salas, Cannon-Bowers, & Stout, 2000; Cooke, Salas, Kiekel, & Bell, 2004). What is known suggests that team research would substantially beneﬁt from the reﬁnement of proved techniques for measuring shared cognition, such as

228 modeling complex systems card sorting, concept mapping, textual analysis, relatedness ratings, laddering interviews, and multidimensional scaling. Perhaps the situation is not so bleak or the outcome of theoretical chaos as of yet predetermined. However, the increasing diffusion without the proper structure in place to deﬁne, guide, and integrate forthcoming advancements will, in our opinion, only contribute to the fragility that currently characterizes the teams domain. In order to curb the ongoing diffusion and other substantive problems noted above, we assert that, where appropriate, new initiatives should build on what is enduring from the past. Future endeavors should be informed by the insights afforded by the models outlined in this chapter. Furthermore, we argue for a return to basics because it is only by beginning anew that we can hope to set the substantive goals and structure that will thrust us forward in an organized, meaningful manner.

Concluding Comments The complexity of contemporary work environments compels a recognition that it is no longer economically viable to navigate current challenges via an exclusive reliance on individual workers (West, Borrill, & Unsworth, 1998). In fact, an accelerating rate of change, driven by the ongoing technological revolution, is sweeping away the last remnants of a business landscape once dominated by an emphasis on colocated individuals (Priest, Stagl, Klein, & Salas, 2006). The lesson learned is clear: organizations as open systems are always growing in differentiation and integration. Thus, organizations must adapt their vision, structure, and human-capital practices in order to avoid entropy and enjoy prosperity in the 21st century. It is our hope that the research effort described in this chapter serves to further the understanding of teams, teamwork, team performance, and team effectiveness by providing common ground or a shared mental model of the science of teams. Furthermore, we eagerly anticipate the day when team researchers everywhere step up to the challenges laid at their feet herein and overcome their differences to collectively deﬁne the future of team research.

229 Team Effectiveness in Organizations

Note The views expressed in this work are those of the authors and do not necessarily reﬂect ofﬁcial army policy. This work was supported by funding from the Army Research Institute under sbir topic #osd02-cr01 and under baa #dasw01-04-r-0001. We would also like to acknowledge Dana Kendall for conducting a thorough literature review.

References Aldag, R. J., & Fuller, S. R. (1993). Beyond ﬁasco: A reappraisal of the groupthink phenomenon and a new model of group decision processes. Psychological Bulletin, 113, 533–552. Alper, S., Tjosvold, D., & Law, K. S. (2000). Conﬂict management, efﬁcacy, and performance in organizational teams. Personnel Psychology, 53, 3, 625–642. Ancona, D. G., & Caldwell, D. F. (1992). Bridging the boundary: External activity and performance in organizational teams. Administrative Science Quarterly, 37, 634–665. Ancona, D., & Chong, C.-L. (1999). Cycles and synchrony: The temporal role of context in team behavior. In R. Wageman (Ed.), Groups in Context, 2 (Research on Managing Groups and Teams, Vol. 2, pp. 33–48). Amsterdam: Elsevier. Annett, J., & Cunningham, D. (2000). Analyzing command team skills. In J. M. Schraagen, S. F. Chipman, & V. L. Shalin (Eds.), Cognitive task analysis (pp. 401-415). Mahwah nj: Erlbaum. Argote, L., & McGrath, J. E. (1993). Group processes in organizations: Continuity and change. In C. L. Cooper & I. T. Robertson (Eds.), International review of industrial and organizational psychology (Vol. 8, pp. 333–389). New York: Wiley. Arrow, H., McGrath, J. E., & Berdahl, J. L. (2000). Small groups as complex systems: Formation, coordination, development, and adaptation. Newbury Park ca: Sage. Avolio, B. J., Kahai, S., Dumdum, R., & Sivasubramaniam, N. (2001). Virtual teams: Implications for e-leadership and team development. In M. London (Ed.), How to evaluate others in organizations (337–358). Mahwah nj: Erlbaum. Baldwin, T. T., & Bedell, M. D. (1997). The social fabric of a team-based M.B.A. program: Network effects on student satisfaction and performance. Academy of Management Journal, 6, 1369–1397. Balkundi, P., & Harrison, D. (2004, August). Networks, leaders, teams and time: Connections to viability and performance. Paper presented at the meeting of the Academy of Management, New Orleans. Barrick, M. R., Stewart, G. L., Neubert, M. J., & Mount, M. K. (1998). Relat-

230 modeling complex systems ing member ability and personality to work-team processes and team effectiveness. Journal of Applied Psychology, 83, 377–391. Barry, B., & Stewart, G. L. (1997). Composition, process, and performance in self-managed groups: The role of personality. Journal of Applied Psychology, 1, 62–78. Beck, A. P. (2002). Work team effectiveness, group efﬁcacy, and the role of informal leaders. International Journal of Group Psychotherapy, 2, 311–315. Belbin, R. M. (2000). Beyond the team. Oxford: Butterworth-Heinemann. Bell, B. F., & Kozlowski, S. W. J. (2002). A typology of virtual teams: Implications for effective leadership. Group and Organization Management, 27, 14–49. Beyerlein, M., Johnson, D., & Beyerlein, S. (2003). Team based organizing: Advances in interdisciplinary studies of work teams (Vol. 9). Oxford: Elsevier Science. Bolman, L. G., & Deal, T. E. (1992). What makes a team work? Organizational Dynamics, 1, 34–44. Borman, W. C., & Motowidlo, S. J. (1993). Expanding the criterion domain to include elements of contextual performance. In N. Schmitt & W. C. Borman (Eds.), Personnel selection in organizations (pp. 71–98). San Francisco: Jossey-Bass. Brandes, W., & Weise, P. (1999). Team performance as a constellation of forces: A general model. Kyklos, 52, 573–590. Brannick, M. T., & Prince, C. (1997). An overview of team performance measurement. In M. T. Brannick, E. Salas, & C. Prince (Eds.), Team performance assessment and measurement: Theory, methods, and applications (pp. 331–356). Mahwah nj: Erlbaum. Brodbeck, F., & Greitemeyer, T. (2000). A dynamic model of group performance: Considering the group members’ capacity to learn. Group Processes and Intergroup Relations, 3, 159–182. Bunderson, J. S., & Sutcliffe, K. M. (2002). Comparing alternative conceptualizations of functional diversity in management teams: Process and performance effects. Academy of Management Journal, 45(5), 875–893. Burke, C. S., Stagl, K. C., Klein, C., Goodwin, G. F., Salas, E., & Halpin, S. M. (in press). What type of leadership behaviors are functional in teams? A meta-analysis. Leadership Quarterly. Burke, C. S., Stagl, K. C., Salas, E., Pierce, L., & Kendall, D. (2006). Understanding team adaptation: A conceptual analysis and model. Journal of Applied Psychology, 91(6), 1189–1207. Campbell, J. P. (1990a). Modeling the performance prediction problem in industrial and organizational psychology. In M. D. Dunnette & L. M. Hough (Eds.), Handbook of industrial and organizational psychology (pp. 687–732). Palo Alto ca: Consulting Psychologists Press. Campbell, J. P. (1990b). The role of theory in industrial and organizational psychology. In M. D. Dunnette & L. M. Hough (Eds.), Handbook of industrial and organizational psychology (pp. 39–74). Palo Alto ca: Consulting Psychologists Press.

231 Team Effectiveness in Organizations Campbell, J. P., Dunnette, M. D., Lawler, E., & Weick, K. (1970). Managerial behavior, performance, and effectiveness. New York: McGraw-Hill. Campbell, J. P., & Kuncel, N. R. (2001). Individual and team training. In N. Anderson, D. S. Ones, H. K. Sinangil, & C. Viswesvaran (Eds.), Handbook of industrial, work and organizational psychology: Vol. 1. Personnel psychology (2nd ed., pp. 272–312). London: Sage. Campion, M. A., Medsker, G. J., & Higgs, A. C. (1993). Relations between work group characteristics and effectiveness: Implications for designing effective work groups. Personnel Psychology, 46, 823–850. Campion, M. A., Papper, E. M., & Medsker, G. J. (1996). Relations between work team characteristics and effectiveness: A replication and extension. Personnel Psychology, 49, 429–452. Cannon-Bowers, J. A., Tannenbaum, S. I., Salas, E., & Volpe, C. E. (1995). Deﬁning competencies and establishing team training requirements. In R. A. Guzzo, E. Salas, et al. (Eds.), Team effectiveness and decision making in organizations (pp. 333–380). San Francisco ca: Jossey-Bass. Choi, J. N. (2002). External activities and team effectiveness: Review and theoretical development. Small Group Research, 2, 181–209. Cohen, S. G. (1994). Designing effective self-managing work teams. In M. M. Beyerlein & D. A. Johnson (Eds.), Advances in interdisciplinary studies of work teams: Theories of self-managing work teams (pp. 67–102). Greenwich ct: jai. Cohen, S. G., & Bailey, D. E. (1997). What makes teams work: Group effectiveness research from the shop ﬂoor to the executive suite. Journal of Management, 23, 239–290. Cohen, S. G., Ledford, G. E., Jr., & Spreitzer, G. M. (1996). A predictive model of self-managing work team effectiveness. Human Relations, 49(5), 643– 676. Cook, T. D., & Campbell, D. T. (1979). Quasi-experimentation: Design & analysis for ﬁeld settings. Chicago: Rand McNally. Cooke, N. J., Salas, E., Cannon-Bowers, J. A., & Stout, R. (2000). Measuring team knowledge. Human Factors, 42, 151–173. Cooke, N. J., Salas, E., Kiekel, P. A., & Bell, B. (2004). Advances in measuring team cognition. In E. Salas & S. M. Fiore (Eds.), Team cognition: Understanding the factors that drive process and performance (pp. 83–106). Washington dc: American Psychological Association. Cooper, C. L., & Robertson, I. T. (Eds.). (2004). International Review of Industrial and Organizational Psychology (Vol. 19). New York: Wiley. Cooper, M. A., Shiﬂett, S., Korotkin, A. L., & Fleishman, E. A. (1984). Command and control teams: Techniques for assessing team performance. Bethesda md: arro. Coovert, M. D., & Dorsey, D. W. (2000). Computational modeling with Petri nets: Solutions for individual and team systems. In D. R. Ilgen & C. L. Hulin (Eds.), Computational modeling of behavior in organizations: The third scientiﬁc discipline (pp. 163–181). Washington dc: American Psychological Association.

232 modeling complex systems Cuevas, H. M., Fiore, S. M., Salas, E., & Bowers, C. A. (2004). Virtual teams as sociotechnical systems. In S. H. Godar & S. P. Ferris (Eds.), Virtual and collaborative teams: Process, technologies and practice (pp. 1–19). London: Idea Group. Cummings, T. G. (1978). Self-regulating work groups: A socio-technical synthesis. Academy of Management Review, 3, 625–634. Dailey, R. C. (1980). A path-analysis of R&D team coordination and performance. Decision Sciences, 11, 356–369. Deeter-Schmelz, D. R., Kennedy, K. N., & Ramsey, R. P. (2002). Enriching our understanding of student team effectiveness. Journal of Marketing Education, 24(2), 114–124. de Jong, R. D., Bouhuys, S. A., & Barnhoorn, J. C. (1999). Personality, self-efﬁcacy and functioning in management teams: A contribution to validation. International Journal of Selection and Assessment, 7(1), 46–49. Denison, D. R., Hart, S. L., & Kahn, J. A. (1996). From chimneys to crossfunctional teams: Developing and validating a diagnostic model. Academy of Management Journal, 39(4), 1005–1023. DeSanctis, G., & Poole, M. S. (1994). Capturing the complexity in advanced technology use: Adaptive structuration theory. Organization Science, 5(2), 121–147. Devine, D. J., & Clayton, L. D. (1999). Teams in organizations. Small Group Research, 30, 678–711. Dickinson, T. L., & McIntyre, R. M. (1997). A conceptual framework for teamwork measurement. In M. T. Brannick & E. Salas (Eds.), Team performance assessment and measurement: Theory, methods, and applications (pp. 19–43). Mahwah nj: nea. Doolen, T. L., Hacker, M. E., & Van Aken, E. M. (2003). The impact of organizational context on team effectiveness. ieee Transactions on Engineering Management, 50(3), 285–296. Drexler, A. B., Sibbet, D., & Forrester, R. H. (1988). The team performance model. In W. B. Reddy & K. Jamison (Eds.), Team building: Blueprints for productivity and satisfaction (pp. 45–61). Alexandria va: National Institute for Applied Behavioral Science; San Diego: Pfeiffer. Driskell, J. E., Hogan, R., & Salas, E. (1987). Personality and group performance. Group Processes and Intergroup Relations: Review of Personality and Social Psychology, 9, 91–112. Driskell, J. E., Radtke, P. H., & Salas, E. (2003). Virtual teams: Effects of technological mediation on team processes. Group Dynamics: Theory, Research, and Practice, 7, 297–323. Driskell, J. E., & Salas, E. (1992). Collective behavior and team performance. Human Factors, 34, 277–288. Druskat, V. U., & Kayes, D. C. (1999). The antecedents of team competence: Toward a ﬁne-grained model of self-managing team effectiveness. Research on Managing Groups and Teams, 2, 201–231. Dunnette, M. D. (1963). A note on criterion. Journal of Applied Psychology, 47, 251–254.

233 Team Effectiveness in Organizations Durham, C. C., & Knight, D. (1997). Effects of leader role, team-set goal difﬁculty, efﬁcacy, and tactics on team effectiveness. Organizational Behavior and Human Decision Processes, 72(2), 203–231. Eby, L. T., & Dobbins, G. H. (1997). Collectivistic orientation in teams: An individual and group-level analysis. Journal of Organizational Behavior, 18, 275–295. Edmondson, A. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44, 350–383. Edmondson, A. (2003). Managing the risk of learning: Psychological safety in work teams. In M. A. West, D. Tjosvold, & K. G. Smith (Eds.), International handbook of organizational teamwork and cooperative working (pp. 235–256). Chichester: Wiley. Edmondson, A., Roberto, M. A., & Watkins, M. D. (2003). A dynamic model of top management team effectiveness: Managing unstructured task streams. Leadership Quarterly, 14(3), 297–325. Emery, F. E., & Trist, E. L. (1960). Sociotechnical systems. In C. W. Churchman & M. Verhulst (Eds.), Management Sciences, Models and Techniques (Vol. 2, pp. 83–97). New York: Pergamon. Entin, E. E., & Serfaty, D. (1999). Adaptive team coordination. Human Factors, 41, 312–325. Erez, A., LePine, J. A., & Elms, H. (2002). Effects of rotated leadership and peer evaluation on the functioning and effectiveness of self-managed teams: A quasi-experiment. Personnel Psychology, 55(4), 929–948. Fleishman, E. A., & Zaccaro, S. J. (1992). Toward a taxonomy of team performance functions. In R. W. Swezey & E. Salas (Eds.), Teams: Their training and performance (pp. 31–56). Norwood nj: Ablex. Flemming, J. L., & Monda-Amaya, L. E. (2001). Process variables critical for team effectiveness: A Delphi study of wraparound team members. Remedial and Special Education, 22(3), 158–171. Flood, P. C., Hannan, E., Smith, K. G., Turner, T., West, M. A., & Dawson, J. (2000). Chief executive leadership style, consensus decision making, and top management effectiveness. European Journal of Work and Organizational Psychology, 9(3), 401–420. Furst, S. A., Reeves, M., Rosen, B., & Blackburn, R. S. (2004). Managing the life cycle of virtual teams. Academy of Management Executive, 18(2), 6– 20. Gersick, C. G. (1988). Time and transition in work teams: Toward a new model of group development. Academy of Management Journal, 31, 9–41. Gibson, C. B., Zellmer-Bruhn, M. E., & Schwab, D. P. (2003). Team effectiveness in multinational organizations: Evaluation across contexts. Group and Organizational Management, 28(4), 444–474. Gist, M. E., Locke, E. A., & Taylor, M. S. (1987). Organizational behavior: Group structure, process, and effectiveness. Journal of Management, 13(2), 237–257. Gittell, J. H. (2000). Organizing work to support relational co-ordination. International Journal of Human Resource Management, 11(3), 517–539.

234 modeling complex systems Gladstein, D. L. (1984). Groups in context: A model of task group effectiveness. Administrative Science Quarterly, 29, 499–517. Glickman, A. S., Zimmer, S., Montero, R. C., Guerette, P. J., Campbell, W. J., Morgan, B. B., & Salas, E. (1987). The evolution of teamwork skills: An empirical assessment with implications for training (Tech. Rep. No. 87-016). Orlando ﬂ: Naval Training Systems Center. Goldstein, I. L. (1993). Training in organizations (3rd ed.). Belmont ca: Wadsworth. Goodman, P. S. (1986). Designing effective work groups. San Francisco: JosseyBass. Goodman, P., Ravlin, E., & Schminke, M. (1987). Understanding groups in organizations. In L. L. Cummings & B. M. Staw (Eds.), Research in Organizational Behavior (Vol. 9, pp. 121–173). Greenwich ct: JAI. Goodwin, G. F., O’Shea, P. G., Driskell, J. E., Salas, E., & Ardison, S. (2004). What makes a good team player? Development of a conditional reasoning test of team orientation. In S. Gustafson (Chair), Making conditional reasoning tests work: Reports from the frontier. Symposium conducted at the 19th Annual Conference of the Society for Industrial and Organizational Psychology, Chicago. Gupta, P. P., Umanath, N. S., & Dirsmith, M. W. (1999). Supervision practices and audit effectiveness: An empirical analysis of gao audits. Behavioral Research in Accounting, 11, 27–49. Guzzo, R. A., & Dickson, M. W. (1996). Teams in organizations: Recent research on performance and effectiveness. Annual Review Psychology, 47, 307–338. Guzzo, R. A., & Shea, G. P. (1992). Group performance and intergroup relations in organizations. In M. D. Dunnette & L. M. Hough (Eds.), Handbook of industrial and organizational psychology (2nd ed., pp. 269–313). Palo Alto ca: Consulting Psychologists Press. Hackman, J. R. (1983). A normative model of work team effectiveness (Technical Report No. 2). New Haven ct: Yale School of Organization and Management. Hackman, J. R. (1987). The design of work teams. In J. Lorsch (Ed.), Handbook of organizational behavior (pp. 315–342). Englewood Cliffs nj: Prentice-Hall. Hackman, J. R. (1990). Groups that work (and those that don’t): Creating conditions for effective teamwork. San Francisco: Jossey-Bass. Hackman, J. R. (2002). Leading teams: Setting the stage for great performances. Boston: hbs. Hackman, J. R., & Oldham, G. R. (1980). Work redesign. Reading ma: Addison-Wesley. Hall, V. (2001). Management teams in education: An unequal music. School Leadership and Management, 21(3), 327–341. Haward, R., Amir, Z., Borrill, C., Dawson, J., Scully, J., West, M., & Sainsbury, R. (2003). Breast cancer teams: The impact of constitution, new cancer workload, and methods of operation on their effectiveness. British Journal of Cancer, 89(1), 15–22.

235 Team Effectiveness in Organizations Heinemann, G. D., & Zeiss, A. M. (2002). A model of team performance. In G. D. Heinemann & A. M. Zeiss (Eds.), Team performance in health care: Assessment and development (pp. 29–42). New York: Kluwer Academic/ Plenum. Helmreich, R. L., & Foushee, H. C. (1993). Why crew resource management? Empirical and theoretical bases of human factors training in aviation. In E. L. Wiener, B. G. Kanki, & R. L. Helmreich (Eds.), Cockpit resource management (pp. 3–45). New York: Academic. Herold, D. M. (1978). Improving the performance effectiveness of groups through a task contingent selection of intervention strategies. Academy of Management Review, 3, 315–325. Hertel, G., Konradt, U., & Orlikowski, B. (2004). Managing distance by interdependence: Goal setting, task interdependence, and team-based rewards in virtual teams. European Journal of Work and Organizational Psychology, 13, 1–28. Higgins, S. E., & Routhieaux, R. L. (1999). A multiple-level analysis of hospital team effectiveness. Health Care Supervisor, 17, 1–13. Higgs, M., & Rowland, D. (1992). All pigs are equal? Management Education and Development, 23(4), 349–362. Hinsz, V. B., Tindale, R. S., & Vollrath, D. A. (1997). The emerging conceptualization of groups as information processors. Psychological Bulletin, 121, 43–64. Hoegl, M., Praveen, P., & Gemuenden, H. G. (2003). When teamwork really matters: Task innovativeness as a moderator of the teamwork-performance relationship in software development projects. Journal of Engineering and Technology Management, 20, 281–302. Hogan, R., Raza, S., & Driskell, J. (1988). Personality, team performance, and organizational context. In P. Whitney & R. B. Ochsman (Eds.), Psychology and productivity (pp. 93–103). New York: Plenum. Hollenbeck, J. R., Ilgen, D. R., Sego, D. J., Hedlund, J., Major, D. A., & Phillips, J. (1995). Multilevel theory of team decision making: Decision performance in teams incorporating distributed expertise. Journal of Applied Psychology, 2, 292–316. Hollenbeck, J. R., Moon, H., Ellis, A. P. J., West, B., Ilgen, D., & Sheppard, L. (2002). Structural contingency theory and individual differences: Examination of external and internal person-team ﬁt. Journal of Applied Psychology, 87, 599–606. Hollingshead, A. B., & McGrath, J. E. (1995). Computer-assisted groups: A critical review of the empirical research. In R. A. Guzzo, E. Salas, et al. (Eds.), Team effectiveness and decision making in organizations (pp. 46–78). San Francisco: Jossey-Bass. Huselid, M. A. (1995). The impact of human resource management practices on turnover, productivity, and corporate ﬁnancial performance. Academy of Management Journal, 83(3), 635–672. Ilgen, D. R. (1999). Teams embedded in organizations. American Psychologist, 54, 129–139.

236 modeling complex systems Ilgen, D. R., Hollenbeck, J. R., Johnson, M., & Jundt, D. (in press). Teams in organizations: From i-p-o models to imoi models. Annual Review of Psychology. Ilgen, D. R., Major, D. A., Hollenbeck, J. R., & Sego, D. J. (1993). Team research in the 1990s. In M. M. Chemers & R. Ayman (Eds.), Leadership theory and research: Perspectives and directions (pp. 245–270). San Diego: Academic. Jackson, S. E., & Ruderman, M. N. (1996). Diversity in work teams: Research paradigms for a changing workplace. Washington dc: American Psychological Association. Janz, B. D., Colquitt, J. A., & Noe, R. A. (1997). Knowledge worker team effectiveness: The role of autonomy, interdependence, team development, and contextual support variables. Personnel Psychology, 50, 877–904. Jung, D. I., Sosik, J. J., & Baik, K. B. (2002). Investigating work group characteristics and performance over time: A replication and cross-cultural extension. Group Dynamics, 6, 153–171. Kahai, S. S., Sosik, J. J., & Avolio, B. J. (1997). Effects of leadership style and problem structures on work group process and outcomes in an electronic meeting system environment. Personnel Psychology, 50, 121–125. Karau, S. J., & Kelly, J. R. (1992). The effects of time scarcity and time abundance on group performance quality and interaction process. Journal of Experimental Social Psychology, 6, 542–571. Katz, D., & Kahn, R. L. (1978). The social psychology of organizations (2nd ed.). New York: Wiley. Kelley, E. (2001). Keys to effective virtual global teams. Academy of Management Executive, 15(2), 132–133. Kirkman, B. L., & Shapiro, D. L. (1997). The impact of cultural values on employee resistance to teams: Toward a model of globalized self-managing work team effectiveness. Academy of Management Review, 22(3), 730–757. Kirkman, B. L., Tesluk, P. E., & Rosen, B. (2004). The impact of demographic heterogeneity and team leader–team member demographic ﬁt on team empowerment and effectiveness. Group and Organizational Management, 29(3), 334–368. Klein, G. (1993). A recognition primed decision (RPD) model of rapid decision making. In G. Klein, J. Orasanu, R. Calderwood, & C. E. Zsambok (Eds.), Decision making in action (pp. 138–147). Norwood nj: Ablex. Klimoski, R., & Jones, R. G. (1995). Stafﬁng for effective group decision making: Key issues in matching people and teams. In R. A. Guzzo & E. Salas (Eds.), Team effectiveness and decision making in organizations (pp. 291–332). San Francisco: Jossey-Bass. Kline, T. J. B. (2001). Predicting team performance: Testing a model in a ﬁeld setting. Journal of Specialists in Group Work, 26(2), 185–197. Kline, T. J. B., & MacLeod, M. (1996). Team effectiveness: Contributors and hindrances. Human Systems Management, 15(3), 183–186. Kolodny, H. F., & Kiggundu, M. N. (1980). Towards the development of a

237 Team Effectiveness in Organizations systems model in woodlands mechanical harvesting. Human Relations, 33, 623–645. Kozlowski, S. W. J. (1998). Training and developing adaptive teams: Theory, principles, and research. In J. A. Cannon-Bowers & E. Salas (Eds.), Making decisions under stress: Implications for individual and team training (pp. 15–153). Washington dc: American Psychological Association. Kozlowski, S. W. J., & Bell, B. S. (2003). Work groups and teams in organizations. In W. C. Borman, D. R. Ilgen, & R. J. Klirnoski (Eds.), Handbook of psychology: Industrial and organizational psychology (Vol. 12, pp. 333–375). London: Wiley. Kozlowski, S. W. J., Brown, K., Weissbein, D., Cannon-Bowers, J., & Salas, E. (2000). A multilevel approach to training effectiveness: Enhancing horizontal and vertical transfer. In K. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research and methods in organization (pp. 157–210). San Francisco: Jossey-Bass. Kozlowski, S. W. J., Gully, S. M., Nason, E. R., & Smith, E. M. (1999). Developing adaptive teams: A theory of compilation and performance across levels and time. In D. R. Ilgen & E. D. Pulakos (Eds.), The changing nature of work and performance: Implications for stafﬁng, personnel actions, and development (pp. 240–292). San Francisco: Jossey-Bass. Kozlowski, S. W. J., & Klein, K. (2000). A multilevel approach to theory and research in organizations: Contextual, temporal, and emergent processes. In K. J. Klein (Ed.), Multilevel theory, research, and methods in organizations: Foundations, extensions, and new directions (pp. 3–90). San Francisco: Jossey-Bass. Krackhardt, D., & Brass, D. J. (1994). Intraorganizational networks: The micro side. In S. Wasserman & J. Galaskiewicz (Eds.), Advances in social network analysis: Research in the social and behavioral sciences (pp. 207–229). Thousand Oaks ca: Sage. Kristof, A. L., Brown, K. G., Sims, H. P., & Smith, K. A. (1995). The virtual team: A case study and inductive model. In M. M. Beyerlein, D. A. Johnson, & S. T. Beyerlein (Eds.), Advances in interdisciplinary studies of work teams (Vol. 2, pp. 229–253). Greenwich ct: jai. Kuo, C. (2004). Research on impacts of team leadership on team effectiveness. Journal of American Academy of Business, Cambridge, 5, 266–277. Leon, G. R., List, N., & Magor, G. (2004). Personal experiences and team effectiveness during a commemorative trek in the high arctic. Environment and Behavior, 36, 386–401. LePine, J. A. (2003). Team adaptation and postchange performance: Effects of team composition in terms of members’ cognitive ability and personality. Journal of Applied Psychology, 88, 27–39. Lewin, K. (1951). Field theory in social science. New York: Harper & Row. Losada, M. (1999). The complex dynamics of high performing teams. Mathematics and Computational Modeling, 30, 179–192. MacMillan, J., Entin, E. E., Entin, E. B., & Serfaty, D. (1994). Structuring and

238 modeling complex systems training high-reliability teams (Technical Report No. 632). Burlington ma: Alphatech. Marks, M. A., Mathieu, J. E., & Zaccaro, S. J. (2001). A temporally based framework and taxonomy of team processes. Academy of Management Review, 26(3), 356–376. Marks, M. A., Zacarro, S. J., & Mathieu, J. E. (2000). Performance implications of leader brieﬁngs and team-interaction training for team adaptation to novel environments. Journal of Applied Psychology, 6, 971–986. Mathieu, J. E., Goodwin, G. F., Heffner, T. S., Salas, E., & Cannon-Bowers, J. A. (2000). The inﬂuence of shared mental models on team process and performance. Journal of Applied Psychology, 85, 273–283. McGrath, J. E. (1964). Social psychology: A brief introduction. New York: Holt, Rinehart & Winston. McGrath, J. E. (1984). Groups: Interaction and performance. Englewood Cliffs nj: Prentice-Hall. McGrath, J. E. (1991). Time, interaction, and performance (TIP). Small Group Research, 22(2), 147–174. McGrath, J. E. (1997). Small group research, that once and future ﬁeld: An interpretation of the past with an eye to the future. Group Dynamics: Theory, Research, and Practice, 1, 7–27. McGrath, J. E., Arrow, H., & Berdahl, J. L. (2000a). The study of groups: Past, present and future. Personality and Social Psychology Review, 1, 95–106. McGrath, J. E., Arrow, H., & Berdahl, J. L. (2000b). A theory of groups as complex, adaptive, dynamic systems. Newbury Park ca: jai. McGrew, J. F., Bilotta, J. G., & Deeney, J. M. (1999). Software team formation and decay: Extending the standard model for small groups. Small Group Research, 30, 209–234. McIntyre, R. M., & Salas, E. (1995). Measuring and managing for team performance: Emerging principles from complex environments. In R. Guzzo, E. Salas, et al. (Eds.), Team effectiveness and decision making in organizations (pp. 149–203). San Francisco: Jossey-Bass. Millitello, L. G., Kyne, M. M., Klein, G., Getchell, K., & Thordsen, M. (1999). A synthesized model of team performance. International Journal of Cognitive Ergonomics, 3(2), 131–158. Mischel, L. J., & Northcraft, G. B. (1997). “I think we can, I think we can . . .”: The role of efﬁcacy beliefs in group and team effectiveness. In B. Markovsky, M. J. Lovaglia, & E. J. Lawler (Eds.), Advances in group processes (Vol. 14, pp. 177–197). Greenwich ct: jai. Morgan, C. P., Dailey, R. C., & Kulisch, W. A. (1976). The composition of effectiveness and ineffectiveness in R&D project teams. Academy of Management Proceedings, 19, 500–504. Morgan, B. B., Jr., Salas, E., & Glickman, A. S. (1994). An analysis of team evolution and maturation. Journal of General Psychology, 120(3), 277–291. Mullen, B., & Cooper, C. L. (1994). The relation between group cohesion and performance: An integration. Psychological Bulletin, 115, 210–227.

239 Team Effectiveness in Organizations Navarro, J. J. (1994). Computer supported self-managing teams. Journal of Organizational Computing and Electronic Commerce, 4, 317–342. Neal, A., & Hesketh, B. (2002). Productivity in organizations. In N. Anderson, D. S. Ones, H. K. Sinangil, & C. Viswesvaran (Eds.), Handbook of industrial, work and organizational psychology: Vol. 2. Organizational psychology (2nd ed., pp. 7–24). London: Sage. Neck, C. P., Connerley, M. L., & Manz, C. C. (1997). Toward a continuum of self-managing team development. In M. M. Beyerlein, D. A. Johnson, & S. T. Beyerlein (Eds.), Advances in interdisciplinary studies of work teams (pp. 193–216). Greenwich ct: jai. Neuman, G. A., & Wright, J. (1999). Team effectiveness: Beyond skills and cognitive ability. Journal of Applied Psychology, 84, 376–389. Nieva, V., Fleishman, E. A., & Reick, A. (1978). Team dimensions: Their identity, their measurement, and their relationships (Contract No. dahc 19-78c-0001). Washington dc: Response Analysis Corp. Oetzel, J. G. (1999). Developing a theoretical model of intercultural small groups: Understanding the effects of culture and cultural diversity on work group processes and outcomes. Psychological Bulletin, 41, 252– 271. Orasanu, J., & Salas, E. (1993). Team decision making in complex environments. In G. Klein, J. Orasanu, R. Calderwood, & C. E. Zsambok (Eds.), Decision making in action: Models and methods (pp. 327–345). Norwood nj: Ablex. Pagell, M., & LePine, J. A. (2002). Multiple case studies of team effectiveness in manufacturing organizations. Journal of Operations Management, 20(5), 619–639. Pearce, J. A., & Ravlin, E. C. (1987). The design and activation of self-regulating work groups. Human Relations, 40, 751–782. Ployhart, R. E., & Schneider, B. (2005). Multilevel selection and prediction: Theories, methods, and models. In A. Evers, O. Smit-Voskuyl, & N. Anderson (Eds.), Handbook of personnel selection (pp. 495–516). Oxford: Blackwell. Poulton, B. C., & West, M. A. (1993). Effective multidisciplinary teamwork in primary health care. Journal of Advanced Nursing, 18, 918–925. Poulton, B. C., & West, M. A. (1999). The determinants of effectiveness in primary health care teams. Journal of Interprofessional Care, 13, 7–18. Priest, H. A., Stagl, K. C., Klein, C., & Salas, E. (2006). Virtual teams: Creating context for distributed teamwork. In C. A. Bowers, E. Salas, & F. Jentsch (Eds.), Making effective work teams with people, machines, and networks (pp. 185–212). Washington dc: American Psychological Association. Rangarajan, D., Chonko, L. W., Jones, E., & Roberts, J. A. (2004). Organizational variables, sales force perceptions of readiness for change, learning, and performance among boundary-spanning teams: A conceptual framework and propositions for research. Industrial Marketing Management, 33, 289–305. Rentsch, J. R., & Hall, R. J. (1994). Members of great teams think alike: A

240 modeling complex systems model of team effectiveness and schema similarity among team members. In M. M. Beyerlein & D. A. Johnson (Eds.), Advances in interdisciplinary studies of work teams: Theories of self-managing work teams (Vol. 1, pp. 223–261). Greenwich ct: jai. Robertson, M. M., Maynard, W. S., Huang, Y. H., & McDevitt, J. (2002). Telecommuting: An overview of emerging macroergonomic issues. Paper presented at the 46th annual meeting of the Human Factors and Ergonomics Society, Baltimore. Rouse, W. B., & Rouse, R. K. (2004). Teamwork in the performing arts. Proceedings of the ieee, 92, 606–615. Ruel, H. J. M. (2000). Reconsidering our team effectiveness models: A call for an integrative paradigm. In M. M. Beyerlein, D. A. Johnson, & S. T. Beyerlein (Eds.), Advances in interdisciplinary studies of work teams: Team development (pp. 173–186). Amsterdam: jai. Salas, E., Burke, C. S., & Cannon-Bowers, J. A. (2000). Teamwork: Emerging principles. International Journal of Management Reviews, 4, 339–356. Salas, E., Burke, C. S., & Stagl, K. C. (2004). Developing teams and team leaders: Strategies and principles. In D. Day, S. J. Zaccaro, & S. M. Halpin (Eds.), Leader development for transforming organizations (pp. 325–355). Mahwah nj: Erlbaum. Salas, E., & Cannon-Bowers, J. A. (2000). Teams in organizations: Lessons from history. In M. M. Beyerlein (Ed.), Work teams: Past, present and future (pp. 323–331). Norwall ma: Kluwer Academic. Salas, E., Dickinson, T. L., Converse, S. A., & Tannenbaum, S. I. (1992). Toward an understanding of team performance and training. In R. W. Swezey & E. Salas (Eds.), Teams: Their training and performance (pp. 3– 29). Norwood nj: Ablex. Salas, E., & Fiore, S. M. (2004). Team cognition: Understanding the factors that drive process and performance. Washington dc: American Psychological Association. Salas, E., Sims, D. E., & Burke, C. S. (2005). Is there a “Big Five” in teamwork? Small Group Research, 36(5), 555–599. Salas, E., Sims, D. E., & Klein, C. (2004). Cooperation and teamwork at work. In C. D. Speilberger (Ed.), Encyclopedia of applied psychology (Vol. 1, pp. 497–505). San Diego: Academic. Salas, E., Stagl, K. C., & Burke, C. S. (2004). 25 years of team effectiveness in organizations: Research themes and emerging needs. In C. L. Cooper & I. T. Robertson (Eds.), International Review of Industrial and Organizational Psychology (Vol. 19, pp. 47–91). New York: Wiley. Schippers, M. D., Den Hartog, D. N., Koopman, P. L., & Wienk, J. A. (2003). Reﬂexivity and diversity in teams: The moderating effects of outcome interdependence and group longevity. Journal of Organizational Behavior, 24, 779–802. Schneider, B. (1987). The people make the place. Personnel Psychology, 40, 437–453. Schneider, B., Smith, D., & Sipe, W. P. (2000). Personnel selection psychol-

241 Team Effectiveness in Organizations ogy: Multilevel considerations. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research, and methods in organizations: Foundations, extensions, and new directions (pp. 3–90). San Francisco: Jossey-Bass. Senge, P. M. (1990). The ﬁfth discipline: The art and practice of the learning organization. New York: Doubleday. Shea, G. P., & Guzzo, R. A. (1987). Groups as human resources. In K. M. Rowland & G. R. Ferris (Eds.), Research in personnel and human resource management (Vol. 5, pp. 323–356). Greenwich ct: jai. Sheard, A. G., & Kakabadse, A. P. (2001). From loose groups to effective teams: The nine key factors of the team landscape. Journal of Management Development, 21(2), 133–151. Sheehan, A. L., & Martin, R. (2003). Understanding diversity to maximize work team effectiveness: Field studies designed to unravel the complex relationship between diversity and team effectiveness. Australian Journal of Psychology, 55, 144. Shiﬂett, S., Eisner, E. J., Price, S. J., & Schemmer, M. F. (1985). The deﬁnition and measurement of small military unit team functions (U.S. Army Research Institute for the Behavioral and Social Sciences Report). Fort Benning ga Fort Benning Field Unit. Smith-Jentsch, K. A., Zeisig, R. L., Acton, B., & McPherson, J. A. (1998). Team dimensional training: A strategy for guided self-correction. In J. A. Cannon-Bowers & E. Salas (Eds.), Making decisions under stress: Implications for individual and team training (pp. 271–297). Washington dc: American Psychological Association. Sonnentag, S., Frese, M., Brodbeck, F. C., & Heinbokel, T. (1997). Use of design methods, team leaders’ goal orientation, and team effectiveness: A follow-up study in software development projects. International Journal of Human-Computer Interaction, 9, 443–454. Spreitzer, G. M., Cohen, S. G., & Ledford, G. E. (1999). Developing effective self-managing work teams in service organizations. Group and Organizational Management, 24, 340–366. Spreitzer, G. M., Noble, D. S., Mishra, A. K., & Cooke, W. N. (1999). Predicting process improvement team performance in an automotive ﬁrm: Explicating the roles of trust and empowerment. In M. A. Neale, E. A. Mannix, & R. Wageman (Eds.), Research on managing groups and teams: Groups in context (Vol. 2, pp. 71–92). Greenwich ct: jai. Stagl, K. C., Salas, E., & Burke, C. S. (in press). Best practices in team leadership: What team leaders do to facilitate team effectiveness. In J. A. Conger & R. E. Riggio (Eds.), The practice of leadership (pp. 172–198). New York: Wiley. Steiner, I. D. (1972). Group processes and productivity. Orlando ﬂ: Academic. Stevens, M. J., & Campion, M. A. (1994). The knowledge, skill, and ability requirements for teamwork: Implications for human resource management. Journal of Management, 20, 503–530. Stevens, M. J., & Campion, M. A. (1999). Stafﬁng work teams: Development and validation of a selection test for teamwork settings. Journal of Management, 25, 207–228.

242 modeling complex systems Stewart, G. L., & Barrick, M. R. (2000). Work team structure and performance: Assessing the mediating role of intrateam process and the moderating role of task type. Academy of Management Journal, 43, 135–148. Stoker, J. I., & Remdisch, S. (1997). Leading work teams: Directions for team effectiveness. In M. M. Beyerlein & D. A. Johnson (Eds.), Advances in interdisciplinary studies of work teams (Vol. 4, pp. 79–96). Greenwich ct: jai. Strasser, D. C., & Falconer, J. A. (1997). Linking treatment to outcomes through teams: Building a conceptual model of rehabilitation effectiveness. Topics in Stroke Rehabilitation, 4(18), 15–27. Sundstrom, E., & Altman, I. (1989). Physical environments and work group effectiveness. Research in Organizational Behavior, 11, 175–209. Sundstrom, E., DeMeuse, K. P., & Futrell, D. (1990). Work teams: Applications and effectiveness. American Psychologist, 45, 120–133. Susman, G. I., & Ray, J. M. (1999). Test of a model of organizational contributors to product development team effectiveness. Journal of Engineering and Technology Management, 16, 223–245. Tambe, M. (1996, July). Teamwork in real-world dynamic environments. Paper presented at aaai Fall Symposium on Plan Execution. Tannenbaum, S. I., Beard, R. L., & Salas, E. (1992). Team building and its inﬂuence on team effectiveness: An examination of conceptual and empirical developments. In K. Kelley (Ed.), Issues, theory, and research in industrial/organizational psychology (pp. 117–153). Amsterdam: Elsevier. Tata, J., & Prasad, S. (2004). Team self-management, organizational structure, and judgments of team effectiveness. Journal of Managerial Issues, 16, 248–265. Tjosvold, D., Wong, A., Nibler, R., & Pounder, J. S. (2002). Teamwork and controversy in undergraduate management courses in Hong Kong: Can the method reinforce the message? Swiss Journal of Psychology, 61, 131–138. Tompkins, T. C. (1997). A developmental approach to organizational learning teams: A model and illustrative research. In M. Beyerlein, D. A. Johnson, & S. T. Beyerlein (Eds.), Advances in interdisciplinary studies of work teams (pp. 281–302). jai. Tubbs, S. L. (1994). The historical roots of self-managing work teams in the twentieth century: An annotated bibliography. Advances in Interdisciplinary Studies of Work Teams, 1, 39–66. Tuckman, B. W. (1964). Personality, structure, group composition and group functioning. Sociometry, 27, 469–487. Tuckman, B. W. (1965). Developmental sequence in small groups. Psychological Bulletin, 63, 384–399. von Bertalanffy, L. (1956). General systems. Yearbook of the Society for Advancement of General Systems Theory, 1, 1–10. Walker, D. H. T., & Vines, M. W. (2000). Australian multi-unit residential project construction time performance factors. Engineering Construction and Architectural Management, 7(3), 278–284.

243 Team Effectiveness in Organizations Werner, J. M., & Lester, S. W. (2001). Applying a team effectiveness framework to the performance of student case teams. Human Resource Development Quarterly, 12(4), 385–402. West, M. A., Borrill, C. S., & Unsworth, K. L. (1998). Team effectiveness in organizations. In C. L. Cooper & I. T. Robertson (Eds.), International Review of Industrial and Organizational Psychology (Vol. 13, pp. 1–48). Chichester: Wiley. West, M. A., Tjosvold, D., & Smith, K. G. (Eds.). (2003). International handbook of organizational teamwork and cooperative working. Chichester: Wiley. Yeatts, D. E., & Hyten, C. (1998). High-performing self-managed work teams: A comparison of theory to practice. Thousand Oaks ca: Sage. Yetton, P. W., & Bottger, P. C. (1982). Individual versus group problem solving: An empirical test of a best-member strategy. Organizational Behavior and Human Performance, 29, 307–321.

Constructive Complexity and Human Change Processes Michael J. Mahoney University of North Texas

In the early 1970s, when clinical psychology was undergoing its cognitive revolution, a young psychologist named Michael Mahoney became a voice for paradigmatic change. One of this volume’s editors, then a graduate student, attended a debate between Mahoney and some of the old-guard ﬁgures of behaviorism. With a style and wit that became legendary, Mahoney showed a homemade ﬁlm to illustrate his position. The ﬁrst scene opened on a bleary-eyed graduate student, working at a typewriter, who paused to stagger to the refrigerator and quaff a soda—uniquely human self-reinforcement? The scene faded to a dog pawing a typewriter. Suddenly the dog stopped, and the camera followed him to the refrigerator, which he opened to lap up one of several puppy treats. He then returned to his task at the typewriter. This was repeated until all the treats had been consumed. As the scene faded, the camera panned to the typewriter to reveal what the dog had been writing: “I type, therefore I think I am.” For the next 30 years, Mahoney delighted the psychology community with such demonstrations and parables. He became a leading ﬁgure in the constructivist movement, a later expression of his early inclinations toward integrative and holistic thinking. His work on chaos theory and related ideas in development and holistic psychotherapy led to his participation in this volume of the Nebraska Symposium on Motivation. A broad scholarly community was deeply saddened when he passed away on May 31, 2006. This volume of the Symposium is dedicated to his memory.

Change and stability are the twin engines of human consciousness. The story gets complex, but so, too, do humans. That story has fascinated many over the course of human history and the short span of our individual lives. Personally, human change processes have long been at the heart of my passions as a scientist. How do people change? What is it that changes when a person changes? Why do some of us seem to change more easily than others? Why is change so difﬁcult for so many? And, ultimately, what should we do?

246 modeling complex systems I shall not presume to offer more than reverent gestures in the direction of responses to such questions. Before offering those gestures, however, let me explain the nontechnical tone of what follows. The growing interest in complexity is reassuring to those of us who have been following its scientiﬁc scent for some time. A common comment made by colleagues, students, and practitioners, however, is that the literature on complexity is too complex. They ﬁnd it difﬁcult to understand. Some of the difﬁculty may have to do with the subject matter, but some of it is also due to the use of technical terms that are unfamiliar. My goal is to use language and concepts that are familiar and accessible. This means, of course, that I will be simplifying, and—as Alfred North Whitehead put it—all simpliﬁcation is oversimpliﬁcation. Technicalities of theory and research evidence can be found among the resources cited in the reference list, a list that has been considerably winnowed. To better serve those whose primary interests may be more practical than theoretical or evidential, I will conclude with a brief discussion of how appreciations of dynamic complexity might inform professional life counseling. Let me begin, however, at the beginning.

Being Boundaries and Knowing: A Brief History It is hardly coincidental that ancient Greek philosophy emerged from a sense of awe regarding the phenomenon of change. Why that sense of awe transformed itself into formal philosophical reﬂection remains unknown. But Aristotle asserts that the original motivation for philosophy was awe. Thales and his followers wondered about a central paradox: How is it that everything changes and yet there is permanence? What is the relation between change and stability? When water changes its form—from ice to rain to steam—what is it within or behind these changing forms that remains the same? An inﬂuential answer was later offered by a young man nicknamed “Broad Shoulders.” We call him Plato. Plato suggested that there was a realm of perfect, permanent, and changeless forms in a world that lay beyond our senses. Pythagoras had anticipated this idea and had suggested that the ultimate language of such a realm would be mathematical. The ultimate truth about reality would be expressed in numbers and their ratios. It is the ratio, incidentally, that serves as

247 Complexity and Human Change Processes the base concept of Western rationality. Thanks to Pythagoras and Plato, Western civilization has been steeped in three assumptions: that stability (order) is more permanent, real, and beautiful than change (disorder); that ideas, numbers, and the intellect are better sources of knowing than experiences, sensations, and the body; and that the mind and the body are separate. These three assumptions are being challenged by complexity studies.

The Evolution of Planetary Life: Edges, Centers, and Cycles of Exchange Suppose one were asked to list the four greatest leaps in the history of life on our planet. First came boundaries. Out of the chaos of primordial soup, something we now call a cell membrane began to form. It was literally the archetypal contrast. What do boundaries do? They separate. Chemical concentrations become different on one side of a membrane than on the other. Boundaries separate, but they also serve to connect. The insides of a cell become more connected. Over time, increasing connection can itself promote an expression of inner order. Some cell parts became specialized by serving various life-support functions. Other parts of the inside begin to serve the coordination of the other parts. We like to call these central ordering processes nuclear. After the ﬁrst edges (boundaries), there came the ﬁrst centers. Nucleation was the second major leap in life on earth. It had some disadvantages, but, by and large, nucleated cells (eukaryotes) were able to adapt themselves to more planetary environments than their less-centered peers (prokaryotes). Coordination and complexity became both more common and more demanding. For a very long time, growth was achieved by means of two kinds: enlargement (hypertrophy) and division (hyperplasticity). Cells enlarged to the limits of their membranes and then divided into new cells. This continued as the only form of reproduction until there was sex. Sex was the third major leap of life on the planet. Sexual reproduction is credited with two contributions to life: diversiﬁcation and death. Life-forms that reproduce asexually do not die. They also do not diversify. They simply replicate themselves through endless divisions. Sexual reproduction involves the exchange of nuclear material. It introduced novel combinations of the centers of living creatures.

248 modeling complex systems The fourth and most recent leap made by life on earth has been the creation and exchange of symbols. Instead of exchanging only their nuclear material, symbol-using creatures exchange information and expressions of their own experiencing. Symbol use occurs only among a few life-forms, all of whom are social. Sociality and symbolic processes seem to be intimately connected. All that we think of as modern is an expression of our accelerating exchange of symbols (e.g., civilized community, arts, science, technology). To summarize this brief history of biological evolution, life as we know it originated in contrasts. Boundaries continue to serve critically important roles in our lives. So do centers. We are life-forms with a legacy of coordinated centering, and we continue to seek and create order in our lives. We diversify and exchange, and it is through our diversity and connectedness that we continue to develop. We become complex, and we change at accelerated rates.

Change and Complexity Those who seek the origins of complexity studies often ﬁnd themselves chasing rabbits that appear and disappear across eras, continents, and cultures. In Western civilization, it is apparent that rates of change in everyday worlds increased sharply during and after the European Renaissance. An important part of what emerged out of that era was the scientiﬁc revolution. The essence of science is connected knowing—a term recently popularized by feminist scholars. Science progresses by means of well-connected networks of clever and careful seekers. The discoveries of Copernicus and Galileo emerged out of communities of inquiry that were becoming increasingly connected. Francis Bacon’s valuable contribution was to organize the support and encouragement of such communities. But the real godfather of change was Charles Darwin. Again reﬂecting the importance of connections, Darwin’s circle of acquaintances encouraged him toward conjectures about change. Archaeology and geology were still infant disciplines. But, if the earth itself showed evidence of change, was it not possible that life itself had changed? And, if life-forms had changed, might not there have been changes in the consciousness that had emerged at the primate end of the spectrum? Darwin’s “dangerous idea” has inﬂuenced much of what has followed.

249 Complexity and Human Change Processes Darwin’s two greatest contributions were his emphasis on mutability (change) and emotionality. In mammals—and particularly in social primates like us—these dimensions are related. Motivation and emotion are based in movement. The words themselves reﬂect this legacy. We are moved to change—put into literal motion—by a complex interaction of processes. Our emotions reﬂect our embodiment, and this realization has resulted in the erosion of mind-body dualism. The “new look” in the cognitive sciences and psychotherapy is one that vigorously embraces embodiment and emotionality. Body and mind can no longer be meaningfully separated. Reason and experience need not be cast only in terms of conﬂict. And order and chaos are generative tensions, not separate species. Before elaborating much more of this story, it might be wise to recall an aphorism attributed to Arthur Schopenhauer. He suggested that there are two kinds of people in the world: those who believe that there are two kinds of people in the world and those who do not. This means, among other things, that we are categorical beings. We tend to classify. Just look at psychology. Why do we classify? Because it is a way of creating order out of chaos. Aristotle’s earliest work was on categories. He spent much of his life classifying (things, beings, areas of study, etc.). He formalized logic, and we use his logic to classify and organize our lives. How do we classify? By creating contrasts. Contrasts and the tensions they reﬂect or create are also at the heart of our planetary life.

Constructivism: A New Expression and Ancient Wisdom Let us now move closer to the issue of how we are structured as human beings and how we change.1 This shift is made easier by a conceptual revolution in which we are now living. Some call it chaos. It certainly does embrace and elevate the role of disorder in development. Others call it complexity studies or dynamic systems theories. It is those as well. An appreciation of complexity and the dynamics of developing systems is now sweeping through many of the sciences and large sectors of the humanities. That appreciation is drawing particular attention in biology, psychology, and neuroscience. But it is not simply a new kid on the block. It is in many ways a new expression of ideas that have been cycling through multiple cultures for

250 modeling complex systems millennia. A popular term for connecting these perspectives is constructivism. In the interest of focus, I shall forgo a digression into its rich legacy in Eastern and Western philosophy and science. In terms of name croppings, however, the following individuals have played roles in the history of constructivism: Lao-Tzu, Buddha, Heraclitus, Giambattista Vico, Immanuel Kant, Arthur Schopenhauer, Johann Herbart, William James, Hans Vahinger, Lev Vygotsky, Alfred Adler, Jean Piaget, Viktor Frankl, Friedrich Hayek, George Kelly, Jerome Bruner, and Albert Bandura. What is constructivism? Succinctly, constructivism is a term now commonly applied to a family of theories—a metatheory—that share a view of human beings as active, complex, and connected lifelong learners. Throughout a fascinating diversity of expression, constructive theorists share ﬁve basic themes: (1) the active agency of the human organism; (2) the importance of order or meaning in human experience; (3) the centrality of an embodied identity—often called a self—in the organization of experiencing; (4) the relational matrix of experiencing, most apparent in our immersion in social and symbolic networks; and (5) lifelong developmental unfolding. With such a breadth of themes, it is, perhaps, not surprising to ﬁnd that former theoretical factions have encountered one another on common conceptual ground. Thus, representatives of cognitive, behavioral, existential-humanistic, and psychodynamic perspectives have all used the term constructivist to describe some of their most recent developments. Given this remarkable conceptual concordance, a closer examination of the ﬁve themes is warranted.

active agency The active agency of the person is a key feature of constructivism. Humans are not simply reactive to the forces of their environments. We are proactive as well. We choose. We change our environments in ways that change us. Bandura has termed it reciprocal determinism. The emphasis is on reciprocity and the efﬁcacy of an agent who is participating in life by inﬂuencing the world by which it is being inﬂuenced. This consideration of the person as an active agent is an important contrast to former portrayals that viewed humans as passive pawns in a mechanical universe.

251 Complexity and Human Change Processes

ordering processes The second theme of constructivism is that much of human activity is devoted to organizing experience. We are all order freaks, some more so than others. We seek to ﬁnd, establish, and maintain a stable order in and from which to live our lives. Recurring patterns in nature have helped. Within their ﬁrst year of life, for example, human infants organize their cycles of sleeping and being awake in congruence with cycles of light and dark (day and night). Biological self-organization is a demanding task of infancy. One of the most powerful expressions of this organization is apparent in early emotional development. Emotions are expressions of life organization that made their ﬁrst appearance with the emergence of mammals about 165 million years ago. Among other things, emotions have contributed to our ﬂexibility in adapting to new challenges and in deepening our capacities to relate to one another. Parental care of offspring is a common example. Most relevant here is the role of emotions in organizing our styles of being. As extensive evidence now reﬂects, what we call personality and personality traits are fundamentally emotional patterns—stylized expressions of our literally feeling our way through life.

identity: the mysteries of a changing stable self The third theme of constructivism is connected to individuality in these expressions. Although there are debates about whether there is a stable psychological entity that could be called the self, the real self, or a true self, there is little debate about the body being the base from which a person experiences. Personal identity is embodied and emotional. It is not coincidental that one of the pioneering classics in the women’s rights movement was titled Our Bodies, Ourselves. We experience from a reference point—an embodied perspective or positionality that is paradoxically stable and yet ever changing throughout life. The paradox of the changing yet stabilizing self is the focus of considerable study. What is clear—and a point that is emphasized in constructivist writings—is that each person experiences his or her world in a unique manner. Note the practical and ethical implications for respecting individual phenomenology.

252 modeling complex systems

relatedness: social symbolic processes But individual selves cannot be separated from their support systems. This is the essence of the fourth theme of constructivism. Not only are we biologically dependent on our early caregivers; we are psychologically relational by our very nature. We live in and from relationships. This is a lifelong fact. Even if we were to live in the isolation of a cave on a remote mountain, we would be living in and from symbolic relationships. Symbols share features that include the re-presentation of that which is currently absent. Skills in the use of symbols have helped us learn from and teach toward generations we can barely imagine. We live in multidimensional webs of relationships that are woven in large part by our symbolic processes. Our inner life—or at least the small part of which we become conscious—is saturated with symbols. Our thinking is not just “in our heads,” however. All of the so-called higher cognitive processes—including the symbolic—are founded in our emotional, relational embodiment.

lifelong development: idiodynamics The complexity of being human is itself awesome. Our warrant for such awe is further ampliﬁed by the ﬁfth and ﬁnal theme of constructivism. In technical terms it is dynamic dialectical development. In more accessible terms, it refers to the fact that we continue to unfold over the course of our lives. We continue to change, whether we want to or not. Indeed, we often change most in the midst of desperate attempts to avoid changing. And this is where the ﬁve themes come full circle. Instead of thinking of them as a linear sequence of stages, imagine them as points on the face of a clock. Activity takes the form of order seeking. Order begins at home, in a base camp of biological embodiment. In humans, a sense of self develops out of formative emotional relationships and expressive symbolic skills. But life keeps coming at us. Situations change. Relationships change. Our bodies change. Our worlds change. We are often thrown off balance. Our life order is challenged. Again and again we make new gestures toward regaining our equilibrium as a dynamic living system. Again and again—for a lifetime—we seek a stability that we never quite achieve.

253 Complexity and Human Change Processes There is an ironic paradox in human complexity. We long for order, yet we live—literally and ﬁguratively—in the essential tensions that mark the edges of chaos. Complex-systems scientists tell us that this is a necessary fact of our existence. This existential fact needs to be addressed with practical compassion. What does all this chaos and complexity mean for us in our everyday lives? Are we doomed to live hypertense, off-balance lives? I do not believe this to be the case. Permit me to translate my limited understanding of these matters in the direction of some practical suggestions for how we counsel professionally.

Human Change Processes How we change is intimately related to how we stay the same (Appendix A). Biologically and psychologically, we are relatively conservative creatures. We tend to change as little as possible to get by. When given a choice, we prefer to change smaller and more superﬁcial dimensions (e.g., clothes, hair, vehicles). But we are sometimes forced toward larger and more fundamental changes. Some of these change-requiring circumstances may happen suddenly—for example, with the loss of a loved one, an accident or injury, or a change in where we work or live. In these instances, there is an identiﬁable event that has challenged us to change. But not all catalysts for change arrive suddenly or from the outside. Sometimes an equally demanding need for change emerges from within. It may take the form of a “crystallization of discontent,” a “tipping point,” or an intense internal “itching.” The point is that there come times when a personal transformation is required.

two kinds of change There are two kinds of change in a person. There are small, gradual adjustments that amount to a kind of remodeling. This is what Gregory Bateson liked to call ﬁrst-order change. It often involves changes in appearances and reﬁnements in serving a present life organization. Contrast this with second order-change, in which there are basic changes to the foundations and operational organization of a life.

254 modeling complex systems A second-order change is often called a transformation or a personal revolution. In some instances people may actually change their name, their occupation, or their life partner and community. When a second-order change happens suddenly, it may be called a quantum shift. But suddenness is not common, and the distinction between kinds of change is more complex than I have here presented it. Perhaps a translation into clinical practice will help clarify that complexity.

problems, patterns, and processes Consider the prototypical circumstances surrounding clients’ presentation for psychological services. They are typically struggling and, perhaps, suffering. They are often puzzled, and they seek professional counsel on what is happening in their life and what they can or should do. More often than not, their self-presentation will focus on one or more current life concerns. Let us call them problems. Broadly deﬁned, a problem is a painfully felt discrepancy between the way things are and the way they “should be.” If this deﬁnition works, then there are two primary ways of solving problems: change the way things are, or change the way you want them to be. Much of contemporary psychotherapy—particularly time-limited and manual-driven psychotherapy—amounts to problem solving. This is often what clients want. But not all problems are simple or soluble, and psychotherapy that begins with a focus on a single problem may not remain so focused. In my own work as a therapist I ﬁnd it useful to imagine three interacting levels of focus. There is, of course, the level of problems. These are often emotionally charged preoccupations. The felt sense of my clients may be that there is something wrong (with them, their world, or both) and that something needs to be done. For some clients at some points in their lives, I serve best by offering suggestions as to what might be done. But rare is the client without multiple recurring problems. Recurring problems may reﬂect patterns. Beyond solving their problem(s), many clients want to understand the patterns they are experiencing. “Why do I repeatedly put myself in abusive relationships? Is there something wrong with me? Does this mean I am crazy? Am I [ﬁll in a dsm classiﬁcation]? Am I this way because [ﬁll in a developmental event or context]?” Their

255 Complexity and Human Change Processes requests for explanations are requests for order. This level of inquiry might be called a pattern level. Clients are searching for meaning. The quest for meaning is a natural, common, and healthy expression of our essence as order-seeking creatures. But this is where—I reluctantly confess—much of psychology’s theorizing has failed to serve constructively.2 To be optimally helpful in developmental processes, order seeking must reﬂect and serve activity. Activity is at the heart of the third level of focus in psychotherapy. I call it the process level. We are, of course, always “in process,” even when we do not acknowledge or label that fact. Process-level work in psychotherapy tends to be more spontaneous, embodied, and emotional than the solution-focused planned interventions of problem solving. Work at the level of process tends to be experiential, exploratory, and experimental.3 Although formal exercises (in session and as homework) can add structure to these ventures, their stochastic movements cannot be completely anticipated or predicted. Process-level work can, therefore, be among the most challenging and rewarding (for both client and therapist). Such work appeals to a wisdom that cannot be localized in the left hemisphere of the neocortex. Note that I am using contrasts to structure (organize) my remarks: two kinds of people, two kinds of change, three levels of focus. Let me introduce another contrast that may further serve an understanding of how we construct orderings to cope with complexity.

Teleological and Teleonomic Order There are also two kinds of order evident in most human activity and development. Teleological order is direction imposed by a speciﬁc destination. If we are at point A and we desire to be at point B, our direction of movement is determined by point B. This deterministic directionality is so familiar to us that it is difﬁcult to imagine any alternative. All rational planning and concrete problem solving reﬂect teleological directionality. But there is also a second kind of orderliness in development. It was not formally recognized or respected until the beginnings of dynamic systems writings among the Scottish moral (social) philosophers of the 18th century. The term teleonomy refers to a spontaneous

256 modeling complex systems order that emerges in complex open systems. Teleology is direction (orderly movement) deﬁned by a speciﬁc destination; teleonomy is direction (orderly movement) without a single or speciﬁc destination. The two most common examples of teleonomy are biological evolution and human personality development. Viewed backward— historically—both reﬂect a remarkable orderliness. Their patterned orders reﬂect the operation of principled processes. But the orders reﬂected in biological evolution and a person’s life-span psychological development are different from planned, rational, teleological orders. An appreciation of that difference is critical to our being able to provide responsible professional counsel on human lives and their conduct (Mahoney, in press). Complex interactive systems exhibit complex expressions of chaos and both kinds of order (teleological and teleonomic). Early Greek political philosophers liked to distinguish between the laws of nature and laws designed by humans to govern their own conduct. We have inherited the burdens and blessings of the Greeks’ early love of rationality and its applications to the design of our social systems. There is, however, a signiﬁcant limit to what can be rationally designed and institutionalized without causing harm. This was one of the central insights of Bernard Mandeville, Adam Smith, Adam Ferguson, and David Hume. They recognized that there is a second kind of order that appears in human conduct. It is the result of human action, but not of rational design. I here call it ﬂow to emphasize the contrast with mechanistic force. Flow reﬂects the order that spontaneously emerges from within the ongoing activity of its constituent agents. Open, complex systems express endless, dynamic exchanges. They follow rules of order that can never be completely speciﬁed, predicted, or controlled. Hayek called this the primacy of the abstract (abstract implies “tacit”). He distinguished between the level of principles and the level of particulars. Concrete particulars are the consequences of multiple interacting forces in a widely distributed and ever-changing network. One can approximate a description of the principles by which complex systems operate, but one can never hope to perfectly predict the particulars. An example in psychotherapy is that of language production. Neither we nor our clients can accurately predict the exact words that may emerge from either of us in the coming moments. Those words are the particulars. But there are undeniable principles that constrain and organize what

257 Complexity and Human Change Processes does emerge (e.g., rules of grammar, the languages we speak, our current concerns). Process-level work in psychotherapy honors this second, more complex and teleonomic kind of order. It respects a kind of wisdom in the distributed yet connected systems that are the developing person. Such work reﬂects an appreciation for evolutionary processes within the individual. Process-level work respects the operation of three essential processes in developmental change: variation, selection, and retention.4 All learning—and, therefore, all development—requires variation or novelty. New forms are a necessary part of adaptive change (whether expressed as organisms, behaviors, or ideas). Variation (diversity) is key. Most new forms will not be viable. They will not be selected (naturally or otherwise) as “keepers” by the environments in which they are expressed. A few, however, will offer novel solutions to adaptive challenges. With a little luck these variants will be retained (through genetic coding, neural structuring, habitual use, social custom, epistemic authorization, etc.). Karl Popper and Don Campbell—both pioneers in evolutionary epistemology—noted that the ﬁrst and third processes in complex developing systems are in dialectical contrast. The variance generators live in essential tension with the order protectors. One family of processes is bent on liberal protection of degrees of freedom in the expression of new forms. This family of liberation is at odds with another, more conservative family. Retention processes strive to conserve life order—the very order that is being challenged. We like to believe that we are being rational in our selection processes, but, as David Hume noted, our emotions are often far more powerful than our intellect would like to admit. What does all this mean, practically speaking? It means, among other things, that we have apparently internalized—literally, incorporated (brought into our bodies)—the very processes from which we (and other earthly life-forms) have emerged. Every person is a live-in developmental laboratory. Yes, we are bounded beings, and we are quite protective of our boundaries. Our insides are teeming with activity. Much of what is inside has become organized into systems and centralized structures. It is worth remembering Sechenov’s discovery—published in Darwin’s time—that the primary activity of the nervous system is inhibitory. We are teeming with excitatory impulses. Their control, voluntary and otherwise, is exercised pri-

258 modeling complex systems marily through inhibitory processes. We live as and within essential tensions (Mahoney & Mahoney, 2001). Level on level, we can see development as a dialectical (contrast-generated) development of lifelong gestures of balance. I believe that our gestures can be reﬁned. Let me therefore turn to some practical strategies that can be useful in psychotherapy (Appendix B).

Practical Implications: Two Kinds of Clients There are two kinds of clients in the world: those who are off balance and those who are stuck. All “two kinds” contrasts are, of course, oversimplifying, but they offer a starting point—a bifurcation, in the technical language of dynamic systems theory. Clients who are off balance often present with self-descriptions that emphasize crisis, incoherence (confusion, distraction, falling apart), and a sense of having lost a sense of their center or life order. They may be experiencing a wide spectrum of emotions. Anxiety, anger, and depression are common. I may serve them best by ﬁrst bearing compassionate witness to their difﬁculties and then helping them develop or regain a sense of center. At the other end of this hypothetical-kinds spectrum are clients who are stuck. There is too much stability in their experiencing. They follow the same ruts or routines of activity even though these routines are dysfunctional or unsatisfying. As one client put it, it is the equivalent of psychological constipation: a painful absence of movement. Such clients may be best served by interactions and exercises that challenge their overly regular stability. Many clients present a mixture of these extremes, and most exhibit oscillations and dynamic cycles of looseness and tightness in their experiencing. As they move through their cycles of expansion and contraction, my professional responsibility is to sensitively attune my counsel toward appropriate mixtures of comforting and challenging them. I often begin therapeutic work by assessing and encouraging reﬁnement of centering skills. These skills become a secure base to which the client regularly returns as our work together moves toward later emphases on new ways of experiencing. Moment by moment (within sessions), and across sessions, my professional tasks are to comfort and challenge as my clients idiodynamically

259 Complexity and Human Change Processes contract and expand. Centering is not just a preliminary emphasis; it is a recurrent skill to be revisited and reﬁned throughout the course of our work together.

Centering: Dynamic Balance Skills There are many different ways to teach and practice centering skills. Examples include highly structured behavioral rituals, progressive muscle relaxation, guided imagery, and breathing exercises. I often use these in combination with some basic meditation exercises. There are many different paths to a sense of center, and there are important individual differences in what works best. Two of the most common skills in centering, however, involve attention and balance. Consciously focusing on a ﬁxed object or a rhythmic pattern, for example, can encourage a sense of stability and order. This is a common component of diverse forms of meditation. People are instructed to focus their gaze on a point or to pay attention to a repetitive sound or sensation (e.g., their breathing). Mantras and repetitive rituals can also induce feelings of order in the midst of chaos. It is important to note, however, that one of the most important aspects of the teaching of centering skills in live iterations with clients is the continued learning and practicing of those skills by the teacher. Therapists who simply recite scripted instructions for relaxation, for example, are less likely to serve their clients as well as therapists who practice a relaxation process in the presence of their clients. In addition to centering meditation exercises, I like to teach centering skills through postural stability. This is an approach that emerged in my early work with Olympic athletes and was later incorporated into embodiment exercises with psychotherapy clients. For the sake of brevity, I defer discussion of biomechanical aspects and laboratory evidence. Practically, a “standing-center” exercise asks an individual to stand and focus attention on the sensations in the soles of the feet. It is a literal “grounding” experience that many people ﬁnd inherently stabilizing. Standing up into stability may also have metaphoric meanings for clients and therapists accustomed to traditionally seated “talking heads” psychotherapy. It invites the risk of a novel experience, which—after initial awkwardness—may invite further explorations of both literal and ﬁgurative

260 modeling complex systems movement (e.g., walking—if only in circles—can positively affect the course of therapeutic process). The standing-center exercise, which may last for only a few minutes, can also illustrate some lessons that are both literal (embodied/concrete) and ﬁgurative (metaphoric/abstract). I may ask clients to close their eyes while standing in order to focus their attention inward and on the sensations in their feet. Eye closure will also increase postural sway—the slight rocking motion by which the body continually ﬁnds and adjusts its balance. We are usually unaware of this movement because it is so basic and habitual. Becoming aware of it can serve to illustrate more general lessons about balance in life: e.g., we are most aware of our center (or balance) when we lose it; we are always moving (even when we are unaware of that fact); our body-brain processes intuitively know well how to ﬁnd their way home; and home (or center) is always there, even when we momentarily can’t ﬁnd it.

Edging: Explorations and Experiments in Experiencing For some clients, centering skills and a comforting presence may be the most important things I can offer in psychotherapy. In time-limited therapy with individuals who are struggling to regain a sense of stability and structure in life, these may be invaluable. Other clients may best be served by an approach to psychotherapy that incorporates an ongoing practice of centering skills with exploratory exercises that encourage an appropriately paced experimentation with novelty. This is particularly the case for clients who present as being more toward the “stuck” (overstabilized) than the unstable end of an order-disorder continuum. What are often termed disorders in psychology are, in fact, overstabilized routines of being in the world. This is most apparent, perhaps, in obsessive-compulsive disorder, which is essentially a disorder of overordering. Many of the other formally diagnosed disorders also reﬂect the hyperstabilization of routine patterns of thought, feeling, and behavior. For persons suffering from such patterns, change often depends on exploratory activity and self-experimentation. Note again that the healthy functioning of the whole person requires a dynamic tension between stabilizing and destabilizing processes. For clients whose current needs are primarily for structure and

261 Complexity and Human Change Processes stability, I emphasize the development and practice of orderly routines. For clients whose current needs are primarily to break out of a rutted order, I emphasize exploration with new possibilities. “Do something different” is a frequent homework assignment heard by my clients. Take a different route home, rearrange your closet, tune in to different channels on your radio or television. The only constraints are that these ventures into novelty be self-caring and socially responsible. The idea is to introduce more variance into their life order. With such individuals a constructively challenging interpersonal style can be balanced with the more comforting style required when they feel destabilized. From this dynamic systems perspective, psychotherapy is a special form of human relationship in which and from which clients can ﬁnd compassionate safety and—each when they are ready—begin to explore alternate ways of experiencing themselves, their worlds, and their possible development. Exercises for exploring and experimenting with new skills and different possibilities for experiencing cover the entire spectrum of techniques employed in psychotherapy and other forms of developmental life counseling (Mahoney, 2003a). Among my favorites are personal journaling, movement meditation (which literally encourages clients to move toward new strengths, skills, and enjoyments), and mirror time. Mirror time is a self-relational exercise that asks the individual to spend brief regular periods of time in front of a mirror. A common format is to suggest alternate minutes with eyes open (looking in the mirror) and eyes closed (focusing internally on bodily sensations, thoughts, and feelings). The fundamental strategy is to encourage selfcompassion, but the process must be paced and structured according to individual starting points and learning styles. Not all clients will take kindly to the exercise or themselves, but laboratory, clinical, and ﬁeld research have suggested that self-relational skills are important components in life quality and well-being.

Cycles and Spirals of Expansion and Contraction In complex open systems like ourselves, development is often expressed in cycles and spirals. Oscillations are always in process. Just as our bodies move back and forth through a gravitational center when we are standing, our complexly whole beings move back and

262 modeling complex systems forth through an ever-changing base of viability. The same is true of our interactions in relationships. Episodes of openness and connection may alternate with episodes of closure and distance. Part of the unique challenge of being a mental health professional is the demand to develop an operating center that is large and ﬂexible enough to safely hold all the destabilized centers being served. Psychotherapy can be a destabilizing experience for therapists, and this should not surprise us. We therapists are changed by our work. The life-span psychological development of psychotherapists is often accentuated and accelerated. Therapists should, therefore, be encouraged to prioritize self-care and to explore contexts and experiences that serve their own unique developmental path and pacing.

Concluding Remarks Appropriately, I hope, I shall close with an opening. I began with a reference to my passions as a scientist fascinated with human change processes. Science and passion have been in a strained relationship for too long. The sources of the strain are less important here than are the means by which we might bring them back into creative balance. As a gesture in that direction, I shall conclude by taking a risk. One ﬁnds very little creative writing in the literature of scientiﬁc psychology, yet it is often through creative expressions—gestures toward novel experience—that clients and therapists best learn their dance. I encourage my clients to write, draw, sing, and sculpt their hearts out. Occasionally, they share some of their private creations with me. One client wrote a poem expressing her feelings as we moved toward the conclusion of our work together. She asked if I would write something she could take home with her. The poem that emerged for her is what I now leave with you: It’s a season of transition and you’re on the move again On a path toward something you cannot disown; Searching for your being in the labyrinths of heart And sensing all the while you’re not alone. Yes, you seem to keep on changing for the better and the worse

263 Complexity and Human Change Processes And you dream about the shrines you’ve yet to ﬁnd; And you recognize your longing as a blessing and a curse While you puzzle at the prisons of your mind. For as much as you seek freedom from your agonies and fears And as often as you’ve tried to see the light, There is still a trembling terror that your liberation nears As you struggle with the edges of your night. For your Reason is a skeptic and rejects what it desires, Playing hard to get with miracles and signs; Till a Witness gains momentum and emerges from within To disclose the patterns well above the lines. Then a window has been opened and you’ve let yourself observe How the fabric of your Being lies in wait; And you want to scream in anger and you want to cry for joy And you worry that it still may be too late. For the roller coaster plummets with a force that drives you sane As you tightly grasp for truths that will abide; Never fully understanding that your need to feel secure Is the very thing that keeps you on the ride. You survive the oscillations and begin to sense their role In a process whose direction is more clear And you marvel as your balance point becomes a frequent home, And your lifelong destination feels like “here.” So with gentleness and wonder, with questions and with quests You continue on the path that is your way; Knowing now that you have touched upon the shores of inner life, And excursions deeper can’t be far away.

264 modeling complex systems There will be so many moments when an old view seems so strong And you question whether you can really change; And yet, from deep within you, there’s a sense of more to come And your old view is the one that now seems strange. Take good care, my friend, and listen to the whispers of your heart As it beats its precious rhythm through your days; My warm thoughts and hopes are with you on your journeys through it all . . . And the paths of life in process ﬁnd their ways. Do be gentle, Process Pilgrim; learn to trust that trust is dear, And the same is true of laughter and of rest; Please remember that the living is a loving in itself, And the secret is to ever be in quest . . .

Appendix A: Human Change Processes, a Synopsis 1. Humans are active participants in organizing their experiences of themselves and their worlds. Dynamic and continuous ordering processes construct, maintain, and revise activity patterns. 2. Active ordering processes are primarily tacit and unique to each individual. 3. Those self-organizing processes most vital to life support and individual functioning—what might be called core ordering processes (cops)—are given special protection against changing. 4. cops organize experiences and activities along interdependent dimensions that include: a) emotionality and valence, b) reality status, c) personal identity, and d) power (control/efﬁcacy/agency). 5. Resistance to change—even desired change—is common, especially when the change is experienced as “too much” or “too

265 Complexity and Human Change Processes quickly.” Such resistance reﬂects basic self-protective processes that serve to maintain the coherence of the living system. 6. Ordering processes always operate in relation to their own contrasts, which are disordering processes. Order and disorder are facets of the same dynamic diamond. 7. The ongoing interaction of ordering and disordering processes creates a simultaneous interplay of both familiar and novel experiences. 8. Novelty is necessary for learning and development. Novelty involves contrast. 9. Familiarity and consistency are essential to life support, systemic coherence, and human well-being. 10. The dynamics between familiarity and novelty lie at the heart of human change processes. 11. Order is necessary for development. 12. All learning and development involve transitional disruptions or perturbations in systemic functioning. At all levels, development feeds on disruptions and then digests them into familiarity. 13. It is in the context of disorder that a living system exhibits both its greatest rigidity and its greatest variability. When rigidity reigns, stereotyped behavior or frozen passivity is common. Variability may emerge, particularly in a safe relationship that encourages ventures into ﬂexibility. From this ﬂexibility in being and these expressions of variability, more fulﬁlling and functional activities may emerge and vie for potential selection in the expression of that person’s life. 14. When novel experiences—contrasts—are deﬁcient relative to an individual’s capacities and developmental needs, stagnation and “hardening of the categories” are likely. 15. When novel experiences greatly exceed the individual’s capacities to balance, feelings of being overwhelmed are common. Episodic or chronic disorder and “breakdown” may result. 16. When new experiences are well timed and suited to the individual’s current developmental capacities and edges, developmental “breakthroughs” may emerge and effect whole-system transformations in experiencing. 17. Although disorder may be experienced and expressed in highly patterned processes of human activity, it is diverse, individually unique, and systemic; we shall advance in our attempts at con-

266 modeling complex systems ceptualization and classiﬁcation only as we are willing to embrace the limits of symbol systems to capture human uniqueness and the ultimate ineffability of complex system dynamics. 18. Persons can become “trapped” in disorder. Many psychiatric disorders are, in fact, rigidly ordered patterns. 19. Change is a nonlinear process. It is neither continuous nor cumulative. Rather, change processes reﬂect many small shifts punctuated by occasional sudden leaps and frequent returns to earlier patterns of activity. 20. Change is often experienced in waves or multirhythmic oscillations. Anchors for dimensions of oscillation are often abstract conceptual polarities, such as life/death, right/wrong, good/bad, real/ fake, and sacred/profane. Descriptions of experiences of oscillation often include references to opening and closing, expansion and contraction, loosening and tightening, or approach and withdrawal. 21. Change emerges from a shifting matrix of competing possibilities. Old, tenured patterns of activity compete with new and experimental possibilities. Like all other forms of evolution and revolution, this competition is never ﬁnished. New patterns, when they gain dominance in the competition, themselves become the “old” and familiar order in contrast with which newer patterns emerge and compete. 22. Ongoing competitions in development are neither “won” nor “lost” in reference to allegedly absolute criteria. Some competitors (i.e., impulses of activity) are selected to assume temporary positions in the “driver’s seat” of the body. The old patterns remain as contenders, and they may “win” occasional episodes of ascendancy in future situations. Old habits are not eliminated completely, but they can be effectively displaced by new ones. 23. Selection processes always include human agency. What “decides” the ongoing competition among activity patterns is a complex dynamic system that emerges out of and expresses a human will. 24. Selection processes must mature into retention processes if a chosen activity is to become an inﬂuential pattern in ongoing self-organization. A change—to become a change that makes an enduring and, therefore, signiﬁcant difference—must be actively practiced. 25. Successful (adaptive) change is facilitated by a rhythmic orchestration of exploratory, selective, and perpetuating activities (i.e., experiments in living, evaluative choices regarding which experi-

267 Complexity and Human Change Processes ments are “working,” and action patterns that serve to perpetuate and elaborate valued experiences). 26. Self-relationships, which emerge in social and symbolic relationships, powerfully inﬂuence life quality and resilience under stress. Awareness, acceptance, and celebration are common quests in self-relating. 27. Interpersonal relationships involving strong emotions are powerful contexts for psychological development. Safe and loving intimacy is an expression of trust, which lies at the core of optimal contexts for development. 28. Symbol systems including language and the arts may offer valuable structure and welcome stimulation in personal development. 29. Conscious practices, both spiritually and secularly pursued, are central to qualities and directions of life experiencing. Intention and action are key (even when the goal is stillness). 30. Psychotherapy should reﬂect an appreciation of the history and motivational power of personal realities, the role of interpersonal, symbolic, and self-relational processes in the maintenance and change of personal realities, and the complex existential agency and responsibility of the socially embedded individual. 31. Love is basic to life. What is basic to life is basic to psychotherapy. Spiritual and wisdom traditions that embrace these insights may be precious sources of comfort, companionship, and direction in the complex processes of life-span personal development.

Appendix B: Recommendations for Constructive Practice 1. Prepare for each session in private reﬂection. Even if only for a few seconds, take the time to nurture a sense of your own center and your intention to serve another developing being. 2. Honor the complexity and uniqueness of each client. Do not presume to know your clients. Let them be themselves, and invite them to share as much of their uniqueness as they are ready. 3. Give yourself and your client permission not to know and not to fully understand. Life is much more than ﬁguring things out; effective life counseling does not require complete understanding or deﬁnitive explanations.

268 modeling complex systems 4. Let the clients set the pace, and honor their process. Some clients will want to move very quickly; some will not want to move at all. Pushing often results in pushing back or digging in. Expansions and contractions tend to alternate. Recognize and respect the importance of timing. 5. Encourage (but do not force) emotional expression. Allow your clients to feel, and to feel freely, but do not convey the message that they must be emotionally expressive if they are to make progress or win your respect or caring. When feelings emerge, invite elaborations and explorations. Seek to understand beyond familiar words and summary labels. Locate felt emotions in bodily sensations. 6. Allow and invite yourself to feel emotional in the process of counseling. Let your heart lead your helping. Open yourself to feeling. Learn your patterns around pain. Cultivate compassion, loving kindness, balance, and trust. Recognize that a primary responsibility of your role as a professional helper is to maintain a spirit of centeredness (balance) large enough to accommodate the combined energies of your client and your self. You will be challenged in your balancing abilities, particularly if you allow yourself to be emotionally alive to your interactions with clients. When you are overwhelmed (e.g., by intense emotions, whether your clients’ or your own), take a moment to breathe deeply and to refocus on your intentions to help the person you are serving. 7. Trust that your clients can endure their pain and be strengthened by the process. You cannot take their pain away, although you might often wish that you could. But you can hold a steady course of conﬁdence in their capacities. 8. Emphasize safety, and offer as much structure as your client needs. Use routines, rhythms, and rituals to create a sense of order and familiarity. Give your clients freedom to create their own path, yet be ready with suggestions and illustrations when they request direction or structure. 9. Afﬁrm and encourage experimentation and exploration. Invite clients to experiment with safe and socially responsible excursions into new patterns of experiencing—especially new patterns of action, interpretation, explanation, or meaningmaking. Afﬁrm the processes and feelings involved; express genuine respect for the challenges of changing, and offer generous encouragement of your clients’ active engagement with those challenges.

269 Complexity and Human Change Processes 10. Teach compassion, forgiveness, and self-care. Help your clients develop compassion for themselves and others. Encourage forgiveness. Teach self-care: set a good example.

Notes These remarks are elaborated in Mahoney (1991, 2003a, 2003b, 2004). To preserve textual ﬂow, additional resources are included in the reference list without being speciﬁcally cited in the text. For their contributions to the development of the ideas here presented, the author is grateful to Albert Bandura, Donald T. Campbell, Vittorio F. Guidano, Friedrich A. Hayek, Thomas S. Kuhn, and Walter B. Weimer. 1. For further information, see www.constructivism123.com, the Web site of the Society for Constructivism in the Human Sciences, or the journal Constructivism in the Human Sciences. 2. Until recently, psychology and psychiatry have aided and abetted strong, culturally transmitted inclinations to “name and blame.” Emphases have been placed on pathologies, deﬁciencies, and culpabilities rather than resources, skills, and appreciated capacities. Problems tend to be associated with pain, and pain ranks high as a motivator of human inquiry. Naming things sometimes helps reduce their frightening or puzzling aspects. Naming people associated with the creation of pain is often involved in a blaming process. We do it every day on both individual and collective levels. Some clients are well served by having a name for their problem and a sense of community in their struggles and suffering. Some clients are temporarily relieved to blame their parents, their partners, and their brains for the pain they are experiencing. Psychology is full of labels and theories that encourage naming and blaming. I have elsewhere elaborated my concerns about the categorical preoccupations of our profession. Most relevant here is the practical consequence that categorical names and no exit/no action explanations often disserve a client. When a diagnostic label results in a client’s identifying with the problem, change may become more (rather than less) difﬁcult. Identiﬁcation with one’s problem(s) results in an added challenge—to change, one must change, not only the problem(s), but also one’s core sense of identity. Likewise, when an explanation suggests that people are a certain problematic way because of their brain chemistry, personality, developmental history, or the like, clients may resign themselves to insight without action. Activity, please recall, is at the heart of our being alive. It is also at the heart of our potential capacities for constructive, developmental change. 3. Historically, mid-20th-century behavior therapists liked to distinguish themselves from the experiential humanistic-existential tradition, which they viewed as less scientiﬁc and experimental. However, both camps were and are “radically empirical” in William James’s sense (Mahoney, n.d.)

270 modeling complex systems because of their emphasis on praxis (action) that takes the learner to the embodied and emotional edges of his or her live experiencing. 4. These are the three essential processes in biological evolution, behavioral development, and evolutionary epistemology (including scientiﬁc progress).

References Anderson, W. T. (1990). Reality isn’t what it used to be. San Francisco: Harper & Row. Anderson, W. T. (Ed.). (1995). The truth about the truth: De-confusing and reconstructing the postmodern world. New York: Putnam. Anderson, W. T. (1997). The future of the self: Inventing the postmodern person. New York: Tarcher/Putnam. Arciero, G., & Guidano, V. F. (2000). Experience, explanation, and the quest for coherence. In R. A. Neimeyer & J. D. Raskin (Eds.), Constructions of disorder: Meaning-making frameworks for psychotherapy (pp. 91–118). Washington dc: American Psychological Association. Bandura, A. (1997). Self-efﬁcacy: The exercise of control. New York: Freeman. Bruner, J. (1990). Acts of meaning. Cambridge ma: Harvard University Press. Bruner, J. (2002). Making stories: Law, literature, life. New York: Farrar Straus Giroux. Collins, R. (1998). The sociology of philosophies: A global theory of intellectual change. Cambridge ma: Harvard University Press. Damasio, A. (1999). The feeling of what happens: Body and emotion in the making of consciousness. New York: Harcourt Brace. Ford, D. H. (1987). Humans as self-constructing living systems: A developmental perspective on behavior and personality. Hillsdale nj: Erlbaum. Freeman, W. (1995). Societies of brains. Hillsdale nj: Erlbaum. Gergen, K. J. (1991). The saturated self. New York: Basic. Gergen, K. J. (1994). Realities and relationships: Soundings in social construction. Cambridge ma: Harvard University Press. Gergen, K. J. (1999). An invitation to social construction. London: Sage. Goldberger, N., Tarule, J., Clinchy, B., & Belenky, M. (Eds.). (1996). Knowledge, difference, and power: Essays inspired by women’s ways of knowing. New York: Basic. Guidano, V. F. (1987). Complexity of the self: A developmental approach to psychopathology and therapy. New York: Guilford. Guidano, V. F. (1991). The self in process. New York: Guilford. Hayek, F. A. (1952). The sensory order. Chicago: University of Chicago Press. Hayek, F. A. (1964). The theory of complex phenomena. In M. Bunge (Ed.), The critical approach to science and philosophy: Essays in honor of K. R. Popper (pp. 332–349). New York: Free Press. Hayek, F. A. (1976). Law, legislation, and liberty: Vol. 2. The mirage of social justice. Chicago: University of Chicago Press.

271 Complexity and Human Change Processes Hayek, F. A. (1978). New studies in philosophy, politics, economics, and the history of ideas. Chicago: Chicago University Press. Johnson, M. (1987). The body in the mind: The bodily basis of meaning, imagination, and reason. Chicago: University of Chicago Press. Kauffman, S. A. (1993). The origins of order: Self-organization and selection in evolution. Oxford: Oxford University Press. Kauffman, S. A. (1995). At home in the universe. Oxford: Oxford University Press. Kegan, R. (1982). The evolving self. Cambridge ma: Harvard University Press. Kegan, R. (1994). In over our heads: The mental demands of modern life. Cambridge ma: Harvard University Press. Kelly, G. A. (1955). The psychology of personal constructs. New York: Norton. Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge ma: mit Press. Kuhn, T. S. (2000). The road since structure. Chicago: University of Chicago Press. Lakoff, G., & Johnson, M. (1999). Philosophy in the ﬂesh: The embodied mind and its challenge to Western thought. New York: Basic. Mahoney, M. J. (1991). Human change processes. New York: Basic. Mahoney, M. J. (2000a). Behaviorism, cognitivism, and constructivism: Reﬂections on persons and patterns in my intellectual development. In M. R. Goldfried (Ed.), How therapists change (pp. 183–200). Washington dc: American Psychological Association. Mahoney, M. J. (2000b). A constructive view of disorder and development. In R. A. Neimeyer & J. D. Raskin (Eds.), Constructions of disorder: Meaning-making frameworks for psychotherapy (pp. 43–62). Washington dc: American Psychological Association. Mahoney, M. J. (2003a). Constructive psychotherapy: A practical guide. New York: Guilford. Mahoney, M. J. (2003b). Pilgrim in process: Collected poems. Plainﬁeld il: Kinder Path. Mahoney, M. J. (2004). Scientist as subject: The psychological imperative. Clinton Corners ny: Eliot Werner. (Original work published 1976) Mahoney, M. J. (2005a). Constructive suggestions for the practical education of professional life counselors. Journal of Clinical Psychology, 61(9), 1179–1184. Mahoney, M. J. (2005b). Suffering, philosophy, and psychotherapy. Journal of Psychotherapy Integration, 15(3), 337–352. Mahoney, M. J. (in press). Power, politics, and psychotherapy: A constructive caution on uniﬁcation. Journal of Psychotherapy Integration. Mahoney, M. J. (n.d.). The heart minding science: A personal history of psychology. Unpublished manuscript. Mahoney, M. J., & Mahoney, S. M. (2001). Living within essential tensions: Dialectics and future development. In K. J. Schneider, J. F. T. Bugental, & J. F. Pierson, (Eds.), The handbook of humanistic psychology (pp. 659–665). Thousand Oaks ca: Sage.

272 modeling complex systems Maturana, H. R., & Varela, F. J. (1987). The tree of knowledge: The biological roots of human understanding. Boston: Shambhala. Neff, K. (2003). Self-compassion: An alternative conceptualization of a healthy attitude toward oneself. Self and Identity, 2, 85–102. Neiman, S. (2002). Evil: An alternative history of philosophy. Princeton nj: Princeton University Press. Neimeyer, R. A., & Mahoney, M. J. (Eds.). (1995). Constructivism in psychotherapy. Washington dc: American Psychological Association. Neimeyer, R. A., & Raskin, J. D. (2000). Constructions of disorder: Meaningmaking frameworks for psychotherapy. Washington dc: American Psychological Association. Núñez, R., & Freeman, W. J. (Eds.). (1999). Reclaiming cognition: The primacy of action, intention and emotion. Tuscon az: Imprint Academic. Petitot, J., Varela, F. J., Pachoud, B., & Roy, J.-M. (Eds.). (1999). Naturalizing phenomenology: Issues in contemporary phenomenology and cognitive science. Stanford ca: Stanford University Press. Radnitzky, G., & Bartley, W. W. (Eds.). (1987). Evolutionary epistemology, theory of rationality, and the sociology of knowledge. LaSalle il: Open Court. Schore, A. N. (1994). Affect regulation and the origin of the self. Hillsdale nj: Erlbaum. Segal, L. (1986). The dream of reality: Heinz von Foerster’s constructivism. New York: Norton. Segal, Z. V., Williams, J. M. G., & Teasdale, J. D. (2002). Mindfulness-based cognitive therapy for depression. New York: Guilford. Stolorow, R. D., & Atwood, G. E. (1992). Contexts of being: The intersubjective foundations of psychological life. Hillsdale nj: Analytic. Stolorow, R. D., Atwood, G. E., & Orange, D. M. (2002). Worlds of experience: Interweaving philosophical and clinical dimensions in psychoanalysis. New York: Basic. Thelen, E., & Smith, L. B. (1994). A dynamic systems approach to the development of cognition and action. Cambridge ma: mit Press. Thompson, E. (Ed.). (2001). Between ourselves: Second-person issues in the study of consciousness. Charlottesville va: Imprint Academic. Van Geert, P. (1998). A dynamic systems model of basic developmental mechanisms: Piaget, Vygotsky and beyond. Psychological Review, 105, 634–677. Van Geert, P. (2000). The dynamics of general developmental mechanisms: From Piaget to Vygotsky to dynamic systems models. Current Directions in Psychological Science, 9, 64–68. Varela, F. J. (1979). Principles of biological autonomy. New York: Elsevier North Holland. Varela, F. J., & Shear, J. (Eds.). (1999). The view from within: First-person approaches to the study of consciousness. Bowling Green oh: Imprint Academic. Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. Cambridge ma: mit Press.

273 Complexity and Human Change Processes Velmans, M. (Ed.). (2000). Investigating phenomenal consciousness: New methodologies and maps. Amsterdam: Benjamins. Watzlawick, P. (Ed.). (1984). The invented reality: Contributions to constructivism. New York: Norton. Weimer, W. B. (1977). A conceptual framework for cognitive psychology: Motor theories of mind. In R. Shaw & J. Bransford (Eds.), Perceiving, acting, and knowing (pp. 267–311). Hillsdale nj: Erlbaum. Weimer, W. B. (1979). Notes on the methodology of scientiﬁc research. Hillsdale nj: Erlbaum. Weimer, W. B. (1987). Spontaneously ordered complex phenomena and the unity of the moral sciences. In G. Radnitzky (Ed.), Centripetal forces in the universe (pp. 257–296). New York: Paragon.

Editors’ Postscript: Modeling Complex Processes in a Rehabilitation Application

In keeping with the theme of balancing theory, research, and modeling approaches with potential practical applications, this, the ﬁnal chapter of this volume is a brief overview of one application of modeling complex processes associated with the practice of rehabilitation. Several levels of analysis, molar to molecular, are addressed in contributions in this volume. Hence, the collection may in some senses be seen as quite eclectic. However, from the perspective of a multifaceted system like that seen in the world of practical clinical rehabilitation for injured and/or ill individuals, it may also be seen as representing complementary areas that together have a great deal to offer with respect to the design and delivery of rehabilitation and to rehabilitation-related research. To illustrate the point, a speciﬁc model developed in Nebraska, the Madonna model—meant to model the quite molar “complex process” of clinical rehabilitation—is presented. Developed as a framework to systematically design and deliver comprehensive, coordinated rehabilitation programs, it is also a model to support design, conduct, and analysis of contextually based, clinical rehabilitation research with an explicitly practical focus. While the speciﬁc model presented here is not derived from other sources, it is to be acknowledged immediately that the individually designated aspects of the model together constitute a set of

276 modeling complex systems foci that would probably be endorsed as relevant to the design and operation of comprehensive rehabilitation programs by most experienced rehabilitation providers and recipients.1 The individual aspects of the model are discussed only in broad outline below, along with brief excerpts drawn from the present collection illustrating the relevance of these areas of research and thought to research and practice in clinical rehabilitation. Most, if not all, facets of the rehabilitation model presented here are addressed in a typical rehabilitation treatment plan or program. The model (see Figure 1) is represented in a circumplex form, intended to emphasize the interdependence of the various facets in the design and delivery of individualized treatment programs for individuals requiring comprehensive rehabilitation. In practice, by carefully evaluating the relations that exist—or are necessary to establish—among these aspects in each unique clinical situation, interdisciplinary rehabilitation teams are most likely to work successfully with the range of stakeholders to create a plan that will optimize functional outcomes, minimize unanticipated complications, and maximize efﬁciency. When these facets are addressed in isolation, outcomes are much more likely to be suboptimal. Although using the model to guide treatment planning naturally introduces some degree of complexity—relative to a simple “critical path,” for example—it is crucial to recognize and respond accordingly to the linked, interdependent nature of these foci. Moreover, the pattern of relations between the aspects of the model, and the relative importance of addressing each aspect, differs among diagnostic categories and, indeed, across patients within the same diagnostic group, owing to differences in patient characteristics and life circumstances. Examination of these relations and incorporation of them in the initial planning process is an important ingredient in designing and delivering a coherent, maximally successful program. Invoking and using the model at the outset of treatment is also a useful vehicle for beginning the process of educating patient and family about the nature of their circumstances and some dimensions of their prospects in the coming months or years. The model can also serve as a coherent guide for research. A rehabilitation research agenda can be expressed as a study of the quantitative and qualitative relations between elements of the model and treatment outcomes. The distinctive focus of such research is the

277 Editors’ Postscript

Figure 1. The Madonna model: an integrative focus guiding treatment and research.

examination of the relative contributions on important functional outcomes of the many separately delivered, but interacting, interventions that a rehabilitation patient receives. For example, many studies have been conducted examining the utility of a range of physical-therapy evaluation and treatment techniques for addressing patients’ mobility outcomes. However, many patients receive comprehensive programs; they are treated by multiple clinicians and receive many different modalities. It is the interaction of all these interventions, along with their own organismic vitality, that produces the ﬁnal result. However, comparatively little is known in speciﬁcs about how these “independent variables” interact to produce the outcomes of interest. Therefore, to maximize functional outcomes, minimize complications, and increase efﬁciency, one needs to look at the combined, joint inﬂuence of these many inﬂuences. As we learn more, we will be able to furnish clinical treatment teams with the

278 modeling complex systems information they need to ensure inclusion of crucial ingredients in the treatment plan and omission of non-value-added components. Ideally, treatment and research endeavors are reciprocal elements of one integrated process.

Model Elements team characteristics The initial component of the model reﬂects the functional primacy of clinical-treatment-team structures and decision-making processes governing the organization and applications of the model components that follow sequentially, reﬂecting the mechanisms through which elements of the later components are applied in the context of an individual treated by the team. Central aspects of this initial component include attention to issues such as interdisciplinarytreatment-team skills and the team culture inﬂuencing issues such as planning, communication, clinician accountability, and team decision-making processes. Research in this domain focuses on issues like ﬁnding ways to optimize collaborative functioning, developing high-performing teams, managing emotional issues at the team level effectively, and developing and maintaining a team social climate that enhances individual team members’ psychological growth and levels of satisfaction and engagement. Although empirical research and, thus, explicit guidelines or recommendations illuminating clinical-treatment-team structures and decision-making processes are underrepresented in the rehabilitation literature, it is important to conceptualize the “team” in an inclusive way: in addition to patients/consumers and families, payers and, ultimately, members of the community are all stakeholders in the outcomes sought. Finding ways to meaningfully include these stakeholders, in real time, is the challenge currently facing rehabilitation workers. The chapters by Lajoie, Musen, Neufeld, and Salas, Stagl, Burke, and Goodwin can all be read proﬁtably as suggesting models of treatment-competency development, service delivery, and practical data representation to aid in decision making, all of which will be increasingly challenging—but crucially important—as rehabilitation moves from conﬁned treatment environments into (complex) community venues.

279 Editors’ Postscript

patient/family engagement This aspect of the model refers to application of processes and procedures to optimize patient and family involvement, motivation, and engagement in the rehabilitation process, without which the effects of the implementation of subsequent components will be suboptimal. From the psychological perspective, this segment of the model includes assessments and treatments designed to minimize trauma associated with injury or illness and to promote positive adaptations to associated challenges. Identiﬁcation of patient and family strengths and resources early in the rehabilitation process is crucial to a successful outcome. Attention to this dimension includes management of a treatment setting’s psychosocial milieu to promote an environment communicating optimism and encouragement, inviting maximum independence for patients and families. Identifying efﬁcient and reliable ways to discover and tap these strengths should be a major research focus. Themes in psychology currently associated with variants of “positive psychology” and “strengths-based” psychology (Buckingham & Clifton, 2001; Clifton & Nelson, 1992; Fredrickson, 2001; Lopez & Snyder, 2003; Seligman, 2002) are promising approaches to systematic and proactive development and application of strategies to optimize this aspect of the rehabilitation model. Michael Mahoney’s chapter in this volume, with its emphasis on “constructivism”—explicitly recognizing “human beings as actively complex, socially-embedded, and developmentally dynamic self-organizing systems” (as the tagline on the cover of Constructivism in the Human Sciences describes his journal’s mission)—offers many avenues to explore in the process of promoting and maintaining patient, family, and social-network engagement. These ideas are explored more fully elsewhere (Mahoney, 2003). Lajoie’s insights and ﬁndings as presented in her chapter in this volume would seem to be as applicable to “educating” patients and families as to training professional personnel.

therapeutics This element of the model refers to the various evaluation and treatment techniques available for use in rehabilitation programs. Instru-

280 modeling complex systems mentation and equipment, range and delivery of therapies, medications, and so on, all fall within the scope of this facet of the model. From a research perspective it represents a focus on improvement and application of speciﬁc forms of patient evaluation and treatment interventions. Clinicians and clinical teams identify promising new approaches or identify particular challenges; research is designed and trials implemented to reﬁne and validate evolving resources and procedures. In addition to individual therapeutic modalities, other research foci should include determination of: • • • •

Optimal timing of treatments following an accident or illness Optimal intensity of treatments Optimal duration of treatments Optimal mix of the interventions provided to a patient, including determination of the relative contributions of the constituent interventions.

All the Symposium speakers provided valuable insights in this area of focus. Each paper suggests innovative approaches that can serve to enrich clinicians’ repertoire of tools to enhance rehabilitation practice, whether in the psychiatric- or the physical-medicine realm.

technology The technology component of the model places an emphasis on development and application of the rapid advances in engineering sciences and derivative technology. Progress in the ﬁelds of electrical engineering, mechanical engineering, industrial engineering, civil engineering, computer sciences, and other domains has tremendous promise with respect to helping individuals with disabilities successfully meet the challenges confronting them in their homes and communities. The rapidly developing area of biomedical engineering reﬂects the increasing importance of technology in rehabilitation as well as in medicine generally. For example, computerized augmentative and alternative communication devices are available now to help individuals with speech, language, and/or cognitive impairments participate more effectively in the activities of daily life and life roles. Laser technology is being employed to help individuals with signiﬁcant motor impairments control their environments and

281 Editors’ Postscript participate more fully in the important domains of their lives. Socalled smart environments that recognize speciﬁc individuals are now being engineered and reﬁned, via the application of electrical sensors, laser-sensitive surfaces, computers, wireless communication technology, global positioning technology, and so on. These environments are designed to be responsive to differences in individuals’ capacities (e.g., cognitive, physical, and communicative) in a given environment (e.g., home, school, workplace), “automatically” altering the environment in ways that best conform to an individual’s unique strengths and needs, thereby increasing prospects for optimal participation in important life activities, enhancing productivity, and maximizing positive quality of life. Implementation of emerging technologies is an important emphasis in the work reported in the chapters by Salas et al., Lajoie, and Musen in particular.

living setting The living-setting facet of the model reﬂects the importance placed on individualized, discharge-destination-speciﬁc rehabilitation-program design. To foster independence and quality of life for many individuals following disabling illnesses or injuries, living settings need to be redesigned, incorporating structural modiﬁcations and/or installation of assistive and alternative technologies. Increased research in this area will facilitate development of rehabilitation and engineering practices that are effective and cost-efﬁcient and that ensure that patient reintegration into his or her living setting is optimized. This aspect of the model is not represented with marked emphasis in this year’s Symposium volume but should be addressed in any comprehensive research or practice efforts, reﬂecting the salience of contextual factors in human experience and functioning. This would seem to be a very useful area to explore in a future Symposium.

participation This element of the model emphasizes the signal importance of maintaining a rehabilitation focus on identiﬁcation of effective and efﬁcient ways of linking individuals and families to resources in

282 modeling complex systems their community, in the effort to optimize reintegration into home, school, vocational, and recreational settings: that is, life roles. This aspect reﬂects the strong emphasis on “activities and participation” articulated in World Health Organization (2001). This dimension of the rehabilitation model is understood to include the individual and his or her family, social network, and community as well as the environment as a system, taken at the broadest of levels. Research in this area, though often appropriately focused on individual change and empowerment, also includes development, examination, and implementation of effective strategies to ensure that community resources are fully accessible as well as efforts targeting timely and adequate education of community members concerning strengths and needs of their disabled members, again with the overarching aim of promoting a successful resumption of signiﬁcant and productive life roles. Operationalization of this facet of the model underscores the importance of rehabilitation-oriented input into community-planning activities. Perspectives and traditions resonant with this component of the model include ecological psychology (e.g., the work of Roger Barker and his colleagues on the concept of behavior settings and the implications of how such settings are “manned”) and community psychology (e.g., Heller & Monahan, 1997; Rappaport, 1977; Sarason, 1988).

program evaluation The program evaluation component of a rehabilitation model is crucial to maintenance of an evidence-based culture, providing timely informational feedback to team members and administrators concerning the outcome of the activities of the program/system. Information regarding improvements in patient functioning, number and pattern of short- and long-term complications associated with recovery processes, and relative efﬁciency of service delivery is developed and made available on a regular basis. Typical measures include • • • • •

Patient outcomes; Program improvement; Program integrity (qa); Cost efﬁciency; Quality of collaborations.

283 Editors’ Postscript Evolving team/program activities include the integration of new protocols as well as the successful implementation of relevant research ﬁndings. Regular dissemination of this information with other organizational teams as well as with colleagues at other rehabilitation settings is an indispensable mechanism for improving rehabilitation practices.

clinical decision support Clinical decision support in a rehabilitation setting is a practical exercise in modeling complex systems. As the amount of information and complexity of treatment technologies increases, and as time and resources available for rehabilitation steadily decrease, clinical-decision-support systems—linked to a guiding model of rehabilitation aspirations and activities—are increasingly important. This highly signiﬁcant component of the model represents the mechanism for translating information derived from program evaluation into practical understanding and effective responding to the relations that exist between all the other elements of the model as they apply to a given individual. It is also a mechanism for gaining an understanding of the interplay of factors inﬂuencing a rehabilitation setting at the level of a treatment program. The development and application of computer-based clinical-decision-support systems, the content of which is derived from program evaluation, should operate to enhance the efﬁciency and effectiveness of patient, family, and clinician decision making during rehabilitation. Simple examples of decisionsupport aids include basic tools such as diagrams and clinical paths. However, more complex procedures are rapidly being developed and incorporated, such as computer-based decision trees, calculated conditional probabilities, and sensitivity analyses to determine the relative impact of changes and interactions between speciﬁc variables on the ﬁnal outcome(s) produced by a combination of several components. Graphic representations of patient performance over time and statistical tools for prediction of patient outcome based on prognostic data are beginning to appear. It is important that the design of such decision-support tools be informed by awareness of the importance of naturalistic contexts (e.g., temporal context, hospital context, community context) in order to develop truly useful decision-support

Figure 2. The context of rehabilitation: illustrative aspects.

285 Editors’ Postscript systems. The chapters by Neufeld, Salas et al., and Musen are all quite relevant to consult when contemplating this aspect of the decision-making process.

model summary The model described here represents an approach to focused rehabilitation treatment and research that is probably compatible with a general rehabilitation philosophy that has been elaborated over the years in a number of settings and that is comprehensive and true to the multifaceted nature of the rehabilitation process. Using this model as a guide, one can work to build and elaborate a successful research program aligned with the essential elements of an effective rehabilitation program. Many of these dimensions are well addressed in the chapters in this volume. As indicated in the schematic of Figure 2, there are a number of elements involved in the rehabilitation process. Indeed, the ﬂow chart presented here is an obvious oversimpliﬁcation. However, one can see many areas of application for the thoughts presented in the chapters in this volume that can be brought to bear on various aspects of the rehabilitation process. Space does not allow description of these applications, but readers are urged to reﬂect on the potential application of these ideas to rehabilitation and psychological practice in general as they read and reﬂect on the wealth of stimulating ideas offered in this volume.

Note 1. The notion of comprehensive rehabilitation is meant to apply to clinical situations in which afﬂictions and impairments are substantial, often involving multiple organismic systems. For example, rehabilitation following stroke frequently requires a comprehensive approach because multiple capacities are affected. At the other end of the spectrum, a bruised shoulder sustained in a softball game might require a few brief sessions of ultrasound therapy, along with a home exercise program, but would not require the use of assistive technology, computer-based clinical decision support, modiﬁcation of the home environment, or a community-reintegration program.

286 modeling complex systems

References Buckingham, M., & Clifton, D. O. (2001). Now, discover your strengths. New York: Free Press. Clifton, D. O., & Nelson, P. (1992). Soar with your strengths. New York: Dell. Fredrickson, B. L. (2001). The role of positive emotions in positive psychology: The broaden-and-build theory of positive emotions. American Psychologist, 56, 218–226. Heller, K., & Monahan, J. (1997). Psychology and community change. Homewood il: Dorsey. Lopez, S. J., & Snyder, C. R. (Eds.). (2003). Positive psychological assessment: A handbook of models and measures. Washington dc: American Psychological Association. Mahoney, M. J. (2003). Constructive psychotherapy: A practical guide. New York: Guilford. Rappaport, J. (1977). Community psychology: Values, research, and action. New York: Holt, Rinehart & Winston. Sarason, S. B. (1988). The creation of settings and the future societies. Brookline ma: Brookline. (Original work published 1972) Seligman, M. E. P. (2002). Authentic happiness. New York: Free Press. World Health Organization. (2001). International classiﬁcation of functioning, disability and health. Geneva.

Contributors

C. Shawn Burke is a research scientist at the University of Central Florida, Institute for Simulation and Training. Her expertise includes teams and their leadership, team adaptability, team training, measurement, evaluation, and team effectiveness. Dr. Burke has presented at 66 peer-reviewed conferences, published 33 articles in scientiﬁc journals and books related to the above topics, and assisted organizations in evaluating aviation-related team-training programs and reducing medical errors. She is currently investigating team adaptability and its corresponding measurement; issues related to multicultural team performance, leadership, and the training of such teams; and the impact of stress on team process and performance. Dr. Burke earned her doctorate in industrial/organizational psychology from George Mason University and is a member of the American Psychological Association, the Society for Industrial and Organizational Psychologists, and the Academy of Management. She serves as an ad hoc reviewer for Human Factors and Quality and Safety in Healthcare. Gerald F. Goodwin is a research psychologist at the U.S. Army Research Institute for Behavioral and Social Sciences, assigned to the Leader Development Research Unit (ldru). He received his ms and PhD in industrial/organizational psychology from Pennsylvania State Uni-

288 modeling complex systems versity. Dr. Goodwin’s current research focus is on leader and team effectiveness issues, particularly with regard to the joint, interagency, and multinational context. He also provides statistical and methodological support within ldru. He was previously employed at the American Institutes for Research, where his project work included test development, employment-litigation support with regard to statistical analysis, training evaluation, and performance modeling. He is a member of the Society for Industrial and Organizational Psychology, the American Psychological Association (apa), and apa Division 19 (Military Psychology). He currently serves as an ad hoc reviewer for the Journal of Applied Psychology and Human Performance and has recently joined the editorial board of Human Factors. Zeno Kupper received his PhD in clinical psychology from the University of Fribourg, Switzerland, in 1998. He was trained in clientcentered and cognitive-behavioral psychotherapy and has extensive clinical experience in the rehabilitation of patients with severe mental illness. He is a research and clinical psychologist at the University Hospital of Psychiatry, University of Bern, Switzerland. In his research he has been engaged in developing analytic techniques to identify and measure complex patterns in the course of mental disorders, for example, using variants of time-series analysis and cluster analysis to identify characteristic patterns of schizophrenia patients’ response to treatment and rehabilitation and modeling stable and unstable states in the course of psychotherapy. This kind of analytic technology aims to translate the insights of dynamic systems theory into clinical research. Susanne P. Lajoie received her doctorate from Stanford University in 1986. In addition to being the James McGill Professor she is chair of the Department of Educational and Counselling Psychology at McGill University. Dr. Lajoie has engaged in a wide array of innovative research and scholarly activities where she applies cognitive theories in the design of computer-based learning environments for classroom and real-world applications. She uses a cognitive approach to skill identiﬁcation and applies her research to the design of computer-coached practice environments in the areas of science, statistics, and medicine. She has numerous publications, including two volumes on computers as cognitive tools published by Erlbaum.

289 Contributors Michael J. Mahoney was a pioneer in the cognitive-behavior-therapy movement in the 1970s, holding public debates with Joseph Wolpe and other classical behaviorists in the newly formed Association for the Advancement of Behavior Therapy. His analyses of the role of verbal representations in behavior acquisition and change were compelling demonstrations of the limitations of “cognition-free” behaviorism. His seminal 1974 Self-Control: Power to the Person (with Carl E. Thoresen) heralded the contemporary age of cognitively oriented, social-learning-theory-based therapies. As cognitive constructs became acceptable and then mainstream in behavior therapy, Mahoney moved on to even more complex models of the psychotherapy enterprise, derived from chaos and self-organizing systems theories. Most recently, Professor Mahoney was a leading thinker in conceptualizing behavior change from a systemic perspective. The “motivation to change” plays a key role in these conceptualizations, but with the systemic view that motivation is a complex interactive process that includes both organismic and environmental components. Mahoney’s application of systemic models to psychotherapy was at the frontier of this area, an endeavor to systematize one of the most complex and difﬁcult-to-quantify decision processes in the human repertoire. At the time of his Nebraska Symposium presentation, Dr. Mahoney was a professor at the University of North Texas. He was also distinguished consulting faculty at Saybrook Graduate School and Research Center, San Francisco. A recipient of many honors, Dr. Mahoney was a fellow of the American Association for the Advancement of Science, the American Psychological Association, and the World Academy of Art and Science. He was also the executive director of the journal Constructivism in the Human Sciences. A collection of his recent poems, Pilgrim in Process, was published in 2003 by Kinder Path. Mark Musen is professor of medicine (biomedical informatics) and computer science (by courtesy) at Stanford University, where he is head of the Stanford Center for Biomedical Informatics Research. Dr. Musen conducts research related to intelligent systems, the Semantic Web, reusable ontologies and knowledge representations, and biomedical decision support. In 1989, he received the Young Investigator Award for Research in Medical Knowledge Systems from the American Association of Medical Systems and Informatics. He received a

290 modeling complex systems Young Investigator Award from the National Science Foundation in 1992. In 2006, he received the Donald A. B. Lindberg Award for Innovation in Informatics from the American Medical Informatics Association. Dr. Musen sits on the editorial boards of several journals related to biomedical informatics and computer science. He is the coeditor of the Handbook of Medical Informatics (Springer, 1997) and the coeditor-in-chief of the journal Applied Ontology. Richard W. J. Neufeld is a professor in the Departments of Psychology and Psychiatry, University of Western Ontario, where he also is a core faculty member of the Program in Neuroscience. He has received the Joey and Toby Tannenbaum Schizophrenia-Research Distinguished Scientist Award, being the ﬁrst psychologist recipient, the Ontario Mental Health Foundation Senior Research Fellowship, and the University of Western Ontario Faculty of Social Science Research Professorship. Professor Neufeld is a past associate editor of the Canadian Journal of Behavioral Science and of Psychological Assessment. Professor Neufeld has authored or edited 7 books and journal special sections. His 155 publications and 21 technical reports have appeared in journals ranging from the Journal of Mathematical Psychology, the British Journal of Mathematical and Statistical Psychology, and the Psychological Review to the Journal of Abnormal Psychology, the Journal of Consulting and Clinical Psychology, and Psychological Assessment. Jeffrey Poland received an ma in clinical psychology from the Southern Connecticut State University in 1982 and a PhD in the philosophy of science from the Massachusetts Institute of Technology in 1983. He is the author of Physicalism: The Philosophical Foundations (Oxford University Press, 1994) and a coauthor (with William Spaulding and Mary Sullivan) of Treatment and Rehabilitation of Severe Mental Illness (Guilford, 2003). He has held academic positions at Colgate University and the University of Nebraska–Lincoln, and he currently teaches in the Department of History, Philosophy, and Social Science at the Rhode Island School of Design and in the Department of Psychology and the Science and Society Program at Brown University. Eduardo Salas is Trustee Chair and professor of psychology at the University of Central Florida. He also holds an appointment as program

291 Contributors director for Human Systems Integration Research Department at the Institute for Simulation and Training. Previously, he was a senior research psychologist and head of the Training Technology Development Branch of navair-Orlando for 15 years. During this period, Dr. Salas served as a principal investigator for numerous researchand-development programs focusing on teamwork, team training, advanced training technology, decision-making under stress, learning methodologies, and performance assessment. Dr. Salas has coauthored over 300 journal articles and book chapters and has co edited 15 books. He currently edits the annual series Advances in Human Performance and Cognitive Engineering Research for Elsevier. He is currently designing tools and techniques to minimize human error in aviation, law enforcement, and medical environments. Bill Shuart is the director of the Institute for Rehabilitation Science and Engineering at Madonna Rehabilitation Hospital in Lincoln, Nebraska. He received his PhD in clinical psychology from the University of Nebraska in 1985, where he retains a clinical afﬁliation with the Department of Psychology. His research interests include modeling team decision making in rehabilitation, with an emphasis on identifying effective applications of information technology to improve medical rehabilitation evaluation and treatment services for children and adults with disabilities. Will Spaulding received his PhD in clinical psychology from the University of Arizona in 1976. He was a postdoctoral fellow in mental health research and teaching, under the mentorship of Rue L. Cromwell, in the Department of Psychiatry of the University of Rochester. After his postdoctoral work, he joined the faculty of the University of Nebraska–Lincoln’s Clinical Psychology Training Program, where he is currently a professor. In addition to training clinical psychologists, he conducts research on the nature of mental illness, clinical decision making, and the outcome of treatment and rehabilitation. He also practices as a clinical psychologist in rehabilitation- and recovery-oriented services for people with severe and disabling mental illness. He has coedited two previous volumes of the Nebraska Symposium on Motivation and is the author, with Mary Sullivan and Jeffrey Poland, of Treatment and Rehabilitation of Severe Mental Illness (Guilford, 2003).

292 modeling complex systems Kevin C. Stagl is an organizational consultant with Assessment Technologies Group and formerly served as a research scientist and research assistant at the University of Central Florida’s Institute for Simulation and Training. Dr. Stagl’s research investigates team performance, leadership, development, and adaptation. The lessons learned and practices distilled from this effort have appeared in scholarly outlets such as the Journal of Applied Psychology, Leadership Quarterly, Organizational Frontiers Series, and International Review of Industrial and Organizational Psychology. Wolfgang Tschacher studied psychology at the University of Tübingen, where he received his PhD in 1990. He also received psychotherapy training in systemic therapy at the Institute of Family Therapy, Munich. After his habilitation in psychology, he received the Venia legendi in 1996 at the University of Bern, Switzerland, where he was awarded a professorship in 2002. He works at the University Hospital of Psychiatry and is currently head of the Department of Psychotherapy. His main interests are in empirical psychotherapy research and experimental psychopathology, with an emphasis on dynamic systems approaches and phenomena of cognitive self-organization.

Subject Index

Page numbers in italics refer to illustrations. active agency, 250 activity, 255, 257–258, 269n2 adaptation, team, 212–214, 213 Adler, Alfred, 250 algorithms, clinical, 165–166 algorithms, numerical, 8, 73n2, 176 algorithms, problem-solving, 171 Amazon.com, 164 American Association for Artiﬁcial Intelligence, 153 American Psychological Association, 3 analytic derivations, 8 anova, 2, 10, 12 apprenticeship, 130, 136–138 Aristotle, 163, 246, 249 artiﬁcial intelligence (ai): A. Newell and H. A. Simon’s research on, 151–153; boom and winter of, 153–154; in ﬁeld of cognitive science, 86; history of, 147; Mark A. Musen’s study of, xii–

xiii; overview of, 145–147; and performance models, 128–129. See also computer-based learning environments (cbles); computer software; technology artiﬁcial neural networks, 175 associative stage of learning, 156 attention skills, 259–260 attitudinal competencies, 191 attractors, 94–99, 95, 115 autonomous stage of learning, 156 avionics trouble-shooting, 126, 127 Bacon, Francis, 248 balance skills, 259–260 balance theory, xx–xxii Bandura, Albert, 250 base distributions, 49–53, 50–51, 58, 61, 69–72 base rates, 39–42, 68 Bateson, Gregory, 253 Baum, F. L.: The Wizard of Oz, 124–125 Bayesian theory: on bold conjecture and falsiﬁability, 8; current

294 modeling complex systems Bayesian theory (cont.) developments of, 67–68; and estimations of base rates, 40; modeling techniques, 9; and parameter distribution, 49, 53–55, 61; and parameter estimation, 25; and performance models, 19–22; posterior distributions, 31–39, 32–38, 54, 55; and process latency models, 27; and statistical property issue, 64–66; theorem, 19–22 behavior: and intelligent system construction, 157–160, 174–176; and orientations, xxxi–xxxiv; Roger Barker on, xxxv–xxxvi; and team performance, 192, 209, 212 behavioral development, 270n4 behavioral ﬁeld, xix–xx behavior settings, xxxviii–xlii, 282 behavior theory, xxii–xxiii, xxv–xxvi Big Five model, 211–214, 212 biological utility, ix biology, 85, 249, 256–257, 270n4 BioWorld (cble), 132–136, 133, 134, 135, 137 blaming, 269n2 blood-oxygen-level-dependent (bold) response, 43 bodies, 249, 251, 252, 257–260 bold conjecture, 8 bootstrap tests, 108 boundaries, 247, 248, 257–258 brain research, 86 Bruner, Jerome, 250 Buchanan, Bruce, 147 Buddha, 250 Campbell, Don, 257 capacity utilization, 55–56, 61, 69–72 Carnegie Mellon University, 158 Case Builder (cble), 138 categorization, item, 24 causality, 114–115 celiacs, 136, 137

cell membrane, 247 centers, 247, 248, 258–260 change processes, human, 245–249, 253–255, 264–267. See also constructivism chaos, Gestalt perception of, 89–90, 89 chaos theory: in constructivism, 249; ﬁngerprints of, 108–109; frameworks of, xiii; and human lifespan, 253; and nonlinearity, 105; and order, 249; schema of, 10–11. See also complexity studies; dynamic systems theory; order; systems theories chess, 158 choice, xxii, 10–11 circular apparent motion (cam), 110–111, 111 circular causality, 88 clinical assessment, 17–21, 18 clinical decision support, 283–285 clinical paths, 283 closure, xviii coaching, 130–131, 140 cognition, shared, 217, 219–220, 227–228 cognitive-apprenticeship model, 130, 140. See also apprenticeship cognitive-dysmetria hypothesis, 92 cognitive orientation study, 101– 102, 102 cognitive processes: and clinical assessment, 17–21, 18; in diagnostic reasoning, 136; formal theory application of, 14–17; as learning tools, 139, 140; and mathematical theory, xi; model frameworks of, 130–132; monitoring changes in, 17; and motivation theories, 124–125; and neurological mechanisms, 43; and psychopathology, xii, 117– 118; and rule-based systems, 151–153; in schizophrenia, 109– 115; and stress susceptibility,

295 Subject Index 45–46; and team performance, 192; and time variance, 11–12 cognitive science: description of, 86; formal theory of, 2–3; stochastic modeling in, 19, 61–66; and treatment-program efﬁcacy, 66–67 cognitive stage of learning, 155–156 cognitive task analysis (cta), xii, 126–128 Columbia (space shuttle), 186 communication, 209–210, 212, 218 communicative acts, xxxi–xxxiv communities-of-learning (col) model, 130–132, 138, 140 communities of practice (cop), 131–132, 140 community. See social situations competencies, team, 191, 209–211, 210, 214, 221, 224 complexity, cellular, 247 complexity, psychological, xxii– xxviii complexity studies: approaches to, ix–x; and change, 248–249; description of, 249; frameworks of, xiii; history of, xiv–xlii; paradox in human, 252–253. See also chaos theory computationalist theory, 86 computer-assisted software-engineering (case) tools, 176 computer-based learning environments (cbles), xii, 127–129, 136–140. See also BioWorld (cble) computers. See artiﬁcial intelligence (ai); computer-based learning environments (cbles); computer software; technology computer software: as engineering tools, 176; engineers and, 185; and intelligent system construction, 158–163; for problem solving, 171, 173; and rehabilitation decision making,

283. See also BioWorld (cble); kads project; mycin (computer program); Protégé (computer system) conceptual models, 160–162 conditional probabilities, 283 conditional-reasoning measurement technique, 227 conditioned responses, xxvii–xxviii congruence, xxxiii connected knowing, 248 connectivity, 90–92 constraint satisfaction, 161 constructive methods, 171 constructivism, xiii–xiv, 140; deﬁnition of, 250; Michael Mahoney and, 245, 279; recommendations for practice of, 267–269; themes of, 249–253 construct representation, 47 construct validity: assessment of, 67; Bayesian extension of, 53–55; and capacity utilization, 55–56; description of, 46–49; for mixing-distribution parameters, 27–30, 57–59; and model parameters, 60–61; in statistical summaries, 59–60 control parameters, 87–89, 92–93, 98 convergence, 57–59 coordination, 247, 248 Copernicus, 248 coping, 17, 45–46 core ordering processes (cops), 264 coupled oscillations, 201–202 creativity, 262 crisis intervention study, 101–102, 102 cryptarithmetic problems, 151, 151 curriculum modeling, 125–128 Dartmouth College conference, 146 Darwin, Charles, 248–249 decisional control, 45–46 decision-and-choice models, 10–11 decision making: through cognitive

296 modeling complex systems decision making (cont.) task analysis, xii, 126–128; and expertise modeling, 130, 135; in Madonna model, 278; in rehabilitation, 283–285; and team effectiveness, 195, 223 decision trees, 283 deductive systems, 3–7, 9, 10 Deep Blue, 158–159 Deep Thought, 158–159 dendral (computer program), 147 design models, 161–162 diagnosis: computer-based learning tools for, 132–136, 137; decision making in, 126; and establish-reﬁne method, 169; of infectious diseases, 147–150; in psychopathology, 117. See also medical students; medicine diagrams, 283 dialectical contrast, 257–258 Dictionary of Occupational Titles, 227 Digital Equipment Corporation, 153 disorder, 108. See also order distribution moments, 48–49, 52, 56, 58, 69–73 Djerassi, Carl, 147 domain knowledge: in intelligent system construction, 181; in mycin, 150; and problem solving, 165, 168–174, 176–178; rules of, 148 dsm-style diagnostics, 117 dual-coding theory, 23 dynamic dialectical development. See lifelong development dynamic systems theory: and chaos theory, 108, 249; classiﬁcation of models of, 10–12; David Krech on, xxix; and deﬁnition of clients, 258; and diagnosis, 117; frameworks of, xiii; limitations of, 116–117; and modeling effects of stress, 7; and models in clinical cognitive science, 19; nonlinear, 105; patterning

in, 90; and psychopathology, 86–87, 96–99, 102–104, 103, 115–118; stochastic models of, 36, 43, 44. See also chaos theory ecology, psychological, xxxiv–xlii, 282 edging, 260–261, 270n3 Educational Researcher, 125 E-E (environment-environment) unit, xxxv–xlii Einstein’s limiting constant, 7 elimination-by-aspects model, 10, 11 emergent social aspects, xxxvi emotions: Charles Darwin on, 249; lifelong development of, 252; as motivational factors, xxviii– xxx; and order seeking, 257; and self-organization, 251 empirical data: and construct validity, 60; and convergence, 57; in psychopathology, 93–115; and statistical property issue, 64–66; and stochastic model distributions, 48, 49; on team-effectiveness models and frameworks, 218 empirical equivalence, 9, 10 encoding process: and cognitive debilities, 46; and estimations of base rates, 40–42; and individualized parameter estimation, 30–31, 32, 33; and parametervalue mixing distributions, 28, 29; and performance-latency model, 25, 26; and posterior problem of group membership, 34–39 engineering, xii–xiii, 175–178, 180, 181, 280–281 entrainment, 201–202 environment: in apprenticeship models, 138; and constructivism, 250; Kurt Lewin’s model of, xx; in model frameworks,

297 Subject Index 131–132; and problem solving, 157–158; and rehabilitation, xxxv, xli–xlii, 98, 280–281; Roger Barker on, xxxiv–xlii; and self-organization, 87, 90, 93; and stress susceptibility, 45– 46; team, 189, 202, 217, 219–220 equations, 13–14 equilibrium, xviii–xxii, xl–xli Erlang distribution: aesthetic appeal of, 13–14; base rates and distributions, 52–53, 58, 58, 70–72; in Bayesian extension, 53; and capacity utilization, 55–56; description of, 12; parameters of, 26, 30; and process-latency model, 24–27; in statistical summaries of performance, 59–60 erp, 20 establish-reﬁne method, 169–171 events, psychological, xxii–xxviii, xxv evolution, biological, 256–257, 270n4 evolutionary epistemology, 270n4 excitement, 96, 96, 99–104 exergy, 93 exogenous variance, 10–13 experiencing, 260–261, 269n3 expertise: in intelligent system construction, 152–157; models of, 124, 126, 130–131, 133–136, 137, 139–140; paradox of, 156; relevant dimensions of, xii; and team effectiveness, 195 expert systems. See intelligent systems explorations, 260–261 falsiﬁability, 8 feature abstraction, 169–170, 169 feedback: in behavioral ﬁeld, xx; diagnostic, 138; in learning process, 140; and team-effectiveness models and frameworks,

206, 209, 210, 214 Feigenbaum, Ed, 147 Ferguson, Adam, 256 ﬁeld (concept), xxxiv ﬁrst-order change, 253 ﬂow, 256 fmri, 20, 39–42 formal theories: aesthetic appeal of, 13–14; assets of, 6–8; barriers to, 16–17; in clinical cognitive science, 2–3; and cognitive processes, 14–17; demarcations of, 3–6; and strength of predictions, 8–10, 73n3 Frankl, Viktor, 250 Galileo, 248 gamma distributions, 27, 28, 30, 50–51, 52, 57–60, 58, 69–73 “general systems” framework, xiii, xiv Gestalt psychology: and connectivity, 91, 92; on evolution of schizophrenia, 109–115; Fritz Heider on, xv, xvii–xviii, xix–xxi; historic role of, xi; and hysteresis effect, 111; vase-face in, 88–89, 88; written word sequence in, 89–90, 89 global market, 185, 189 goal rule, 150 good ﬁgure concept, xvii good form, xviii–xxi goodness-of-ﬁt tests, 64–66 Gram-negative rod, 170 Granger-causal models, 116. See also time-series method graphic representations, 283 groups: in apprenticeship models, 138–140; cognitive functioning in schizophrenic, 17, 19–22, 25–27, 34–39; Gladstein model of effectiveness of, 203, 203; likelihood of schizophrenic symptoms in, 39–42; and model frameworks, 131–132; and psy-

298 modeling complex systems groups (cont.) chometric measures, 62–65. See also communities-of-learning (col) model; team effectiveness; team performance; teams; teamwork The Growth of the Mind (Koffka), xviii Grundy (computer system), 170 Hayek, Friedrich, 250, 256 Hebb’s law, 91 Heider, Fritz: On Perception, Event Structure, and the Psychological Environment, xv Heraclitus, 250 Herbart, Johann, 250 heroin, 103–104 heuristic classiﬁcation, 2–3, 169–172, 169, 179, 181, 188, 217 hierarchies, 165, 166, 169, 170, 193, 195 horizontal apparent motion (hsam), 111–112 human models, 129, 136–138 Hume, David, 256, 257 hypotheses test, 133–136 hysteresis, 90, 111–112, 114 ibm, 158 ideas, history of, xiii–xlii inattentional blindness experiments, 92 individuals: in behavior settings, xl–xli; cognitive functioning in schizophrenic, 17, 19–22, 25, 30–34, 32, 33, 39–42; communicative behavior between, xxxi– xxxiv; perception and behavior of, xxxv–xxxvi; statistical summaries of performance of, 59–60; and team effectiveness, 203–204, 206, 217–218, 221 infection: diagnosis of, 147–150; inference patterns of, 169–172; and mycin system, 172–174, 173 inference engines, 148–150, 153

informal theories, 8, 73n4 input-process-output (ipo) models: and team adaptation framework, 213, 214–215; and team effectiveness, 200, 205, 206; and team performance measurement, 224; and team processes framework, 210–211, 215–222; uses of, xiii Intel, 185 Intellicorp, 154 intelligent systems: advances of using Internet, 179–180; cognitive processes of developers of, 162–163; deﬁnition of, 146–147; engineering of, 175–178; and expertise transfer, 152–157; history of construction of, 174–176; and ontologies, 165–167; and problem solving methods, 168–171, 173; psychological applications of, 146–147, 155–159, 162, 174–178, 180–181; rule-based, 148; and software programs, 158–163; stages of development of, 160–162, 160. See also artiﬁcial intelligence (ai); Protégé (computer system) interaction, xxxi International Association of Knowledge Engineers, 153 International Joint Conference on Artiﬁcial Intelligence (1985), 153–154 Internet, 164, 177–180 intersensory binding, 113–115, 114 investigator role, 16–17 isomorphism, xix, xxiii–xxv item encoding, 24 item-response theory (irt), 64, 74n8 James, William, 250, 269n3 kads project, 159–162 Kant, Immanuel, 250 Kasparov, Gary, 158

299 Subject Index Kelly, George, 250 knowledge: competencies, 191; as human aspect, 145, 155–156; and intelligent system development, 146–147, 154–159, 162– 171, 178, 180; in production systems, 152–153; tacit, 156, 157. See also domain knowledge knowledge, skills, and attitudes (ksas), 190–191, 211, 221, 224 knowledge, skills, attitudes, and other skills (ksaos), 190–191 knowledge-acquisition system, 167, 176 knowledge-based systems. See intelligent systems knowledge-building tools, 139 knowledge industry, 153–154 Koffka, Kurt: The Growth of the Mind, xviii Kohler, Wolfgang: Physical Gestalten, xvii Landsberg order, 105–107 language production, 256–257 Lao-Tzu, 250 latency distributions: and Bayes’s theorem, 21–22, 24; in cognitive process measurement models, 19–21; in individualized parameter estimation, 31–34, 32–34, 39–42; in parametervalue mixing distributions, 30; and stress, 46. See also processlatency model latency variance, 11–12 leadership, 209–212, 217, 219–221 learning: model construction for, 140; and real-world situations, 132–140, 137; Roger Barker on psychological principles of, xxxvii–xxxviii; stages of, 155–156 learning curves, xxvii–xxviii learning theories, 125–129, 131–132 Lederberg, Joshua, 147

lifelong development, 250, 252–253 life space, xix, xx, xxxiv Likert scales, 227 limit cycle attractors, 95 lisp programming language, 148, 153 living settings, 281 L-mode, 5 local independence, 10 Los Angeles ca, 153–154 Madonna model, 275–285, 277; elements of, 278–285; interdependence of facets of, 276 Mahoney, Michael, 245 Mambac, 5 management tools, 139 Mandeville, Bernard, 256 maple (computer-algebra program), 16 mappings, 172, 176–177, 180, 181 mathematica (computer-algebra program), 16 mathematical theory, xi, 16, 73n5, 87, 94 matlab Optimization Toolbox, 41 Maxcov, 5 Maxeig, 5 Maxslope, 5 McCarthy, John, 146 mean latency models, 7–8 medical students, 135–139, 137 medicine, xii–xiii, 103–105, 147–150, 165–167, 165 memory search paradigms, 22–24 memory-trace theory, 24 metacognition, 124 method ontology, 171–172 mind-body dualism, 249, 250 mirror time, 261 mixture models: base distributions in, 49; design of, 31; latency distributions in, 57–59, 69–73; parameters of, 27–30, 34, 37, 42, 46, 49, 50–51, 52; and performance measurement, 19, 20;

300 modeling complex systems mixture models (cont.) and posterior problem of group membership, 35–39, 36–38; statistical property issue in, 64–66 mle(v), 54–55 modeling: computer v. human, 128–129, 136–138; curriculum of, 125–128; diagnostic reasoning in, 132–136; in intelligent system construction, 155–157, 160–162, 180; and ontologies, 163–167; and performance, 123–124, 128–132; and rulebased systems, 151–153; social context of, 138–140; subjects for, 124–125, 127, 139–140; for team effectiveness, 186–188, 195, 199–202 molar learning theory, xiv–xv molecular learning theory, xiv–xv morphine, 103–104 motion-induced blindness (mib), 112–113, 113 motivation: and behavior settings, xxxix; characterization of, xii; and cognitive processes, 124– 125; deﬁnition of, 92; Edward L. Walker’s theory of, xxii; and E-E units, xxxvi; emotional and perceptual factors of, xxviii– xxx; in self-organization, 92–93 motivational processes, 86 mri, 43–44 multidimensional scaling, 12, 24 mycin (computer program): conceptual model of, 160–161; design model of, 161–162; domain knowledge in, 172–174, 173; inﬂuence of, 152, 153; problem solving in, 169–170, 169; as rule-based system, 147–150, 149 naming, 269n2 naturalistic settings, xiii, 283–285 Nebraska Symposium, xi, xii, xiv–xlii

neural nets, simple and complex, xxiv Neural Process Complexity, xxiii– xxv neurological mechanisms, 16, 17, 20–21, 43, 73n5 neuropsychological symptom patterns, 100–101 neuroscience, 90–92, 117–118, 174– 175, 249 Newell, A., 151–153, 158 Newton’s universal law of gravity, 6–7, 16 nonlinearity, 94, 105–116; ﬁngerprints of, 108–109, 115–116 nonsummativity, 91 nucleation, 247 nurses, 97, 126–128 nurses’ observation scale (noise), 97 obsessive-compulsive disorder, 260–261 Occidental’s Piper Alpha platform, 186 off-balance clients, 258 Ohio State University, 168–169 O*Net, 227 On Perception, Event Structure, and the Psychological Environment (Heider), xv ontologies, 163–168; and Internet, 179–180; and problem solving, 171–174; protocols of, 165–166, 165, 176; and user interfaces, 166–167, 167 open systems theory, 200 opioids, 103–104 opportunities, xxxviii optimal complexity, xxiv–xxviii order: and centering skills, 259–260; and chaos theory, 249; excessive, 258, 260–261; history of, 246–248; in human change processes, 250, 251, 264–267; in lifelong development, 252; in nonlinear systems, 105–108; in

301 Subject Index schizophrenia treatment and rehabilitation, 106–108, 107; seeking of in problem solving, 255; teleological and teleonomic, 255–258. See also chaos theory order, Gestalt perception of, 89–90, 89 order effect, 107, 107 order parameters, 88, 92 organization of behavior, xxii orientations, xxxi–xxxiv, 209–210, 212, 218, 219 parameters: construct validity of distribution, 46–61; deﬁnition of, 19, 73n1; estimation of in diagnostic-group level analysis, 25; estimation of in individuals, 30–31; mixing-distribution, 27– 30, 42, 57–59; of process-latency models, 47–49. See also control parameters participation, xv–xvi, xxxv, 279, 281–282 Pascal distribution, 58–59 patterning, ix, xvii–xviii; in dynamic systems theory, 90; and nonlinearity, 105–115; and selforganization, 86, 88–90; systematic interrelations in, 99–104; trajectories and attractors in, 94–99; transitions in, 104 pattern levels, 255 perception, xvi–xviii, xxviii–xxx, xxxv–xxxvi, 109–115 performance: construct validity paradigms affecting, 61; and expertise studies, 126; and fmri technology, 43–44; in mixing distributions, 57–58; model selection and function, 123–124, 128–132; models of cognitive, 17–21, 61–66, 139–140; parameters of in process-latency models, 47–49; and parametervalue mixing distributions, 27;

sample of in Bayes’s theorem, 21–22, 25, 74n7; of schizophrenics, 110; statistical summaries of prototypical, 59–60. See also team performance performance-latency distributions, 17, 19 performance-model parameters, 17 personality, 251, 256–257. See also self personal revolution, 253–254 person model, xx person-object relationships, xxxi– xxxiv pharmacology, 85 phase transitions, 90, 91, 104, 105, 107 Physical Gestalten (Kohler), xvii physical-symbol-systems premise, 86 Piaget, Jean, 250 Plato, 246–247 plug-ins, 177 points attractors, 95 Poisson distributions, 28–29, 30, 50, 52, 55–56, 57, 70–73 Popper, Karl, 257 Positive and Negative Symptom Scale (panss), 110–114 posterior latency distributions, 31, 34–36, 66 posterior probability, 34–42 Postman-Bruner hypotheses, xxxii p-o-x system, xxi predictions, 3–9 primacy of the abstract, 256 probabilities: cumulative and multiple distributions of, 43–45, 44; estimation of in schizophrenia, 39–42; for selected base distributions, 70; in stochastic models, 10–11 problems, 254 problem solving: A. Newell and H. A. Simon’s research on, 151– 153; and artiﬁcial neural

302 modeling complex systems problem solving (cont.) networks, 175; and BioWorld, 133–136, 137; and cognitive task analysis, xii, 126–127; components of, 171; in computerbased learning environments, 140; and domain ontologies, 176–178; and environment, 157–158; and intelligent system construction, 155–157, 163, 168– 174, 179–181; and kads project, 159; medical tutorials of, 136– 139; methods of in infectious organism-identiﬁcation, 173; and model selection, 129; and psychotherapy, 254–255 process-completion potential, 56, 57 process-latency model, 25–27; base distributions in, 49–53, 50–51, 58–59, 58, 69–73; Bayesian extension of, 53–55; in construct validity, 61; parameters of, 47–49; and posterior problem of group membership, 35–39, 36–38; and statistical summaries of performance, 59–60 process level, 255, 257 production rules, 155 production systems, 150, 152–153, 180–181 proﬁciency models, 124 Prospector (computer system), 153, 170 Protégé (computer system), 165, 166–167, 167, 176–178, 180, 181 protocols: analysis of in rule-based systems, 151; and ontologies, 165–166, 165, 176 prototypical dynamics, 94–94, 95 psychiatry, 90, 269n2 Psychological Issues, xv psychology: classiﬁcation in, 249; and complexity studies, 249; description of ﬁeld of, 85; diagnosis and treatment efﬁcacy in, 134; and dynamic systems

theory, 90; ecological and community, xxxiv–xlii, 282; and intelligent system development, 146–147, 155–159, 162, 174–178, 180–181; and molar and molecular learning theories, xiv–xv; “name and blame,” 269n2; Paul Meehl on, 1, 2; “positive,” 279; and team performance, 192, 219–220 psychometric validity, 46, 53–54, 62–65, 74n8. See also construct validity psychopathological impairment, 94–99 psychopathology: connectivity in, 92; diagnosis of, 117; empirical research in, 93–115; and Gestalt psychology, xi, 114; inconsistencies in research of, 115–116; and motivation, 92–93; nonlinear patterns in, 105–115; scope of ﬁeld of, 85–86; symptom patterns of, 100–101; systematic interrelations in, 99–104; therapy, 116–117; trajectories and attractors, 94–99 psychopharmacology, 103–104 psychosocial crises, 101 psychotherapy: and applications of dialectical development strategies, 258–262; and dynamic systems theory, 90; language production in, 256–257; and order in pattern formation, 105–108; and problem solving, 254–255; and time-series methods, 116–117 psychotic episodes, 99–104 psychoticity, 96, 96, 99–104 punctuated-equilibrium model (pem), 205, 205 punishments, xxxii pure death system with a linear death rate, 51, 52 Pythagoras, 246–247

303 Subject Index qualitative research methodology, 3, 6, 9 quantum shift, 254 reciprocal determinism, 250 rehabilitation: and complex-process modeling, 275–285; comprehensive, 285n1; decision making in, 283–285; efﬁcacy assessment of, 66–67; and environment, xxxv, xli–xlii, 98, 280–281; illustrative aspects of, 284; and living settings, 281; and Madonna model team characteristics, 278; and order, 106–107, 107; participation in, xv–xvi, xxxv, 279, 281–282; patient and family engagement in, 279, 281–282; practical model for, xi; program evaluation of, 282–283; of schizophrenics, 97–98, 98; task affordances in, 98; and technology, 280–281, 283–285; and therapeutics, 279–280; and time-series methods, 116–117 repeated activation, xxv response parameters, 18, 74n6 responsibilities, xxxviii retention, 257, 260n4 rewards, xxxii Rich, Elaine, 170 rule-based systems: as cognitive models, 151–153; difﬁculties associated with, 174; rise of, 147–150; successful applications of, 153–154, 173, 180 schizophrenia: and Bayes’s theorem, 21; cognitive patterns in, 109–115; connectivity in, 92; and individualized parameter estimation, 31, 32, 33; likelihood estimation of symptomgroup base rates of, 39–42; measurements of progress in, xii; and memory-search para-

digms, 22–24; and motivation, 92–93; nonlinearity of, 108–109; order in treatment of, 106–108, 107; and parameter-value mixing distributions, 27–29; and posterior problem of group membership, 34–39; stimulus-encoding dysfunction in, 17–46; and stochastic modeling, 2–3, 61–62; symptom patterns of, 99–104, 110–115; systematic interrelations in, 99–104; trajectories and attractors in, 94–99 Schizophrenia Process Study, 100–101 Schlumberger, 153 Schopenhauer, Arthur, 249, 250 science, 262 science instruction, 132–136 Sechenov, Ivan, 257–258 second-order change, 253–254 selection, 257, 260n4 self: behavior regulation of, xxii; deﬁnition of, 250; relationship to, 261; stability of, 251 self-monitoring, 124, 128, 131, 134–135 self-organization: biological, 251; connectivity in, 90–92; control parameters of, 92–93; description of, 87; in dynamic systems theory, 87; motivation in, 92–93; nonlinear, 105–115; parameters of, 87–89; pattern formation by, 86, 88–90, 94–115; phase transitions in, 104 Semantic Web, 179, 180, 181 sensitivity analyses, 283 sexual reproduction, 247 Sherlock (cble), 127 Shortliffe, Ted, 147 sicun (computer tutor), 127–128 Simon, H. A., 151–153, 158 simplicity, xvii–xviii, xxi situated action, 157–158 Smith, Adam, 256

304 modeling complex systems social network analysis, 218–220 social processes: in rehabilitation, xli–xlii, 279, 282; Roger Barker on, xxxv–xxxvi; symbols in, 248, 250, 252 social situations, xxx–xxxiv, 124– 125, 138–140 sociology, 85, 138 sociotechnical systems theory, 200 solution reﬁnement, 170 spatialized psychology, xx sri International, 170 stability. See order standard state, xviii, xx, xxi standing-center exercises, 259–260 Stanford University, 147 Stark, uss, 186 statistical tools for prediction, 283 statistics: in formal theory, 4, 7; of performance samples, 59–60; and stochastic modeling, 64–66 stimulus complexity, xxiii, xxvi stimulus-encoding process. See encoding process stochastic models: characteristics of, 10–12; in clinical cognitive science, 19, 61–66; of cognitive performance, 17, 20; distributions of, 48, 49, 61; and effects of stress, 7; and formal theory, 4; multiple processes in, 44–45; and neurological measurement, 36, 43, 44; and statistical property issue, 64–66; in study of schizophrenia and psychological stress, 2–3 stress, psychological: and empirical performance conﬁgurations, 60; measuring of, 17, 27; predictive models of, 10; and stochastic modeling, 2–3, 7; susceptibility to, 45–46 stroboscopic apparent motion (sam), 111–112, 112 structural-equation causal modeling (sem), 4–5

structure: of cognitive process models, 19; concept of, xvi; and Kurt Lewin’s models, xxi stuck clients, 258, 260–261 superadditivity, 7, 9 Switzerland, 103 Symbolics, 154 symbols, 248, 250, 252 synergetics, 87, 91, 205. See also selforganization systematic interrelations, 94, 99–104 systems theories: characteristics of, xiv; and evolution of mental diseases, 94, 98; and ipo models, 200; multisectored, xxxvi. See also chaos theory target levels of inference, 18, 74n6 task affordances, 88–89, 98 task performance: in kads project, 160–161; modeling of, 129, 131–132; and problem solving methods, 168; in schizophrenia, 110. See also cognitive task analysis (cta) tasks, 204, 206–211, 214, 215, 217– 218 taskwork, 209, 221 team effectiveness: conceptualizations of models and frameworks of, 202, 203–215, 203; deﬁnition of, 193–196; dynamics of, 195, 215, 220–221; integrative framework of, 205–206, 206, 215–222, 216; integrative research of, 226–228; and ipo models, 200; mental models of, 195, 212, 217, 219–220; and model focus, 202; normative model of, 204; overview of models and frameworks of, 186–188, 196–199, 197–198, 215; and performance, 193, 223–224; and rhythm of task accomplishment model, 210–211, 211; selection of models and

305 Subject Index frameworks for study of, 199– 200; and skill competencies, 209–210, 210; and stafﬁng, 225; synthesized model of, 206–209, 207, 219; team-evolution andmaturation model of, 208, 209; and training team members, 224–225. See also team performance team performance: appraisal systems of, 196; in Big Five model, 212; conceptual models of, 202, 203–215, 203; deﬁnition of, 192–193; integrative framework of, 206, 217; measuring of, 223–224, 227; norms of, 191; overview of models of, xiii; punctuated-equilibrium model of, 205, 205; in team-adaptation model, 214–215; and team effectiveness, 194. See also team effectiveness team processes, 205, 210–212, 211, 215–222, 224 teams: characteristics of, 217–218, 278; deﬁnition of, 189; history and nature of, 185–186; science of, xiii; stafﬁng of, 225; viability of, 194 teamwork: and competencies, 224; deﬁnition of, 189–191, 209; dynamic interdependence in, 191; in integrative team-effectiveness framework, 221, 222; research on, 186; skills of, 190–191 technology, xii, 280–281, 283–285. See also artiﬁcial intelligence (ai); computer-based learning environments (cbles); computer software Teknowledge, 154 teleological order, 255–258 teleonomic order, 255–258 tension, xx Thales, 246

therapists, 262, 267–269, 269n3 therapy. See psychotherapy; rehabilitation therapy-session reports, 105–106 Thurstone scaling, 227 Thurstonian discriminal-difference size dispersions, 23 time: and probability distributions, 11–12, 15; and psychological events, xxii–xxviii; and team performance measurement, 223, 224. See also latency distributions time-series method: attractors and trajectories in, 95–98; model of, 100; and process attributes of schizophrenia, 108; and systematic interrelations, 99–104; and therapy, 116–117 Today’s Evaluation of Psychopathology (tep), 100 training, 206, 209, 224–225 trajectories, 94–99 transformation, 253–254 translational research, x–xi U.S. military, 153 Vahinger, Hans, 250 variation, 257, 260n4 vector autoregression (var) approach, 99–101, 104, 115, 116 verbal reasoning, 3–4, 6 vertical apparent motion (vsam), 111–112 Vico, Giambattista, 250 Vincennes, uss, 186 visualization tools, 139 visual processing, xvii–xix vocational rehabilitation, 106–107 Vygotsky, Lev, 250 waterfall models, 161–162 Weibull distribution, 51, 52, 55, 58, 58, 70–72

306 modeling complex systems Whitehead, Alfred North, 246 withdrawal, 96, 96, 99–104 The Wizard of Oz (Baum), 124–125 work structure, 216, 218, 219

World Health Organization, xxxv World Wide Web. See Internet Yahoo!, 164

Author Index

Page numbers in italics refer to illustrations. Acton, B., 190 Aguirre, G. K., 43 Aleong, P., 132 Alexander, P. A., 124, 125 Alley, W. E., 127 Almond, R. G., 129 Ancona, D., 186, 201 an der Heiden, U., 117 Anderson, J. A., 156 Anderson, J. R., 125, 130 Andreasen, N. C., 92 Ardison, S., 218 Arnold, M., 90 Arpaia, J. P., 94 Ashby, F. G., 5, 11, 18, 19, 27, 52, 56, 69 Averill, J. R., 45 Azevedo, R., 126 Bachant, J., 174 Bacon, Francis, 1 Bailey, D. E., 221

Baker, A. J., 5 Baker, T. B., 5 Balakrishnan, N., 11 Bamber, D., 60 Bandettini, P. A., 43 Bandura, A., 73n4, 123, 250 Banerjee, S., 105 Barab, S. A., 131 Barker, Roger, xiv, xv, xxxiv–xlii, 282 Barrick, M. R., 225 Batchelder, W. H., 9, 19, 40, 60, 68 Batsell, R. R., 10 Baum, F. L., 124 Baur, N., 102 Beard, R. L., 187, 206 Bélair, J., 117 Belbin, R. M., 186 Bell, B., 227 Bell, B. F., 186 Bell, B. S., 187, 192 Benjamins, V. R., 179 Benn, K. D., 59 Berger, J. O., 19 Berners-Lee, T., 179 Best, C., 178

308 modeling complex systems Beyerlein, M., 186 Beyerlein, S., 186 Blanco, C., 92 Bleuler, E., 92 Boksman, K., 8, 43 Bollen, K. A., 10 Bonneh, Y. S., 112 Borman, W. C., 214 Borowski, E. J., 73n1 Borrill, C. S., 228 Borwein, J. M., 73n1 Bottger, P. C., 194 Bowker, G. C., 164 Boynton, G. M., 43 Braff, D. L., 108 Braithwaite, R. B., 3, 5, 6 Brannick, M. T., 223 Brass, D. J., 218 Braun, C., 90 Braun, H. A., 12 Brenner, C. A., 62 Breyer, F. J., 129 Broga, M. I., 62, 64 Brooks, F. P., 159 Brown, A. L., 130, 131 Brown, E., 12 Brown, J. S., 130 Brown, Joe, xlii Brown, K., 225 Browne, M. W., 64 Brunswik, Egon, xvi, xix, xxxv, xxxviii, xl, xli, xlii Bryk, A., 5 Buchanan, B. G., 147, 148 Buckingham, M., 279 Burke, C. J., x Burke, C. Shawn, xiii, 125, 186, 190, 212, 213, 214, 215, 220, 221, 223, 278 Busemeyer, J. R., 7, 10, 11, 19, 74n6 Campbell, D. T., 46, 188 Campbell, J. P., 187, 192, 193, 221, 225, 226 Campion, M. A., 187, 190, 193, 206– 209, 215, 219, 225

Campione, J. C., 130 Cannon-Bowers, J. A., 131, 187, 190, 191, 223, 224, 225 Carbotte, R. M., 62 Carter, J. R., 8, 12, 20, 24, 46, 59, 60, 62, 68, 74n5 Carter, R., 86 Casti, J. L., 10, 15 Chabris, C. F., 92, 112 Chandrasekaran, B., 163, 168 Chapman, J. P., 62 Chapman, L. J., 62 Chechile, R. A., 9 Checile, R. A., 60 Chi, M. T. H., 124, 126 Chong, C.-L., 186, 201 Chudowsky, N., 125 Ciompi, L., 90, 108 Clancey, W. J., 157, 158, 163, 169, 170 Clark, A., 86 Clark, J. M., 14 Clifton, D. O., 279 Cobb, P., 125 Cohen, J., 62, 64, 194 Cohen, S. G., 193, 221 Collins, A., 130 Colonius, H., 11 Confrey, J., 125 Conrad, K., 109 Converse, S. A., 189 Cook, T. D., 188 Cooke, N. J., 227 Cooper, C. L., 186, 218 Cooperman, A., 112 Cordova, D. I., 125 Cramer, R. D., 10 Cronbach, Lee J., xi, 46, 47 Crubézy, M., 166, 168, 172 Cuesta, M. J., 95 Dailey, R. C., 218 Dauwalder, J.-P., 86, 117 Davis, J., 129 Davison, M., 105 de Jong, T., 129 Delucchi, K. L., 68

309 Author Index DeMeuse, K. P., 194 De Roure, D., 179 Derry, S. J., 128 D’Esposito, M., 43 Dickinson, T. L., 189, 190, 191, 201, 209, 210, 212–214, 215, 220 Dickson, M. W., 189, 201 Dirac, P. A. M., 13 diSessa, A., 125 Dobbins, G. H., 218 Doob, J. L., 4 Draper, S. W., 178 Dreyfus, H. L., 152, 155 Driskell, J. E., 218, 225 Dubouloz, P., 110 Duda, R. O., 153, 170 Duffy, T. M., 131 Dunnette, M. D., 192 Eby, L. T., 218 Edmondson, A., 186, 187 Eggan, G. M., 126, 127 Eisner, E. J., 201 Ellery, S., 130 Embretson, W. S., 46, 47 Emery, F. E., 200 Emrich, H., 117 Engel, S. A., 43 Ericsson, K. A., 123, 124, 126, 151 Eriksson, H., 168 Ernst, N., 178 Evans, M., 8, 12, 28, 49 Faremo, S., 129, 135–136, 138 Farr, M. J., 124 Feigenbaum, E. A., 147, 152 Feller, W., 52, 70 Fergerson, R. W., 166, 178 Fiﬁc, M., 9, 19, 49 Fiore, M. C., 5 Fiore, S. M., 227 Fisher, Ronald, xi Fiske, D. W., 46 Fitts, P. M., 155 Fleishman, E. A., 190, 201, 202, 203, 220

Fleiszer, D., 126 Fodor, J. A., 156 Fokas, A. S., 43 Forster, M. R., 64 Fouladi, R. T., 2 Frackowiak, R. S. J., 43 Fredrickson, B. L., 279 Freedman, D. Z., 13 Friston, K., 43 Futrell, D., 194 García Pérez, M. A., 66, 68 Garcia-Toro, M., 92 Gardner, R. C., 28, 46, 64 Geddes, K. O., 70 Gennari, J. H., 166, 171, 172 George, L., 23 Gersick, C. G., 186, 205, 208, 209, 215 Gheorghiu, V., 110 Gibbon, J., 52 Gibson, James, xvi, xlii Giel, R., 94 Gilden, D. L., 13 Gilmore, G. C., 110 Gladstein, D. L., 203, 206, 215 Glaser, R., 124, 125 Glass, D. C., 7 Glass, L., 117 Gleick, J., 13 Glickman, A. S., 186, 208, 209 Globus, G. G., 94 Glover, G. H., 43 Gluck, K., 127 Goberson, T., 129 Golden, R. R., 5, 63 Goldstein, I. L., 201 Gonzalez, A., 92 Goodman, P. S., 203 Goodwin, Gerald F., xiii, 125, 218, 278 Gott, S. P., 127, 130 Grawe, K., 90, 102, 105 Gray, C. M., 90 Greeno, J. G., 125 Greer, J. E., 132

310 modeling complex systems Gregson, R. G., 10 Grochowski, S., 110 Grosso, W. E., 166 Guarino, N., 164 Guerrera, C., 132 Gully, S. M., 202 Guzzo, R. A., 189, 201 Hackman, J. R., 186, 187, 194, 200, 204, 215, 219, 221 Haemmig, R., 103 Haken, H., 87, 91, 93, 111 Hamilton, V., 60 Harris, B., 52 Harrison, L., 43 Hashimoto, Y., 108 Hastings, N., 8 Haynes, S. N., 27, 46, 53, 57, 60 Heeger, D. H., 43 Heider, Fritz, xiv, xv–xxii, xxxiv, xxxviii, xlii Heller, K., 282 Hendler, J. A., 179 Herold, D. M., 218 Herrenkohl, L., 129 Herrnstein, R. J., 16 Higgs, A. C., 187, 206–209 Highgate-Maynard, S., 22, 24 Hinsz, V. B., 220 Hinton, G., 175 Hintzman, D. L., 6 Hockley, W. E., 23 Hoffmann, H., 97, 101, 106 Hogan, R., 218 Hohenschutz, C., 117 Hollenbeck, J. R., 186, 200, 225 Holmes, P., 12 Homan, S. M., 40 Huber, M. T., 12 Huberman, A. M., 6 Huettel, S. A., 43 Huselid, M. A., 225 Hyman, R. B., 110 Ilgen, D. R., 186, 187, 200 Jackson, S. E., 225

Jacobshagen, N., 101, 103 James, W., 124 Jetté, J., 8, 46, 56 Johnson, D., 186 Johnson, L., 129 Johnson, M., 200 Johnson, M. A., 132 Johnson, N. L., 11, 49 Johnson, P. E., 155, 156, 174 Jonassen, D. H., 129 Jones, G., 127 Jones, R. G., 225 Jones, T., 129 Jorenby, D. E., 5 Joshi, S. W., 59 Jundt, D., 200 Junghan, U., 110, 113 Kahn, R. L., 200, 225 Kahneman, D., 174 Kant, Immanuel, 1 Karabatsos, G., 36 Katz, D., 200, 225 Kay, J. J., 93 Keeping, E. S., 71 Kelso, J. A. S., 87, 90 Kendall, D., 213, 214 Kenny, J. F., 71 Kerzel, D., 87 Kiekel, P. A., 227 Kirk, R. E., 10 Klein, C., 190, 228 Klein, G., 195 Klein, George S., xv, xvi Klein, K., 192, 193 Klimoski, R., 225 Kline, M., 6 Kline, R. B., 5 Knapp, B., 60 Knight, R. A., 62, 64 Knoblich, G., 87 Koçak, H., 10, 73n2 Koch, S., ix Koffka, Kurt, xvi, xvii, xix, xx Köhler, W., xvii, 91 Kotz, S., 11

311 Author Index Koyfman, A., 62 Kozlowski, S. W. J., 186, 187, 192, 193, 202, 224, 225 Kozma, R. B., 129 Krackhardt, D., 218 Krech, David, xxix Krieg, J. C., 12 Krueger, C. W., 162 Kruse, P., 110, 111 Kuhn, T. S., 1 Kukde, M. P., 10 Kuncel, N. R., 187, 221, 225 Kupper, Zeno, xi, 94, 95, 97, 99, 101, 106, 114, 134 Kushmerick, N., 158 Lajoie, Suzanne P., xii, 124, 126, 127, 128, 129, 132, 133, 135, 136, 138, 139, 278, 279, 281 Landsberg, P. T., 105 Lane, R. D., 43 Lanius, R. A., 43 Lassila, O., 179 Lave, J., 130 Lavigne, N. C., 132, 133–134, 135 Lawler, E., 192 Lawson, A. E., 132 Lederberg, J., 147 Leeper, Robert Ward, xiv, xv, xxviii–xxx Lees, M. C., 45 Lefebvre, L. A., 10 Lehrer, R., 125 LePine, J. A., 218, 225 Lepper, M. R., 124, 125 Lesgold, A., 124, 126, 127, 129 Lewin, Kurt, xvi, xix, xx, xxi, xxix, xxxiv, 211 Lewkowicz, D. J., 113 Lieb, R., 94 Liebowitz, J., 145 Lindenmayer, J.-P., 110 Lindsay, R. K., 147 Link, S. W., 6 Lloyd, B. B., 117 Logan, D., 126, 127

Logan, Frank, xxviii Lopez, S. J., 279 Lu, J., 139 Luce, R. D., 11 Luke, D. A., 40 Lütkepohl, H., 99 Lyons, W., 156 Lysaker, P. H., 62 MacCallum, R., 5 Mach, Ernst, xvii Mahoney, Michael J., x, xiii, 90, 256, 258, 261, 269n3, 279 Mahoney, S. M., 258 Major, D. A., 186 Mandelbrot, Benoit B., 13 Manifold, V., 60 Marinakis, V., 43 Marks, M. A., 186, 190, 191, 201, 210, 211, 215, 220, 221, 224 Marley, A. A. J., 11 Marr, D., 74n6 Marx, N., 129 Mathieu, J. E., 186, 190, 211 Matussek, P., 109 Mayer, R. E., 125, 129 Maze, J., 105 McCarthy, G., 43 McCarthy, J., 146 McCarty, T., 46, 59 McClelland, J., 73n5 McDermott, J, 153, 168, 171, 174 McFall, R. M., 4, 7, 9 McGill, W. J., 52 McGrath, J. E., 200, 202, 215, 218 McIntyre, R. M., 190, 191, 201, 209, 210, 212–214, 215, 220, 224 McPherson, J. A., 190 Mechsner, F., 87 Medsker, G. J., 187, 193, 206–209 Meehl, Paul E., xi, 1, 2, 4, 5, 8, 9, 46, 47, 63 Meier, R., 110 Menzies, T., 158 Metz, K. E., 132 Michotte, A., 114

312 modeling complex systems Miles, M. B., 6 Miller, C. M., 10 Miltner, W., 90 Milton, J., 117 Minsky, M. L., 146 Mislevy, R. J., 129 Monahan, J., 282 Mooijaart, A., 68 Moreno, R., 129 Morgan, B. B., Jr., 186, 208, 209, 215 Morrison, D. G., 72 Morrison, M. S., 10, 46 Mothersill, K., 20 Motowidlo, S. J., 214 Mount, M. K., 225 Mullen, B., 218 Munsie, S. D., 132 Murdock, B. B., Jr., 23 Musen, Mark A., xii, 129, 166, 168, 171, 172, 173, 177, 178, 278, 281, 285 Myung, I. J., 64 Nakamura, C., 138 Nason, E. R., 202 National Academy of Science, 125 Nelson, P., 279 Neubert, M. J., 225 Neufeld, Richard W. J., xi, 5, 7, 8, 9, 10, 11, 12, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 38, 45, 46, 52, 53, 55, 56, 59, 60, 62, 64, 66, 68, 74n5, 278, 285 Neuman, G. A., 225 Newborn, M., 158 Newcomb, Theodore, xiv, xv, xx, xxx–xxxv Newell, A., 86, 151, 158, 168 Newman, S. E., 130 Newton, Isaac, 16 Nichols, P., 127 Nicholson, I. R., 24, 25, 38, 46 Nicolis, G., 87 Nienhuis, F. J., 94 Nieva, V., 202, 203, 215 Nisbett, R. E., 156

Nissen, H. W., ix Norman, D. A., 178 Nosofsky, R. M., 12 Noy, N. F., 166, 178 Nozawa, G., 45 O’Brien, W. O., 46, 60 O’Donnell, B. F., 62 O’Leary, D. S., 92 Orasanu, J., 189 O’Shea, P. G., 218 Paivio, A., 14, 23 Palincsar, A. S., 129 Papper, E. M., 193 Paradiso, S., 92 Parzen, E., 12 Patil, G. G., 59 Paulus, M. P., 108 Pavlekovic, B., 110 pdp Research Group, 73n5 Pea, R. D., 129 Peacock, B., 8 Pearson, Karl, xi Pellegrino, J., 125 Penny, W., 43 Peralta, V., 95 Perkins, D. N., 129 Pezard, L., 109 Pfeifer, R., 86 Pﬁster, H., 94 Phillips, W. A., 92, 110 Piasecki, T. M., 5 Pierce, L., 213, 214 Place, E. J., 110 Plantinga, E. P. O., 155 Ployhart, R. E., 225 Pokorny, R., 127 Polking, J. C., 10 Polkinghorne, J., 13 Prendergast, K. A., 145, 153 Pressing, J., 10 Price, S. J., 201 Priest, H. A., 228 Prigogine, I., 87 Prince, C., 223

313 Author Index Prinz, W., 87 Puerta, A. R., 168 Ramsay, J. O., 74n8 Rappaport, A., 10 Rappaport, J., 282 Raudenbush, S. W., 5 Reder, L., 125 Reeves, T. C., 129 Regoczei, S., 155 Reick, A., 202, 203 Reiman, E. M., 43 Reynolds, M. L., 12 Riefer, D. M., 9, 60, 68 Robertson, I. T., 186 Rochester, N., 146 Roe, R. M., 11, 20 Rosch, E., 117 Ross, S. M., 49, 52, 70 Rössler, O. E., 95 Rothenﬂuh, T. E., 171 Rubin, E., 88 Ruderman, M. N., 225 Ruel, H. J. M., 226 Rumelhart, D., 73n5 Ruscio, A. M., 5 Ruscio, J., 5 Russell, J., 129 Sagi, D., 112 Salas, Eduardo, xiii, 125, 131, 186, 187, 189, 190, 191, 206, 208, 209, 211, 212, 213, 214, 215, 218, 220, 223, 224, 225, 227, 228, 278, 281, 285 Salomon, G., 129 Salva, J., 92 Santor, D. A., 74n8 Sarason, S. B., 282 Schauble, L., 125 Scheier, C., 86, 105, 108, 113 Schemmer, M. F., 201 Schiffman, S. S., 12 Schneider, B., 225 Schneider, E. D., 93 Schreiber, G., 159, 161

Schuler, D., 113 Schuster, P., 94 Scott, T. I., 70 Sego, D. J., 186 Seligman, M. E. P., 279 Senge, P. M., 195 Sergent, J., 12 Shahar, Y., 168 Shannon, C. E., 146 Shiﬂett, S., 201, 220 Shimojo, S., 113 Shiner, J. S., 105 Shortliffe, E. H., 147, 148, 153, 171 Shute, V. J., 127 Sibbald, P. R., 105 Silverstein, S. M., 62, 64, 92, 109, 110 Simon, H. A., xiii, 86, 125, 126, 151, 168 Simons, D. J., 92, 112 Sims, D. E., 190, 212 Singer, J. E., 7 Singer, W., 90 Sipe, W. P., 225 Slooff, C. J., 94 Smith, D., 225 Smith, E. M., 202 Smith, K. G., 187 Smith, P. L., 44 Smith, S. S., 5 Smith-Jentsch, K. A., 190 Smolin, L., 13 Snyder, C. R., 279 Soloway, Elliott, 129 Sommerville, I., 161, 176 Song, A. W., 43 Spaulding, W. D., 100 Staab, S., 164 Staddon, J. E. R., 2, 6, 8, 73n4 Stadler, M., 110, 111 Stagl, Kevin C., xiii, 125, 186, 213, 214, 220, 223, 228, 278 Star, S. L., 164 Steﬁk, M., 145 Steiger, J. H., 2 Steinberg, L. S., 129 Steiner, I. D., 218

314 modeling complex systems Sternberg, S., 12, 18, 22 Stevens, M. J., 190, 225 Stewart, G. L., 225 Storey, M.-A., 178 Stout, J. C., 19 Studer, R., 164 Suchman, L. A., 157 Sundstrom, E., 194, 200 Sutherland, J. L., 147 Takane, Y., 12 Tannenbaum, S. I., 187, 189, 190, 194, 200, 205, 206, 215, 219 Taub, E., 90 Tindale, R. S., 220 Tjosvold, D., 187 Tollenaar, N., 68 Tolman, E. C., xxix Tomarken, A. J., 5 Townsend, J. T., 4, 6, 8, 9, 10, 11, 13, 18, 19, 27, 28, 44, 45, 49, 52, 56, 69, 74n6 Trist, E. L., 200 Tschacher, Wolfgang, xi, 86, 93, 94, 95, 99, 101, 102, 103, 105, 106, 108, 110, 113, 114, 117, 134 Tu, S. W., 168, 171 Tuckman, B. W., 208, 209 Tversky, A., 10, 174 Uhlhaas, P. J., 109, 110 Unsworth, K. L., 228 van Joolingen, W. R., 129 Van Petten, C., 43 Van Zandt, T., 66 Varela, F. J., 90 Viken, R. J., 4 Vinacke, W. E., x Vollick, D., 8, 19, 24, 27, 66 Vollrath, D. A., 220 Volpe, C. E., 190

von Bertalanffy, L., 193, 200 Vygotsky, L. S., 132 Wagenmakers, E., 64 Waldorp, L., 64 Walker, Edward L., xiv, xv, xxii– xxviii Waller, N. G., 5 Wang, Y., 20 Weick, K., 192 Weinberg, G. M., 178 Weinstein, N. D., 7, 60 Weissbein, D., 225 Wenger, E., 130, 131 Wenger, M. J., 6, 9, 27, 45, 52, 56, 69 Wertheimer, Max, xvii West, M. A., 187, 228 Wicker, A. W., xlii Wiersma, D., 94 Wiggins, J. S., 47 Wilkie, T. V., 132 Williams, S. M., 130 Williamson, P., 9, 24, 25, 27, 66 Wilson, T. D., 156 Wilt, A., 62 Winston, P. H., 145, 153 Wiseman, J., 129, 136, 138 Wittchen, H. U., 94 Witte, H., 90 World Health Organization, xv, xxxv, 282 Wright, B. D., 23 Wright, J., 225 Yetton, P. W., 194 Young, F. W., 12 Young, R. M., 152 Zaccaro, S. J., 186, 190, 201, 211, 220 Zarahn, E., 43 Zeisig, R. L., 190 Zuboff, S., 145