COMPUTATIONAL

SOCIAL PSYCHOLOGY

Computational Social Psychology showcases a new approach to social psychology

that enables theorists and researchers to specify social psychological processes in
terms of formal rules that can be implemented and tested using the power of
high-speed computing technology and sophisticated software. This approach
allows for previously infeasible investigations of the multidimensional nature of
human experience as it unfolds in accordance with different temporal patterns
on different timescales. In effect, the computational approach represents a redis-
covery of the themes and ambitions that launched the field over a century ago.
This book brings together social psychologists with varying topical interests
who are taking the lead in this redirection of the field. Many present formal
models that are implemented in computer simulations to test basic assumptions
and investigate the emergence of higher-order properties; others develop models
to fit the real-time evolution of people’s inner states, overt behavior, and social
interactions. Collectively, the contributions illustrate how the methods and tools
of the computational approach can investigate, and transform, the diverse land-
scape of social psychology.

Robin R. Vallacher is Professor of Psychology at Florida Atlantic University, a

Research Associate in the Center for Complex Systems, University of Warsaw,
Poland, and a Research Affiliate in the Advanced Consortium on Cooperation,
Conflict, and Complexity at Columbia University. He has been a visiting scholar
at University of Texas at Austin, University of Bern (Switzerland), Max Planck
Institute for Psychological Research (Germany), and University of Montpellier
(France). Dr. Vallacher has authored or edited seven professional texts, and has
published over 100 book chapters and journal articles on a wide range of topics
in social psychology, including self-concept, self-regulation, social judgment, close
relationships, prejudice and discrimination, sport psychology, social justice, and
intergroup conflict. In recent years, he and his colleagues have adapted concepts
and methods from the study of nonlinear dynamical systems in the natural sci-
ences to the investigation of personal, interpersonal, and societal processes.

Stephen J. Read is Professor of Psychology at the University of Southern

California, head of the social psychology area, and a Fellow of the Association
for Psychological Science and the Society for Experimental Social Psychology.
Dr. Read has edited three books and published over 100 journal articles and
book chapters on person perception, causal reasoning, attachment theory,
decision-making, use of interactive media (DVD and game) for changing risky
sexual behavior, personality, and the neurobiology of risky decision-making. He
has developed and published neural network models of person perception, causal
reasoning, cognitive dissonance, personality and motivation, and risky decision-
making. He has also worked on computational models of personality in intelligent
agents and models of the role of narrative representations in military decision-
making. Recently, he has focused on integrating work on the neurobiological
bases of risky decision-making with neural network models of the neural systems
involved in risky decision-making.

Andrzej Nowak is Professor of Psychology at Florida Atlantic University,

University of Warsaw, Poland, where he directs the Center for Complex Systems,
and Warsaw University of Humanities and Social Sciences. He is also a Fellow
at the European Center for Living Technologies, and a Research Affiliate in the
Advanced Consortium on Cooperation, Conflict, and Complexity at Columbia
University. Dr. Nowak has authored or edited 16 professional texts, and has pub-
lished over 100 book chapters and journal articles on a wide range of topics in social
psychology, including social influence, social change, self-concept, self-­regulation,
social judgment, sport psychology, intergroup conflict, and psychological aspects
of new media.Working with his colleagues, he has adapted concepts and methods
from the study of complex systems and nonlinear dynamics in the natural sciences
to the investigation of personal, interpersonal, and societal processes. He publishes
in psychology, physics, and interdisciplinary journals.
Frontiers of Social Psychology
Series Editors:
Arie W. Kruglanski, University of Maryland at College Park
Joseph P. Forgas, University of New South Wales

Frontiers of Social Psychology is a series of domain-specific handbooks. Each volume

provides readers with an overview of the most recent theoretical, methodological,
and practical developments in a substantive area of social psychology, in greater
depth than is possible in general social psychology handbooks. The editors and
contributors are all internationally renowned scholars whose work is at the cutting
edge of research.
Scholarly, yet accessible, the volumes in the Frontiers series are an essential resource
for senior undergraduates, postgraduates, researchers, and practitioners, and are suitable
as texts in advanced courses in specific subareas of social psychology.

Published Titles
Social Metacognition, Briñol & DeMarree
Goal-Directed Behavior, Aarts & Elliot
Social Judgment and Decision Making, Krueger
Intergroup Conflicts and Their Resolution, Bar-Tal
Social Motivation, Dunning
Social Cognition, Strack & Förster
Social Psychology of Consumer Behavior, Wänke

Forthcoming Titles
For continually updated information about published and forthcoming titles in
the Frontiers of Social Psychology series, please visit: https://www.routledge.com/
psychology/series/FSP
COMPUTATIONAL
SOCIAL PSYCHOLOGY

Edited by Robin R. Vallacher, Stephen J. Read,

and Andrzej Nowak
First published 2017
by Routledge
711 Third Avenue, New York, NY 10017
and by Routledge
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
Routledge is an imprint of the Taylor & Francis Group, an informa business
 2017 Taylor & Francis
The right of Robin R.Vallacher, Stephen J. Read, and Andrzej Nowak to be
identified as the authors of the editorial material, and of the authors for their
individual chapters, has been asserted in accordance with sections 77 and 78
of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or
utilized in any form or by any electronic, mechanical, or other means, now
known or hereafter invented, including photocopying and recording, or in
any information storage or retrieval system, without permission in writing
from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered
trademarks, and are used only for identification and explanation without
intent to infringe.
Library of Congress Cataloging in Publication Data
Names: Vallacher, Robin R., 1946- editor. | Read, Stephen J., editor. |
Nowak, Andrzej (Andrzej Krzysztof ), editor.
Title: Computational social psychology / edited by Robin R. Vallacher,
Stephen J. Read, and Andrzej Nowak.
Description: New York : Routledge, 2017. | Series: Frontiers of social
psychology | Includes bibliographical references and index.
Identifiers: LCCN 2017000667| ISBN 9781138951648 (hb : alk. paper) |
ISBN 9781138951655 (pb : alk. paper) | ISBN 9781315173726 (ebook)
Subjects: LCSH: Social psychology.
Classification: LCC HM1033 .C636 2017 | DDC 302—dc23
LC record available at https://lccn.loc.gov/2017000667

ISBN: 978-1-138-95164-8 (hbk)

ISBN: 978-1-138-95165-5 (pbk)
ISBN: 978-1-315-17372-6 (ebk)

Typeset in Bembo
by Swales & Willis Ltd, Exeter, Devon, UK
CONTENTS

List of Contributors x
Prefacexii
Acknowledgmentsxvi

Overview1

  1 Rethinking Human Experience: The Promise of

Computational Social Psychology 3
Robin R. Vallacher, Andrzej Nowak, and Stephen J. Read

PART I
Intrapersonal Dynamics 13

  2 Virtual Personalities: A Neural Network Model of

the Structure and Dynamics of Personality 15
Stephen J. Read, Vita Droutman, and Lynn C. Miller

  3 Using Connectionist Models to Capture the Distinctive

Psychological Structure of Impression Formation 38
Brian M. Monroe, Tei Laine, Swati Gupta, and Ilya Farber
viii Contents

  4 The Whole Elephant: Toward Psychological Integration

of the Individual as a Complex System 61
Jennifer Rose Talevich

  5 Computational Modeling of Health Behavior 81

Mark G. Orr and Daniel Chen

PART II
Interpersonal Dynamics 103

  6 Interaction-Dominant Dynamics, Timescale Enslavement,

and the Emergence of Social Behavior 105
Brian A. Eiler, Rachel W. Kallen, and Michael J. Richardson

  7 Computational Temporal Interpersonal Emotion Systems 127

Emily A. Butler, Jinyan Guan, Andrew Predoehl,
Ernesto Brau, Kyle Simek, and Kobus Barnard

  8 Mapping Co-Regulation in Social Relations through

Exploratory Topology Analysis 144
Jonathan E. Butner, Arwen A. Behrends, and Brian R. Baucom

  9 Dynamics of Synchrony and Joint Action 178

Kerry L. Marsh

10 From Interaction to Synchronization in Social Relations 208

Robin R. Vallacher and Andrzej Nowak

PART III
Collective Dynamics 229

11 Simulating the Social Networks in Human Goal Striving 231

James D. Westaby and DaHee Shon

12 Computational Models of Social Influence and

Collective Behavior 258
Robert J. MacCoun
 Contents  ix

13 Modeling Cultural Dynamics 281

Yoshihisa Kashima, Michael Kirley, Alexander Stivala,
and Garry Robins

PART IV
Transforming Social Psychology 309

14 Models Are Stupid, and We Need More of Them 311

Paul E. Smaldino

15 Big Data in Psychological Research 332

David Serfass, Andrzej Nowak, and Ryne Sherman

16 The Future of Computational Social Psychology 349

Andrzej Nowak and Robin R. Vallacher

Index368
CONTRIBUTORS

Kobus Barnard, University of Arizona

Brian R. Baucom, University of Utah

Arwen A. Behrends, University of Utah

Ernesto Brau, Boston College

Emily A. Butler, University of Arizona

Jonathan E. Butner, University of Utah

Daniel Chen, Biocomplexity Institute of Virginia Tech

Vita Droutman, University of Southern California

Brian A. Eiler, Center for Cognition, Action and Perception, University of Cincinnati

Ilya Farber, Institute of High Performance Computing (IHPC): Agency for Science,
Technology, and Research (A*STAR), Singapore
Jinyan Guan, University of Arizona

Swati Gupta, Callaghan Innovation, New Zealand

Rachel W. Kallen, Center for Cognition, Action and Perception, University of Cincinnati

Yoshihisa Kashima, University of Melbourne

Michael Kirley, University of Melbourne

Tei Laine, Université Grenoble Alpes

Kerry L. Marsh, University of Connecticut

 Contributors  xi

Robert J. MacCoun, Stanford University Law School

Lynn C. Miller, University of Southern California

Brian M. Monroe, University of Alabama

Andrzej Nowak, Warsaw University and Warsaw School of Social Sciences and Humanities

Mark G. Orr, Biocomplexity Institute of Virginia Tech

Andrew Predoehl, University of Arizona

Michael J. Richardson, Center for Cognition, Action and Perception, University of

Cincinnati
Stephen J. Read, University of Southern California

Garry Robins, University of Melbourne

David Serfass, Florida Atlantic University

Ryne Sherman, Florida Atlantic University

DaHee Shon, Columbia University

Kyle Simek, University of Arizona

Paul E. Smaldino, University of California, Merced

Alexander Stivala, University of Melbourne

Jennifer Rose Talevich, University of Southern California

Robin R. Vallacher, Florida Atlantic University

James D. Westaby, Columbia University

PREFACE

Human behavior is complex and often difficult to understand or predict, making

its meaning a preoccupation throughout history for lay people and scholars alike.
Some of the ambiguities and mysteries are localized, referring to specific actions
that can presumably be traced to momentary motives or immediate external influ-
ences. At other times, the focus is on the bigger picture—the driving forces in
human nature responsible for action generally. Insight into both the local causes
and general features of thought and behavior underwent a dramatic transforma-
tion with the advent of the scientific method and its adaptation to the unique
subject matter of human experience. The history of scientific social psychol-
ogy is brief compared to that of other disciplines such as physics and chemistry,
however, and it is not entirely clear that the field has hit upon the adaptation of
science that is best suited to the challenge the field faces.
This book examines this issue, noting the limitations of the approach that
has dominated theory construction and research for most of social psychology’s
existence as a field of scientific inquiry. Rather than lamenting this state of affairs,
we showcase an alternative approach that has emerged in recent years because of
significant advances in computer technology and its increased availability. These
developments enable theorists and researchers to specify social psychological pro-
cesses in terms of formal rules that can be implemented and tested using the
power of high-speed computing technology and sophisticated software. Because
of its reliance on computer technology and its expression in formal rules, we refer
to this approach as computational social psychology.
Computational social psychology casts a very broad net within which spe-
cific theories and research platforms find expression. This state of affairs is to
be expected at this stage in its development. Whereas the methods and tools
 Preface  xiii

for traditional research (e.g. analyses of variance, significance testing) have been
available for over a century, the powerful computers, sophisticated software, and
electronic communications technology defining the computational approach
have been around for a mere generation and have only recently found their way
into the conduct of research in social psychology. Tremendous progress has been
made in reframing social psychological phenomena in computational terms, but
this work is still in its nascent stage, so it is unclear at this point how the selec-
tive pressures of theoretical and applied gain will sort out the winners and losers.
To showcase the computational approach and demonstrate its added value,
we have brought together social psychologists with varying topical interests who
are taking the lead in this redirection of the field. Operating within the computa-
tional framework, each contributor brings to bear a unique method focusing on a
distinct facet of human experience. Collectively, the contributions illustrate how
the diverse topical landscape of social psychology can be investigated with the
benefits of the methods and tools associated with the computational approach.
In so doing, the book goes beyond providing new strategies for how to investi-
gate social processes to provide direction for what should be investigated. Human
experience is inherently complex and dynamic, defining features that were
emphasized in the formative years of social psychology. Because complexity and
dynamism have proven difficult to investigate with traditional methods, how-
ever, social psychological theory and research have tended instead to focus on
simple causal relations and one-step processes. Computational social psychology,
by contrast, is explicitly concerned with the multidimensional nature of human
experience that unfolds in accordance with different temporal patterns on differ-
ent timescales. In effect, the computational approach represents a rediscovery of
the themes that launched the field over a century ago.

Overview of Chapters
The contributions we have assembled each provide a unique way of showcasing
the computational approach to social psychology. Most present formal mod-
els that are implemented in computer simulations to test basic assumptions and
investigate the emergence of higher-order properties, but some develop models
to fit the real-time evolution of people’s inner states, overt behavior, and social
interactions. The diversity of methods is essential at this stage in the development
of computational social psychology, but it is also testament to the flexibility of
this approach in tackling how psychological systems behave.
In an introductory chapter, we provide an overview of the computational
approach and its contrast to the traditional approach that has defined social psy-
chological research for decades. We emphasize that this emerging paradigm not
only provides precise tools for assessing social processes, it also shines the light on
important features of human experience that have gone largely unexamined in
xiv Preface

traditional approaches—features that were recognized as crucial in the early years

of social psychology but were not investigated for want of appropriate tools and
methods.
The next three sets of chapters are organized to reflect expanding levels of
experience, from processes that reflect intra-individual dynamics to those that
operate at the group and societal level. In Part I, the emphasis is on intrap-
ersonal dynamics. Read, Droutman, and Miller (Chapter 2) present a neural
network model that captures in parsimonious fashion the interplay of personal-
ity structure and dynamics. Monroe, Tei Laine, Gupta, and Farber (Chapter 3)
illustrate how connectionist modeling can represent the basic factors at work in
impression formation. Talevich (Chapter 4) discusses the integrative potential
of complex systems generally, and demonstrates this potential for modeling dif-
ferent attachment styles. Orr and Chen (Chapter 5) show how computational
modeling can be used to represent how people manage (and mismanage) their
health behavior.
The chapters comprising Part II offer different, though complementary, means
of modeling interpersonal dynamics. Eiler, Kallen, and Richardson (Chapter 6)
illustrate a central feature of interaction-dominant dynamical systems: how local
interactions between social agents give rise to global patterns that subsequently
feed back to constrain future micro- and macroscopic social processes. Butler,
Guan, Predoehl, Brau, Simek, and Barnard (Chapter 7) investigate the emer-
gence of shared emotions among interacting individuals and present a model
that captures the essence and detail of this temporal trajectory. Butner, Behrends
and Baucom (Chapter 8) employ topological analysis to capture how individu-
als come to regulate one another in their social relations. Marsh (Chapter 9)
presents evidence showing how the multidimensional context in which thought
and behavior are embedded promotes the emergence of synchronization and
joint action among interacting individuals. Vallacher and Nowak (Chapter 10)
use coupled dynamical systems to model the synchronization of overt behavior
and internal states that underlies the evolution of social relations in the course of
social interaction.
Part III consists of three chapters that illustrate the added value of compu-
tational models for providing insight into collective dynamics. Westaby and Shon
(Chapter 11) present a simulation model that shows how individual goal-striving
is expressed in the context of dynamic social networks. MacCoun (Chapter 12)
shows how many classic principles and phenomena in collective behavior can be
represented as various special cases of a single computational model. Kashima,
Kirley, Stivala, and Robins (Chapter 13) show how basic dimensions of culture
variation can be modeled by the iteration of a few simple formal rules.
Part IV addresses the potential of the computational approach for transform-
ing social psychology. Smaldino (Chapter 14) discusses how the very simplicity of
computational models is what enables them to capture the essential features that
 Preface  xv

give rise to the complexity of human experience at all levels of psychological reality.
Serfass, Nowak, and Sherman (Chapter 15) discuss the recent emergence of Big
Data as a tool for understanding social processes, and consider the expanding
role that this approach is destined to have in the years to come. In a concluding
chapter, Nowak and Vallacher (Chapter 16) forecast future developments in the
computational approach and suggest how these emerging trends are likely, in
combination with traditional approaches, to provide a new paradigm for social
psychology.
ACKNOWLEDGMENTS

To an important extent, this book has its genesis in the pre-conference, Dynamical
Systems and Computational Modeling in Social Psychology, that was initiated at the
annual meeting of the Society of Personality and Social Psychology (SPSP) in
2012. This pre-conference has attracted increased attention since that time and is
now a regular feature of the annual meeting of the SPSP. Many of the contribu-
tors to the pre-conference are represented in this book.
We wish to thank Joseph Forgas and Arie Kruglanski, the co-editors of the
Frontiers of Social Psychology series, for the invitation to prepare this volume. They
appreciated the potential of the computational approach to become a dominant
paradigm for the field, and they felt that the time was right to showcase the vari-
ety of strategies that have emerged in recent years. They entrusted us to assemble
the set of contributions to provide this showcase. Their foresight, support, and
trust in this endeavor are greatly appreciated.
Andrzej Nowak acknowledges the support of a grant from the Polish
Committee for Scientific Research [DEC-2011/02/A/HS6/00231].
Overview
1
RETHINKING HUMAN
EXPERIENCE
The Promise of Computational
Social Psychology

Robin R. Vallacher, Andrzej Nowak,

and Stephen J. Read1

Social psychology adopted the scientific method over a hundred years ago, and
in that time has generated an enormous literature concerning every conceivable
aspect of human experience, from internal mechanisms of mind and emotion to
basic principles that underlie societal functioning. Although the field consists of
numerous theories, some of which are mutually contradictory, researchers for
the most part have not called into question the means by which such theories
are generated. After all, for psychological research to be published in the field’s
top journals, it must meet stringent standards of scientific practice. The results
emanating from psychology labs over the past decades are thus beyond reproach
and can be assumed to capture important features of how people think, feel, and
behave. If there is conflict among theoretical interpretations, all one need do is
perform further research to settle the conflict. Like every other scientific field,
psychology is built upon the resolution of such conflicts, and through this dia-
lectic provides progressive understanding, with increases in both the nuance and
generality of the phenomena it is designed to investigate.
Or so we thought. In recent years, the meaningfulness of effects generated
in psychology labs, and hence the value of theories based on such efforts, has
come under intense scrutiny. Most notably, the findings of the Open Science
Collaboration (OSC, 2015) have raised concerns that what are presented as solid
generalizations about psychological processes may instead be unreliable effects
observed under idiosyncratic laboratory conditions, often analyzed with inad-
equate statistical methods. These concerns have generated considerable discussion
and created a crisis of confidence among many in the field. Not surprisingly,
there has been pushback against the conclusions of the OSC (e.g., Gilbert, King,
Pettigrew, & Wilson, 2016). But even if the OSC findings are eventually quali-
fied, one can ask whether the standard paradigm of psychological research is
4  Vallacher, Nowak, and Read

sufficient for generating important insights into the complexity and dynamism
at the heart of human experience in real-world contexts. People’s internal states
and overt actions are embedded in a multidimensional context, unfold over time
in accordance with temporal patterns on various timescales, and often display
reciprocal rather than unidirectional causality. Standard experimental designs that
isolate specific relationships and concentrate on the immediate effect of inde-
pendent variables are not designed to capture these defining features of human
experience.
In recent years, a viable alternative to the traditional approach has begun to
gain acceptance in social psychology. This approach is made possible by the
explosive growth in computer technology and sophisticated software that enables
researchers to identify fundamental principles that incorporate the complexity
and dynamism of human thought, emotion, and behavior. Because this approach
uses computers and rule-based algorithms to model and quantify psychological
processes, we refer to it as computational social psychology.2 The contributions to
this volume, though different in their respective topics of interest and specific
research strategies, are all built on the assumption that formal models can be
generated to provide quantitative rather than merely qualitative understanding of
social psychological phenomena.

Overview of the Chapter

Our aim in this chapter is to describe the computational approach and high-
light its added value for theory construction and verification. The first section
describes the qualitative changes in research that have emerged in recent years as a
result of new technologies and statistical methods. The new approaches currently
co-exist alongside traditional research methods in social psychology, but there
is reason to suspect that the shift toward computational methods will accelerate
as they become increasingly familiar to psychologists, particularly to those who
have grown up in an era of high-speed computers, electronic communication,
and social networks.
The second section describes significant changes in theoretical focus engen-
dered by the computational approach. As in other areas of science, the tools that
are available constrain the phenomena that can be investigated. Traditional social
psychological methods have served a useful purpose in addressing many impor-
tant topics but they are not designed to capture some defining features of human
experience, many of which, ironically, were the focus of social psychology in its
formative years last century—well before the advent of electronic calculators, let
alone high-speed computers. The technology and statistical methods available
today allow psychologists to gain insight into the complexity and dynamism of
human experience that figured so prominently in the insights of such luminaries
as Kurt Lewin and William James.
 Rethinking Human Experience  5

In a concluding section, we describe the integrative potential of computa-

tional social psychology. This approach is clearly heuristic, generating new lines
of research and opening the door on a host of important phenomena. But the
approach also holds potential for identifying basic properties that underlie top-
ics that have traditionally been conceptualized in very different terms. Indeed, a
guiding vision for computational researchers is the development of a foundational
science consisting of a small set of basic principles that provide conceptual inte-
gration for the fragmented topical landscape that characterizes social psychology.
The fragmentation issue has been a persistent sore point in the field, fueling
a crisis in the 1970s that was never resolved and which continues to provoke
consternation among theorists and researchers today.

Methods of Computational Social Psychology

At one level, computational social psychology simply means gaining knowledge
about the subject matter of social psychology using computers. This potential
is manifest in several ways. Modern computer technology, first of all, has made
available new and extensive sources of data. The internet, Big Data, social media,
and smart phones provide an unimaginable amount of information concerning
people’s attitudes and values, the structure of social contacts and social relations,
and the spread of ideas, rumors, fads, and awareness of events. This frees research-
ers to go beyond simple laboratory studies to extract large volumes of data from
the real world. This source of information, moreover, is becoming increasingly
representative of people’s spontaneous thoughts, attitudes, desires, and openness
to social influence as computers, smart phones, and social media have become the
personal tools of the majority of people worldwide.
Beyond providing a wealth of data concerning people, the growing availability
of computer software enables social psychologists to formulate their theories in
terms of rules expressed in computer code, to check the consistency of their theo-
retical assumptions, to observe how the model behaves over time, and to examine
results for unexpected and emergent properties. Computers also offer unprec-
edented analytical power for advanced statistics and automatic model building.
Finally, because of their visualization capability, computers enable psychologists
to see patterns in complex data sets that might not be apparent with recourse to
statistics or verbal descriptions (Nowak, Rychwalska, & Borkowski, 2013).
In more general terms, computational social psychology refers to a revolu-
tionary new way of doing research. Traditionally, theories in social psychology
have been formulated in natural language, stating the qualitative relations among
variables of interest—which variables influence other variables, and the direc-
tion of such influence. Natural language, however, does not offer the precision
necessary to describe the quantitative nature of the relationship among vari-
ables, especially when the relations involve complex interactions and nonlinear
6  Vallacher, Nowak, and Read

dependencies. In contrast, describing social theories in terms of computational

models enables researchers to combine many factors and to capture the complex
relations among them.
Computer simulations are especially useful in this regard because there is
virtually no limit to the number of factors that can be investigated, nor is there a
limit to the number and complexity of relationships among such factors that can
be assessed (see, e.g., Gilbert & Troitzsch, 2011; Liebrand, Nowak, & Hegselman,
1998; Nowak & Vallacher, 1998; Read & Miller, 1998; Smith & Conrey, 2007).
Computer simulations provide an ideal compromise between the freedom but
imprecision of natural language and the daunting requirements of mathematical
formulas. For this reason, computer simulations have been described as the third
pillar of the scientific method, alongside experimentation and theory building
(cf. Ostrom, 1988).
Within the computational framework, of course, empirical data are essential to
provide assumptions for a model and to test a model’s predictions. In contrast to
traditional empirical methods, which are designed to show that a relationship is
probably not due to chance, the computational approach describes the relation-
ship among variables in quantitative terms, and can do so with a vast number of
cases rather than with small convenience samples. In this endeavor, the compu-
tational approach uses statistics in a fundamentally different way. Instead of t-tests
and ANOVAs designed for significance testing, there is an emphasis on advanced
methods for model-fitting, including structural equation modeling, log linear
models, and multilevel modeling. The computational approach is also reflected in
new statistics, such as boot-strapping methods and machine learning.
These advanced tools are proving useful in time series methodology, which
can be employed to characterize the temporal patterns in thought, emotion,
and behavior (e.g., Boker & Wenger, 2007; Butner, Gagnon, Geuss, Lessard, &
Story, 2014; Freeman & Ambady, 2010; Hollenstein, 2013; Kuppens, Oravecz, &
Tuerlinckx, 2010; Spivey & Dale, 2006; Vallacher, Van Geert, & Nowak, 2015).
This strategy differs from the computer simulation approach in a fundamen-
tal way: whereas the strategy of computer simulations is to formulate rules and
observe their operation as they are iterated over time, the strategy in time series
analyses is to discover the rules that produce the structure and temporal patterns
in empirical data. Clearly, the two strategies work together synergistically. The
rules discovered in time series can provide the basis for the rules implemented in
computer simulations, and the results of simulations can provide insight into the
properties to look for in time series of empirical data.

Theories in Computational Social Psychology

Computational social psychology is more than a recipe for how research should
be conducted. In a fundamental sense, it enables researchers to address features of
social life that were largely off limits within traditional experimental designs—in
 Rethinking Human Experience  7

effect, it is a blueprint for what can be investigated. For one thing, more complex
theories can be formulated and tested. Theorists no longer need to be bound
by independent, moderating, mediating, and dependent variables. Instead, they
can formulate theories in terms of multiple feedback loops, reciprocal causality,
complex interactions, and temporal changes in these properties.
It is particularly noteworthy that the topics and issues rendered open to inves-
tigation within the computational approach represent central concerns that were
at the forefront during the field’s formative years (e.g., Cooley, 1902; James,
1890; Lewin, 1936; Mead, 1934). Psychologists have known all along that human
experience is embedded in a multidimensional field of forces and displays inter-
nally generated change on various timescales, but they did not have the tools
necessary to capture the complexity and dynamism of mind and action. Largely
due to the sophistication and availability of computer technology and software,
we can now observe behavior unfold in ecologically meaningful settings and we
can implement the rules for such behavior in computer simulations that reveal
the consequences of these rules on relevant time scales. In a concrete sense, the
availability of modern technological advances enables psychologists to finally
address questions that have gone unanswered since they were raised in the early
20th century.
This potential has been realized in recent years in the adaptation of complex-
ity science and nonlinear dynamical systems to social processes (e.g., Guastello,
Koopmans, & Pincus, 2009; Vallacher, Read, & Nowak, 2002). This perspec-
tive introduces such concepts as self-organization, emergence, attractors, fractals,
and nonlinearity, which are central to the study of complex systems in the nat-
ural sciences (e.g., Schuster, 1984; Strogatz, 1994; Waldrop, 1992; Wolfram,
2002; Weisbuch, 1992), to theory construction and research in social psychol-
ogy (Nowak & Vallacher, 1998; Read & Miller, 1998; Vallacher & Nowak,
1994, 1997, 2007). The methods and tools made available by the computational
approach have been indispensable in this endeavor.
Computer simulations employing agent-based modeling, for example, have
shown how higher-order structures emerge in both mental systems and social
systems through the self-organization of the system’s basic elements. With respect
to mental systems, computer simulations have shown how the mutual influence
among specific thoughts and feelings give rise to higher-order mental states such
as attitudes (e.g., Read, Vanman, & Miller, 1997; Monroe & Read, 2008), self-
concepts (e.g., Nowak, Vallacher, Tesser, & Borkowski, 2000), stereotypes (e.g.,
Queller, 2002), and enduring moods (Thagard & Nerb, 2002). With respect to
social systems, simulation work has revealed how the local interactions among
individuals in a society promote the emergence of public opinion (e.g., Nowak,
Szamrej, & Latané, 1990), cooperation versus competition in social dilemmas
(e.g., Messick & Liebrand, 1995), social norms surrounding mating (e.g., Kenrick,
Li, & Butner, 2003), societal change (e.g., Nowak & Vallacher, 2001), and the
survival versus collapse of minority opinions (e.g., Jarman et al., 2015).
8  Vallacher, Nowak, and Read

In another simulation approach, social relationships are modeled as the progres-

sive synchronization of individuals’ internal states (e.g., temperament, personality).
Individuals are represented as separate dynamical systems (logistic equations) with
unique control parameters signifying their internal states (Nowak, Vallacher, &
Zochowski, 2005). The degree of coupling (mutual influence) between individuals
and the similarity in the initial settings of their respective control parameters
are varied, and the individual systems are given the chance to synchronize the
dynamics of their overt behavior over many trials (iterations of the program).
After synchronization is achieved, the coupling is reduced to zero and the program
assesses how well the two systems remain synchronized.
The results show that synchronization in dynamics is achieved very quickly
under strong coupling, but that the systems diverge when mutual influence is
broken. Under moderate coupling, synchronization is achieved more slowly, but
it persists after the coupling is broken because the systems converge on a common
control parameter (internal state) to compensate for the lack of strong mutual
influence. In recent years, synchronization has emerged as a prominent way to
conceptualize social relations (e.g., emotional coordination); the computer simu-
lations of this process identify the optimal degree of influence that facilitates the
development of lasting synchrony between individuals.
Time series methodology, meanwhile, has been employed to identify dynam-
ical properties in the stream of thought (e.g., Delignières, Fortes, & Ninot,
2004; Vallacher et al., 2015). James (1890) clearly saw value in focusing on the
flow of thought, feeling that it provided a better characterization of mental
process than did the central tendency of thought collapsed over time (e.g., an
average judgment), but he pretty much left it at that. The tools for measuring
the dynamical properties of thought simply did not exist in his time. Because
of the concepts and tools adapted from nonlinear dynamical systems in recent
years, researchers can now extract meaningful information from the temporal
trajectories of thought.
Research employing the mouse paradigm (cf. Vallacher, Nowak, & Kaufman,
1994) illustrates this strategy. Participants first verbalize, in private, their thoughts
and feelings regarding a topic (e.g., a relationship partner, a social issue, them-
selves) for several minutes and then listen to a recording of their narrative. As they
hear themselves talk, they use the computer mouse to adjust the position of the
cursor on the monitor to indicate the evaluation of the topic conveyed at each
point in the narrative. Because the cursor’s position is tracked on a moment-
to-moment basis, participants’ stream of thought can be displayed as a temporal
trajectory of evaluation. Analytical tools are then applied to the trajectory to reveal
the dynamics of participants’ thinking.
This strategy has proven useful in capturing basic dynamical properties,
including attractors, repellers, multistability, and fractals in the temporal patterns
of thought, as well as individual differences in these properties. For example,
individual differences in self-concept clarity (Campbell, 1990) are reflected in
 Rethinking Human Experience  9

the landscape of attractors and repellers in people’s self-evaluative narratives

(Wong, Vallacher, & Nowak, 2016). People with higher clarity (signaling a
well-integrated self-concept) have positive self-evaluation attractors and weaker
self-esteem repellers in their trajectories of self-evaluation, whereas those with
lower clarity have less positive attractors and stronger repellers when they reflect
on themselves. Fractal structure also has been observed in people’s self-evaluation
narratives (Wong, Vallacher, & Nowak, 2014) and in their stream of thought
regarding close friends (Vallacher et al., 1994).

The Integrative Potential of Computational

Social Psychology
The current consternation regarding the state of social psychology is some-
what reminiscent of the so-called “crisis” in social psychology in the 1970s
(e.g., Gauld & Shotter, 1977; Gergen, 1978; Harré & Secord, 1972). The con-
cerns at that time were partly philosophical in nature, questioning whether
human behavior was deterministic and whether logical positivism in general,
and reductionism in particular, were appropriate for understanding psychologi-
cal phenomena. But concern was also voiced regarding the proliferation of
localized “mini-theories” devoted to different topics and that were devoid of
contact with one another. The field was fragmented, lacking an agreed-upon
conceptual or empirical foundation. Lacking a way out of the dilemmas that
were identified, researchers maintained the status quo in their approach to psy-
chological phenomena. The crisis thus eventually abated without any change in
the conduct of “normal” psychological research.
This is not surprising in view of how normal science operates. As Kuhn (1970)
observed, established paradigms are highly resistant to challenges, maintaining
their status when problems are exposed. In social psychology, as in other dis-
ciplines, the implicit attitude seems to be that “a bad theory is better than no
theory at all.” For a paradigm shift to occur, an alternative to normal science must
be forwarded that resolves the problems of the existing paradigm and provides a
better depiction of the phenomena of interest.
Computational social psychology holds promise as the needed alternative.
The properties that provide the underpinnings and focus of computational social
psychology—intrinsic dynamics, self-organization, coordination, attractors and
repellers—characterize all levels of psychological reality. At the intrapersonal
level, thoughts and emotions display patterns of change on fairly short timescales
and tend to become progressively more coherent due to their mutual influence.
At the interpersonal level, the dynamics of interacting individuals tend to become
progressively coordinated, enabling them to co-regulate one another’s feelings,
thoughts, and actions. At the collective level, group dynamics are characterized
by the synchronization of moods, attitudes, norms, and behavior—both in local
circumstances (e.g., deindividuation, groupthink) and with respect to societal
10  Vallacher, Nowak, and Read

norms, preferences, and values. And at each level, a shared tool kit is available for
identifying and modeling the phenomenon of interest.
By viewing social processes through the lens of computational models, then, it
may be possible to establish commonalities among topics as seemingly distinct as
attitudes, close relations, and social change. Beyond providing redirection for the
field, computational social psychology holds potential for integrating a field that
is acknowledged as fragmented and in need of such integration. We are not there
yet, but there is reason to be optimistic that the scalability of the computational
approach will someday create the foundational science that the field has lacked
all these years.

Notes
1 Andrzej Nowak acknowledges the support of a grant from the Polish Committee for
Scientific Research [DEC-2011/02/A/HS6/00231].
2 Nowak, Vallacher, and Burnstein (1998) coined this term to describe the approach of
computer simulations in investigating social processes. “Computational social psychology”
as employed in this book encompasses a wider range of approaches.

References
Boker, S. M., & Wenger, M. J. (2007). Data analytic techniques for dynamical systems in the
social and behavioral sciences. Mahwah, NJ: Lawrence Erlbaum.
Butner, J., Gagnon, K. T., Geuss, M. N., Lessard, D. A., & Story, T. N. (2014). Using
topology to generate and test theories of change. Psychological Methods, 20, 1–25.
Campbell, J. D. (1990). Self-esteem and clarity of the self-concept. Journal of Personality and
Social Psychology, 59, 538–549.
Cooley, C. H. (1902). Human nature and the social order. New York: Scribner.
Delignières, D., Fortes, M., & Ninot, G. (2004). The fractal dynamics of self-esteem and
physical self. Nonlinear Dynamics in Psychology and Life Sciences, 8, 479–510.
Freeman, J. B., & Ambady, N. (2010). MouseTracker: Software for studying real-time
mental processing using a computer mouse-tracking method. Behavior Research Methods,
42, 226–241.
Gauld, A., & Shotter, J. (1977). Human action and its psychological investigation. London:
Routledge & Kegan Paul.
Gergen, K. J. (1978). Toward generative theory. Journal of Personality and Social Psychology,
36, 1344–1360.
Gilbert, D., King, G., Pettigrew, S., & Wilson, T. (2016). Comment on “Estimating the
reproducibility of psychological science.” Science, 351, 1037a–1037b.
Gilbert, N., & Troitzsch, K. G. (2011). Simulation for the social scientist (2nd ed.). New York:
Open University Press.
Guastello, S., Koopmans, M., & Pincus, D. (Eds.) (2009). Chaos and complexity in psychology:
The theory of nonlinear dynamical systems. New York: Cambridge University Press.
Harré, R., & Secord, P. F. (1972). The explanation of social behavior. Oxford: Blackwell.
Hollenstein, T. (2013). State space grids: Depicting dynamics across development. New York:
Springer.
James, W. (1890). Principles of psychology. New York: Holt.
 Rethinking Human Experience  11

Jarman, M., Nowak, A., Borkowski, W., Serfass, D., Wong, A., & Vallacher, R. R.
(2015). The critical few: Anticonformists at the crossroads of minority survival and
collapse. Journal of Artificial Societies and Social Simulation, 18(1), 6. Retrieved from
http://jasss.soc.surrey.ac.uk/18/1/6.html
Kenrick, D. T., Li, N. P., & Butner, J. (2003). Dynamical evolutionary psychology:
Individual decision-rules and emergent social norms. Psychological Review, 110, 3–28.
Kuhn, T. S. (1970). The structure of scientific revolutions (2nd ed.). Chicago, IL: University
of Chicago Press.
Kuppens, P., Oravecz, Z., & Tuerlinckx, F. (2010). Feelings change: Accounting for
individual differences in the temporal dynamics of affect. Journal of Personality and Social
Psychology, 99, 1042–1060.
Lewin, K. (1936). Principles of topological psychology. New York: McGraw-Hill.
Liebrand, W., Nowak, A., & Hegselman, R. (Eds.) (1998). Computer modeling of social
processes. New York: SAGE.
Mead, G. H. (1934). Mind, self, and society. Chicago, IL: University of Chicago Press.
Messick, D. M., & Liebrand, V. B. G. (1995). Individual heuristics and the dynamics of
cooperation in large groups. Psychological Review, 102, 131–145.
Monroe, B. M., & Read, S. J. (2008). A general connectionist model of attitudes and
attitude change: The ACS (Attitudes as Constraint Satisfaction) Model. Psychological
Review, 115, 733–759.
Nowak, A., Rychwalska, A., & Borkowski, W. (2013). Why simulate? To develop a
mental model. Journal of Artificial Societies and Social Simulation, 16(3), 12. Retrieved
from http://jasss.soc.surrey.ac.uk/16/3/12.html
Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitude to public opinion: A
dynamic theory of social impact. Psychological Review, 97, 362–376.
Nowak, A., & Vallacher, R. R. (1998). Dynamical social psychology. New York: Guilford Press.
Nowak, A., & Vallacher, R. R. (2001). Societal transition: Toward a dynamical model
of social change. In W. Wosinska, R. B. Cialdini, D. W. Barrett, & J. Reykowski
(Eds.), The practice of social influence in multiple cultures (pp. 151–171). Mahwah, NJ:
Lawrence Erlbaum.
Nowak, A., Vallacher, R. R., & Burnstein, E. (1998). Computational social psychology:
A neural network approach to interpersonal dynamics. In W. Liebrand, A. Nowak, &
R. Hegselman (Eds.), Computer modeling and the study of dynamic social processes
(pp. 97–125). New York: SAGE.
Nowak, A., Vallacher, R. R., Tesser, A., & Borkowski, W. (2000). Society of self: The
emergence of collective properties in self-structure. Psychological Review, 107, 39–61.
Nowak, A., Vallacher, R. R., & Zochowski, M. (2005). The emergence of personality:
Dynamic foundations of individual variation. Developmental Review, 25, 351–385.
Open Science Collaboration (2015). Estimating the reproducibility of psychological science.
Science, 349(6251), aac4716–aac4716. http://doi.org/10.1126/science.aac4716
Ostrom, T. M. (1988). Computer simulation: The third symbol system. Journal of
Experimental Social Psychology, 24, 381–392.
Queller, S. (2002). Stereotype change in a recurrent network. Personality and Social
Psychology Review, 6, 295–303.
Read, S. J., & Miller, L. C. (Eds.) (1998). Connectionist models of social reasoning and social
behavior. Mahwah, NJ: Lawrence Erlbaum.
Read, S. J., Vanman, E. J., & Miller, L. C. (1997). Connectionism, parallel constraint
satisfaction processes, and Gestalt principles: (Re)introducing cognitive dynamics to
social psychology. Personality and Social Psychology Review, 1, 26–53.
12  Vallacher, Nowak, and Read

Schuster, H. G. (1984). Deterministic chaos. Vienna: Physik Verlag.

Smith, E. R., & Conrey, F. R. (2007). Agent-based modeling: A new approach for theory
building in social psychology. Personality and Social Psychology Review, 11, 87–104.
Spivey, M. J., & Dale, R. (2006). Continuous dynamics in real-time cognition. Current
Directions in Psychological Science, 15, 207–211.
Strogatz, S. (1994). Nonlinear dynamics and chaos. Cambridge, MA: Perseus Books.
Thagard, P., & Nerb, J. (2002). Emotional Gestalts: Appraisal, change, and the dynamics
of affect. Personality and Social Psychology Review, 6, 274–282.
Vallacher, R. R., & Nowak, A. (Eds.) (1994). Dynamical systems in social psychology. San
Diego, CA: Academic Press.
Vallacher, R. R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 4, 73–99.
Vallacher, R. R., & Nowak, A. (2007). Dynamical social psychology: Finding order in the
flow of human experience. In A. W. Kruglanski & E. T. Higgins (Eds.), Social psychology:
Handbook of basic principles (pp. 734–758). New York: Guilford Press.
Vallacher, R. R., Nowak, A., & Kaufman, J. (1994). Intrinsic dynamics of social judgment.
Journal of Personality and Social Psychology, 67, 20–34.
Vallacher, R. R., Read, S. J., & Nowak, A. (Eds.) (2002). The dynamical perspective in
social psychology. Personality and Social Psychology Review, 6, 264–388.
Vallacher, R. R., Van Geert, P., & Nowak, A. (2015). The intrinsic dynamics of psychological
process. Current Directions in Psychological Science, 24, 58–64.
Waldrop, M. M. (1992). Complexity: The emerging science at the edge of order and chaos.
New York: Schuster.
Weisbuch, G. (1992). Complex system dynamics. Redwood City, CA: Addison-Wesley.
Wolfram, S. (2002). A new kind of science. Champaign, IL: Wolfram Media.
Wong, A., Vallacher, R. R., & Nowak, A. (2014). Fractal dynamics in self-evaluation reveal
self-concept clarity. Nonlinear Dynamics, Psychology, and Life Sciences, 18, 349–370.
Wong, A. E., Vallacher, R. R., & Nowak, A. (2016). Intrinsic dynamics of self-evaluation:
The role of self-concept clarity. Personality and Individual Differences, 100, 167–172.
PART I

Intrapersonal Dynamics
2
VIRTUAL PERSONALITIES
A Neural Network Model of the
Structure and Dynamics of Personality

Stephen J. Read, Vita Droutman,

and Lynn C. Miller

There is a long-standing split in personality between structural and dynamic

approaches to understanding personality (for recent discussions, see Funder,
2001; Mischel & Shoda, 1998). These two approaches typically rely on different
theories and constructs. The structural approach typically focuses on determin-
ing the psychometric structure of personality tests (e.g., Lee & Ashton, 2004;
McCrae & Costa, 1999) or the everyday language used to describe person-
ality (Digman, 1997; Goldberg, 1981; John & Srivastava, 1999), and a major
(although not the only) focus is on understanding stable personality disposi-
tions. In contrast, dynamic approaches, such as Mischel’s Cognitive-Affective
Systems theory (Mischel & Shoda, 1995, 2008), Atkinson’s Dynamics of Action
model (Atkinson & Birch, 1970) or McAdams’ narrative approach to person-
ality (McAdams, 2008; McAdams, Josselson, & Lieblich, 2006) typically focus
on uncovering and understanding the psychological mechanisms that underlie
individual differences and changes in behavior across situations and time, and
are less interested in understanding stable, broad dispositions. Although some
personality researchers have suggested that there is no need to try to bring these
two approaches together, many researchers have become increasingly concerned
with doing so.
In the current chapter we argue that modeling personality in terms of the
behavior of organized motivational systems can help unify the two approaches.We
present a neural network model of personality in terms of structured motivational
systems (Read et al., 2010; Read & Miller, 2002) that both demonstrates how the
structure of personality, such as the Big Five, can arise from these motivational
systems and how the dynamics of everyday behavior can arise from the interac-
tion between these structured motivational systems and the changing affordances
of situations over time and the bodily states of individuals. Our goal is to show
16  Read, Droutman, and Miller

how personality structure and personality dynamics can arise from the same psy-
chological architecture in interactions with the social and physical environment.
People differ considerably in their behavior and how they respond to the same
situations. Personality psychologists who take the psychometric approach to per-
sonality often try to understand these individual differences by asking people to
respond to items on a personality test, which typically ask people to characterize
themselves in terms of typical behaviors, beliefs, emotions, and sometimes moti-
vation. The response to these items across a large sample of individuals is factor
analyzed, and the result is typically something like the Big Five (John, Naumann, &
Soto, 2008; John & Srivastava, 1999) or similar structures such as the six-factor
HEXACO model (Ashton & Lee, 2007; Lee & Ashton, 2004), which essentially
adds an Honesty/Humility factor to the Big Five. The Big Five (often referred to
by the acronym OCEAN) consists of five major dimensions of human personality:
Openness to experience, Conscientiousness, Extraversion, Agreeableness, and
Neuroticism.
Differences between individuals can then be considered in terms of different
individuals’ relative standing on each of the Big Five dimensions. Each individual
can be thought of as having a relative profile across each dimension that describes
their standing relative to others on the various dimensions.
But something is typically missing in the various structural approaches to per-
sonality: an explanation of why this particular structure is found and what are the
underlying mechanisms or processes. We have argued that this particular structure is
the result of underlying structured motivational systems.
Our neural network model of personality (Read et al., 2010; see also Read &
Miller, 1989) argues that personality traits, and specifically the Big Five, arise from
the interaction between structured motivational systems and the goal affordances
of situations. We have argued that a number of relatively specific brain systems
manage a variety of different motivational domains and their related behavior,
and that these specific systems are then organized into two higher-level Approach
and Avoidance systems that integrate over the lower-level systems. The structure of
these motivational systems is responsible for the structure of human personality. But
additionally, the dynamics of personality can be understood in terms of the ways
in which these structured motivational systems interact with the affordances of
different situations that the individual encounters and with the individual’s cur-
rent bodily state.
As noted above, personality researchers have largely addressed the structure
and dynamics of personality independently. However, researchers have become
increasingly interested in how those two important approaches can be related.
Here we use our neural network model of personality to show how the structure
and dynamics of personality can arise from the dynamics of structured motiva-
tional systems.
In the first section of the chapter, we demonstrate how such structured moti-
vational systems can transform variability in chronic motivations across individuals
 Virtual Personalities  17

into patterns of behavior from which we can recover the Big Five. We will do
this by constructing a neural network model of a set of structured motivational
systems, and then training the network to associate different clusters of behavior
with different underlying motives. This network will then be used to simulate the
behavior of a large number of different individuals as they respond to a variety
of different situations. We will then factor analyze the resulting behavior. The
result of this factor analysis will map onto the underlying motivational structures
in the model.
In the second section of the chapter, we will show how both within- and
between-person variability in personality-related behavior can be understood
in terms of the dynamics of the interaction between individual’s motives, the
affordances of situations, and current bodily states. In a series of recent experience-
sampling studies of everyday behavior, Fleeson has shown that while there are
clear individual differences, consistent with the Big Five, in trends in personality
related behavior over time, at the same time the within person variability over
situations is at least as large as the between person differences (Fleeson, 2012;
Fleeson & Jayawickreme, 2015).
In the final section of the chapter, we will discuss the implications of our com-
putational model for thinking about person–situation interactions.

General Theoretical Background

The Virtual Personalities model draws upon a wide range of research from per-
sonality, neuroscience, developmental psychology, and evolutionary approaches to
human behavior that have helped us identify the basic motivational structures that
underlie human behavior. Here we briefly review some of the relevant research.

The Big Five Structure

A large body of research examining the structure of trait scales (e.g., Lee &
Ashton, 2004; McCrae & Costa, 1999; Wiggins & Trapnell, 1996) and another
body of research that has examined the structure of trait language used in every-
day language (e.g., Digman, 1997; Goldberg, 1981; Peabody & De Raad, 2002;
Saucier & Ostendorf, 1999) has demonstrated that the overall structure of human
personality can be described in terms of five broad factors, typically referred to as
the Big Five, or OCEAN, as explained above. Other research, such as HEXACO
(Ashton & Lee, 2007; Lee & Ashton, 2004), has argued for six factors, most of
which largely overlap with the Big Five.

Traits as Goal-Based Structures

We and others have argued for a Motive/Goal-based model of traits: the central
idea is that personality traits can be viewed as configurations of chronic goals,
18  Read, Droutman, and Miller

plans, resources, beliefs, and styles (e.g., Fleeson, 2012; Fleeson & Jayawickreme,
2015; Miller & Read, 1987; Read & Miller, 1989; Yang et al., 2014). Probably
most central is the role of motives and goals: at the center of most traits is some-
thing people want or something they want to avoid.

Evolutionary Tasks
A number of researchers (e.g., Bugental, 2000; Kenrick & Trost, 1997) have
noted that there are a wide range of problems and tasks that all humans must solve
if they are to survive and successfully reproduce, and they have identified a num-
ber of motivational systems that have evolved to handle these tasks (e.g., Maslow,
1943; Murray, 1938). Among the various tasks that must be solved and the rel-
evant system are mating, nurturance of young, affiliation and bonding with peers,
establishing dominance hierarchies, insuring attachment to caregivers, avoidance
of social rejection, and avoidance of physical harm. Over the last 15 years we have
developed two extensive motive taxonomies from which we draw in our work
(Chulef, Read, & Walsh, 2001; Talevich, Read, Walsh, Chopra, & Iyer, 2015).

Biological Temperament
In addition, work from neuroscience and work on biological temperament has
demonstrated that there are at least three major biologically based dimensions
of personality or temperament that map onto three dimensions of the Big Five:
Extraversion, Neuroticism, and Conscientiousness. Each of these three dimensions
corresponds to a major aspect of motivation. A number of researchers have argued
that there is a general Approach (Clark & Watson, 1999; Rothbart & Bates, 1998) or
Behavioral Approach System (BAS) (Gray, 1987; Gray & McNaughton, 2000) that
governs sensitivity to cues signaling rewards and, when activated, results in an active
approach. This maps onto Extraversion. There is considerable evidence that the
neurotransmitter dopamine plays a major role in this Approach or Reward system.
In addition, there is a general Avoidance or Behavioral Inhibition System
(BIS) (Carver & White, 1994; Clark & Watson, 1999; Gray, 1987; Gray &
McNaughton, 2000) that governs sensitivity to cues of punishment and loss
and manages avoidance of threatening situations. This maps onto Neuroticism.
A third broad dimension concerns a General Disinhibition/Constraint System
(Derryberry & Rothbart, 1997; Rothbart & Bates, 1998) which is related to execu-
tive function and inhibitory control. This probably maps onto Conscientiousness.

Overview of the Simulations and Their Implications

The section below on “The Dynamics of Motivation” details how we go from a
structured motivational model to how resulting motivational dynamics across hun-
dreds of virtual personalities produce emergent outcomes resembling the Big Five.
 Virtual Personalities  19

The section on “Within-Person versus Between-Person Variability” describes

a simulation of how our neural network model captures motivational dynamics
and changes in behavior over time.
The section on “Person–Situation Interactions” discusses the implications of
this computational model for thinking about person–situation interactions.

Overview of Neural Networks

A neural network model consists of nodes and weighted links between them.
Processing in the network proceeds by the passing of activation among nodes
along these weighted links. The activation of a node is a function of the inputs
it receives from the other nodes with which it is linked. Typically, the input is a
weighted average (or sum) of the activations from other nodes, where the activa-
tions are weighted by the strength of the link:

•• Nodes—These represent various features or concepts, analogous to neurons

or systems of neurons. Their level of activation represents their importance
and their readiness to fire and send activation to other nodes with which
they are linked.
•• Layers—Nodes are organized in layers, corresponding to systems (Situational
Input, Approach and Avoidance Motives, Behavior) of nodes that interact
in terms of excitatory and inhibitory relations among them. In the neural
architecture that we use in the current simulations (Leabra; Aisa, Mingus, &
O’Reilly, 2008), nodes within a layer compete with each other for activa-
tion. The degree of competition is a parameter that can be manipulated in
Leabra.
•• Links—These represent the relations among nodes. The strength and direc-
tion of weight represents degree and direction of influence. Nodes can be
connected either unidirectionally or bidirectionally, depending on the type
of model. When there is a unidirectional connection between two nodes,
then activation can only flow in one direction and only the first node can
influence the subsequent one. However, when nodes are bidirectionally
connected, then activation can flow in both directions and the nodes can
mutually influence one another.

Current Neural Network Model

The current networks are modeled in the emergent (Aisa et al., 2008; O’Reilly &
Munakata, 2000) neural network modeling system, using the Leabra architec-
ture. Leabra (Local, error-driven and associative, biologically realistic algorithm)
is a specific architecture developed by O’Reilly that attempts to use biologically
realistic algorithms, based on actual neural systems, to capture both the activation
of nodes and the process of learning the weights between nodes. A learning rule
20  Read, Droutman, and Miller

allows us to model changes in association between nodes as a function of their

frequency of co-occurrence.
Figure 2.1 shows the version of the Virtual Personality model used for the
Big Five simulations. There are five meaningful layers. The Input or Situational
Features layer is the initial input layer to the network that represents informa-
tion about the different situations or situational features to which the network
is exposed. There are 30 input nodes, representing 30 different situations.
Processing in the network starts by applying activation from the available fea-
tures to these nodes and then having activation spread across the weighted links
to subsequent nodes.
The Approach and the Avoid(ance) layers organize specific approach-related
and avoidance-related motives. Each node represents a different motive. The
Approach layer has two nodes, corresponding to two different potential factors,
each with a corresponding motive, whereas the Avoid layer has three nodes,
corresponding to three different motives and factors. The specific motives are
described in the following. Nodes within a layer compete for activation, and the
amount of competition can be manipulated, which influences how many nodes
can be active. This allows us to control how many motives can be active at any
one time and trying to influence behavior. In the current network, competition
is set so that only one motive within a layer will tend to be active at any one time.
The Motivation Level layer is used to represent chronic individual differences
in specific motive activation. Each node in this layer has a one-to-one connec-
tion with a node in the Approach and Avoid layers. By setting activations in the
Motivation Level nodes, we can control the baseline activation for the motive
nodes in the Approach and Avoid layers, which influences the likelihood that the
motive nodes will be active.

30 behaviors,
Behaviors
6 for each motive

Hidden_3

Motivation Level Approach Avoid 5 motives,

with corresponding
motivation level

Hidden
30 situations

Situational Features

FIGURE 2.1  Version of the Virtual Personality model used for the Big Five simulations.
 Virtual Personalities  21

The Behaviors layer represents the personality-related behaviors, and in this

model there are 30 different behaviors. (For this simulation, the behaviors are
abstract—e.g., Behavior 1, Behavior 2, etc.—since the concrete nature of a
behavior is not important here.) Competition is set in this layer so that only one
node (behavior) at a time can be active, to capture the obvious idea that only one
action can be enacted at any given time.
The two Hidden layers allow the network to learn complex or conjunctive
combinations of features that represent a more abstract concept. For example,
in the visual system, hidden layers enable the visual system to learn a hierarchy
of visual representations, such as from dots on the retina to lines, from lines to
shapes, and then from shapes to entire objects.

Activation of Node and Output Activation

The activation level of a node is calculated as a function of its current activation
plus its input activation. In Leabra, input activation is calculated as the sum of
three terms: (1) excitatory input from other neurons, (2) inhibitory input from
other neurons, and (3) leak current, which is a constant inhibitory term. The sum
of these three terms provides the membrane potential Vm of the node. Whether
and the extent to which a node will fire or send activation is a function of the
extent to which the current activation of the node exceeds a threshold value that
is a function of the degree of inhibitory input and leak current.1
The excitatory input to a node is a function of two terms: (1) the number of
ion channels which can be potentially open ( g e), and (2) the proportion of ion
channels that are actually open ( ge). As we discuss later, the number of ion chan-
nels that are potentially open can serve as a gain or sensitivity parameter for the
node. Higher sensitivity results in a steeper activation function. The proportion
of ion channels that are currently open ge can be calculated as:
1
g e (t ) = ∑ xi w i (2.1)
n i
where xi is the activity of a particular sending neuron indexed by the subscript i,
w i s the synaptic weight strength that connects sending neuron i to the receiving
neuron, and n is the total number of channels of that type (in this case, excitatory)
across all synaptic inputs to the cell. This equation is essentially the same equa-
tion that is used in all neural network models to calculate the input activation as
a function of the activation being sent by input nodes, multiplied by the weights
between the input and output nodes. A similar equation applies to inhibitory
inputs. From this information, the overall levels of excitation, inhibition, and leak
current can be calculated:
•• excitatory conductance:
g e ge (t )
22  Read, Droutman, and Miller

•• inhibitory conductance:
g i gi ( t )

•• leak conductance:
gl

(The leak current does not have a time varying component, it is a constant.)
From these terms one can calculate the membrane potential, with the conduct-
ance terms explicitly broken out into the “g-bar” constants and the time-varying
“g(t)” parts. The E terms correspond to what is called the “driving potential” for
the particular kind of channel:

g e ge (t ) g i gi ( t )
vm = Ee + Ei 
g e g e ( t ) + g i gi ( t ) + g l g e g e ( t ) + g i gi ( t ) + g l

gl
+ El (2.2)
g e g e ( t ) + g i gi ( t ) + g l

The membrane potential can then be used to calculate the threshold geθ , which is
the threshold that the total excitatory conductance must exceed in order to fire:

gi ( E i − Θ ) + gl ( E l − Θ )  (2.3)
geΘ =
Θ− Ee

The output activation of the node is a sigmoidal or S-shaped function of the

extent to which the excitatory input to the node exceeds the threshold for firing.
1
y=
 
1 + 1  (2.4)
 γ  ge − ge  + 
 Θ
 
y is the output value, ge is excitatory input, geθ is the threshold for firing, γ is
the gain parameter, which governs the steepness of the slope, and the + indicates
that only a positive value will be returned.

Learning
Weights between nodes are modified to capture two different kinds of informa-
tion. One kind of learning is Hebbian learning, which captures the covariance
or correlation among the activation of nodes. In Hebbian learning, the weight
between two nodes will increase if they are active at the same time, whereas they
will decrease if one is active and the other is not. A second kind of learning is
error-correcting learning, where weights will change to capture the extent to
 Virtual Personalities  23

which activation of the input node successfully predicts activation of the output
node. Error-correcting learning changes the weights to reduce the degree of
error in predicting an outcome.2

Processing
Processing starts by applying situational or input features to the Input layer.
Activation then flows from there through the hidden layer and then to the
Approach and Avoidance layers. Activation also flows from the Motivation level
to the Approach and Avoidance layers to set the baseline or chronic activation of
each motive. Activations of the motives are a function of: (1) the current situa-
tion, (2) individual differences in chronic motive activations, set by the Chronic
Motive level, and (3) individual differences in the sensitivities or gains of the
Approach and Avoidance systems (a parameter that can be set in Leabra), which
captures individual differences in sensitivity to reward and punishment.
Activated motives in the Approach and Avoidance layers then compete with
one another for activation. Competition takes place independently in each of
the two layers. The amount of competition, and thus the number of nodes that
can be active, can be manipulated in Leabra. The relevant parameter is set so
that only one or so motives can be active in each layer. Motive nodes that are
active then send activation to behaviors that can help to satisfy those motives.
Behaviors in the Behavior layer then compete for which behavior will be enacted.
Competition is set within the Behavior layer so that only one behavior node can
be active at a time, capturing the idea that multiple behaviors cannot typically be
enacted at the same time.
In the simulation of the Big Five, two important parameters will be manipu-
lated: (1) the conductances (or gains) for the Approach and Avoidance layers,
which influence the sensitivity or gain of all nodes within the relevant layer, and
(2) the baseline or chronic activation of individual motives within the relevant
layer, which is set by activation from the Chronic Motive Level layer.

Simulation of the Structure of Personality:

The Big Five
If personality traits, and specifically the Big Five, arise from the interaction
between organized motivational systems and the goal affordances of situations,
then we should be able to use this neural network model to simulate a pat-
tern of behavior across situations and different individuals that will result in the
Big Five. Specifically, if we: (1) model several hundred individuals (or Virtual
Personalities—VPs) by varying the parameters of appropriately structured
underlying motivational systems, (2) measure the resulting behavior from these
hundreds of VPs, and (3) factor analyze that behavior, then the factor analysis will
result in something like the Big Five.
24  Read, Droutman, and Miller

For the current simulation, we have structured the motivational systems as

follows. We have divided the underlying motivational systems into two broad
systems, corresponding to the Approach and Avoidance layers, in the network.
Within each of these two layers or systems, we represent multiple motives, corre-
sponding to multiple Big Five factors. Following DeYoung’s (2015) work on the
structure of the two broad factors underlying the Big Five, in the Approach system
we have Extraversion and Openness to Experience, whereas in the Avoidance
system we have Neuroticism, as well as Agreeableness and Conscientiousness.
In this simple simulation, we have five nodes, two in the Approach layer and
three in the Avoidance layer. Each node represents a plausible motive for the
corresponding factor.
The Big Five factors and plausible underlying motives are:

•• Extraversion—relating and belonging; being with others;

•• Openness to Experience—appreciating beauty; enjoying intellectual
experiences;
•• Neuroticism—avoiding rejection by others;
•• Agreeableness—helping others; wanting others to like you;
•• Conscientiousness—doing my duty; being disciplined.

Training of the Model/Learning

This instantiation of the model has 30 input nodes, corresponding to 30 differ-
ent situations, five motive nodes, two in the Approach layer and three in the
Avoidance layer, and 30 different behavior nodes, corresponding to 30 different
behaviors. The network is trained to learn the following patterns of associations
among the different nodes by repeatedly presenting it with 150 different instances
that are structured as follows.
Each of the 30 situations co-occurs equally often with each of the five motives,
creating 150 instances. This results in roughly equal weight strengths between
each situation and each motive. Thus, each motive is equally likely to be acti-
vated by a given situation.
Within these 150 instances, each of the five motives co-occurs equally often
(five times) with each behavior in a cluster of six behaviors. That is, each
motive is highly related to six behaviors, but unrelated to the other 24. So
when a particular motive is activated, each of the six associated behaviors will
tend to be equally activated, on average. Thus, activation of a specific motive
will tend to activate a set of six behaviors. This will lead to correlations among
the activations of the behaviors. However, because random noise leads to
slightly different levels of activation for each behavior, the competitive dynam-
ics among the behaviors ensure that on any given trial only one behavior will
end up being enacted.3
 Virtual Personalities  25

Testing
To test whether we could get something like the structure of the Big Five
from this network, we first generated 576 instances of the neural network (576
“individuals”) with different patterns of individual differences, by doing the
following: (1) We generated all 32 possible combinations of the five binary (0,1)
motive levels in the Chronic Motive layer (2 × 2 × 2 × 2 × 2 = 32), (2) generated
all possible combinations of three levels of sensitivity for both the Approach and
Avoidance layers (3 × 3 = 9), and (3) trained two different networks, with dif-
ferent random starting weights. This gave us a total of 576 unique “individuals”
(9 × 32 × 2 = 576). We could get many more instances by more continuously
varying the parameters. However, we felt that 576 individuals would be enough
to test our predictions.
Once we had the 576 individuals, we then tested each one on five passes
through the 30 situations (150 total test situations). Each of the 30 nodes in the
Situation layer was treated as a different situation. Because of random noise in the
Input or Situation layer, each time a situation was presented, it should have had
a slightly different activation.
For each of the 150 “situations” presented, we recorded the net activation
across each of the 30 behavior nodes and then averaged the net activation for
each node across the 150 test cases. We can think of the net activation as the
tendency to enact that behavior. Thus, for each individual we have a profile of
the net activation or behavioral tendency across the 30 behaviors. The 30 behav-
ior nodes can be thought of as 30 items on a scale. And because of the way the
behavior nodes are linked to the five motives, there are six items for each motive.
Essentially we have the scores for 576 “individuals” on 30 different items in a
scale, with six items measuring each of the Big Five. From this, we can calculate a
correlation matrix among the 30 behaviors and then factor analyze the correlation
matrix to look for evidence of the Big Five structure.
We then did a Principal Factor Analysis on the correlation matrix. Figure 2.2
shows the Scree plot, which has a very clear bend between five and six factors,
indicating that there are five factors.
Figure 2.3 shows the factor loadings for a five-factor solution, with Varimax
(orthogonal) rotation. The figure is set to only display loadings that are 0.4 or
greater. There are five clear factors, corresponding to the five clusters of items
associated with the five motives. With one minor exception, each item is very
highly loaded on only one factor and the pattern of loadings corresponds to the
relationship of the behaviors to the five motives.
One way to think about these results that might be helpful is to think of this as
analogous to a Latent Variable Factor model, with specific motivations as latent
variables and behaviors as manifest variables that are indicators of the latent, moti-
vational variable. Different behaviors that are influenced by the same motivation
will tend to occur frequently within the same individual.
FIGURE 2.2  S cree plot of the eigenvalues of the factors in a Principal Factors
solution. There is a very clear bend between 5 and 6, indicating that
five factors is probably the best number for this data set.

FIGURE 2.3   actor loadings for a five-factor solution, after Varimax (orthogonal)
F
rotation.
 Virtual Personalities  27

Interim Conclusions
Read and Miller (e.g., Miller & Read, 1991; Read & Miller, 1989) and oth-
ers, such as Fleeson (e.g., Fleeson & Jayawickreme, 2015) in his Whole Trait
Theory, argue that goals and motives play a fundamental role in the nature of
personality traits, and that personality structure can be understood in terms of
underlying motivational structures. Here we have used a computer simulation to
show how the Big Five could arise from suitably organized motivational systems.
This further suggests how stable underlying motivational systems (structure) can
nevertheless produce considerable variability in behavior across individuals.

The Dynamics of Motivation: Changes in

Behavior over Time and Situation
Recently, various researchers have raised the question of how the structure and
dynamics of personality can be related. Among the questions that have been
raised are: Can stable personality structures result in dynamic, changing behavior
over time and situations? If so, how? What are the mechanisms by which under-
lying stable structures can result in dynamic behavior?
Although there are strong differences between individuals in their dispositions
or general levels of different kinds of behavior, behavior is highly dynamic and
varies considerably over time and situations. The same individual may be highly
talkative at a party or a meeting, but quiet when in church or studying at the
library. A major source of those behavioral dynamics is motivational dynamics—
changes in what we “want” over time and situations. Motivation changes—in
the short term—due to changes in opportunities in the environment for motive
or goal-related behavior (Gibson, 1979 called this “affordances”) and our current
bodily or interoceptive state.
As a result of changes in our motivation and the motivational affordances of
situations, our behavior will vary over time and situations. In this section, we
show how our personality model, based on structured motivational systems can
also capture the variability of individuals’ behavior over time and situations. Our
approach to motivational dynamics is influenced by Atkinson and Birch’s (1970)
Dynamics of Action (DOA), which presented a model of how the ebb and flow
of competing motives over time and situations would lead to major changes in
behavior. It is also influenced by Revelle and Condon’s (2015) recent reparam-
eterization of Atkinson and Birch’s model. In particular, Revelle and Condon
suggest that the DOA model can actually be modeled as a neural network model.
Although our model does not precisely follow their parameterization, it does
have some similarities with their description.
In both the DOA and Revelle and Condon’s reparameterization, they treat the
strength of current motivation as a function of previous motivation, instigating
forces and consummatory forces resulting from actions such as eating or drinking
that reduce the strength of motivation. In their approach, instigating forces are
28  Read, Droutman, and Miller

solely a function of the strength or attractiveness of a cue. However, Berridge and

colleagues (e.g., Berridge, 2007; Zhang, Berridge, Tindell, Smith, & Aldridge,
2009) argue that motivation for a behavior or the strength of “wanting” is also
strongly influenced by current bodily or interoceptive state. Specifically, they argue
that the strength of “wanting” is a multiplicative function of both cue strength
(attractiveness of approach cue and aversiveness of avoidance cue) and current
bodily state (e.g., Hunger, Thirst, etc.). For example, Food is more wanted when
it is highly attractive and when one is very hungry. In terms of the DOA model,
this view of the nature of wanting argues that instigating forces are a joint function
of cue strength and bodily state, rather than just a function of cue strength. We
follow this idea in the current model, and we add a layer that represents current
interoceptive or bodily state to our simulation of the dynamics of motivation.

Not a Cybernetic Model

One important point to note is that our model is not a cybernetic or homeostatic
model of control. There is no set point or comparator, such as a thermostat on a
heating system. Instead, it is an open control model (see Bolles, 1980), in which
the level of motivation is the result of the balance between instigating forces and
consummatory forces.

Structure of the Computational Model and

Flow of Processing
In the current simulation, we start out with a simple model system to simulate
behavioral dynamics over time and situations. The simple model system is a rat,
with basic motives, bodily states, and situational affordances. Model systems are
typically used in biology to provide a simple and clearly understood system for a
phenomenon that still captures the central characteristics of the system. In previ-
ous modeling work, we have started with much more complex models and have
ended up spending much of our time trying to understand the complexities of
the model, rather than the central principles. However, we believe that once we
understand this model system well, it should be fairly easy to scale up to more
complicated systems.
A picture of the model, as represented in the Emergent software system can
be seen in Figure 2.4. The Environment layer represents cues in the environ-
ment that have strong motivational affordances, here Food, Water, and a Cat.
The Bodily State layer represents the strength of current interoceptive or bodily
states: Hunger, Thirst, and Fear.
The Environment layer and the Bodily State layer jointly send activation to
the Approach and Avoidance layers. Both the Approach and Avoidance layers
are implemented as Multiplicative layers, which take the inputs from two lay-
ers and multiplies them together. Thus, the activation of each motive node is a
multiplicative function of the corresponding Environment cues and Bodily State.
 Virtual Personalities  29

Eat Drink Run

Behavior

Hunger Thirst Fear

Approach Avoidance

Food Water Cat Hunger Thirst Fear

Environment Bodily_state
Network_1 Value: act_eq

FIGURE 2.4  A screenshot of the Emergent network for the rat model of Motivation.
Note that one important implication of this is that typically neither a strong
cue nor a strong bodily state is sufficient by itself to strongly activate wanting.
However, in subsequent work (Read, Smith, Droutman, & Miller, in press) we
have begun to investigate how to model the possibility that a highly activated
bodily state will lead the animal to start looking for objects that can satisfy that
state. So high levels of hunger, with no food present in the current environment,
would instigate animals to start looking for food.
In general, consistent with a large body of work suggesting organisms are
more sensitive to potential negative events than to potential positive events (e.g.,
Cacioppo & Berntson, 1994; Cacioppo, Gardner, & Berntson, 1997), the net-
work is more sensitive to Threat cues than to Approach cues, so Threat will
typically override Approach-related motives. Thus, even with strong affordances
and strong Bodily State, Threat (Cat) will override the other systems. The strong
impact of the Threat cue is implemented by giving a higher weight scaling to the
weight from the Avoidance layer (Fear) to the Run node in the Behavior layer.
The Motive nodes then send activation to the Behavior layer, where the three
behaviors of Eat, Drink, and Run compete with each other for activation. When
a behavior is activated, it then can reduce the level of its corresponding bodily
state. Thus, engaging in a consummatory behavior, such as eating or drinking, or
30  Read, Droutman, and Miller

running away will reduce the relevant bodily state and thus reduce the likelihood
of the relevant behavior. (In the current version of the model, this direct effect
on bodily state is not currently implemented, but instead is simulated by changing
the input to the bodily state from the Inputs. In ongoing work, we have imple-
mented this direct effect: Read et al., in press).
An important implication of the fact that motives and behaviors compete with
each other for activation is that the likelihood of a behavior being activated is not
a function solely of the strength of the relevant motive, but rather is a function
of the strength of wanting something compared to the strength of wanting other
things. For example, if there are both food and water around and we are moder-
ately hungry but are also very thirsty, we will drink. However, in the absence of
thirst, the same level of hunger and food will lead to eating.

Factors Influencing Dynamics

Behavior from this network is highly dynamic and will shift over time as a func-
tion of a variety of different factors: (1) changes in situational affordances—as
affordances appear and disappear as situations change, behavior will change even
with consistent bodily state; (2) Changes in current bodily state—as bodily state
waxes and wanes, behavior will change even with consistent affordances, and
behaviors will influence motivation, and thus future behavior, by influencing both
bodily state and situational affordances. For example, eating will reduce the bodily
state cue associated with hunger, and by reducing bodily state allows motivation
or “wanting” to be higher for another behavior, and thus allow another behavior
to be enacted. Behaviors can also influence situational affordances, by moving the
organism into different situations that have different affordances or by removing
affordances, such as eating all the food or drinking all the water. All of these com-
ponents acting together over time and situations will result in a highly dynamic
situation in which the organism’s behavior will change across time and place.

Simulation
In the following simulations, we provide an example of how the model behaves
as both affordances and bodily state change over time. The columns in the fig-
ures below show the activation values of the different layers and their nodes over
time. The Environment and Bodily State columns represent the Inputs to the
network, the Motivation layer represents the intervening Motivational state, and
the Behavior layers represent the outputs.

Simulation 1
In Figure 2.5a, the three columns labeled “Environment” indicate the strength
of the activation of three cues in the Environment—Food, Water, and Cat—at
 Virtual Personalities  31

each time point. The strength of the cues stays constant, and the strength of the
Food cue is stronger than the strength of the Water cue. The second column,
“Bodily State,” indicates the strength of the activation of each Bodily State at
each time point. Over time, the Hunger state decreases and the Thirst state
increases. The third column, “Motivation”, represents the multiplicative value of
the Environment activations and the Bodily State activations. The final column,
“Behavior,” represents the activation of each behavior after the competitive
dynamics have played out in this layer. The most strongly activated behavior is
the one that will be enacted.
As can be seen in Figure 2.5a, with a constant set of environmental cues to
Food and Water, changes in bodily state over time lead to a shift in motivation

(a)
Bodily State
change Environment Bodily State Movaon Behavior
Time Food Water Cat Hunger Thirst Fear Hunger Thirst Fear Eat Drink Run
1 0.80 0.50 0.00 0.80 0.50 0.00 0.64 0.25 0.00 0.89 0.00 0.00
2 0.80 0.50 0.00 0.70 0.50 0.00 0.56 0.25 0.00 0.86 0.00 0.00
3 0.80 0.50 0.00 0.60 0.60 0.00 0.48 0.30 0.00 0.79 0.00 0.00
4 0.80 0.50 0.00 0.50 0.60 0.00 0.40 0.30 0.00 0.71 0.00 0.00
5 0.80 0.50 0.00 0.40 0.70 0.00 0.32 0.35 0.00 0.08 0.58 0.00
6 0.80 0.50 0.00 0.30 0.70 0.00 0.24 0.35 0.00 0.00 0.68 0.00
7 0.80 0.50 0.00 0.30 0.80 0.00 0.24 0.40 0.00 0.00 0.74 0.00

(b)
Environment
change Environment Bodily State Movaon Behavior
Time Food Water Cat Hunger Thirst Fear Hunger Thirst Fear Eat Drink Run
1 0.80 0.50 0.00 0.80 0.50 0.00 0.64 0.25 0.00 0.89 0.00 0.00
2 0.70 0.50 0.00 0.80 0.50 0.00 0.56 0.25 0.00 0.86 0.00 0.00
3 0.60 0.60 0.00 0.80 0.50 0.00 0.48 0.30 0.00 0.79 0.00 0.00
4 0.50 0.60 0.00 0.80 0.50 0.00 0.40 0.30 0.00 0.71 0.00 0.00
5 0.40 0.70 0.00 0.80 0.50 0.00 0.32 0.35 0.00 0.08 0.58 0.00
6 0.30 0.70 0.00 0.80 0.50 0.00 0.24 0.35 0.00 0.00 0.68 0.00
7 0.30 0.80 0.00 0.80 0.50 0.00 0.24 0.40 0.00 0.00 0.74 0.00

(c)
Cat Environment Bodily State Movaon Behavior
Time Food Water Cat Hunger Thirst Fear Hunger Thirst Fear Eat Drink Run
1 0.80 0.60 0.00 0.80 0.50 0.00 0.64 0.30 0.00 0.88 0.00 0.00
2 0.80 0.70 0.00 0.80 0.50 0.00 0.64 0.35 0.00 0.87 0.00 0.00
3 0.80 0.70 0.50 0.80 0.50 0.50 0.64 0.35 0.25 0.00 0.00 0.72
4 0.80 0.70 0.50 0.80 0.50 0.50 0.64 0.35 0.25 0.00 0.00 0.72
5 0.80 0.70 0.50 0.80 0.50 0.50 0.64 0.35 0.25 0.00 0.00 0.72
6 0.80 0.70 0.50 0.80 0.50 0.50 0.64 0.35 0.25 0.00 0.00 0.72
7 0.80 0.80 0.50 0.80 0.50 0.50 0.64 0.40 0.25 0.00 0.00 0.69

FIGURE 2.5   ables representing the Environment and Bodily State inputs, and the
T
resulting Motivation and Behavior activations for the rat model. (a) With a
constant set of environmental cues to Food and Water, changes in bodily
state over time lead to a shift in motivation and a sudden shift in behavior.
(b) Bodily State remains constant over time, but changes in the strength
of Environmental cues lead to nonlinear shifts in Motivation and the
corresponding behavior. (c) Because of the stronger weights from Fear to
Behavior (representing greater fear sensitivity), relatively weak Fear cues
can override much stronger Environment and Bodily State cues.
32  Read, Droutman, and Miller

and a sudden shift in behavior. Moreover, this shift from Eating to Drinking can
occur even though the strength of the Food cue is consistently stronger than the
strength of the Water cue.

Simulation 2
The results of the simulation in Figure 2.5b show that when Bodily State remains
constant over time, changes in strength of Environmental cues leads to a shift in
Motivation and a corresponding shift in Behavior. Moreover, the shift is highly
nonlinear.

Simulation 3
The results of the simulation in Figure 2.5c simply show that because of
the stronger weights from the Fear motivation to Behavior and because of the
competitive dynamics in the Behavior layer, even relatively weak Fear cues can
override much stronger Environment and Bodily State cues. This captures the
idea that the Avoid- or Threat-related motivational system is more sensitive to
Environmental cues than is the Approach system. Again, the change in behavior
is highly nonlinear.
Taken together, these three simulations demonstrate how behavior can change
quite strongly over time and situations, in a highly nonlinear fashion, as a func-
tion of changes in bodily state and changes in the strength of environmental cues.
This shows how a stable structure can nevertheless lead to considerable variability
over time and space as the inputs to the structure change over time and with
exposure to different situations.

Within-Person versus Between-Person

Variability
Both within- and between-person variability can be understood in terms of motive
dynamics. Fleeson has observed, over a number of studies (Fleeson, 2001, 2012;
Fleeson & Jayawickreme, 2015), that within-person variability in personality-
related behaviors over time and situations can be as large as between-person
variability in personality-related behavior. But how can this be if there are
stable individual differences, and more specifically, how can this be if there
are stable indi­vidual differences in the structure of motivation?
We argue that this pattern makes sense because both within- and between-
person variability in personality-related behavior can be understood in terms
of motivational dynamics: specifically, the dynamics of the interaction among
individuals’ chronic motive activations, motive affordances of situations, and
individuals’ current “bodily” or interoceptive state.
 Virtual Personalities  33

Within-person variability in personality-related behavior can be understood in

terms of changes in motive affordances as people move through different situ-
ations and changes in current “bodily” state over time (e.g., due to satiation,
deprivation, etc.).
Between-person variability in personality-related behavior can be understood in
terms of between-person differences in chronic motive activations, between-
person differences in sensitivity to situations, and between-person differences in
choice of situations with different affordances.
The preceding suggests that the ratio of between-person variability to within-
person variability in personality-related behavior depends on:

ratio of between-person variability in chronic activation of different

motives
to
within-person variability in momentary motive activation as individuals
move through time and situations

Thus, within-person variability can easily be equivalent to typical between-person

variability as long as the motive affordances of situations vary strongly over time
and there is strong variability in bodily state.

Person–Situation Interactions
Related to the above discussion on within- versus between-person variability,
a key question in personality over the years has been how stable personality
dispositions can nevertheless result in different kinds of behaviors in different
situations. In the preceding simulations, we have shown how this question can
be addressed through the role of structured motivational systems interacting with
the motive affordances of situations. Specifically, we have argued that behavior is
a joint function of individual differences in chronic motives, motive affordances
of the specific situations an individual is in, and current bodily state. Thus, an
individual’s behavior in a situation will be the result of the interaction of the
motive affordances of the situation with both chronic and temporary differences
in motivational state within the individual. As a result, stable dispositions based
on chronic motives can lead to very different behaviors as the nature of the situ-
ation and the current bodily states change.
In the current chapter, we have focused on the role of motive affordances
in explaining why people may behave quite differently in different situations.
However, we acknowledge and have discussed extensively in other work (e.g.,
Miller & Read, 1991; Read & Miller, 1989) how behavior is strongly influenced
by other aspects of situations, such as roles, and rules and scripts for appropriate
and typical behavior. In this current work, we are attempting to see how much
34  Read, Droutman, and Miller

of the variability across situations can be explained by the concept of affordances.

But other aspects of situations could be implemented in an expanded version of
this model. For example, we could simulate individual differences in situational
knowledge or reinforcement history in different situations.

Conclusion and Discussion

This chapter has outlined how the structural and dynamic approaches to per-
sonality can be integrated and understood in terms of the interaction between
structured motivational systems within the individual, and affordances for goal
pursuit within the situation. In the first simulation, we showed how structured
motivational systems can result in personality structures, such as the Big Five,
when we look at the behavior of a large number of individuals. In the second
simulation, we showed how such structured motivational systems can result in
personality dynamics over time and situations, such that the variability in behav-
ior within individuals can be large, and potentially as large as the variability in
behavior between individuals. Following on this simulation, we have discussed
how person–situation interactions can be understood in terms of the interaction
between underlying motivational structures and the varying affordances for goal
pursuit that characterize different situations.
In sum, our models and the simulations using them demonstrate that the same
kind of underlying motivational systems can be responsible for both the structure
and dynamics of personality. Although the field has tended to approach personal-
ity structure and personality dynamics independently and to view them as very
different types of things, we have shown how they can arise out of the same
psychological architecture in interactions with the environment.

Notes
1 For a detailed description of how activation is calculated in Leabra, see the chapter
“CCNBook/Neuron” in the online wiki textbook Computational Cognitive Neuroscience
(2nd ed.), Boulder, CO: CCNLab, University of Colorado. Retrieved from https://grey.
colorado.edu/CompCogNeuro/index.php/CCNBook/Neuron.
2 For a detailed description of the learning rule in Leabra, see the chapter “CCNBook/
Learning” in Computational Cognitive Neuroscience (2nd ed.). Retrieved from https://grey.
colorado.edu/CompCogNeuro/index.php/CCNBook/Learning.
3 This only looks at the impact of behaviors depending on common motives. It does not
examine the impact of situation–goal links.

References
Aisa, B., Mingus, B., & O’Reilly, R. (2008). The Emergent neural modeling system.
Neural Networks, 21(8), 1146–1152. http://doi.org/10.1016/j.neunet.2008.06.016
Ashton, M. C., & Lee, K. (2007). Empirical, theoretical, and practical advantages of the
HEXACO model of personality structure. Personality and Social Psychology Review,
11(2), 150–166. http://doi.org/10.1177/1088868306294907
 Virtual Personalities  35

Atkinson, J. W., & Birch, D. (1970). The dynamics of action. New York: John Wiley.
Berridge, K. C. (2007). The debate over dopamine’s role in reward: The case for incentive
salience. Psychopharmacology, 191(3), 391–431.
Bolles, R. (1980). Some functionalistic thoughts about regulation. In F. M. Toates &
T. Halliday (Eds.), Analysis of motivational processes (pp. 77–102). London: Academic
Press.
Bugental, D. B. (2000). Acquisition of the algorithms of social life: A domain-based
approach. Psychological Bulletin, 126(2), 187–219. http://doi.org/10.1037/0033-
2909.126.2.187
Cacioppo, J. T., & Berntson, G. G. (1994). Relationship between attitudes and evalua-
tive space: A critical review, with emphasis on the separability of positive and negative
substrates. Psychological Bulletin, 115(3), 401.
Cacioppo, J. T., Gardner, W. L., & Berntson, G. G. (1997). Beyond bipolar conceptual-
izations and measures: The case of attitudes and evaluative space. Personality and Social
Psychology Review, 1(1), 3–25.
Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral activation, and
affective responses to impending reward and punishment: The BIS/BAS scales. Journal
of Personality and Social Psychology, 67, 319–333.
Chulef, A. S., Read, S. J., & Walsh, D. A. (2001). A hierarchical taxonomy of human
goals. Motivation and Emotion, 25(3), 191–232.
Clark, L. A., & Watson, D. (1999). Temperament: A new paradigm for trait psychology.
In L. A. Pervin & O. P. John (Eds.), Handbook of personality: Theory and research (2nd ed.)
(pp. 399–423). New York: Guilford Press.
Derryberry, D., & Rothbart, M. K. (1997). Reactive and effortful processes in the organi-
zation of temperament. Development and Psychopathology, 9(04), 633–652. http://doi.
org/10.1017/S0954579497001375
DeYoung, C. G. (2015). Cybernetic Big Five Theory. Journal of Research in Personality, 56,
33–58. http://doi.org/10.1016/j.jrp.2014.07.004
Digman, J. M. (1997). Higher-order factors of the Big Five. Journal of Personality and Social
Psychology, 73(6), 1246–1256. http://doi.org/10.1037/0022-3514.73.6.1246
Fleeson, W. (2001). Toward a structure- and process-integrated view of personality:
Traits as density distributions of states. Journal of Personality and Social Psychology,
80(6), 1011.
Fleeson, W. (2012). Perspectives on the person: Rapid growth and opportunities for
integration. In K. Deaux & M. Snyder (Eds.), The Oxford handbook of personality and
social psychology (pp. 33–63). New York: Oxford University Press.
Fleeson, W., & Jayawickreme, E. (2015). Whole trait theory. Journal of Research in
Personality, 56, 82–92.
Funder, D. C. (2001). Personality. Annual Review of Psychology, 52(1), 197–221. http://
doi.org/10.1146/annurev.psych.52.1.197
Gibson, J. J. (1979). The ecological approach to visual perception. Boston, MA: Houghton
Mifflin.
Goldberg, L. R. (1981). Language and individual differences: The search for universals
in personality lexicons. In L. Wheeler (Ed.), Review of personality and social psychology
(Vol. 2, pp. 141–165). Beverly Hills, CA: SAGE.
Gray, J. A. (1987). The psychology of fear and stress (2nd ed.). New York: Cambridge
University Press.
Gray, J. A., & McNaughton, N. (2000). The neuropsychology of anxiety: An enquiry into the
functions of the septo-hippocampal system (2nd ed.). New York: Oxford University Press.
36  Read, Droutman, and Miller

John, O. P., Naumann, L. P., & Soto, C. J. (2008). Paradigm shift to the integrative Big
Five trait taxonomy. In P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of
personality: Theory and research (3rd ed.) (pp. 114–158). New York: Guilford Press.
John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History, measure-
ment, and theoretical perspectives. In L. A. Pervin & O. P. John (Eds.), Handbook of
personality: Theory and research (2nd ed.) (pp. 102–138). New York: Guilford Press.
Kenrick, D. T., & Trost, M. R. (1997). Evolutionary approaches to relationships. In
S. Duck (Ed.), Handbook of personal relationships: Theory, research, and interventions
(pp. 151–177). Chichester: Wiley.
Lee, K., & Ashton, M. C. (2004). Psychometric properties of the HEXACO Personality
Inventory. Multivariate Behavioral Research, 39(2), 329–358. http://doi.org/10.1207/
s15327906mbr3902_8
Maslow, A. H. (1943). A theory of human motivation. Psychological Review, 50, 370–396.
McAdams, D. P. (2008). Personal narratives and the life story. In O. P. John, R. W. Robins, &
L. A. Pervin (Eds.), Handbook of personality: Theory and research (3rd ed.) (pp. 242–262).
New York: Guilford Press.
McAdams, D. P., Josselson, R. E., & Lieblich, A. E. (2006). Identity and story: Creating self
in narrative. Washington, DC: American Psychological Association.
McCrae, R. R., & Costa, P. T., Jr., (1999). A five-factor theory of personality. In
L. A. Pervin & O. P. John (Eds.), Handbook of personality: Theory and research (2nd ed.)
(pp. 139–153). New York: Guilford Press.
Miller, L. C., & Read, S. J. (1987). Why am I telling you this? Self-disclosure in a goal-
based model of personality. In V. J. Derlega & J. Berg (Eds.), Self-disclosure: Theory,
research, and therapy. Plenum.
Miller, L. C., & Read, S. J. (1991). On the coherence of mental models of persons and
relationships: A knowledge structure approach. In G. J. O. Fletcher & F. Fincham
(Eds.), Cognition in close relationships (pp. 69–99). Hillsdale, NJ: Erlbaum.
Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality:
Reconceptualizing situations, dispositions, dynamics, and invariance in personality structure.
Psychological Review, 102(2), 246–268. http://doi.org/10.1037/0033-295X.102.2.246
Mischel, W., & Shoda, Y. (1998). Reconciling processing dynamics and personality
dispositions. Annual Review of Psychology, 49(1), 229–258. http://doi.org/10.1146/
annurev.psych.49.1.229
Mischel, W., & Shoda, Y. (2008). Toward a unified theory of personality: Integrating
dispositions and processing dynamics within the cognitive-affective processing system.
In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality: Theory and
research (3rd ed.) (pp. 208–241). New York: Guilford Press.
Murray, H. A. (1938). Explorations in personality: A clinical and experimental study of fifty
men of college age. Oxford: Oxford University Press. http://psycnet.apa.org/psycinfo/
1938-15040-000
O’Reilly, R. C., & Munakata, Y. (2000). Computational explorations in cognitive neuroscience:
Understanding the mind by simulating the brain. Cambridge, MA: MIT Press.
Peabody, D., & De Raad, B. (2002). The substantive nature of psycholexical personality
factors: A comparison across languages. Journal of Personality and Social Psychology, 83(4),
983–997. http://doi.org/10.1037/0022-3514.83.4.983
Read, S. J., & Miller, L. C. (1989). Inter-personalism: Toward a goal-based theory of
persons in relationships. In Lawrence A. Pervin (Ed), Goal Concepts in Personality and
Social Psychology (pp. 413–472). Hillsdale, NJ: Lawrence Erlbaum,
 Virtual Personalities  37

Read, S. J., & Miller, L. C. (2002). Virtual personalities: A neural network model of
personality. Personality and Social Psychology Review, 6(4), 357–369.
Read, S. J., Monroe, B. M., Brownstein, A. L., Yang, Y., Chopra, G., & Miller, L. C.
(2010). A neural network model of the structure and dynamics of human personality.
Psychological Review, 117(1), 61–92.
Read, S. J., Smith, B., Droutman, V., & Miller, L. C. (in press). Virtual personalities:
Using computational modeling to understand within-person variability. Journal of
Research in Personality.
Revelle, W., & Condon, D. M. (2015). A model for personality at three levels. Journal of
Research in Personality, 56, 70–81. http://doi.org/10.1016/j.jrp.2014.12.006
Rothbart, M. K., & Bates, J. E. (1998). Temperament. In N. Eisenberg (Ed.), Handbook of
child psychology, Vol. 3: Social, emotional, and personality development (5th ed.) (pp. 105–176).
New York: Wiley.
Saucier, G., & Ostendorf, F. (1999). Hierarchical subcomponents of the Big Five
personality factors: A cross-language replication. Journal of Personality and Social
Psychology, 76(4), 613–627. http://doi.org/10.1037/0022-3514.76.4.613
Talevich, J. R., Read, S. J., Walsh, D. A., Chopra, G., & Iyer, R. (2015). A comprehensive
taxonomy of human motives. Unpublished manuscript, University of Southern California.
Wiggins, J. S., & Trapnell, P. D. (1996). A dyadic-interactional perspective on the five-
factor model. In J. S. Wiggins (Ed.), The five-factor model of personality: Theoretical
perspectives (pp. 88–162). New York: Guilford Press.
Yang, Y., Read, S. J., Denson, T. F., Xu, Y., Zhang, J., & Pedersen, W. C. (2014). The
key ingredients of personality traits: Situations, behaviors, and explanations. Personality
and Social Psychology Bulletin, 40(1), 79–91.
Zhang, J., Berridge, K. C., Tindell, A. J., Smith, K. S., & Aldridge, J. W. (2009). A
neural computational model of incentive salience. PLoS Computational Biology, 5(7),
e1000437.
3
USING CONNECTIONIST MODELS
TO CAPTURE THE DISTINCTIVE
PSYCHOLOGICAL STRUCTURE OF
IMPRESSION FORMATION
Brian M. Monroe, Tei Laine, Swati Gupta,
and Ilya Farber

Introduction
In this chapter we present a novel connectionist architecture for modeling first
impression formation, the process whereby a person makes judgments about
another person based on limited observational data with no prior knowledge.
Our purposes are twofold. First, as scientists interested in the process of impres-
sion formation, we are interested in seeing whether it is possible to develop
a computational model which captures more of the subtleties and distinctive
features of this process than the existing, relatively straightforward models have
done, thus potentially underpinning an enhanced understanding of the process
more broadly. Second, as computational modelers, we here describe and attempt
to show the value of a particular strategy for creating connectionist models of
psychological processes: rather than employing a general-purpose network struc-
ture and relying on connection weights to provide all the structure, we start by
identifying key features of the target phenomenon and find ways to represent
those directly in the architecture of the network, before training occurs. As we
will demonstrate, this makes it possible to efficiently create a network model
which captures more of the distinctive properties and dynamics of impression
formation than has been possible with more conventional network architectures.

Neural Network Models

Connectionist-type models have been the most commonly adopted by person
perception researchers (e.g., Read & Miller, 1993; Kunda & Thagard 1996; see
Chapter 2 in this volume for a more extended introduction in the context of
personality processes; see also Read, Vanman, & Miller, 1997 and Smith, 1996
for more general discussions of their uses; for explorations of learning effects, see
 Impression Formation  39

Van Overwalle, 2007). The interplay among different abstract concepts in these
explanatory networks using a parallel constraint satisfaction framework has shown
how stimuli can be flexibly interpreted, which allows coherent inferences to
become active simultaneously while inhibiting inconsistent cognitions—a hall-
mark of social psychology theory (cf. Abelson et al., 1968) and an approach more
powerful than simple algebraic approaches to integration (cf. Anderson, 1968,
1981). While these networks typically do not represent memories of specific
episodes or people, other connectionist networks have shown that experience
with individuals could lead to abstractions being produced and subsequently
represented (Smith & DeCoster, 1998), which allows the networks to exhibit
properties like generalization to new stimuli. Another advantage of neural net-
works is their ability to handle parallel processing; findings suggest that different
person inferences can spontaneously become simultaneously activated (Uleman &
Moskowitz, 1994; Ham & Vonk, 2003; Freeman & Ambady, 2009), and it is
not clear how a serial model or a model that makes forced categorizations at an
early stage (e.g., Srull & Wyer, 1989; Fiske & Neuberg, 1990) would handle
this. Findings showing that forcing people to use deliberative (versus automatic)
judgment strategies decreases accuracy in a person perception task (Ambady &
Rosenthal, 1993; Patterson & Stockbridge, 1998) also suggest, albeit indirectly,
that serial processes make it hard to integrate information that is routinely used,
and that parallel integration is more natural and efficient.

Extensions to a Cognitive Neural Network Model

We have mentioned that connectionist models addressing impression formation
already exist and seem like a good candidate to explain relevant processes. So
what’s left to explain? The current chapter suggests how, by focusing on improv-
ing the fidelity of connectionist networks in reproducing human judgments, we
can refine and make more accurate a psychological theory of how impression
judgments are made. Without accounting for the abilities and biases that people
demonstrate in making their judgments, computational models simply will fall
short of embodying a valid explanation for these processes. The specific psycho-
logical features, and corresponding additions to the computational model that we
focus on are:

1. Prior probabilities—how one makes judgments when there is a lack of specific

diagnostic information; base rates or prior probabilities in belief for the attrib-
utes in the network (e.g., the likelihood of being a serial killer is small in the
absence of other information) can be useful in making these judgments.
2. Asymmetric connections—there is often an asymmetry in the nature of learned
associations between pairs of attributes (such that P(A|B) ≠ P(B|A)).
Consequently, we have incorporated asymmetry into the bi-directional con-
nections between attributes in the network.
40  Monroe, Laine, Gupta, and Farber

3. Evaluation feedback—the process of evaluating the target runs automati-

cally and concurrently to the attribute inference process; this suggests a
feedback loop between the inferences and the evaluation that influences
the representation.
4. Limited attention—the impact of differing levels of available processing
capacity (attention) to devote to the judgment task is an important deter-
minant of the output judgment; we implement this as a variable limit on
the activation available for the impression formation task.

Of course, the lack of inclusion of anything on the above list does not preclude
its potential importance for understanding the impression formation process. The
aim is to build a framework essentially from the ground up, and to test the idea
of incorporating these features into the structure of the model itself. If successful,
it could form a solid foundation that could in turn hopefully incorporate more
complicated processing as needed.
We will use a combination of thought experiments and demonstration simu-
lations with a limited version of our model in order to argue for its validity. In
the end, though, the question is if a model of this type can reproduce human
judgments. It is worth noting that none of the published models presented have
actually tried to reproduce actual human data by using a training process derived
from human data. Instead, they have used hand-wired connections and attempted
only to replicate general patterns like we do in our preliminary demonstrations.
So we then present an abbreviated set of results from the human studies we con-
ducted with a larger version of the model.

The Present Model

Our concern is with how the representation of person-relevant information
when combined with some basic processing mechanisms might lead to the types
of actual human judgments we see. A conceptual outline is presented here.
Because of the potential complexity of perceiving and interpreting each type
of information that could potentially be input into the person judgment pro-
cess (e.g., interpreting the meaning of language or figuring out the intention
of the actor in a behavior for which one doesn’t have an existing schema), we
stick to a level of analysis where schematic knowledge already exists and can be
integrated—a level that we believe can help us understand critical aspects of this
process without being overwhelming. Like many previous connectionist models
of social cognition, our instantiation of a network (Figure 3.1) is intended to
represent contents of schematic knowledge at a level that is somewhat abstracted
from individual experiences or memories. We also do not attempt to model
explicit mechanisms for acquisition of knowledge through experience (but see
Smith & Decoster, 1998; Van Overwalle & Labiouse, 2004). While not exhaus-
tive, hopefully it is rich enough to represent a mental model of another person.
 Impression Formation  41

Each node in the network represents one of two things: either (1) an attribute
that a person might have, including but not limited to physical features, per-
sonality traits, values, and other enduring mental states like beliefs, goals, or
interests;1 or (2) an observation, of a behavioral cue or other feature, that could
be predictive of these attributes.
Our connectionist network architecture, a modified autoassociator (cf.
McClelland & Rumelhart, 1981), is separated into two layers: the input layer that
consists of nodes representing observations and have only feed-forward connec-
tions to the attribute nodes, and an inference layer that consists of attribute nodes,
which can be fully and recurrently connected with excitatory and inhibitory links.
The nodes corresponding to observations about a person are externally activated,
and the activation feeds to the attribute nodes, and among them, until the net-
work settles in some stable configuration of activations. The final activation of
each attribute node represents the degree of belief that the target person has the
given attribute. The activation levels also can be positive or negative, representing
a range of belief from likely to unlikely. (See the appendix for extended details.)

Base Rates/Prior Probabilities

Imagine the situation where you are waiting in line at your regular coffee place
and you find yourself in casual banter with one of the other waiting customers.

FIGURE 3.1  S chematic depiction of the network model. An inferred attribute

(within the box) can potentially be connected to any other inferred
attribute, but these connections are not shown, to retain simplicity.
42  Monroe, Laine, Gupta, and Farber

You must to some degree make assumptions about what they know or might
be interested in, in order to carry on a simple conversation. You might be able
to predict some of these things merely from their appearance, and the fact that
they are in a coffee joint to begin with. But some of the assumptions you make
about them might be based on the fact that they are a female whose age can
be approximated visually, rather than on the aforementioned. This represents
knowledge averaged over a larger pool of experience. And still other assumptions
might be based merely on overall experience with people, the largest possible
pool. Another term that can describe this most general inference is a normative
expectation about people in general.
We reason that if none of the possible predictive cues for a judgment of
interest is present and available, or one merely doesn’t notice any of the cues
that could be predictive, one must rely on what the other customer knows, in
the absence of knowledge of these predictors. Importantly, there is no single
cue that is always present, but as people, that doesn’t stop us from making
assumptions about the target. We take advantage of the fact that a person must
at some level recognize that the entity they are perceiving is a person. Therefore
it makes sense to include a node that represents this “person” feature of the
judgment object. It also makes sense that this node will always be active, since
that defines the domain of the model. This node has the power to account
for inferences that are not accounted for by other nodes in the network, and
which it will have acquired through experience with every person who has
been encountered. Through associative learning, this connection would natu-
rally have built up to form stronger links for common attributes, and weaker
links for rarer attributes, making common attributes more likely to be recruited
in forming an impression. It is important to note that this base rate implemen-
tation also influences the impression even when the other predictive cues are
present, in combination with them.

Base Rate Simulations

In our simulation of this first thought experiment (Figure 3.2), one can imagine
that if an arbitrary feature such as having earned a college degree is not diagnostic
of lawfulness, but rather is the normative expectation of people in general that
leads to this conclusion, it is important for a connectionist model to be able to
represent this, resulting in the same conclusion whether the arbitrary feature is
there or not. A model that lacked this would either not be able to make a predic-
tion at all, or misattribute the predictive value to the arbitrary feature, in essence
leading the knowledge structure to inaccurately represent what the perceiver has
learned through experience.
In the second thought experiment (Figure 3.3), we can see what hap-
pens when we combine a base rate with another attribute that is diagnostic.
 Impression Formation  43

FIGURE 3.2   ase rates between conditions. Numbers displayed inside nodes are
B
settled activations for inferred attributes, or external activations for
input observations. Numbers accompanying arrows are the connection
weights. (a) Observed attribute is only “College graduate” (plus
“Non-diagnostic appearance,” not shown). (b) Observed attribute
is only “Appearance” (not shown). Grayed-out nodes are inactive,
representing observations that are not currently present, but are kept
for comparison purposes.

Inferences can be equally sensitive to diagnostic inputs, even though they

have different levels of likelihood in absolute terms. This is easily illustrated
by contrasting an inference to an attribute that is relatively common with an
inference to an attribute that is rare. Here we assume that being intelligent is
likely to increase judgments of friendliness and also judgments of being a scam
artist, both slightly. If one did not take into account their overall prevalence
level, the model would make predictions that each was equally likely to be
true. But since that does not match the intuition that people are not likely
to assume someone is a scam artist based on the observation of being a smart
person, we show that adding the base rate node allows the model to make the
correct prediction.

Asymmetric Relations between Nodes

The model we are presenting is ultimately a model of inference about person
qualities that have not been directly observed, based on those things that are
observed. A common way for inferences under uncertainty to be expressed is
in terms of conditional probabilities. There is no requirement that P(A|B) =
P(B|A); that would be a special case. These probabilities usually represent
degrees of belief in a certain proposition. Even when using subjective beliefs
as opposed to objective frequentist probabilities, observation of A can be more
44  Monroe, Laine, Gupta, and Farber

FIGURE 3.3  Base rates within-condition.

revealing about B than vice versa. Especially when A and B are related by a
deductive rule, it is likely that the inference is stronger in one direction than
the other. When making predictions of continuous variables, regression coef-
ficients (as well as semipartial correlations) are asymmetric when used among
the same set of variables in the alternate roles of predictor and outcome. These
asymmetric conditional probabilities and regression coefficients are the quan-
tities that are learned by standard network learning algorithms (O’Reilly &
Munakata, 2000).
Considering this situation, it is somewhat surprising that other models of
person inference have exclusively used symmetric relationships between nodes
(e.g., Read & Miller, 1993; Kunda & Thagard, 1996). It is possible these models
did not explore cases where precise discrimination was necessary, and so could
get away with only symmetric relationships. Thagard (2004) argued that using
(unidirectional) conditional probabilities as instantiated in a Bayesian network,
instead of symmetric weights, did not improve the quality of prediction in
certain inference networks. We believe the focus of Thagard’s explanation net-
works is typically of facts of physical occurrences that either can both happen
or are mutually exclusive, for which “consistent” or “inconsistent” symmetrical
coherence relationships might be more appropriate. The judgments those mod-
els produce are also interpreted as either accepting or rejecting a proposition
or explanation. Impression judgments, in contrast, do not require complete
acceptance or rejection, but instead seem to involve combining predictions
of the diagnostic value of available pieces of information. Even if a symmetric
model is able to produce data that match human coherence judgments well
when they are interpreted dichotomously, it is less certain that it will work
for more graded strength-of-prediction judgments. The base rate exploration
 Impression Formation  45

suggests a case where it is important to judge attributes on the full continuum

from likely to unlikely, rather than only allowing extremely likely or extremely
unlikely judgments.
Regardless of the above, one could imagine that a well-trained network will
eventually learn to predict correctly based on symmetric inference weights, plus
the weights coming from the input nodes (in the latter case only the weight
going from observation to inference matters because input activations are fixed.)
In the previous impression formation models with symmetric weights, the num-
ber of judgments that were correctly made was fewer than the number of free
parameters (the connection strengths) in the network, so it was always possible
to use a set of parameters in combination to get those judgments to come out
correctly (a situation that could be described as an overfit model). The issues this
raises are that symmetric weight patterns might not work when a much larger
set of predictions is attempted, and subsequently this casts doubt that it mean-
ingfully reproduces the psychological representation. For all the above reasons,
we believe using asymmetric connections instead is an appropriate solution to
accomplishing the types of inference one must make for impression judgments.
This is a case where the result is an increase in the number of parameters in
models that are often already overfit, but we feel it is justified.

Asymmetry Simulation
The advantage is demonstrated in the simulation of the following thought
experiment (Figure 3.4). The two inferences of interest are whether someone
is an ignorant person, and whether someone is an intolerant person. In the first
case, the perceiver observes the target person pepper spraying the participants at
a pride parade for a minority group (a hateful act). In the second case, it arises
in conversation that someone doesn’t know a relatively common fact. These
are fairly straightforward unambiguous inferences, to intolerance in the first
case, and ignorance in the second case. What is of importance is how, given
one attribute, the perceiver judges the likelihood of the other (ignorance in the
first case, and intolerance in the second). If a network with symmetric-only
connections is used, and the observations are equally strong predictors of their
respective traits, the judgments in the two cases would be identical. In contrast,
in a model with asymmetric connections,2 the person who pepper sprays is
judged to be ignorant more than the person who doesn’t know the common
fact is judged to be intolerant. This is because intolerance has a higher cue value
for ignorance than vice versa. This can be thought of as a knowledge struc-
ture where ignorance is a requirement for intolerance, but intolerance is not a
requirement for ignorance.
46  Monroe, Laine, Gupta, and Farber

FIGURE 3.4   symmetry. Grayed-out nodes show inactive inputs. Base rates for this
A
simulation were assumed to be similar, and were not included in the
network. (a) Observation of intolerant behavior. (b) Observation of
ignorant behavior.

Summary Evaluation and Valence Feedback

When people judge entities in their world, evaluation on a positive–negative
dimension is one of the most fundamental aspects of that. Studies of abstract
concepts, items, or adjectives show that evaluation accounts for much of the
variance in their cognitive representation (e.g., Osgood, Suci, & Tannenbaum,
1957; Heise, 1969). In person perception research, positive qualities are typically
correlated with each other, as are negative qualities (e.g., Asch 1946; Nisbett &
Wilson, 1977; Cooper, 1981; Anderson & Sedikides, 1991; Eagly, Ashmore,
Makhijani, & Longo, 1991). Further, trait qualities that are evaluatively tinged
seem to be judged less reliably than neutral items (John & Robins, 1993). This
suggests that valence is an indispensible part of the representation of a person, and
further, that somehow the evaluation of someone actively influences the process
of forming a judgment of them or their qualities.
These findings make sense when looking at the role of evaluation in an agent
who must then act on their judgments. Echoing the premise that “thinking is for
doing” (cf. Ferguson & Bargh, 2004), it has been theorized that systems exist with
the specific function of resolving conflicts and removing ambivalence from judg-
ments so we can act decisively (Harmon-Jones, Harmon-Jones, Fearn, Sigelman, &
Johnson, 2008; Harmon-Jones, Amodio, & Harmon-Jones, 2009). The evalua-
tion system serves to inform the agent whether they should approach or avoid
the object of interest—a critical decision-based mechanism for an individual.
Consistent with this mechanism, it has more specifically been suggested that once
 Impression Formation  47

an evaluation has formed, it shapes subsequent information processing (Srull &

Wyer, 1989), and also that there is a bi-directional pathway of influence between
evaluation and beliefs as examined from perspectives in decision modeling
(Wagar & Thagard, 2004), emotional appraisal coherence (Nerb, 2007), social
neuroscience (Cunningham & Zelazo, 2007), and attitude modeling (Monroe &
Read, 2008; Ehret, Monroe, & Read, 2014). Previous modeling efforts have also
recognized that influences from valenced emotional responses can account for
judgments made from available evidence (Thagard, 2003). This general valence
mechanism in some form could plausibly account for a broad range of biased
reasoning phenomena (e.g., Kunda, 1990).
In order to account for this, we have implemented connections to two valence
nodes (positive and negative)3 that not only read out the evaluative components
of the other inferences, but their own activations influence the activation of the
other nodes by means of a bi-directional connection. We have little basis to
assume the exact strength of this feedback connection, so in the current exercise
the connections between attribute nodes and valence nodes are symmetric; a
looser condition that we would agree with only requires that the feedback con-
nections be positively correlated with the respective connections from attribute
to valence. In this model, the strength of the connections from the attribute
nodes to the valence nodes is fixed.
The valence nodes in our model provide a functional-level implementation
of this dynamic inhibition without representing a commitment to any particular
hypothesis about the underlying brain mechanisms. Theories of how valenced
representations directly affect propositional representations have not been tested
directly (but see Thagard & Aubie, 2008 for one proposed mechanism that spe-
cifically includes emotion appraisals) and there are several possibilities; we have
chosen to implement the minimum essential element that is able to highlight
the advantages of putting in a valence loop. The advantage of incorporating this
element into the model is best illustrated through simulations.
An empirical finding that evidences the operation of a valence-driven pro-
cess is one that suggests impression consistency ratings are far more sensitive
to valence than they are to semantic relations. In experiments by Roese &
Morris (1999), subjects were given several hypotheses to judge simultaneously
about both momentary and dispositional attributes that could be suggested by
the behavior of a target person in a scenario. When rating the likelihood that
certain attributes or behavioral explanations were compatible with each other,
the judgments of valence were far more important than merely how the expla-
nations covaried, or what other causal mechanisms they might share. So, for
example, if someone is known to be humorous, the ratings of their conscien-
tiousness tend to be higher than ratings of their dishonesty, and the most critical
factor in these ratings across different combinations of attributes (as determined
by multiple regression analyses) is the consistency of valence of those attributes
48  Monroe, Laine, Gupta, and Farber

rather than a perceived covariation of the attributes or even the other personality-
behavioral mechanisms they entail. These experiments and corresponding
simulation also allow us to illustrate how valence effects are important within
the same judgment episode.

Valence Simulation
In our thought experiment and simulation, inspired by the Roese & Morris (1999)
experiments (Figure 3.5), we externally activate the default explanation (humorous,
which was the attribute given to the participants in the example above). Activation
spreads to both the inferred attributes equally, but also to the valence nodes. The
pattern of weights only among the attributes is effectively symmetric, and so on its
own would result in both attributes being equally active. However, the activation
of the valence nodes sends excitatory input to the valence-compatible attribute,
conscientious, and inhibits the valence-incompatible attribute, dishonest. A test
for the ratings of likelihood of the secondary explanations, which is represented by
the resulting activation value of the relevant attribute or behavior inference, con-
firms the pattern. The simulations show that the valence-compatible attributes are
favored, while the valence-incompatible attributes are suppressed, which is consist-
ent with the ratings of the likelihood that the conjunction of attributes is present as
indicated in the results from the experiment.
One could imagine similar results possible, without the process being mediated
by a valence pathway, as long as the relevant attributes were already significantly
positively correlated or negatively correlated in their schematic knowledge or
shared common personality-behavioral mechanisms. But this only pushes the
question back a step, where one has to ask where the consistent positive and
negative correlations between same and oppositely valenced items come from

FIGURE 3.5   alence feedback within-condition. Base rates were assumed to be

V
similar and not included in the network.
 Impression Formation  49

in the first place. The sheer difference in size of the effects of Roese & Morris
(1999) between covariation and valence also calls into question the potential for
genuine covariation to account for the valence effect. Therefore we would argue
the mechanism we have demonstrated is a much more parsimonious explanation
for this pattern of results.

Cognitive Capacity
Cognitive capacity, or the amount of available attentional capacity one devotes to
the target task, is a central variable of interest in many social cognitive models of
information processing. In addition to being prominent in cognitive theories of
attitude change (Petty & Cacioppo, 1986; Kruglanski & Thompson, 1999), person
perception researchers have also noted that the amount of attention or processing
one devotes to person stimuli can have significant effects on the conclusions one
draws about the target (Brewer & Feinstein, 1999; Fiske, Lin, & Neuberg, 1999).
Low attention promotes reliance on well-learned associations like stereotypes
(Macrae, Hewstone, & Griffiths, 1993; Wigboldus, Sherman, Franzese, & Van
Knippenberg, 2004) and insufficient consideration of situational constraints on
behavior (Gilbert, Pelham, & Krull, 1988; Trope & Alfieri, 1997). High attention
promotes consideration of individuating information and more detailed represen-
tations of the target person (Fiske, Neuberg, Beattie, & Milberg, 1987; Pendry &
Macrae, 1994). Knowing that even when presented with the same information
people can come to completely different characterizations merely based on the
amount of attention paid to the target, we believe it is important to incorporate
a mechanism to allow this type of processing into the model.

Implementation of Cognitive Capacity

While it is very important to understand how perceptual features are identified
and fit together to build up a conceptual interpretation of an observation and
how processing capacity might affect this, it is beyond the scope of the current
model. So for this exploration we bypass the issue of low-level perception and
focus only on how high versus low capacity differentially affects concept activa-
tion. Therefore we only put a hard ceiling on the amount of activation available
for processing in the combined pool of downstream inferred concepts.4 Perhaps
most consequential of the potential factors that can interact with capacity in these
limited circumstances is the order in which information is received through time,
which we explore at length here.

Order Effects
It is well documented that when people are given pieces of information in
series about a target person, the overall impression is sensitive to the order,
even though the overall pool of information is the same (e.g., Asch, 1946;
50  Monroe, Laine, Gupta, and Farber

Anderson, 1965; Dreben, Fiske, & Hastie, 1979). Aside from artificial lab
manipulations of order, the manner in which observations about a person
are made in a natural environment suggests that it is important to address
this aspect of impression judgments. The cliché of saying that one should not
judge a book by its cover reflects the inescapability of perceiving some pieces
of information before others, and the potential for the initially encountered
stimuli to drive the judgment so strongly that in some cases it prevents the
perceiver from even attending to further information.
A network seems appropriate to model this because its mechanisms match
those proposed on the basis of empirical findings in impression formation.
Specifically, the activation of categorical and/or trait schema representations has
been shown to have a marked effect on resulting judgments (Srull & Wyer, 1979;
Bodenhausen, 1988), and it is the interpretation of subsequent incoming stimuli via
this temporarily established influence that leads to primacy effects (e.g., Jaccard &
Fishbein, 1975; cf. Kunda & Thagard, 1996). However, it is important to note
that this tendency doesn’t always hold: when placed under sufficient compet-
ing processing demands, recency effects can be seen such that given a series of
person attributes, the ones given later tend to dominate the overall judgment
(Lichtenstein & Srull, 1987).

Simulation: Primacy and Recency Effects

In these simulations (Figures 3.6 and 3.7), the network received a series of four
behaviors, in counterbalanced order. The network consists of several different
related attributes, intended to represent related parts of a schema. Related parts
of the schema are activated, which likely activate associated attributes to some
degree. In this schema, we have represented the two focal attributes (intelligent
and honest) as well as two additional ones that are associated with each of them to
some degree, for a total of six attributes. We have also included as the inputs to the
network one behavior related to each of the two focal attributes, and correspond-
ing negative behaviors related to the same traits. We activated the first behavior
and allowed the network to settle in its own time due to the dynamic evolution
of activation; then the observation nodes were turned off (but preserving the state
of the rest of the network) and the new observation was activated, and this was
repeated until the series was complete. When the cognitive capacity was sufficient
for the schema associated with the behaviors to become sufficiently active (which
was then preserved between introductions of new observations in a continuous
stream), this schema constrained and inhibited new information that was incon-
sistent with it—so the traits that became active earlier remained active (a primacy
effect). When the cognitive capacity was reduced, the activation of any given
schema was not strong enough to produce this effect, so the recency advantage of
the later information allowed it to influence the final pattern of activation more
(a recency effect). This matches the pattern seen in Lichtenstein & Srull (1987).5
FIGURE 3.6   etwork for cognitive capacity simulation. Connections between
N
inferred attributes are present but not shown.

FIGURE 3.7  S imulation results for cognitive capacity. (a) Full capacity. (b) Low
capacity. Activation in trait nodes (not shown) shows a similar pattern
to the valence nodes. Each subsequent observation was introduced after
the network settled into a stable configuration due to its own dynamics;
the number of time steps varied in each case (roughly, more disruption
requires more time to resolve).
52  Monroe, Laine, Gupta, and Farber

FIGURE 3.7  (continued)

Empirical Test of the Model

The ability of the constructed models to produce the intended outcomes pro-
vides an initial form of validation. Ultimately, however, the real test of the value
of the structural features described above is whether a model that has them can
more accurately replicate human performance than a model that lacks them. In
a large-scale project, we collected human data on inferences of traits. People
provided their judgments, given either (1) single observed characteristics, or
(2) combinations of a few characteristics together in mini-profiles. Practical con-
cerns dictated choices for the large-scale model, including data collection volume
(and thus the size of the network), and we did not manipulate attention or cog-
nitive load during exposure to the stimuli. In the end, the large-scale model for
which we report results had 47 different characteristics as possible observed inputs
(plus base rate if applicable), and 52 traits as possible inferences. The observed
characteristics included things like gender, age, ethnicity, profession, and hob-
bies. Instead of setting weights by hand, it was trained using the human data for
single observed characteristics, with single input–inference pairs at a time and
then performance-tested on the profiles containing multiple characteristics. The
performance during training (and testing) was based on how close the model’s
settled activation for a given node was to the human data, re-scaled to fit the acti-
vation range of the model. Any discrepancy was considered an error with respect
 Impression Formation  53

to the training data. Critically, all models were trained to the same mean-square-
error (MSE) standard on the training corpus.
The main comparison of interest for us is against a model like the one used by
Kunda & Thagard (1996) (hereafter KT), which is a connectionist network, but one
that does not contain any of the critical features we have discussed up to this point.
Rather than testing a new model with all of these features, it is important to selectively
add or subtract one at a time the feature critical to demonstrating a particular effect so
one could determine if it in fact caused any possible improvements in performance.
A full analysis and description of these results will be the subject of future work
and is beyond the scope of this chapter. Here we present only a few key observa-
tions based on an analysis of simulations using our present model and comparison
models via human judgments,6 to illustrate some of the ways in which empirical
data can be used to assess the significance of the particular architectural choices
presented in this chapter.7

1. Adding a base rate only to a model like KT allows the model to match
human judgments on a one-to-one basis much better than a KT model. This
is especially so when more observed characteristics were present in a profile,
where the KT model’s estimates were systematically more extreme. This
suggests the model without the base rate was misattributing the diagnostic
value to the observed characteristics, rather than normative expectations of
people in general.
2. Adding asymmetric weights allows a model to perform better than a KT
model. This suggests the asymmetry is in fact important for replicating
human inferences.
3. Adding valence nodes allows a model to perform better than a KT model,
and the strength of this advantage is positively correlated with the extremity
of the evaluation of the target profile. This suggests that the valence nodes
are in fact important, and influence human inferences in a causal fashion
above and beyond the trait schemas that are active.
4. Adding multiple components to the model improved performance incre-
mentally over models with just single additional components.

Current Limits and Future Paths

In this chapter we have demonstrated processing features that have not previously
been included in models of impression formation, and which we believe are criti-
cal to account for because they capture psychological features relevant to forming
these judgments. We have focused more for the time being on explaining how and
why these features are necessary. But these concerns are ultimately equally impor-
tant because accounting for them will result in a framework that would be more
successful in accurately reproducing human judgments of target persons.We believe
that existing models of impression formation, both verbal and computational, do
54  Monroe, Laine, Gupta, and Farber

not meet these criteria. The model is fairly limited in its scope, though. We recog-
nize there are many aspects of impression formation that are important, which this
model is not explicitly designed to account for.
For example, this model does none of the upstream processing that would
be necessary to understand people in real time, such as facial expression
recognition, language understanding, or parsing and interpreting actions. We
have limited the processing in the network to the social meaning level, to
simplify the task and examine it at a level that is manageable, but recognize
that in order to understand a fuller complement of phenomena, one cannot
overlook these phases of processing. For example, the inferences that are drawn
from physical appearance that can be perceived immediately might typically
precede inferences drawn from behaviors that are observed. Thus the inferences
that have been activated first from appearance will then interact (and possibly
compete) with the inferences from behavior. In this case we believe it will be
important to account for the temporal order, especially when the observations
are not strongly diagnostic themselves and depend on combining evidence from
several sources to make a significant inference. This is an additional concern
that suggests the importance of incorporating temporal differences, as we have.
It is worth mentioning that a model could be extended in a manner entirely
consistent with the current architecture by implementing multiple-layer
representations, each layer hierarchically building up more complex or abstracted
inputs from simple ones in appearance or language or behavior comprehension
(cf. Read & Miller, 1998, 2005; Freeman & Ambady, 2011), each of these layers
also connecting to the common pool of inferred attributes.

Conclusion
Much of what we know about impression formation is based on experiments
where one factor is manipulated to understand how it affects another variable.
What is still lacking is an integrative framework of how an impression is formed.
There are descriptions of the what of impression formation that are impressively
comprehensive (Carlston, 1994), but this needs to be combined with the how.
In order to address this, we have attempted to abide by a few guiding principles:
(1) in making a computational model in particular, we are forced to be more pre-
cise than existing verbal models (e.g., Brewer, 1988; Srull & Wyer, 1989; Fiske &
Neuberg, 1990) in specifying how information gets combined; (2) avoiding
assumptions or distinctions in representation or processing unless the existing
framework is arguably unable to account for the phenomena; and (3) advancing
existing network architectures by noting where the deficiencies arise in their abil-
ity to replicate human judgments. It is our hope that by doing these things, the
resulting model will have explanatory value as both a parsimonious, coherent, and
broadly applicable model of certain aspects of social cognition, and also one that
will be able to actually predict human judgments in real circumstances.
 Impression Formation  55

Towards the third aim especially, we have gone beyond the KT model, and
suggested that a network without the features we have described throughout this
chapter fails to capture important aspects of person judgments. Prime examples of
this are: (1) the ability of the current model to make inferences that discriminate
between common and rare attributes, which suggests that it is very important for
the researcher to pay attention not just to what is observed about the target, but
also what is not observed (in the sense that base rate expectations derived from
observations containing information that is not present in the current observa-
tion will continue to influence judgments) when modeling person inferences;
(2) the ability for multiple levels of abstraction or construal to simultaneously
influence perceptions, such as the most general person representation (i.e., base
rate) combining with the gender or ethnicity of a person; one does not need to
choose only one level of abstraction from a hierarchy to construe a person, as each
might have independent predictive value; and (3) allowing for rule-based infer-
ences that are likely asymmetric.
The shift in perspective that comes out of this modeling exercise should not
be overlooked. While we do not view our results as embodying a full theory
of person judgments, we have tried to use an approach that has the potential to
accommodate a comprehensive account of the process and outcome. Instead of
positing a domain-specific (and ultimately unexplained) fixed sequence of pro-
cessing, as do many entrenched verbal theories, the sequence of processing is
assumed to conform to a more general perceptual-cognitive set of rules. It has
been shown that sequences of judgments can often seemingly be reversed from
that of prevailing theories (e.g., Krull, 1993), which suggests that those theories
take too narrow a view. Specifically, impression judgments are highly depend-
ent on the schematic knowledge structures that people have (represented in the
current model by the pattern of connections and weights between nodes)—one
cannot form an impression without such knowledge, and so the schema should
be a major determinant of the flow of processing, along with the attention paid
to the task at hand, personal motivations, and the spatiotemporal configuration
in which the information about the target is observed. By taking a more general
schematic view, one can, for example, start to generate hypotheses about why
certain social categories seem to have the operational properties they do, rather
than just assume that they rigidly result in reduced attention to other attributes as
is the case in dual-process models (e.g., Brewer, 1988; Fiske & Neuberg, 1990).
One can also hopefully be on firmer ground to explore more automatic versus
more deliberate processing and how different judgments are produced if that
processing must either be assimilated or contrasted with a mechanism that has
been shown to match human judgments in a wide set of cases. More generally,
in being forced to specify processing mechanisms and relationships, it can sug-
gest gaps in our knowledge of the impression formation process, and force us to
re-examine existing theories (cf. McClelland, 2009; Sun, 2009). A model like this
one could be the basis for a more complete account of impression processing and
56  Monroe, Laine, Gupta, and Farber

judgments, and progress in understanding can indeed be made through a cycle of

adequate testing and extension of the model.

Appendix
The net input η to a node j is the sum of the activation-weight products across
all nodes it is connected to. Since weights are potentially asymmetric, the model
uses the incoming weight.
η j = ∑xi w ij
i  (3.1)
And this evolves over time steps t with a leaky integration process, becoming

η j (t ) = (1 − δt ) η j (t − 1) + δt ( ∑xi wij )  (3.2)

i

with a fractional integration parameter δ, activations x, and weights w.

The outgoing activation x of a node j is the net input transformed by a sigmoi-
dally shaped function to bound it between -1 and 1.
ηj
x j = tanh( ) (3.3)
2
Processing in the models starts with observation nodes having their activation
values set externally. Processing in the model stops when the change in activation
over successive time steps over all nodes in the network falls below a specified
limit.
For the large model with learning implemented, the learning algorithm combines
error-correcting learning with a form of purely associative learning. The error-
correcting learning was implemented using contrastive Hebbian learning

∆w ij = [ xi+ y +j − xi− y −j ]  (3.4)

where the superscripts denote the plus and minus phases of activation
(O’Reilly & Munakata, 2000) and the associative learning used Oja’s rule:

∆w ij =  ( xi y j − y 2j w ij )  (3.5)
using the minus phase activations to calculate the weight change. The proportion
of associative learning was set at 0.1% of the total weight change, with the
remainder due to error-correcting learning.

Notes
1 It is important to emphasize that not all attribute nodes need be abstract personality
traits. For example, a central tenet of correspondent inference theory (Jones & Davis,
1965) is that when a behavior is observed, the perceiver naturally makes the inference
that the actor desired the specific outcome(s) their behavior achieved. This inference
 Impression Formation  57

of present desires can then be connected to more stable personality attributes through
schematic knowledge (this connection is highlighted, e.g., in Read, Jones, & Miller,
1990), but the present desire inference is an important and potentially necessary inter-
mediate step.
2 We limit ourselves presently to cases in which the subjective base rates are roughly the
same to illustrate this point in a simple demonstration, although it should apply as well
even in cases when the base rates are different.
3 For present purposes these function similarly to a single bipolar node. For demonstra-
tions of positive and negative nodes functioning independently, see Ehret, Monroe, &
Read (2015).
4 For more discussion of the nature of the capacity mechanism, see Monroe & Read
(2008).
5 We note that our model suggests that sufficient activation is required among the inferred
attributes to achieve a primacy effect, even under full capacity, because if the pattern of
activation is not strong, the more recent inputs will have more influence. Thus there are
multiple factors in the model that determine whether primacy forces are strong enough
to control the judgment.
6 B. M. Monroe, T. Laine, S. Gupta, & I. Farber, unpublished raw data, 2016.
7 Except where specifically noted, performance measures used a combination of within-
profile correlations across the entire set of profiles, as well as MSE comparisons, all
comparing the human ratings of the 52 traits per profile to the model outputs on the
traits.

References
Abelson, R. P., et al. (Eds.) (1968). Theories of cognitive consistency: A sourcebook. Chicago,
IL: Rand-McNally.
Ambady, N., & Rosenthal, R. (1993). Half a minute: Predicting teacher evaluations from
thin slices of nonverbal behavior and physical attractiveness. Journal of Personality and
Social Psychology, 64(3), 431–441.
Anderson, C. A., & Sedikides, C. (1991). Thinking about people: Contributions of a
typological alternative to associationistic and dimensional models of person perception.
Journal of Personality and Social Psychology, 60(2), 203–217.
Anderson, N. H. (1965). Averaging versus adding as a stimulus-combination rule in
impression formation. Journal of Experimental Psychology, 70(4), 394–400.
Anderson, N. H. (1968). Application of a linear-serial model to a personality-impression
task using serial presentation. Journal of Personality and Social Psychology, 10(4), 354–362.
Anderson, N. H. (1981). Foundations of information integration theory. London: Academic
Press.
Asch, S. (1946). Forming impressions of personality. Journal of Abnormal and Social
Psychology, 41(3), 258–290.
Bodenhausen, G. V. (1988). Stereotypic biases in social decision making and memory:
Testing process models of stereotype use. Journal of Personality and Social Psychology,
55(5), 726–737.
Brewer, M. B. (1988). A dual process model of impression formation. In T. K. Srull &
R. S. Wyer (Eds.), Advances in Social Cognition (Vol. 1, pp. 1–36). Hillsdale, NJ:
Lawrence Erlbaum.
Brewer, M. B., & Feinstein, A. S. H. (1999). Dual processes in the cognitive representa-
tion of persons and social categories. In S. Chaiken & Y. Trope (Eds.), Dual-process
theories in social psychology (pp. 255–270). New York: Guilford Press.
58  Monroe, Laine, Gupta, and Farber

Carlston, D. E. (1994). Associated Systems Theory: A systematic approach to cognitive

representations of persons. In T. K. Srull & R. S. Wyer (Eds.), Advances in Social
Cognition (Vol. X, pp. 1–78). Hillsdale, NJ: Lawrence Erlbaum.
Cooper, W. H. (1981). Ubiquitous halo. Psychological Bulletin, 90(2), 218–244.
Cunningham, W. A., & Zelazo, P. D. (2007). Attitudes and evaluations: A social cognitive
neuroscience perspective. Trends in Cognitive Sciences, 11(3), 97–104.
Dreben, E. K., Fiske, S. T., & Hastie, R. (1979). The independence of evaluative and
item information: Impression and recall order effects in behavior-based impression
formation. Journal of Personality and Social Psychology, 37(10), 1758–1768.
Eagly, A. H., Ashmore, R. D., Makhijani, M. G., & Longo, L. C. (1991). What is
beautiful is good, but . . .: A meta-analytic review of research on the physical attrac-
tiveness stereotype. Psychological Bulletin, 110(1), 109–128.
Ehret, P. J., Monroe, B. M., & Read, S. J. (2015). Modeling the dynamics of evaluation:
A multilevel neural network implementation of the Iterative Reprocessing Model.
Personality and Social Psychology Review, 19, 148–176.
Ferguson, M. J., & Bargh, J. A. (2004). Liking is for doing: The effects of goal pursuit on
automatic evaluation. Journal of Personality and Social Psychology, 87(5), 557–572.
Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). The continuum model: Ten years later.
In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 231–254).
New York: Guilford Press.
Fiske, S. T., & Neuberg, S. L. (1990). A continuum of impression formation, from
category-based to individuating processes: Influences of information and motivation
on attention and interpretation. Advances in Experimental Social Psychology, 23, 1–74.
Fiske, S. T., Neuberg, S. L., Beattie, A. E., & Milberg, S. J. (1987). Category-based and
attribute-based reactions to others: Some informational conditions of stereotyping and
individuating processes. Journal of Experimental Social Psychology, 23(5), 399–427.
Freeman, J. B., & Ambady, N. (2009). Motions of the hand expose the partial and parallel
activation of stereotypes. Psychological Science, 20(10), 1183–1188.
Freeman, J. B., & Ambady, N. (2011). A dynamic interactive theory of person construal.
Psychological Review, 118(2), 247–279.
Gilbert, D. T., Pelham, B. W., & Krull, D. S. (1988). On cognitive busyness: When
person perceivers meet persons perceived. Journal of Personality and Social Psychology,
54(5), 733–740.
Ham, J., & Vonk, R. (2003). Smart and easy: Co-occurring activation of spontaneous
trait inferences and spontaneous situational inferences. Journal of Experimental Social
Psychology, 39(5), 434–447.
Harmon-Jones, E., Amodio, D. M., & Harmon-Jones, C. (2009). Action-based model of
dissonance: A review, integration, and expansion of conceptions of cognitive conflict.
Advances in experimental social psychology, 41, 119–166.
Harmon-Jones, E., Harmon-Jones, C., Fearn, M., Sigelman, J. D., & Johnson, P. (2008).
Left frontal cortical activation and spreading of alternatives: Tests of the action-based
model of dissonance. Journal of Personality and Social Psychology, 94(1), 1–15.
Heise, D. R. (1969). Some methodological issues in semantic differential research.
Psychological Bulletin, 72(6), 406–422.
Jaccard, J. J., & Fishbein, M. (1975). Inferential beliefs and order effects in personality
impression formation. Journal of Personality and Social Psychology, 31(6), 1031–1040.
John, O. P., & Robins, R. W. (1993). Determinants of interjudge agreement on personality
traits: The Big Five domains, observability, evaluativeness, and the unique perspective of
the self. Journal of Personality, 61(4), 521–551.
 Impression Formation  59

Jones, E. E., & Davis, K. E. (1965). From acts to dispositions: The attribution process in
person perception. Advances in experimental social psychology, 2, 219–266.
Kruglanski, A. W., & Thompson, E. P. (1999). Persuasion by a single route: A view from
the unimodel. Psychological Inquiry, 10(2), 83–109.
Krull, D. S. (1993). Does the grist change the mill? The effect of the perceiver’s inferential
goal on the process of social inference. Personality and Social Psychology Bulletin, 19(3),
340–348.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3),
480–498.
Kunda, Z., & Thagard, P. (1996). Forming impressions from stereotypes, traits, and behav-
iors: A parallel-constraint-satisfaction theory. Psychological Review, 103(2), 284–308.
Lichtenstein, M., & Srull, T. K. (1987). Processing objectives as a determinant of the
relationship between recall and judgment. Journal of Experimental Social Psychology,
23(2), 93–118.
Macrae, C., Hewstone, M., & Griffiths, R. J. (1993). Processing load and memory for
stereotype-based information. European Journal of Social Psychology, 23(1), 77–87.
McClelland, J. L. (2009). The place of modeling in cognitive science. Topics in Cognitive
Science, 1(1), 11–38.
McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context
effects in letter perception I: An account of basic findings. Psychological Review, 88,
375–407.
Monroe, B. M., & Read, S. J. (2008). A general connectionist model of attitude structure
and change: The ACS (Attitudes as Constraint Satisfaction) model. Psychological Review,
115(3), 733–759.
Nerb, J. (2007). Exploring the dynamics of the appraisal–emotion relationship: A constraint
satisfaction model of the appraisal process. Cognition and Emotion, 21(7), 1382–1413.
Nisbett, R. E., & Wilson, T. D. (1977). The halo effect: Evidence for unconscious
alteration of judgments. Journal of Personality and Social Psychology, 35(4), 250–256.
O’Reilly, R. C., & Munakata, Y. (2000). Computational explorations in cognitive neuroscience:
Understanding the mind by simulating the brain. Cambridge, MA: MIT Press.
Osgood, C. E., Suci, G. J., & Tannenbaum, P. H. (1957). The measurement of meaning.
Oxford: University of Illinois Press.
Patterson, M. L., & Stockbridge, E. (1998). Effects of cognitive demand and judg-
ment strategy on person perception accuracy. Journal of Nonverbal Behavior, 22(4),
253–263.
Pendry, L. F., & Macrae, C. (1994). Stereotypes and mental life: The case of the motivated
but thwarted tactician. Journal of Experimental Social Psychology, 30(4), 303–325.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of persuasion.
Advances in experimental social psychology, 19(1), 123–205.
Read, S. J., Jones, D. K., & Miller, L. C. (1990). Traits as goal-based categories: The
importance of goals in the coherence of dispositional categories. Journal of Personality
and Social Psychology, 58(6), 1048–1061.
Read, S. J., & Miller, L. C. (1993). Rapist or “regular guy”: Explanatory coherence in
the construction of mental models of others. Personality and Social Psychology Bulletin,
19(5), 526–540.
Read, S. J., & Miller, L. C. (1998). On the dynamic construction of meaning: An
interactive activation and competition model of social perception. In S. J. Read &
L. C. Miller (Eds.), Connectionist models of social reasoning and social behavior (pp. 27–68).
Mahwah, NJ: Lawrence Erlbaum.
60  Monroe, Laine, Gupta, and Farber

Read, S. J., & Miller, L. C. (2005). Explanatory coherence and goal-based knowledge
structures in making dispositional inferences. In B. F. Malle & S. D. Hodges (Eds.),
Other minds: How humans bridge the divide between self and others (pp. 124–139). New
York: Guilford Press.
Read, S. J., Vanman, E. J., & Miller, L. C. (1997). Connectionism, parallel constraint
satisfaction processes, and Gestalt principles: (Re)introducing cognitive dynamics to
social psychology. Personality and Social Psychology Review, 1(1), 26–53.
Roese, N. J., & Morris, M. W. (1999). Impression valence constrains social explana-
tions: The case of discounting versus conjunction effects. Journal of Personality and Social
Psychology, 77(3), 437–448.
Smith, E. R. (1996). What do connectionism and social psychology offer each other?
Journal of personality and social psychology, 70(5), 893.
Smith, E. R., & DeCoster, J. (1998). Knowledge acquisition, accessibility, and use
in person perception and stereotyping: Simulation with a recurrent connectionist
network. Journal of Personality and Social Psychology, 74(1), 21–35.
Srull, T. K., & Wyer, R. S. (1979). The role of category accessibility in the interpretation
of information about persons: Some determinants and implications. Journal of Personality
and Social Psychology, 37(10), 1660–1672.
Srull, T. K., & Wyer, R. S. (1989). Person memory and judgment. Psychological Review,
96(1), 58–82.
Sun, R. (2009). Theoretical status of computational cognitive modeling. Cognitive Systems
Research, 10(2), 124–140.
Thagard, P. (2003). Why wasn’t O.J. convicted? Emotional coherence in legal inference.
Cognition and Emotion, 17(3), 361–383.
Thagard, P. (2004). Causal inference in legal decision making: Explanatory coherence vs.
Bayesian networks. Applied Artificial Intelligence, 18(3–4), 231–249.
Thagard, P., & Aubie, B. (2008). Emotional consciousness: A neural model of how
cognitive appraisal and somatic perception interact to produce qualitative experi-
ence. Consciousness and Cognition, 17(3), 811–834.
Trope, Y., & Alfieri, T. (1997). Effortfulness and flexibility of dispositional judgment
processes. Journal of Personality and Social Psychology, 73(4), 662–674.
Uleman, J. S., & Moskowitz, G. B. (1994). Unintended effects of goals on unintended
inferences. Journal of Personality and Social Psychology, 66(3), 490–501.
Van Overwalle, F. (2007). Social connectionism: A reader and handbook for simulations. Hove:
Psychology Press.
Van Overwalle, F., & Labiouse, C. (2004). A recurrent connectionist model of person
impression formation. Personality and Social Psychology Review, 8(1), 28–61.
Wagar, B. M., & Thagard, P. (2004). Spiking Phineas Gage: A neurocomputational
theory of cognitive-affective integration in decision making. Psychological Review,
111(1), 67–79.
Wigboldus, D. H., Sherman, J. W., Franzese, H. L., & Van Knippenberg, A. (2004).
Capacity and comprehension: Spontaneous stereotyping under cognitive load. Social
Cognition, 22(3), 292–309.
4
THE WHOLE ELEPHANT
Toward Psychological Integration of
the Individual as a Complex System

Jennifer Rose Talevich1

As people, we intuitively understand that we are not our environment, our

emotions, perceptions, motivation, or even our behavior. We have the sense
that who we are is something greater than just the sum of these parts. That is,
although we may not have the technical vocabulary, we intuitively understand
ourselves as complex systems.
Complex systems are ubiquitously embedded throughout the human experi-
ence. In every moment, we are served by systems within ourselves (our brains,
our immune system). We depend upon and contend with both natural systems
(climate systems and ecological systems, etc.) and social systems (our network
of friends and family, governance systems, and the economy, to name a few).
Throughout the physical and social sciences, from anthropology to physics, the
research on these systems is abundant.
The basic component of models in the social sciences and humanities is indi-
vidual humans. It is their behavior, and the complex interactions between them,
that give rise to societies, cultures, and economic systems. In these models, typi-
cally, simple agents, outfitted with a simple pre-determined map of behaviors
determined by basic decision-making rules, represent the individual human.
However, decision-making and behavior, in realistic human beings (in contrast
to these simple agents), are the nonlinear product of the motivations, feelings and
experiences of an individual’s mind or psyche. Yet within complexity science,
research on the complexity of this foundational component of social systems, the
individual mind, is scarce.
This gaping oversight receives little attention because the field that best
understands the psyche and human behavior—psychology—is not, itself, work-
ing on the entire problem. Although cognitive science and cognitive psychology
have worked to develop cognitive architectures, they have largely neglected the
62  Jennifer Rose Talevich

emotional, motivational, and most importantly, the social aspects of the mind
and behavior. The social mind is the bridge between the individual and the col-
lective. How cognitive, emotional, and motivational processes interact with the
social world is the particular focus of social psychology.
The purpose of this chapter is to argue for the mutually beneficial impor-
tance of psychology’s full participation in complexity science. Psychology’s key
contribution should be integrated psychological models of the individual mind
situated in the social world. The chapter will introduce an integrated social and
developmental psychological model as one example of how to apply complex
systems principles to integrate psychological theory.

The Missing Field in Complexity Science

One of the foremost institutes devoted to complexity research is the Santa
Fe Institute. Their most popular massive online course, “Introduction to
Complexity,” enrolled 4,275 students in one semester (Complexity Explorer,
2015). Of particular interest are Week 7, which discusses models of biologi-
cal self-organization, and Week 8, which models cooperation in social systems.
Another week, “Scaling in Biology and Society,” moved from a discussion of
metabolism to cities, with nothing to connect the two. To this psychologist,
the lack of connection between biological and social complexity was striking.
A review of the research and curriculums of several other programs in complex
systems science show the same gap. It would seem that complexity science does
not consider people to be complex—or, if it does, does not think understanding
the individual person is critical to understanding society.
Why this oversight? Consider the following quote recently published by a
physicist in the LA Times: “That’s right. Psychology isn’t science. Why can we
definitively say that? Because . . . [p]sychologists can’t use a ruler or a microscope,
so they invent an arbitrary scale” (Berezow, 2012)—an amusing commentary
from the field in which “everything is relative” and invisible particles are theoretically
inferred for decades, if not centuries, before they can be measured. According to
this definition, it would seem that among the physical sciences, psychology has
the most in common with physics. So why this disconnect between the physical
sciences and psychology—and why is the disconnect not true, at least to the same
extent, for the other social sciences?
One thing that sociology, anthropology, and economics share with the physi-
cal sciences, and which much of psychology does not, is a strong emphasis on
computational models. Much work in the modern natural sciences is computa-
tional in nature. Computational models require an exacting level of specification,
are entirely quantifiable; conditions in experimental simulations are completely
controlled, reproducible, predictable, and testable. A favored type of computa-
tional model in both these social and the physical sciences are multi-agent systems
in which the interactions of numerous simple agents result in highly complex
 The Whole Elephant  63

phenomena. In physics, we find the simplest of all agents: atoms! To those who
study atoms, simple agents are familiar and, as such, appear intrinsically reason-
able and accurate. The misunderstandings and inaccuracies of modeling people as
simple would not be apparent to them.
This is a problem because the people who study atoms are numerous among
the people who decide what research projects receive funding. This is because the
social sciences are comparatively young. Our older siblings, the “hard sciences,”
were already being supported by private and public funds (and sitting on review
committees) before we came on the scene. This translates into a lot of influ-
ence and, thereby, money. Consider, for instance, the Superconducting Super
Collider project, for which the United States government, in 1992, approved
$8.25 billion ($13.5 billion today) for a single project (Appell, 2013).
The National Science Foundation (NSF) supports fundamental research and
education in all the non-medical fields of science and engineering. It is the major
source of federal backing in mathematics, computer science, and the social sci-
ences. The NSF grants five times as much money to anthropology, sociology,
and economics ($160 million) as it does to the behavioral sciences ($31 million)
(Yamaner, 2014). That is, it provides five times the support to the social sciences
that utilize and contribute to complex systems science.
Complexity science is a space in which physics, and the computational social
and biological sciences, all interact to discover commonalities and share meth-
ods of both experimentation and analysis. Complexity scientists are particularly
concerned with the common properties of systems considered complex in
nature, society and science. With that, complexity has become not just a research
approach, but a movement—a movement of unprecedented interdisciplinary
collaboration among the natural and social sciences.
Although psychology is not yet participating sufficiently to garner a real presence
in complexity science, there is evidence its contribution would be most welcome.
For instance, two of the editors of this book, Vallacher and Nowak, have been
invited to give a talk on dynamical systems psychology at the Santa Fe Institute.
The professor featured in the Santa Fe Institute’s massive online introductory course
mentioned above, Dr. Melanie Mitchell, began her scientific career with a search for
an understanding of the mind (Mitchell, 2011). She has published on psychological
topics. The evidence suggests the exclusion of psychology from complex systems
science is not only a matter of discrimination, but also a failure of our own field to
fully participate.This chapter will argue that this absence is rooted in a basic incom-
patibility between traditional psychological research and the study of the mind as a
complex system: it comes down to the study of the whole versus study of the parts.

Holism in Complexity Science

The special focus of complexity science is on holism. A fundamental understand-
ing is that the whole is not necessarily greater, but rather often different from the
64  Jennifer Rose Talevich

sum of the parts, and different from the parts themselves. There is no sentience
in a neuron, no tornado in a raindrop, no financial collapse in a single dollar bill.
The whole is not found among the parts, but arises as something different, and
often impossible to predict, from its individual components.
Integration of theory and sub-systems is a particular challenge for psychology,
as discussed directly above. As with other fields, what separates the complex
systems approach from the disciplines that contribute to it is a special focus on
holism: the whole that arises from many interacting parts.

“System Nestedness”, “System Embeddedness”

Wholes are but parts in even larger systems. Thus, in complexity science, there is
also an emphasis on the multiplicity of systems and the effect that one sub-system
has upon the behavior of another. This is the concept of system embeddedness or
system nestedness. Individual components self-organize into hierarchical levels.
Biological complexity is hierarchically nested. Cells are complex systems in
which the simple agents are molecules. Cells form tissues, which form organs,
and organs form the body-wide organ systems of which the organism is com-
posed. The organism, then, is a complex system composed of many biological
agents that are each simple relative to the whole organism.
Social complexity is also hierarchically nested. The world economy is com-
posed of national economies, and national economies are composed of state
economies. “Human” culture is composed of a great variety of cultures found
throughout the world and can be studied by identifying the universals found
among them. In turn, these unique cultures often include many nations, each
of which are composed of a plethora of groups and organizations. Some models
include groups as the basic unit, but most often individual people self-organize
into hierarchically nested social groups and structures.
The individual, then, is the basic unit of social complexity. This is why
psychological integration at the level of the individual is how we can most signifi-
cantly contribute to complexity science. As the individual at the biological level
is a complex system composed of multiple body-wide complex systems, the indi-
vidual mind is a complex system composed of multiple psychological complex
systems. Complex systems scientists want to know what the overall system looks
like: the whole cell, the whole society. A complex systems psychologist, then,
would want to know what the person, as a whole, looks like: how psychologi-
cal parts (which may themselves be hierarchically nested psychological systems)
interact to give rise to an emergent psyche and the behavior it produces.

Holism in Psychology
At its founding, the purpose of the field of psychology was not to understand psy-
chological phenomena, but to understand people.The fundamental understanding
 The Whole Elephant  65

of Gestalt psychology held that the mind could only be understood as an irreduc-
ible whole, something “other than” its parts. Unfortunately, the Second World War
interrupted this German school of psychology and it, thereafter, became largely
circumscribed to the study of perceptual illusions. Yet Gestalt psychology was
poured into the very foundation of modern social psychology by theorists such as
Asch, Heider, Lewin, and by Festinger (1950) (Read,Vanman, & Miller, 1997). For
instance, Heider and Simmel’s animation of “interacting” geometric shapes (1944)
and Asch’s impression formation (1946) showed that social stimuli, like perceptual
illusion, are not processed in parts, but as a whole.
Early theories were “Big Theories” such as Lewin’s force-field theory (1935),
Bowlby’s behavioral systems theory (1969), and others such as Murray’s com-
prehensive system of needs (1959), Maslow’s psychology of being (1968), as well
as work by Jung (1939) and Freud (1927 [1923]). These theorists all developed
comprehensive frameworks that attempted to integrate the entire psyche. But
how does one test big theory? There really has been no way to do this except to
break it down, carve it up, and manipulate the pieces in lab experiments. And so,
the field moved towards reduction. Now we know a lot about emotions, motiva-
tion, perception, and the like—but have we made equal strides in understanding
the person, the mind, the psyche? No. The individual person isn’t just the level of
scale missing in complexity science, it is also the missing level of scale in modern
psychological science.
Psychology is a highly fractured field. We are split into several basic sub-
disciplines (e.g., social psychology) and, within each sub-discipline, scholars work
as artisans crafting their own theories and carving out their highly individualized
niches. Psychology, as a field, does not often combine sub-systems or levels of
analysis. The result is highly detailed information about specific psychological
phenomena without much integration. Nowhere in modern psychology is there
a movement to integrate existing psychological knowledge and theories so as to
represent the individual person holistically.
This is unfortunate because, as will be discussed further below, the individual
person is the basic unit of social complexity in anthropological, economic, and
sociological models. Holism at the level of the individual human is the bridge
between psychology and the ability of our field to contribute significantly to
complexity science. Nonlinear dynamical psychologists, like those whose con-
tributions you will read in other chapters of this book, are doing the important
work of understanding different psychological processes as complex systems. The
authors in this book are psychologists working to change the field. We study
nonlinear psychological dynamics, chaos, catastrophe, and complexity.2 The next
step is to begin to integrate multiple dynamical systems into broader psycho-
logical systems and, eventually, to integrate these systems so as to represent the
individual person in a holistic manner. Reduction was the best science possible at
one time. It is no longer. The infeasible pipe-dreams of early psychologists have,
with the invention of computational methods, become a manageable feat.
66  Jennifer Rose Talevich

Contributing the Individual

A key problem in studying the psyche as a whole is that, well, it’s complicated!
What is meant by “whole” is not “entire,” but “integrated”. Any one model
need only include that which has been known to influence the particular social
behavior of interest. Even so, it is difficult to build a model that integrates mul-
tiple psychological theories and phenomena (each complex systems in their own
right), and nearly impossible to track, or even imagine in any detail, how these
non-additively interact.
Yet this is the key problem, across fields that gave rise to complexity as a sci-
ence and to common solutions. Psychologists can now draw on this new body of
knowledge in the attempt to integrate their field. But to join complexity science,
psychology needs to not just draw from it, but to contribute solutions. What
problems, as newcomers to complexity, can psychologists solve?

The Limitations of Multi-Agent Systems

Many complex systems models are multi-agent models in which individual
agents, be they ants or people, are programmed with simple decision rules. From
the interactions of these simple agents arise sophisticated collective behavior. This
has led to the general consensus that only simple representations of people are
necessary to model social complexity (Boero, Castellani, & Squazzoni, 2008). If
the goal is to predict what collective behavior may arise given certain conditions
and assumptions, then this approach is quite useful.
But if the goal is to understand social change, then simple-agent models are
inadequate—particularly if the goal is to simulate the possible effects of social
interventions such as campaigns designed to sway public opinion, strategies to
prevent crime, or predicting the effects of government policy changes on the
economy (Boero, 2015). The source of change in social models is individual
people, and behavioral change in an actual human is fundamentally nonlinear.
Simple-agent decision rules are pre-programmed. But changes in the envi-
ronment such as new information, new opportunities, or new obstacles can
fundamentally change individuals and their decision rules.
The human response to an event is to think, perceive, and feel. Behavior
emerges from the pattern of perceptions, motivation, and emotions that become
activated in response to an event. These patterns are highly individualized because
they reflect the personal history of the individual. That is, personal histories
become ingrained as patterns of association among psychological components. A
lost opportunity is easily dismissed by one person, cause for fear in another, and
cause for anger in yet another, based on their past histories. Of those who feel
fear, different histories will associate different behavioral responses to fear urg-
ing fight, flight, or freeze. Each change in the environment, given time, will be
encoded as new thoughts and feelings and decision-making rules.
 The Whole Elephant  67

Essentially, the modeling of social change requires the transversal of multi-

ple hierarchically nested levels of complexity. From the complex interactions of
human behavior emerge sociological phenomena. These emergent phenomena
of macro models, society, culture, and the economy, are, in turn, environmental
inputs to the complex psychological models from which individual behaviors
emerge.
One unfortunate byproduct of the general failure to acknowledge psychologi-
cal complexity is a dearth of available technologies with which to build multi-agent
models with sophisticated agents (cf. Boero, 2015; Marsella, Pynadath, & Read,
2004). So even nonlinear dynamical systems psychologists, today can only build
simple-agent multi-agent models. The fact that they do so, nonetheless, is mean-
ingful. The collective dynamics chapters in this book demonstrate the fact that,
ultimately, understanding human behavior requires the transversal of many scales
of complexity.
As complexity science extends its reach to solve more and more complex
social issues, it will become more apparent that they are missing a critical level of
scale. Then one of two things will happen: psychologists will provide the miss-
ing level of scale—the complex individual—or computer scientists will reverse
engineer psychology. The choice is ours.

The Complex Behavioral System

This section describes a complex psychological system. It takes a classic psy-
chological theory, integrates modern theories and findings within the classic
framework, and then implements the final model as a neural network.
Behavioral systems theory was developed by John Bowlby in the 1950s and
1960s (Bowlby, 1969). In his early career as a physician at an orphanage, Bowlby
observed that child abandonment initially caused anxiety and crying, followed by
anger and acting out, then finally despair and detachment. Children responded
to their situation through a predictable process—an environmental cause and
behavioral effect that could not be accounted for by the psychological theories
of his day, such as Freud’s psychic energy model. So Bowlby looked outside of
psychology to engineering and ethology. Then-new control systems theory dealt
with the behavior of dynamical systems with inputs and studied how their behav-
ior is modified by feedback. Animal reflex and fixed-action patterns (e.g., mating
dances of birds) in ethology linked environmental cues to the appropriate behav-
ior with advantageous speed. Pulling these together, Bowlby envisioned a kind
of psychological control system, one in which psychological phenomena such as
perception, motivation, and emotion were organized so as to link typical human
circumstances to adaptive responses with a kind of immediacy approximating that
found elsewhere in the animal kingdom.
The neural network presented herein fills in Bowlby’s sketch of how psycho-
logical experiences link events and behavior with modern findings and theories.
68  Jennifer Rose Talevich

Most prominent among those integrated are Smith and Lazarus’s (1990) pri-
mary appraisal process, in which the adaptive function of emotion is to signal
motivational assessments of well-being that trigger behavior. This is then inte-
grated with the neuroscientific finding that approach and avoidance motivation
occurs via different neural pathways (Gray, 1991) and they are learned through
different experiential processes (Gable, 2006). The model presents Bowlby’s
framework as a complex system in which environmental inputs activate multiple,
networked, psychological sub-systems. These sub-systems interact dynamically
to form associations, through self-organized learning, to produce the behavioral
patterns observed by Bowlby. These behaviors and the underlying psychological
experiences are then fed back into the system such that the patterns become so
consistent as to be trait-like: that is, the network forms a personality.

Attachment Theory
Attachment theory, and behavioral systems theory more broadly, is both an evo-
lutionary and developmental psychological theory (Bowlby, 1969). Humans are
proposed to have evolved several behavioral systems, each with a specialized
function for responding to species-typical situations such as needing protection
from predators or to protect those resources necessary for survival and thriving.
Bowlby proposed five systems, including the attachment, caregiving, affiliation,
sexual, and exploration systems (Mikulincer & Shaver, 2012). Each of these sys-
tems overlaps fairly well with domains of adaptive problems proposed by modern
evolutionary psychology (Bugental, 2000; Kenrick, Li, & Butner, 2003). In
line with that literature, power has also been delineated as a behavioral system
(Shaver, Segev, & Mikulincer, 2011).
Attachment is by far the best-known and most well-researched of these
systems. The specialized function of the attachment system, in particular, is to
obtain help in times of need. Human babies are incapable of meeting their own
most basic survival needs: feeding, protecting, and cleaning themselves. To obtain
this care, babies have evolved to perform certain behaviors, such as crying out,
when in need of assistance. The attachment bond is a motivational and behavioral
system that directs children to seek proximity to a familiar caregiver, and pair-
bonded adults to seek out their significant others, with the expectation that they
will receive protection and emotional support. The strength and quality of an
attachment bond, at any age, depends on how much one can trust that another
will be there for them in times of need.

Modeling Psychological Complexity with Neural Networks

The model presented herein is abstracted from the neuro-architecture in order
to represent the attachment system as a complex system. By communicating
 The Whole Elephant  69

through chemical interaction, neurons self-organize into a biological neural

network which gives rise to “the brain.” By analogy, psychological compo-
nents self-organize in an artificial neural network by adjusting their connection
weights. Patterns of connections across a psychological network give rise to
“the mind.”
Connections between psychological systems in the network are adjusted
according to biologically realistic learning algorithms. Not all neural net tech-
nologies are capable of modeling complex properties like self-organized learning.
In feed-forward neural networks, activation flows in one direction and error
signals are unnaturally back-propagated (sent backwards) against the flow of acti-
vation. This is criticized as a form of executive rather than decentralized control.
However, the model herein is a recurrent neural network that is bi-directionally
connected. Activations flow in a directed cycle which allows the network to have
an internal state that can change over time. This, in turn, allows the network
to exhibit dynamic temporal behavior in which activation states are mutually
influenced by one another and settling over time leads to states that satisfy the
constraints of the weights in the network. Constraint satisfaction is an emergent
computational property of recurrent neural nets.
The current network was built in the Emergent software program and the
Leabra (Local, error-driven and associative, biologically realistic algorithm) archi-
tecture (Aisa, Mingus, & O’Reilly, 2008). Learning in this architecture utilizes
the extended contrastive attractor learning (XCAL) equation, which is a balance
of Hebbian, associative learning, and contrastive Hebbian learning (CHL) for
targeted learning. In CHL, the network learns the difference between a sequence
of activation states: an expectation and then a subsequent outcome. To use a bas-
ketball analogy, back-propagation is similar to a coach saying, “You missed the
basket, aim more to the left.” In contrast, CHL is similar to seeing the difference
between where you expected the ball to land and where it actually landed, and self-
adjusting your aim accordingly.
In this network, the CHL algorithm adjusts weights to reduce the difference
in activation between the expectation and outcome. That is, the weighted value
of a connection (conceptually analogous to beta weights in a structural equation
model) is adjusted so that the connection between the experience and the cor-
rect output is strengthened, and the connection to the wrong one is lessened on
each trial. The “compute” or hidden layers are necessary to this error-correcting
component.3
Finally, Leabra utilizes k-Winners-Take-All (kWTA) sparse distributed repre-
sentations. This is a form of competitive learning in which k number of nodes in
a given layer become active. Nodes in the network mutually activate and inhibit
each other, and after some time, only the most highly activated (k-number) of
nodes will be active in the output layer. Learned patterns (or representations) are
composed of the winners of this competition.
70  Jennifer Rose Talevich

The Attachment Neural Network

The attachment neural network models the psychological circuit that has evolved
to meet the demands of the species-typical requirement for social assistance to
survive and thrive as a human being. Life experience tunes this evolved system to
develop the set of attachment strategies most effective in the environment. This
is referred to as one’s attachment style. The purpose of the model is to show how
the evolved circuit is represented in the mind, how these relationship processes
are learned and eventually ingrained as a personality or chronic relationship style
that maintains social behavior. It is a holistic psychological architecture (including
social perception, emotion, motivation, and behavior) that demonstrates how,
with learning over time, individual differences develop.
The network structure (Figure 4.1) begins with inputs representing features
of the situation. These features are those identified as being most important to
attachment in the literature: who is present, how they are behaving (e.g., are
they being attentive to you?), their past behavior, and whether or not the cur-
rent situation is threatening. These features of the situation activate a schema, or
“working model,” as it is referred to in the attachment literature. This hidden
layer nonlinearly transforms the environmental inputs into a dynamic pattern to
represent the situation as a whole.
The schema activates the motivation sub-system including a behavioral
approach (BAS) and behavioral avoidance (BIS) motivation system. There are

FIGURE 4.1  The attachment neural network.

 The Whole Elephant  71

also several congruence layers, one for each motive, connected to the Situation
Compute. These assess the impact that the situation will have upon one’s most
important goals. Activations from the BIS and BAS motive layers and each
congruence layer connect to an additional hidden layer called the Goal Impact
Compute. This layer nonlinearly transforms the inputs from all the motive layers
into a single, dynamic pattern, to represent the state of the network’s motivation
system. This architecture is consistent with the primary process in appraisal mod-
els of emotion (Scherer, Schorr, & Johnstone, 2001; Smith & Lazarus, 1990) and
has been formalized with human data studies in previous work by the author and
colleagues (Talevich, Read, & Walsh, 2014).
The output of the motivation system, the Goal Impact Compute, then acti-
vates specific emotions in the emotions layer. Both emotions and motivation
layers then converge on yet another hidden layer intended to capture the state
of the network (the Internal State Compute) which will then activate behavior.
This layer is analogous to Bowlby’s model of the self (Bowlby, 1969), and is
necessary to capture goal-corrected behavior. In the last step, the Internal State
Compute and the behaviors layer are both copied and fed back into the network
as inputs for the next time step.
With computational tools, it is possible to model within-person processes over
time. Temporal contingencies among relevant cues and the behaviors they evoke
can be simulated moment-by-moment. This allows for hypothesis testing of the
dynamic social cognitive processes related to relationship formation, growth, and
change at a level of detail that is otherwise impossible. Competing activations in
the network feed forward and back to find balance throughout the entire system
by satisfying the constraints of all the psychological forces. The final state of the
system, once it has settled, is copied along with the behavior it evoked and fed
back into the system on the next iteration. In this way, each experience happens in
the context of “what happened last time.” With learning over time, this informa-
tion builds up to create a context of “what usually happens.” This process will be
discussed in more detail below under “Dynamics in the Psychological Network.”

Hierarchy and Scales of Complexity

Complexity is hierarchically nested. Each level of scale is a complex system
whose emergent behavior is the input or comparatively “simple” agent of the
next level of scale. While the lower level remains a complex system, its emergent
behavior is simple as compared to the whole of the higher-level system. In neural
networks, the “agents” are nodes analogous to neuron groups. For computational
feasibility, these are the smallest units in an artificial neural network. Nodes are
organized into layers analogous to functional brain regions. As a psychological
network (Figure 4.1), the nodes represent concrete psychological variables (e.g.,
fear) while the layers represent nonlinear dynamical psychological sub-systems
(e.g., emotions).
72  Jennifer Rose Talevich

Decentralized Learning
Another key feature of complex systems is decentralization. There is no executive
control. The psychological network presented herein is a decentralized system in
which the individual nodes competitively learn to associate with other nodes and
whether or not to activate based on what other nodes are doing. Decision rules
arise through experience as connection weights strengthen as the model learns
to make associations between variables. For instance, over time, the “happiness”
node of the attachment model will associate with certain motivation activations,
the weights between them will increase, and happiness will learn to activate if the
output from the motivation system is positive (i.e., the outlook for active goals
looks good). Next, the “explore” node will learn to activate if the happiness node
is active and the fear node is not.
Thus, psychological phenomena are connected, or networked, and represen-
tations in an artificial (or biological) neural network are distributed. Note that,
although this is a model of attachment formation, there is no node or layer for
anxious, avoidant, or secure attachment styles in Figure 4.1. That is because per-
sonality and chronic behavioral patterns like attachment are emergent—they arise
from the collective actions of, and interactions between, emotions, motivation,
and other psychological components.

Dynamics in the Psychological Network

Inspired by biological brains, neural networks can model dynamics such as non-
linear activation functions, thresholding, changing connections (weights), and
inhibition between psychological variables.
The Leabra neural network architecture allows for the modeling of impor-
tant psychological dynamics such as the development of chronic associations,
differences in sensitivities to attractive or aversive stimuli, and the approach or
avoidance motives they activate, etc. (For a more detailed explanation of these
functions with the Leabra architecture, see Read et al., 2010.)
Like biological neurons, computational nodes can have different sensitivities
to inputs (conductance and gain parameters). Greater sensitivity in one neuron/
node can result in higher output in response to the same input. This is one way
in which individual differences in psychological variables can be manipulated.
Furthermore, each node has a firing threshold, and input to the network must
exceed that threshold before the node will “fire.” This plays out at the psycho-
logical level in the attachment network. For instance, the motivational appraisal
that the situation looks bad for one’s most important goals (a pattern represented
in the Goal Impact layer) will activate different emotions based on the severity of
the computed situation. While the “fear” node may learn to respond to even a
mildly poor goal appraisal, the severity of the situation must reach a particularly
high level before the “sadness” node will activate. This is consistent with the
 The Whole Elephant  73

finding that a fear response is generally faster and has a lower threshold than posi-
tive responses for the adaptive purpose of focusing attention on goals.
Another fundamental aspect of Leabra is a general mechanism for inhibition.
This is particularly important for modeling motivation. The active pursuit of
one goal should inhibit competing goals because conflicting signals about what is
important lead to a kind of behavioral freeze.

Adaptation
Dynamics constitute adaptation if the changes occurring within the system are in
response to an external environment. In order to evolve, a complex system must
find a way to extract information from its environment and then adapt to it. If
the response improves the system’s fitness in that environment, then complex
systems scientists call the adaptation Darwinian evolution. This is a much more
basic or general sense of the word “evolution” than is used in common parlance,
which refers more specifically to evolution of a species. The complex system that
is a human mind develops in accordance with the constraints and affordances of
biology and culture. Consistent with common parlance, evolutionary psychol-
ogy theorizes about adaptations at the species level. However, several disciplines
in psychology study adaptive processes: cultural psychology studies adaptation
at both the societal level (e.g., cultural differences) and the species level (e.g.,
“human universals” such as gender differences in power). Social psychology
studies adaptation more broadly and from the perspective of the immediate
environment: how diverse individual human beings commonly respond (adapt)
to similar situations (similar environments). Developmental psychology studies
adaptation at the individual level across the lifespan. As complex systems science
defines it, if these social and developmental psychological adaptations improve
the individual’s fitness in their social environment or culture, the adaptation is a
form of Darwinian evolution.
The introduction to attachment theory at the beginning of this section was
an evolutionary argument. Bowlby’s behavioral systems theory theorizes about
the adaptive problem that each functional system evolved to solve at the species
level. The predictive power of attachment theory lies in how individuals adapt
the system to maximize fitness in their social environments. These adaptations
result in highly predictable patterns of behavior.
The dynamics by which individual variables within a psychological system
adjust in response to one another were discussed above. But how does the system,
as a whole, adapt to its environment? That is, how does an emergent psyche, or
at least aspects of it, evolve?
The attachment system is designed to respond to both threats and the amount
of social support currently available in the environment. The system responds
to threats by seeking aid from close others—those known to the system to be
more reliable sources of aid than other potential sources (a secure situation).
74  Jennifer Rose Talevich

If, however, aid is not forthcoming, the system adjusts to “gear up” help-seeking
behaviors by protesting inadequate aid (an anxious situation). But if this too fails,
the system suppresses help-seeking in order to turn efforts toward self-help (an
avoidant situation).
The key factor in determining the type of attachment style that develops is a
function of attachment figure responsiveness in these times of need. The network
formalizes this function as schedules of reinforcement where responsiveness of
the attachment figure is the reward. Over time, the network forms expectancies
regarding the quality of care one is likely to receive. Attachment behaviors are
strategies for dealing with this expected outcome.
When responsiveness is positively reinforced, the network develops a secure
attachment style. Weights throughout the network adjust to associate threat with
fear. However, when responsiveness is reinforced, threat also becomes associated
with the “get help” motive and its positive congruence with the situation. This
is the internal state of the network: a pattern of expecting a positive response.
Expecting its needs will be met, the secure network adaptively implements
help-seeking or independence depending on whether there is or is not a threat,
respectively.
When care is delivered on a partial reinforcement schedule, an anxious attach-
ment style develops. Weights throughout the network adjust in expectation of
having to “work for it,” and the anxious system pre-emptively over-implements
help-seeking and protests rejection. An attachment-avoidant network develops
when care is delivered on an extinction schedule. Weights throughout the net-
work adjust in expectation of going unaided, and independence is automatic
rather than in direct response to the situation at hand.
The adjustment of these connection weights is a unique feature of neural net-
work programming, designed to replicate the strength of connections between
neurons and neuronal pathways. This is one way in which neural networks are
ideally suited for the modeling of complex psychological systems.
This process of internal, nonlinear dynamics, aimed toward adapting the sys-
tem to its environment, is how experiences become ingrained as a personality or
chronic relationship style that maintains social behavior.

The Emergence of Behavioral Patterns:

Attachment Styles
In Figure 4.1 we have what seems to be a great omission: an attachment model
without attachment styles. One might reasonably expect an “attachment layer”
with three nodes for secure, anxious, and avoidant attachment. Certainly, these
appear in any statistical model dealing with attachment variables. However, as
with infants and adults alike, the behavior of the network is diagnostic of its
attachment style. The attachment style representation is distributed in patterns of
activation throughout social perceptions, motivations, emotions, and behavior.
 The Whole Elephant  75

In the above we discussed how these patterns are learned. This section shows
how attachment bonds emerge from these patterns.
After training, each network was put in a simulation of the Ainsworth Strange
Situation: an experiment in which mothers bring their infants into the lab and,
in a series of episodes, leave the child to play alone or with a stranger for short
periods of time. Human infants respond to these situations with reliable patterns
of behavior that can be used to identify their attachment style (Ainsworth, Blehar,
Waters, & Wall, 1978). A child is classified as secure if he or she plays contentedly
with toys (exploration) when his or her mother is present (Episodes 1 and 2), but
then seeks proximity to her after she returns from an absence (Episodes 5 and 8).
However, if the child does not explore, and either ignores his or her mother or
resists her greeting upon her return, the child is classified as avoidant or anxious,
respectively.
The output of each network was coded using the Strange Situation code
book (Ainsworth et  al., 1978). Each neural network’s behavior was consist-
ent with the behavior of differently attached human infants in the Ainsworth
Strange Situation. As hypothesized, networks trained with reinforced respon-
siveness behaved in accordance with the secure classification (seeking proximity
only in need). The partially reinforced network behaved in accordance with the
resistant (i.e., anxious) classification (marked by protest/resistance). Finally, the
network receiving extinction schedules behaved in accordance with the avoidant
classification (marked by dismissive behavior). The behavior of the network is
consistent with the behavior of human infants.
The network output is, like infant behavior, diagnostic of its attachment style.
Unlike actual human beings, we can look directly into the “minds” of these net-
works: we see that the attachment style representation is distributed throughout
that “mind” as different patterns of learned associations between social percep-
tions, motivations, and emotions.

Conclusion
Presented herein was a complex system in which psychological phenomena are
networked to connect human circumstances with human behaviors. Through
self-organized learning, psychological sub-systems (e.g., emotions, motivation)
come to be associated and behavioral patterns emerge. Distributed among these
sub-systems, and represented by the patterns of interaction that form among
psychological components, an individual, a “self” as relates to close others,
arises as long histories are encoded and the “self” system adapts to its external
environment.
Emergent patterns of chronic behavior in known contexts allow us to iden-
tify the underlying personality representation. Identifying this representation
then allows us to predict the individual’s future behavior in additional con-
texts. Therefore, in order to change behavior, we must modify the underlying
76  Jennifer Rose Talevich

representation, the interactions among psychological components. That is,

to predict people’s responses to changing circumstances accurately requires a
complex model that can simulate the numerous interactions of the many psy-
chological components from which the ever-changing individual “self” arises.
As a complex system, the psyche as a whole will constrain the behavior of the
individual psychological components we study in ways that we cannot yet com-
prehend. The difficulties with replication that currently plague psychology are
not due to a lack of scientific rigor—they are due to using reductionist science to
model complex phenomena. Each field currently engaged in complexity science
hit a wall in its work. That we have come to the limits of reductionist psychol-
ogy is a sign that our science has matured. This is reason for congratulation and
excited anticipation for the next frontier.
There is much to learn from the new field of complexity science. It has arisen
from the participation of multiple disciplines, and is being applied to aid discov-
ery at every (non-psychological level) of human (and non-human) existence from
cell biology to brain networks to social life. The greatest troubles of our time are
not technological, but social. Yet funding for the social sciences lags behind that
of the physical sciences. Further, funding for psychology as a basic science lags
behind the social sciences that are currently engaged in complex systems research
such as multi-agent modeling.
We have seen here that the individual is a complex system from which pre-
dictable behavior patterns (in complex systems terminology, “decision rules”)
are emergent phenomena. Given this, any model designed to simulate and accu-
rately predict human social change needs to include psychologically complex, not
simple, individuals. Yet most anthropological, sociological, and economic com-
putational models are multi-agent models using simple agents pre-programmed
with simple decision rules. It is fairly ironic that the individual person is the most
important yet least understood level of social complexity. One might argue that
behavioral science is not just a missing level of scale, but also the most important.
While the study of emergent macro social behavior is fascinating, the ultimate
utility of these models is to predict social changes such as the effects of public pol-
icies, interventions, or calamities. Collective behavior emerges from the behavior
of individual decision-makers. Yet the social environment shapes the individuals
that make those decisions. The social variable, “inputs,” to which an individual
human responds changes the decision-making rules of the individual such that
the behavior they output is nonlinearly related to environmental inputs. The
human mind is not simple, and simple models of the human mind are not going
to be accurate—particularly when predicting social change over time.
The absence of psychology, particularly social psychology, at complex sys-
tems research centers indicates there is much work yet to do in convincing this
community of the importance of the complex individual. However, it is a wor-
thy goal, and one that has received substantial financial backing in recent years.
A National Institute of Health funding opportunity entitled “Modeling Social
 The Whole Elephant  77

Behavior (R01)” (PAR-13-374, 2013–2017), involving several behavioral and

mental health programs, calls for computational approaches. It puts particular
emphasis on understanding the system as a whole, and further specifies that
research must traverse multiple scales of social complexity—examinations at sin-
gle levels of scale are not fundable.
In 2011, the John Templeton Foundation funded the massive online courses
in complexity discussed in the introduction. This educational outreach was a
major component of a three-year $5 million grant to the Santa Fe Institute to
illuminate many hidden regularities in the biological and social systems (universal
patterns in the emergence of complex societies) by capitalizing on the opportuni-
ties presented by recent advances in computational power.
In 2012, the John Templeton Foundation offered another $3 million in grants,
each proposal to win approximately $250,000, in a funding competition aimed
at psychologists and social scientists “designed to promote integration of existing
lines of research and to generate and test new hypotheses emerging from such
integration.” An expansion to the attachment neural network described in this
chapter was proposed. It was the only proposal for a computational model entered
into the competition. This gave it a decided advantage—not because it was com-
putational, but because only computational methods can harness the integrative
potential of complex systems: holism and nestedness. The project was one of the
winners of the competition and is currently being funded. An express hope, by
the grantor and grantee, is that the work will encourage more integrative compu-
tational research, to the benefit of both psychology and complexity science alike.

Summary
This chapter has argued for the importance to psychology of more fully partici-
pating in complexity science—not only its methods, but also its intellectual focus
on holism and the transversal of multiple levels of scale. It has further argued that
the greatest value that psychology can contribute is to integrate psychological sci-
ence. Integrated models would include all the major psychological phenomena
such as cognition, emotion, and motivation, and consider them in developmental
and socially situated systems. When modeling the psyche rather than its parts,
social-behavioral phenomena can be more easily scaled up in collaboration with
the collective-oriented social sciences.
For instance, attachment behaviors influence a surprising number of social
interactions: not only the parent–child bond, but also mating bonds (Miller,
Christensen, Pedersen, Putcha-Bhagavatula, & Appleby, 2013), attachments to
the supernatural (Kirkpatrick, 1997), and even different product brands (Park,
MacInnis, & Priester, 2007). Thus, the model of attachment presented herein
would be an appropriate complex agent in multi-agent models of changing family
and marriage patterns, religious organizations, species-level evolution of gender
differences, and even some consumer economic models.
78  Jennifer Rose Talevich

Thus, an integrated psychological science would result in realistically com-

plex human agents. As technology and processing power increase, these complex
human agents can be integrated into multi-agent models. That technological
future is not too far off, but psychology has quite a bit of integration to do before
it will be ready to take part in such an endeavor.
Yet it is only by uniting the behavioral and collective social sciences, mod-
eling both the individual decision-maker who drives social change and the social
environment that shapes the decision-maker, that we will accurately predict the
impact of social policy and intervention. This is the only way to understand
change: how we came to be the people and peoples that we are, and who we
may become.
Some might consider striving towards an integrated model of the complex
human psyche as too ambitious—even mythic. But as the field of physics has
demonstrated, sometimes it pays to think really big. What is psychology’s—or,
better yet, the social sciences’—Large Hadron Collider project? Self and societal
well-being and improvement are the domain of the social sciences. Let physics
have the God particle, ours is the domain of human salvation.

Notes
1 My thanks to Stephen J. Read for reading and editing many versions of this chapter, for
financial support by the John Templeton Foundation, and for the shepherding of our
program manager, Nicholas Gibson.
2 For an introduction to nonlinear dynamics and complexity from a psychological
perspective, see the following: Guastello, Koopmans, & Pincus (2009); Nowak &
Vallacher (1998); Read et al. (1997); Vallacher & Nowak (1994, 1997, 2007); Wiese,
Vallacher, & Strawinska (2010).
3 To learn more about these learning mechanisms, see O’Reilly and Munakata (2000).

References
Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of attachment:
A psychological study of the strange situation. Hillsdale, NJ: Lawrence Erlbaum.
Aisa, B., Mingus, B., & O’Reilly, R. (2008). The Emergent neural modeling system.
Neural Networks, 21(8), 1146–1152.
Appell, D. (2013). The supercollider that never was. Scientific American, October 15.
Retrieved from www.scientificamerican.com/article/the-supercollider-that-never-was/
Asch, S. E. (1946). Forming impressions of personality. Journal of Abnormal and Social
Psychology, 41(3), 258.
Berezow, A. B. (2012). Why psychology isn’t science. Los Angeles Times, July 13.
Retrieved from http://articles.latimes.com/2012/jul/13/news/la-ol-blowback-
pscyhology-science-20120713
Boero, R. (2015). Behavioral computational social science. Hoboken, NJ: John Wiley & Sons.
Boero, R., Castellani, M., & Squazzoni, F. (2008). Individual behavior and macro social
properties: An agent-based model. Computational and Mathematical Organization Theory,
14(2), 156–174.
Bowlby, J. (1969). Attachment (Kindle ed., Vol. 1). New York: Basic Books Classics.
 The Whole Elephant  79

Bugental, D. B. (2000). Acquisition of the algorithms of social life: A domain-based

approach. Psychological Bulletin, 126(2), 187.
Complexity Explorer (2015). Course statistics from fall, 2014. Retrieved from www.
complexityexplorer.org/news/15-course-statistics-from-fall-2014
Festinger, L. (1950). Informal social communication. Psychological Review, 57(5), 271.
Freud, S. (1927 [1923]). The ego and the id. London: Hogarth.
Gable, S. L. (2006). Approach and avoidance social motives and goals. Journal of Personality,
74(1), 175–222.
Gray, J. A. (1991). The neuropsychology of temperament. In J. Strelau & A. Angleitner
(Eds.), Explorations in temperament: International perspectives on theory and measurement
(pp. 105–128). New York: Plenum.
Guastello, S. J., Koopmans, M., & Pincus, D. (2009). Chaos and complexity in psychology:
The theory of nonlinear dynamical systems. Cambridge: Cambridge University Press.
Heider, F., & Simmel, M. (1944). An experimental study of apparent behavior. American
Journal of Psychology, 57, 243–259.
Jung, C. G. (1939). The integration of the personality. New York: Farrar & Rinehart.
Kenrick, D. T., Li, N. P., & Butner, J. (2003). Dynamical evolutionary psychology:
Individual decision rules and emergent social norms. Psychological Review, 110(1),
3–28.
Kirkpatrick, L. A. (1997). A longitudinal study of changes in religious belief and behavior
as a function of individual differences in adult attachment style. Journal for the Scientific
Study of Religion, 36(2), 207–217.
Lewin, K. (1935). A dynamic theory of personality. New York: McGraw-Hill.
Marsella, S. C., Pynadath, D. V., & Read, S. J. (2004). PsychSim: Agent-based modeling of
social interactions and influence. In Proceedings of the International Conference on Cognitive
Modeling (pp. 243–248). Mahwah, NJ: Lawrence Erlbaum.
Maslow, A. H. (1968). Toward a psychology of being. Floyd, VA: Sublime Books.
Mikulincer, M., & Shaver, P. R. (2012). Attachment theory expanded. In K. Deaux &
M. Snyder (Eds.), The Oxford handbook of personality and social psychology (pp. 467–492).
New York: Oxford University Press.
Miller, L. C., Christensen, J. L., Pedersen, W. C., Putcha-Bhagavatula, A., & Appleby, P. R.
(2013). Attachment fertility theory: Complex systems of mechanisms simplify sex,
mating, and sexual risks. Psychological Inquiry, 24(3), 211–220.
Mitchell, M. (2011). Complexity: A guided tour (1st ed.). Oxford: Oxford University Press.
Murray, H. A. (1959). Preparations for the scaffold of a comprehensive system. Psychology:
A Study of a Science, 3, 7–54.
Nowak, A., & Vallacher, R. R. (1998). Dynamical social psychology. New York: Guilford
Press.
O’Reilly, R. C., & Munakata, Y. (2000). Computational explorations in cognitive neuroscience:
Understanding the mind by simulating the brain. Cambridge, MA: MIT Press.
Park, C. W., MacInnis, D. J., & Priester, J. R. (2007). Beyond attitudes: Attachment and con-
sumer behavior (SSRN Scholarly Paper No. ID 961469). Rochester, NY: Social Science
Research Network.
Read, S. J., Monroe, B. M., Brownstein, A. L., Yang, Y., Chopra, G., & Miller, L. C.
(2010). A neural network model of the structure and dynamics of human personality.
Psychological Review, 117(1), 61–92.
Read, S. J., Vanman, E. J., & Miller, L. C. (1997). Connectionism, parallel constraint
satisfaction processes, and Gestalt principles: (Re)introducing cognitive dynamics to
social psychology. Personality and Social Psychology Review, 1(1), 26–53.
80  Jennifer Rose Talevich

Scherer, K. R., Schorr, A., & Johnstone, T. (Eds.) (2001). Appraisal processes in emotion:
Theory, methods, research. New York: Oxford University Press.
Shaver, P. R., Segev, M., & Mikulincer, M. (2011). A behavioral systems perspective
on power and aggression. In P. R. Shaver & M. Mikulincer (Eds.), Human Aggression
and Violence: Causes, Manifestations, and Consequences (pp. 71–87). Washington, DC:
American Psychological Association.
Smith, C. A., & Lazarus, R. S. (1990). Emotion and adaptation. In L. A. Pervin (Ed.),
Handbook of personality theory and research (609–637). New York: Guilford Press.
Talevich, J. R., Read, S. J., & Walsh, D. A. (2014). Goal impact: A goal systems domain-
general prediction tool, applied to voluntary job turnover. Basic and Applied Social
Psychology, 36(1), 35–50.
Vallacher, R. R., & Nowak, A. (Eds.) (1994). Dynamical systems in social psychology. San
Diego, CA: Academic Press.
Vallacher, R. R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 8(2), 73–99.
Vallacher, R. R., & Nowak, A. E. (2007). Dynamical social psychology: Finding order
in the flow of human experience. In A. W. Kruglanski & E. T. Higgins (Eds.), Social
psychology: Handbook of basic principles (2nd ed.) (734–758). New York: Guilford Press.
Wiese, S. L., Vallacher, R. R., & Strawinska, U. (2010). Dynamical social psychology:
Complexity and coherence in human experience. Social and Personality Psychology
Compass, 4(11), 1018–1030.
Yamaner, M. (2014). Federal funding for basic research at universities and colleges essen-
tially unchanged in FY 2012. National Science Foundation, NSF 14-318, September.
Retrieved from www.nsf.gov/statistics/infbrief/nsf14318/#fn1
5
COMPUTATIONAL MODELING
OF HEALTH BEHAVIOR
Mark G. Orr and Daniel Chen1

The field of health psychology is broad, covering a variety of psychological

mechanisms (e.g., stress/coping), contexts (extended hospital visits), and appli-
cations (e.g., in-clinic interventions). The distinct field of health behavior
(the study of behaviors related to health), largely influenced by psychology,
is equally broad, but in a different way—it addresses behavior embedded in a
variety of levels of scale, from the individual to the community to the popula-
tion. In this chapter, we will focus exclusively on the issues of health behavior,
a perspective that puts a premium on articulating how psychological processes
are implicated across levels of scale. Thus, our interest lies in modeling both
an individual’s behavior and a collection of individuals simultaneously. This
goal clearly invokes a systems approach that leverages computational models of
individual-level health behavior that can be embedded in computational social
systems (e.g., social networks).
Central to the study of health behavior is the dynamics of change—in atti-
tudes, beliefs, self-efficacy, and behavior—where the principal challenge is to
explain how social and environmental contexts drive change in a person’s mental
representations and processes. This challenge is captured well by the notion that
behavior at any point is always driven by the individual and his/her situation and
context (Mischel & Shoda, 1995; Monroe & Read, 2008; Shoda, LeeTiernan, &
Mischel, 2002; Shoda & Mischel, 1998). A primary context in the health behavior
field is, naturally, other persons—something that is well captured in the social
network literature.
The social network literature extends and formalizes what we call the person–
context system. Here, a person’s context is composed of other people, providing
an account of the behavior of both individuals and populations of individuals
simultaneously. Furthermore, some social network effects provide convergent
82  Mark G. Orr and Daniel Chen

evidence that behaviors are socially determined, evidence that comes largely from
outside of the bounds of traditional experimental paradigms used in psychology
and, in effect, offers a greater degree of ecological validity (for an excellent exam-
ple, see Muchnik, Aral, & Taylor, 2013). However, a key limitation of the social
network approach is the difficulty in interfacing with psychological processes that
are informed by psychological theory (for a rare exception, see Bhattacharyya &
Ohlsson, 2010)—that is, the approach is too heavy on context and too light on
person.
This limitation has not been reconciled in the health behavior field. Here,
when social networks are considered, it is as an auxiliary piece of information for
understanding individuals’ behavior, and not as an integrated part of the psycho-
logical processes related to behavior—for example, to understand the degree to
which individuals should be targeted based on their position in a network (for a
thorough review, see Valente, 2010). Thus, the health behavior field, although
deeply wedded to the idea of people embedded in social systems, has not yet
provided the theoretical infrastructure to support models of health behavior that
are truly dynamic and integrated with respect to social context.
The central problem, then, is that each approach focuses too much on one
level of scale—social networks on networks, health behavior on individuals.
What is needed to remedy this situation is the development of computational
models of health behavior. This would offer the possibility of building compu-
tational models of social systems that capture both levels of scale simultaneously.
This idea is embodied well by the agent-based modeling approach ( for an intro-
duction to this approach, see Wilensky & Rand, 2015).
The remainder of this chapter will be guided by the goal of building person–
context systems of health behavior, but with equal attention to both the person
and context. We will review the nascent field of what we have coined the com-
putational health behavior modeling approach (Orr & Plaut, 2014), a subfield
of health behavior that emphasizes the use of computational and mathematical
modeling for representing health behavior theory. We offer the criteria in Table 5.1
as a way to organize our thoughts around computational models of health behav-
ior. On the whole, these criteria define the desired qualities of a useful health
behavior theory that can promote the unification of social networks and health
behavior to scale.

Past and Related Work

The field of computational health behavior modeling (Orr & Plaut, 2014), to
date, has only used two formalisms: dynamical systems and artificial neural net-
works. Aligning with this dichotomy is the fact that only two labs have produced
computational models of health behavior. The dynamical systems approach,
led by Daniel Rivera at Arizona State University, provides a control systems
engineering perspective on health behavior. The neural networks approach to
 Computational Modeling of Health Behavior  83

TABLE 5.1  Criteria for useful computational models of health behavior.

Criterion name Description

Dynamic Is it sensitive to changing contexts? Does it capture past

behavior as part of its current state?
Learning Is there ability for permanent change of model aspects,
parameters, etc.? Is it important for well-established
theory?
Network-ready Is there a reasonable interface with the environment
and/or with the social context?
Novel mechanisms Does it pose novel theoretical entities that are testable?
Empirical grounding Has the model been compared to empirical data? Have
novel predictions been tested? Have novel mechanisms
been tested? Have the social aspects (network-ready)
been empirically tested?
Social psychological theory Is it grounded in principles of social psychology? Does
the model incorporate constructs from existing theory?
Cognitive science theory Is the model grounded in principles of cognitive science?

modeling health behavior, developed by our lab (Orr is lab leader), was highly
influenced by work in social cognition (for a compendium, see Read & Miller,
1998), an area of research that encapsulated the first sketch of the potential prom-
ise of computational modeling of health behavior (see Shoda & Mischel, 1998).
We will review each formalism in turn.

Modeling Health Behavior as a Dynamical System

The work by the Rivera Lab used a dynamical systems approach to represent the
dynamics of health behavior (Martin et al., 2014; Navarro-Barrientos, Rivera, &
Collins, 2011). Dynamical systems, a field with roots in both physics and control
systems engineering, affords modeling how a set of variables change over time
in relation to one another, with an emphasis on the system as a whole. Some of
the common systems that can be modeled using the dynamical systems approach
are the heating of a home and the spread of a virus in a population. Much of the
time, the solutions of real-world applications or complex theoretical relationships
are numerically estimated using computational algorithms, but for some simpli-
fied cases, analytic solutions are available (e.g., identification of equilibrium). In
psychology, the dynamical systems approach is most dominant in understand-
ing issues related to embodied and situated cognition and perception (Schoner,
2008), to development (Smith & Thelen, 1994), and to interpersonal dynamics
(Marsh, Richardson, & Schmidt, 2009; Vallacher, Read, & Nowak, 2002).
The approach in the Rivera Lab has been to reconceptualize both the Theory
of Planned Behavior (TPB) (Ajzen, 1991) and Social Cognitive Theory (SCT)
84  Mark G. Orr and Daniel Chen

(Bandura, 1986), two core psychological theories that have been widely co-opted
by the health behavior field, into a dynamical systems model in which each varia-
ble is considered as either an inventory (think number of units in a warehouse) or
as a flow (think deliveries into and out of a warehouse). To gain some insight into
this modeling method, imagine the following. A warehouse, with a capacity of
100 widgets, receives ten widgets at the open of business and sends out 10% of its
inventory by the close of business that same day, that is: Xend = (Xbegin+10)*0.90.
To run a simulation over a course of a month on this very simple system, you
would first specify the number of units in the warehouse at the opening of the
first day of the simulation. Then you would calculate the inventory of the ware-
house at the close of business each day in the simulation. To gain insight into this
simple system, it would be useful to construct a simple table of corresponding
Xbegin and Xend values under two starting conditions: Xbegin = 0 and Xbegin = 100.
The systems developed by the Rivera Lab use the same principle, whereby
flows and inventories represent psychological constructs. Figure 5.1 shows their
effort to reconceptualize the TPB (Navarro-Barrientos et  al., 2011). Some of
the TPB constructs are represented as inventories (attitude, subjective norms,
perceived behavioral control, intention, and behavior) and some as flows (ξ1, ξ2,
and ξ3 correspond to outcome belief, normative beliefs, and perceived control
beliefs, respectively). To see how this system works, consider the attitude stock.
Its inflow, as per TPB, is outcome beliefs (ξ1), which is scaled by γ11 and time (the
θ1 represents delays, so that as θ1 increases, the inflow decreases) and is perturbed
by ζ1 (functioning as noise). Its outflow is split in two—some flows to the inten-
tion inventory scaled by β41, and the remaining flows out of the system (inversely
proportional to the flow to intention). Each of the inventories acts in this way.
As per the TPB, the only external input to the system is from ξ1 to ξ3. That is,
external modification of the system operates strictly through changes in beliefs.
Another slightly later effort by the Rivera Lab was to build a dynamical systems
model of Social Cognitive Theory (Martin et al., 2014).
The Rivera Lab used the TPB and SCT models in two ways. First, they con-
ducted a series of what are called counterfactuals to explore and compare the
effects of different health behavior intervention approaches. The simulations using
the TPB system focused on: (a) changes in both temporal (e.g., time lag of flow
between intention and behavior) and structural relations between inventories
(e.g., more or less flow between inventories), and (b) the efficacy of an interven-
tion as a function of the temporal ordering of the components of said intervention
(e.g., intervene on attitude prior to subjective norms). The simulations using
SCT centered around comparing how initial values of the inventories affect the
dynamics of behavior, looking for potential nonlinear dynamics (e.g., if the initial
levels of self-efficacy were large enough, would the system be self-sustaining and
maintain itself without further external cuing from a simulated intervention).
Second, the Rivera Lab attempted to estimate the parameters of the SCT
system given data from the Mobile Interventions for Lifestyle Exercise and Eating at
 Computational Modeling of Health Behavior  85

FIGURE 5.1   dynamical systems model of the Theory of Planned Behavior

A
(Fishbein & Ajzen, 2010) (reproduced with permission from
Navarro-Barrientos et al., 2011). See text for details.

Stanford. This data set contained dynamic data (daily, over an eight-week period)
regarding some of the core inputs to and outputs of SCT (for details of the data
set, see King et al., 2013). The estimation approach used gray-box system identi-
fication (Ljung, 1999), an approach developed specifically to estimate parameters
of mathematical models of physical systems; it leverages statistical estimation and
prior knowledge of the structure of a dynamical system. Gray-box system identi-
fication may prove to be useful for integrating dynamic empirical data on health
behavior into a computational modeling framework. See Bohlin (2006) for an
accessible introduction to gray-box system identification.
How does this approach stand up to the criteria offered in Table 5.1? The
approach is dynamic in the sense that it can directly capture changing contexts
and contains feedback loops among inventories; it is not directly network-
ready, but could easily accommodate this need; these models assume several
novel processes (most of the parameters in Figure 5.1). However, the Rivera
Lab approach is weaker in terms of the empirical grounding of its theoretical
assumptions. In our estimation, the data-driven gray-box fitting algorithm used
to model the SCT had many free parameters and should be considered one of
several approaches for empirical grounding—experimentation will be needed.
Finally, although the use of dynamical systems has precedence in both social
psychology and cognitive science as a theoretical framework, the specifics of the
Rivera approach are not well grounded theoretically. Of principal concern are:
(1) the conceptualization of attitudes as stocks and of causal relations as flows
86  Mark G. Orr and Daniel Chen

with rates, and (2) the assumption that information flows forward only (e.g.,
from beliefs to attitudes to intention to behavior, but not the reverse)—an
assumption that does not accord with the well-established notion of constraint
satisfaction as a mechanism underlying several social psychological phenomena
(Read & Miller, 1998).
In summary, the work from the Rivera Lab is of interest because it brings
an engineering/control systems perspective to health behavior modeling using a
well-established psychological theory as its foundation. For the health behavior
field, this effort supports the larger goal of developing intervention systems that,
similar to a thermostat in a heating system, control behaviors related to health. To
realize this goal, we suggest that the next steps in this work focus on strengthen-
ing the approach with respect to the criteria in Table 5.1.

Modeling Health Behavior Using Artificial Neural Networks

The work of Lowe and colleagues (Lowe, Bennett, Walker, Milne, & Bozionelos,
2003) represents, to our knowledge, the earliest effort to integrate computational
modeling with health behavior theory. This effort revolved around the question
of whether a feed-forward pattern associator (a specific type of artificial neural
network) could represent well the mapping between intention and a pattern of
beliefs across a corpus of real human subjects’ data. The empirical measures were
designed in terms of the TPB.
Figure 5.2 shows the structure of the model. It was trained, successfully, to
learn the mappings between each subject’s set of beliefs and his or her intention.
Lowe et  al. (2003) used the hidden layer representation to classify subjects as
more or less similar to each other. Using this information, the subjects were clas-
sified into three types of high-intenders, one type of moderate intender and two
types of low-intenders. This was useful in that it provided a more nuanced analy-
sis of the differences between subjects (or really, groups of subjects) that would be
difficult to detect using more traditional statistical techniques.
Although this was an interesting and fruitful approach for understanding the
given data in a novel way, it does not serve as an example of a dynamic compu-
tational model of health behavior. The manner in which the model was analyzed
and interpreted makes it clear that their approach was a data-analytic approach
that was designed to reveal different types of subjects (which it did successfully).
This is an approach to the analysis of complex data that is commonly used in the
machine learning and statistical learning literatures.
More recently, over the past few years, our lab has developed a dynamic con-
straint satisfaction model of the Theory of Reasoned Action (TRA) (Fishbein &
Ajzen, 1975) that was strongly influenced by the criteria listed in Table 5.1 (Orr &
Plaut, 2014; Orr, Thrush, & Plaut, 2013). The TRA, the precursor to the TPB,
assumes that behavior is driven by one’s intention to perform the behavior;
intention is driven by norms, attitudes, and perceptions of behavioral control
 Computational Modeling of Health Behavior  87

Output layer (one unit)

Hidden layer (three units)

Three-section
input layer
Control beliefs Normative beliefs Behavioral beliefs
(4 units) (8 units) (14 units)

FIGURE 5.2   schematic structure of an artificial neural network (reproduced with

A
permission from Lowe et al., 2003) that was used as a statistical learning
approach for classifying types of subjects in a human subjects study.
See text for details.

which are driven by beliefs related to the behavior. Our model’s core innova-
tion was that it represents the dynamics of what we call intention formation—a
notion of how intention is generated on-the-fly (for the origin of this notion, see
Conrey & Smith, 2007).
In our model, intention formation is dynamic in a particular way. It accounts
for both what has been learned about a behavior (in terms of beliefs) by social
learning from past experience and the immediate influence from the current
social context, simultaneously. As an extreme analog, imagine this situation. A
Democrat is at a cocktail party hosted by the Republican National Convention,
and the topic of voting for the Democratic presidential candidate comes up in
conversation. Our model attempts to capture the on-the-fly intention forma-
tion (to vote for the Democratic candidate) of the Democratic constituent given
his/her past experiences and the current conversation at the cocktail party. In
terms of health behavior, we can imagine other relevant examples, such as a col-
lege student studying abroad. The learned beliefs about a behavior might vary
significantly across international borders.
Figure 5.3 depicts the formal structure of our model. A belief is represented
as a set of memory units that are linked; one unit represents positive valence, and
the other represents negative valence (e.g., two circles labeled with numeral 1
represent the positive and negative valence of one belief, and the fixed horizontal
lines connecting these two units represent a bidirectional inhibitory function
(called inter-bank connections in Figure 5.3). The coupling between valence
units is, by design, inhibitory such that everything else equal, the activation of
one valence will inhibit the potential for activation of the other valence within a
belief. Conceptually, this is a reconceptualization of the belief structure proposed
88  Mark G. Orr and Daniel Chen

by the Theory of Reasoned Action in which a belief is represented as (using out-

come beliefs as the example) b*e, where b is strength of the belief (the subjective
probability that the outcome of performing a behavior will come true), and e is
the evaluation of the behavioral outcome (its valence). The inhibitory coupling
within belief valence units in our model does not exclude the potential for both
belief units having high activation simultaneously, but only attempts to reduce
this possibility. Memory units are constrained to vary in activation from not
active to fully active (represented as 0 and 1, respectively).
A key assumption of our model is that the memory units are activated (or
cued) by exposure to belief valence analogs in social situations and contexts in
which relevant belief are present—that is, these units are hypotheses that a belief
and its associated valence is present or absent in a social situation or context. In
the simulations described below, a social context is operationalized by clamp-
ing the relevant memory units that map onto the belief valences present in the
social context. Once exposed to a social context, the system settles into a local
equilibrium that is constrained by (1) the presence of the cued belief valences in
the current social context, and (2) the set of connections between memory units.
Formally, the equilibrium state defines the on-the-fly intention state as the mul-
tidimensional point in unit activation space. In the simulations that follow, we
simplify this by aggregation of the activations at equilibrium of the positive and
negative valence units separately. In doing so, intention is relative to which set of
valences (positive or negative) has a higher average value.
The set of like-valenced units (either positive or negative) are fully connected
by a set of modifiable connections that allow for encoding the patterns of belief
structures in the social contexts across time (called the intra-bank connections in
Figure 5.3). That is, the model has the ability to capture statistical regularities over
a set of social contexts separated in time (e.g., the simplest regularity is frequency of
exposure to a belief valence; a complex regularity would be a nonlinear combina-
tion such as the positive belief valence 1 occurs only when either the positive belief
valence 2 or the positive belief valence 3 is present, but not when both are present).
These connections are depicted in Figure 5.3 as the intra-bank connections, where
the term “bank” refers to the grouping of units by valence.
It is important to realize that our modeling assumptions (the formal struc-
ture and processes) should not be thought of as a contradiction of the TRA,
but instead as an attempt to flesh out some critical details when considering
TRA as a dynamic theory (for details on this issue, see Orr & Plaut, 2014).
The choices that dictated our model architecture were theoretical. We rep-
resented single beliefs as opposing valences, an idea we borrowed from prior
work (Shultz & Lepper, 1996, 1998) and work on ambivalence of attitudes
(De Liver, Van der Pligt, & Wigboldus, 2007; Thompson & Zanna, 1995).
Moreover, our desire to capture the notion of an attitude being a state rather
than a thing (borrowed from Conrey & Smith, 2007), naturally led to a recurrent
(constraint satisfaction) network.
FIGURE 5.3   depiction of an artificial-neural network model of the Theory of
A
Reasoned Action (Fishbein & Ajzen, 2010) (reproduced with permission
from (Orr et al., 2013). Each belief (the numbered circles where number
indexes belief) is composed of two processing units, one representing
negative valence and the other representing positive valence. The
constraints within each bank (depicted as curved arrows on the right
and left of the rows of belief units) represent the relations among beliefs;
learning operates by modifying these constraints. The constraints between
the banks are fixed and reflect the theoretical commitment that beliefs
are unidimensional (i.e., valence units within a belief should inhibit one
another). Model operation starts with an exposure to beliefs from a social
context (the external input) and ends in an intention state through the
process of constraint satisfaction (the processing units are directly clamped
by external input; the output of the processing units is directly taken as
each unit’s activation). See text for details.
90  Mark G. Orr and Daniel Chen

Using this basic model, we addressed two separate issues that are of high rel-
evance for health behavior theory. The first issue centers on a key driver of the
heterogeneity in behavior across individuals with respect to intention formation
(Orr & Plaut, 2014). The second issue deals with how a person mitigates the
effects of past experience (what one has learned over time, e.g., from parents and
peers) and the effects of his/her immediate social context (“immediate” is a rela-
tive term, but to mean something like within a few days) when forming health
behavior intentions (Orr et al., 2013). In what follows, we summarize this prior
work (for details, please consult the original works).

Drivers of Heterogeneity in Health Behaviors

Health behavior theory explains heterogeneity within a population in a linear
way. In terms of dynamics, this means, roughly, that changes in the inputs to a
system will produce similarly scaled outputs. However, an alternative explanation
was offered by so-called quantum behavior change theory—a characterization
of health behavior as prone to sudden and dramatic shifts (e.g., quitting and
re-uptake of cigarettes) (Resnicow & Page, 2008). Such sudden shifts are by
nature very difficult to predict precisely because the drivers of behavior interact in
a nonlinear, so-called quantum fashion. By this account, nonlinear processes may
drive the heterogeneity across individuals. Our work, described next, puts forth
the first formal computational model of this process and provides a set of simula-
tions to serve as proof of concept (Orr & Plaut, 2014). The overarching goal of
the simulations was to test whether our TRA neural network model would act
characteristically as a nonlinear complex system. This is both a practical concern
(How do we intervene when the system may act chaotically?) and a theoretical
concern (a formalization of the notion of quantum behavior change put forth by
Resnicow & Page, 2008 to include sensitivity to initial conditions, path depend-
ence and tipping points).
The simulation was designed to capture, in the abstract, the dynamic nature
of intention given a series of changing social contexts. (This was a theoretical
exercise, so the model was not supposed to represent a particular behavior.) It is
useful to imagine the time-course of intention states as a person goes through a
significant period of time (this could be several weeks or months) during which
each time point represents exposure to another social context. At every time
point, the model computes an intention state using its intention state from the
last time point mixed with the intention states in the immediate social context
(similar to neighbors in a social network). In this simulation, the model was not
“allowed” to learn because our interest was in whether the system, even without
learning, would show complexity. In fact, we purposely set up the intra-bank
connections such that all were positive; thus, small amounts of activation within
a valence bank had the potential for rapidly increasing the activation of other
units in the same valence bank during the settling process for a single exposure to
 Computational Modeling of Health Behavior  91

a social context (see the original source for more details on the model structure:
Orr & Plaut, 2014).
During a single simulation, on each of 600 time steps, the model was ran-
domly exposed to one of three types of social contexts (biased towards intending,
biased towards not intending, and ambivalent with respect to intention). Each
of the three contexts was presented exactly 200 times. The biasing was formal-
ized as inputs that activated all of the units within only one of the valence banks;
ambivalence activated an equal number of units within both valence banks (40%
of the inputs). During the process of the simulation, the activation levels of all
belief valence units were recorded upon exposure to each social context, yielding
a time series (length = 600) of the intention state of the model. That is, the state
of the model at equilibrium after exposure to a social context was read out as the
raw output of all of the units of the model individually; the valence of each out-
put unit was tagged so that the overall valence could be seen graphically.
We ran 20 separate simulations; the difference between each was the order-
ing of the sequence of social contexts (each was a unique, random order). The
purpose of running multiple simulations was to capture the variety of behaviors
of the model under the condition that the only difference between the 20 simula-
tions was the ordering of the inputs.
The operation of each simulation was as follows. The initial state of the model
was driven by exposure to the first social context. For example, it may have
been exposed to the ambivalent context. Subsequently, the model dynamics were
driven by the series of social contexts. Importantly, for all of the simulation,
except at the very beginning, the state of the model at any point in time was
also dependent on its immediate (t-1) past intention state. This secondary feature
forced the model to mitigate its very recent past with its current social context.
Figure 5.4 shows four of the 20 simulations. The key behaviors of the model
were represented by four distinct dynamic profiles. The first profile, shown in
panels A and B, shows the positive valence bank (to “intend”) winning out early
on, followed by maintenance of this state throughout the simulation; the negative
valence bank (to “not intend”) was reduced to zero activation. The second row
shows the model rapidly changing from “intend” to “not intend.” Row 3 exhib-
its a persistent ambivalence towards the behavior. Both the positive and negative
valence banks were strongly activated throughout the simulation. The final row
is characterized by switching between “not intend” and ambivalence.
Remember, the overarching purpose of this simulation effort was to judge
to what degree our computational model exhibited complex, quantum-like
behavior. The complex systems literature, including catastrophe theory, offers
some well-established qualitative criteria for this purpose (Miller & Page, 2007;
Witkiewitz & Masyn, 2008). Using these criteria, our model displayed key sig-
natures of complex systems. First, the statistical signature of the different contexts
(1/3 favored intend, 1/3 not intend, and 1/3 ambivalent), did not present in time
series in the simulations—that is, we would expect to see 1/3 of the time series
92  Mark G. Orr and Daniel Chen

FIGURE 5.4   he timecourse of the activation of the positive and negative valence
T
units over four separate simulations, each represented by a row
(reproduced with permission from Orr & Plaut, 2014). The first
column represents the first 100 social contexts; the second column
shows the full time series of social contexts (600 contexts). The positive
valence units are depicted as black circles, the negative as gray crosses.
See text for details on the simulation procedure.

show intention states of not intend, 1/3 intend, and 1/3 ambivalent, but this was
not the case. This feature, that the output was not linearly related to the input, is a
classic signature of complex systems. Panel B of Figure 5.4 illustrates this behavior
in the extreme—the time series of intention states of the model did not show any
indication of negative or ambivalent social contexts even though 2/3 of the input
contexts were of this nature. Second, panels B, D, and F in Figure 5.4 clearly
demonstrate path dependence, another signature of complex systems whereby
the system remains in an attractor state. Third, in panel C, we see a bifurcation,
probably driven by a self-organizing process, in which the system changes states
rapidly—another clear signature of complex systems. Fourth, panels E and G
show oscillatory behavior, something that is common in complex systems.
 Computational Modeling of Health Behavior  93

Also, our model shows some of the key qualitative properties of a catastrophic
system, called “catastrophe flags” (Gilmore, 1981). For example, we see what are
called sudden jumps—a rapid transition from one behavioral state to another (see
Figure 5.4, panels C and H); we also see multimodality—having only a few pro-
nounced behavioral states. The valence units were mostly either of an activation
between 0.60 and 0.80 or of very low activation across the 20 simulations. Finally,
we see what is called inaccessibility—behavioral states that were very unstable.
The valence units did not have resting values other than near zero activation or
between 0.60 and 0.80.
In summary, this modeling effort served as a proof of concept that our com-
putational model of the TRA could exhibit behavior as if it were a complex
system. Thus, we succeeded in unifying TRA, a traditional health behavior the-
ory, with quantum health behavior theory via instantiation of a computational
model. Furthermore, these results beg the question: What if quantum-like health
behaviors were networked? We will address this question below, in the section
“Emerging Work on Psychologically Plausible Social Systems.”

Mitigating Past Experience with Immediate

Social Context
The second simulation explored the role of how learning from past experience
can change the way current social context is interpreted. This simulation used
human subjects data, on sexual behavior in a cohort of adolescents in a US school
district, that were developed specifically with consideration of the measure-
ment constructs of the Theory of Reasoned Action (these data are described in
Gillmore et al., 2002). The simulation explored what happens when the model
learns from a social context in which beliefs about having sex are less positive,
and then, after learning is complete, switches to a context that is more positive.
(The original work and the modeling details can be found in Orr et al., 2013.)
To this end, we first trained the model, using the generalized delta-rule, to
learn a belief structure from exposure to the beliefs of 10th-grade females who
never had sex. There were 105 10th-grade females in the data set, each of which
was transformed into a binary vector; 200 epochs of the 105 inputs were used
for training. Then we tested the model via exposure to 12th-grade females who
had never had sex and 12th-grade females who had had sex. During testing,
we constrained the model so that it could no longer learn. By this manipula-
tion, the simulation probed the degree to which the formation of intention was
constrained by both past (exposure to 10th-grade non-experienced females)
and current contexts (sexually experienced females); please refer the section
“Modeling Health Behavior Using Artificial Neural Networks” for a reminder
of how constraint satisfaction works in the model. Simply put, we explored
how the intention towards sexual behavior of a 10th-grade female virgin
would emerge when she was put in a social situation with 12th-grade females.
94  Mark G. Orr and Daniel Chen

(We did not choose the term “virgin,” but simply follow the lead of the authors
of the original empirical study: Gillmore et al., 2002. Please refer to the original
work for further details.)
Figure 5.5 shows the key findings in terms of the mean activation of each
of the sets of valence units (positive = intend; negative = not intend). From left
to right, the first condition (F10V) illustrates that the model exhibits a strong
negative response when tested against the same input that was used when it was
learning. This makes sense given that the data for F10V were largely negative in
terms of sexual behavior. This condition serves as a baseline for comparison to
the other two conditions, F12V and F12NV, in which we see a decline in the
activation of the negative valence units and a corresponding increase for the posi-
tive valence units. Also, notice that for the F12NV condition, the response of the
model was still more negative than positive.
The remaining two conditions, P50 and P75, reflect purely simulated data,
but provide further insight into the model’s behavior during testing. The
P50 condition represents completely neutral inputs. For this condition, we
would expect the activation for both the positive and negative valence units
to be equal if the prior training context did not bias the system. However,
the negative valence units win out, reflecting the prior training context. The
same bias was found for the P75 condition, which represents largely posi-
tive inputs. Although the positive valence wins out, it is highly dampened
(we would expect greater activation in this condition without bias from past
experience).
In summary, the current social context influenced but did not override
what was learned from past experience. Specifically, exposure to relatively
more positive beliefs about sexual behavior in the current social context did
increase intention, but this tendency was dampened from the learning in
prior social contexts with less positive beliefs about sexual behavior. In short,
this notion is highly congruent with the needs of the health behavior field—a
formal and explanatory account of how the present and the past are integrated
to generate an intention state. We can imagine that the distribution of biases
from past experience would greatly impact the spread of intention along a
social network.

Synopsis of the Neural Network Approach

to Health Behavior
Our prior work presented above on the use of artificial neural networks and
health behavior owes equally to prior work in both social cognition and social
networks. In fact, we designed the model to serve as the vertices of a large-scale
social network, and thus, by our estimation, it meets all of the criteria, by design,
listed in Table 5.1.
FIGURE 5.5  S imulations of the interdependence between learning from past
experience and the current social context on intention formation in
the Theory of Reasoned Action (reproduced with permission from
Orr et al., 2013). The x-axis represents the condition during testing
(F10V = female 10th-grade virgins; F12V = female 12th-grade virgins;
F12NV = female 12th-grade non-virgins). The other two conditions,
P50 and P75, were highly structured, simulated data (P50 inputs
were 50 percent positive; P75 inputs were 75 percent positive). The
mean activation of the valence banks is represented on the y-axis
(dark gray = positive; light gray = negative). The dashed horizontal
lines show the mean bank activation of values 0.25, 0.50, and 0.75.
96  Mark G. Orr and Daniel Chen

Emerging Work on Psychologically Plausible

Social Systems
Our prior work using artificial neural networks of health behavior, as described
above, was designed in part for integration into the social networks framework.
This idea, for us, originated from work on the dynamics of dyadic relations
(Shoda et al., 2002): work that suggested, theoretically, that the mental state of
an individual depended upon the presence of social interaction—without it, an
individual ended up in state A; with it, the same individual, everything else equal,
could end up in state B. This was intriguing from a psychological point of view
because it put the person–context system in a formal computational model with
an equal emphasis on both person and context. The preliminary work we present
next was driven by a different but closely related question, one that is crucial for
the health behavior field: To what extent would the dynamics of the diffusion
of behavior in a social network depend on the parameters of the computational
model of individuals’ behavior?
Both of the simulations from our lab presented above are suggestive. Imagine
a set of agents connected in a social network with the heterogeneity shown in
Figure 5.4 (each of the rows in Figure 5.4 represents a separate individual). Would
this heterogeneity persist if the individuals were influencing one another (e.g., the
input of one individual would be the output of another)? Consider the notion that
learning from past experience biases the relation between input to output (from
Figure 5.5). Do such biases, at the individual level, have an effect on the diffusion of
intention in a population? These and similar considerations led our lab to develop
the Multi Agent Neural Network (MANN) platform, a new simulation platform
designed to study the effects of different vertex architectures (“architecture” refers
to the type of computational models used to represent individuals’ behavior) on the
diffusion of attitudes, beliefs, and behaviors on relatively large-scale social networks
(n = 10^3). (For the link to the code base, see Chen & Orr, 2016.)
What is presented next is a proof of concept that variation in parameters of
the vertex architecture does, in fact, affect the diffusion of intention on a social
network even when the generating parameters of the social network are fixed.
In general terms, the diffusion paradigm works as follows. First, it sets all verti-
ces to one type of state (the simple case is “off”), then seeds a small fraction of
the vertices to have a different state (the simple case is “on”). The seed vertices
then spread the different states throughout the network in a way defined by
assumptions about how the state can be transferred between vertices joined by
a social network tie. A real-world analog is the spread of a virus in a school or
community.
For this exercise, we based the vertex architecture on the neural network
presented above from our prior work (Orr & Plaut, 2014). Our approach was
to span the parameter space generated by the full factorial of two ranges of the
inter- and intra-bank connections. (See Figure 5.3 and its description above for
 Computational Modeling of Health Behavior  97

details of these two parameters.) For each cell of the factorial we ran approxi-
mately ten simulations using the diffusion paradigm to simulate the diffusion of
intention. The work presented here is a replication of a similar parameter sweep
that showed similar results using the MANN platform (Orr, Ziemer, & Chen,
in press).
Each simulation began with the generating of a small-world social network (as
specified in Watts & Strogatz, 1998), fixed with the following parameter values:
250 vertices, ten nearest neighbors, a 0.02 rewiring probability, and bidirectional
edges). All vertices were initialized to have the same approximate values of the
inter- and intra-bank connections. The diffusion process began by exposing five
of the vertices to an input, randomly chosen, that reflected either strong inten-
tion, strong not intention, or ambivalence. Diffusion was allowed to progress
for 100 time steps. Each time step updated all of the vertices, during which,
roughly, each vertex was given one of its neighbors’ intention state vector (its
output) at t-1 as its input. The intention state of each vertex was calculated as the
average value of the positive valence bank minus the average value of the nega-
tive valence bank. Figure 5.6 provides a comparison between the early diffusion
process and its later stage in a single simulation.
Figure 5.7 shows the full results of the full factorial of the inter- and intra-
bank connection parameters (20 levels of inter-bank connections and 30 levels
of intra-bank connections for a total of 600 cells in the factorial). Each cell in
Figure 5.7 represents the average value over approximately ten simulations. (So,
in total, Figure 5.7 represents approximately 6000 separate diffusion simulations.)
The measurement represented in Figure 5.7 is the assortativity with regard to

FIGURE 5.6   n example diffusion of intention simulation on a small-world social

A
network (N = 250). The vertex architecture is a recurrent neural
network that models intention formation per the Theory of Reasoned
Action (Fishbein & Ajzen, 2010). Vertex states are represented by the
gray scale; black is “not intending” and white is “intending.” The plots
show the diffusion process at (a) an early and (b) a late point in the
simulation.
98  Mark G. Orr and Daniel Chen

intention state at the end of each simulation—a measure that represents the
amount of clustering of intention given the typology of the social network. For
reference, Figure 5.6b depicts strong clustering with regard to intention and a
high degree of assortativity.
These results provide clear evidence that, in the abstract, the parameters of
the vertex architecture alone can have a dramatic effect on the social network
diffusion process. Given the theoretical nature of these simulations, a natural
set of questions arise with regard to the implications of an idea—the idea of

FIGURE 5.7   he average network assortativity across approximately ten diffusion

T
of intention simulations (per cell). The two axes represent parameters
of the recurrent neural network of the Theory of Reasoned Action
(Fishbein & Ajzen, 2010) (20 levels of inter-bank connections and 30
levels of intra-bank connections for a total of 600 cells in the factorial).
For all simulations, the generating parameters of the social network
were identical. See text for further details.
 Computational Modeling of Health Behavior  99

embedding computational psychological models into social networks; this ques-

tion is equally important for both psychology and the field of health behavior.
(We co-opted the notion of thinking in terms of the implication of an idea
with regard to the use of computational modeling for understanding cognitive
processes from McClelland, 2009.) The primary implications, as we see it, are as
follows. First, the link to applications in the practice of public health and health
behavior is dependent on successfully bridging theory (presented above) that
is closely grounded in data (not presented above) which can eventually drive
interventions and prevention efforts in human populations by virtue of causal
what-if scenario modeling. Public health, in particular, has had some success
leveraging agent-based modeling for the purpose of intervention and preven-
tion work that is tied to data and used to think about potential intervention/
prevention efforts (e.g., Hashemian, Qian, Stanley, & Osgood, 2012; Hennessy
et al., 2016), so there may be use in borrowing this approach. Another possibil-
ity may be the investigation of how the interaction between vertex architecture
and social network topology drives how an agent’s inputs vary over time and
how inputs vary across agents within a system. Distribution of inputs, as a broad
construct, is important for many psychological processes (e.g., learning, devel-
opment, skill acquisition). A third, admittedly vague, implication is that there
may be potential for gaining new insights into how cognitive processes have
evolved in evolutionary time (see McLeod, Plunkett, & Rolls, 1998 for early
work in this area in cognition, and also Goldberg & Holland, 1988 for informa-
tion on the related work on genetic algorithms). A system like MANN allows
for, in theory, manipulating environment (e.g., social network topology), cog-
nitive mechanisms, and the interdependencies between both simultaneously.
This seems, to a first approximation, minimally sufficient for beginning to study
biological evolution related to cognition and social context. Although many
challenges arise in these types of endeavors, something like the MANN platform
should prove useful, at least as a starting point.

Future Directions
The criteria in Table 5.1 lay out a good deal of the work that is needed over
the next several years in order to work towards computational social systems
of health behavior. The more difficult activities will be the iteration between
model development, theory generation, and empirical grounding. In this vein,
we think a fruitful approach will use a mixture of tightly controlled experi-
ments and large-scale social experiments. For example, our lab is leading an
NSF-funded project (with collaborators at Virginia Tech, Carnegie Mellon
University, Sandia National Laboratory: NSF #1520359) that will soon run a
controlled large-scale (N = 1000) online social experiment in which participants’
social behavior (something like posting and reading tweets on Twitter) is col-
lected over several weeks. In addition to this, we plan on capturing changes in
100  Mark G. Orr and Daniel Chen

attitudes at several points throughout the online experiment using an adaption of

the evaluative priming experimental procedure (see Fazio & Olson, 2003). Thus,
we will combine a loosely controlled social experiment with a tightly controlled
attitude measurement procedure—ecological validity at the social level with high
construct validity at the individual level.
A central issue for all of the modeling approaches we reviewed above, includ-
ing those in the section “Emerging Work on Psychologically Plausible Social
Systems,” is that the models are of limited (or focused) scope. Our neural net-
work of TRA is really just a dynamic model of attitude formation. The work
by the Rivera Lab is similarly constrained in scope. Neither approach represents
decision-making in the larger sense, something for which production systems—
for example, ACT-R (Anderson, 2007; Anderson & Lebiere, 1998)—and other
larger scoped cognitive architectures are designed. We think a fruitful and poten-
tially necessary direction for the computational health behavior modeling field
will lie in this direction.

Note
1 The production of this chapter was supported in part by NSF Award #1520359.

References
Ajzen, I. (1991). The theory of planned behavior. Organizational behavior and human
decision processes, 50, 179–211.
Anderson, J. R. (2007). How can the human mind occur in the physical universe? Oxford:
Oxford University Press.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Mahwah, NJ: LEA.
Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood
Cliffs, NJ: Prentice Hall.
Bhattacharyya, S., & Ohlsson, S. (2010). Social creativity as a function of agent cognition
and network properties: A computer model. Social Networks, 32, 263–278.
Bohlin, T. (2006). Practical grey-box process identification: Theory and applications. London:
Springer-Verlag.
Chen, D., & Orr, M. G. (2016). Multi Agent Neural Network (MANN). Retrieved from
https://github.com/chendaniely/multi-agent-neural-network
Conrey, F. R., & Smith, E. R. (2007). Attitude representation: Attitudes as patterns in
a distributed, connectionist representational system. Social Cognition, 25(5), 718–735.
De Liver, Y., Van der Pligt, J., & Wigboldus, D. (2007). Positive and negative associations
underlying ambivalent attitudes. Journal of Experimental Social Psychology, 43, 319–326.
Fazio, R. H., & Olson, M. A. (2003). Implicit measures in social cognition research: Their
meaning and use. Annual Review of Psychology, 54, 297–327.
Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention and behaviour: An introduction to
theory and research. Reading, MA: Addison-Wesley.
Fishbein, M., & Ajzen, I. (2010). Predicting and changing behavior: The reasoned action approach.
New York: Psychology Press.
 Computational Modeling of Health Behavior  101

Gillmore, M. R., Archibald, M. E., Morrison, D. M., Wilsdon, A., Wells, E. A., Hoppe,
M. J., Nahom, D., & Murowchick, E. (2002). Teen sexual behavior: Applicability of
the theory of reasoned action. Journal of Marriage and the Family, 64(4), 885–897.
Gilmore, R. (1981). Catastrophe theory for scientists and engineers. New York: Wiley.
Goldberg, D. E., & Holland, J. H. (1988). Genetic algorithms and machine learning.
Machine Learning, 3, 95–99.
Hashemian, M., Qian, W., Stanley, K. G., & Osgood, N. D. (2012). Temporal aggre-
gation impacts on epidemiological simulations employing microcontact data. BMC
Medical Informatics and Decision Making, 12(132).
Hennessy, E., Ornstein, J. T., Economos, C. D., Herzog, J. B., Lynskey, V., Coffield, E., &
Hammond, R. A. (2016). Designing an agent-based model for childhood obesity inter-
ventions: A case study of ChildObesity180. Preventing Chronic Disease: Public Health
Research, Practice, and Policy, 13, 150414.
King, A. C., Hekler, E. B., Grieco, L. A., Winter, S. J., Sheats, J. L., Buman, M. P.,
Banerjee, B, Robinson, T. N, & Cirimele, J. (2013). Harnessing different motiva-
tional frames via mobile phones to promote daily physical activity and reduce sedentary
behavior in aging adults. PLoS One, 8(4), e62613.
Ljung, L. (1999). System identification: Theory for the user. Englewood Cliffs, NJ: Prentice Hall.
Lowe, R., Bennett, P., Walker, I., Milne, S., & Bozionelos, G. (2003). A connectionist
implementation of the theory of planned behavior: Association of beliefs with exercise
intention. Health Psychology, 22(5), 464–470. doi:10.1037/0278-6133.22.5.464
Marsh, K. L., Richardson, M. J., & Schmidt, R. C. (2009). Social connection through
joint action and interpersonal coordination. Topics in Cognitive Science, 1, 320–339.
Martin, C. A., Rivera, D. E., Riley, W. T., Hekler, E. B., Buman, M. P., Adams, M. A., &
King, A. C. (2014). A dynamical systems model of social cognitive theory. Paper pre-
sented at the 2014 American Control Conference, Portland, OR.
McClelland, J. L. (2009). The place of modeling in cognitive science. Topics in Cognitive
Science, 1, 11–38.
McLeod, P., Plunkett, K., & Rolls, E. T. (1998). Introduction to connectionist modelling of
cognitive processes. New York: Oxford University Press.
Miller, J. H., & Page, S. E. (2007). Complex adaptive systems: An introduction to computational
models of social life. Princeton, NJ: Princeton University Press.
Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality:
Reconceptualizing situations, dispositions, dynamics and invariance in personality
structure. Psychological Review, 102(2), 246–268.
Monroe, B. M., & Read, S. J. (2008). A general connectionist model of attitude
structure and change: The ACS (Attitudes as Constraint Satisfaction) Model.
Psychological Review, 115(3), 733–759.
Muchnik, L., Aral, S., & Taylor, S. J. (2013). Social influence bias: A randomized
experiment. Science, 341(6146), 647–651.
Navarro-Barrientos, J. E., Rivera, D. E., & Collins, L. M. (2011). A dynamical model
for describing behavioural intentions for weight loss and body composition change.
Mathematical and Computer Modelling of Dynamical Systems, 17(2), 183–203.
Orr, M., & Plaut, D. C. (2014). Complex systems and health behavior change: Insights
from cognitive science. American Journal of Health Behavior, 38, 404–413.
Orr, M. G., Thrush, R., & Plaut, D. C. (2013). The Theory of Reasoned Action as paral-
lel constraint satisfaction: Towards a dynamic computational model of health behavior.
PLoS One, 8(5), e62409.
102  Mark G. Orr and Daniel Chen

Orr, M. G., Ziemer, K., & Chen, D. (in press). Systems of behavior and population
health. In A. M. El-Sayed & S. G. Galea (Eds.), Systems science and population health.
New York: Oxford University Press.
Read, S. J., & Miller, L. C. (Eds.). (1998). Connectionist models of social reasoning and social
behavior. Mahwah, NJ: Lawrence Erlbaum.
Resnicow, K., & Page, S. E. (2008). Embracing chaos and complexity: A quantum change
for public health. American Journal of Public Health, 98(8), 1382–1389. doi:10.2105/
ajph.2007.129460
Schoner, G. (2008). Dynamical systems approaches to cognition. In R. Sun (Ed.), The
Cambridge handbook of computational psychology. New York: Cambridge University Press.
Shoda, Y., LeeTiernan, S., & Mischel, W. (2002). Personality as a dynamical system:
Emergence of stability and distinctiveness from intra- and interpersonal interactions.
Personality and Social Psychology Review, 6(4), 316–325.
Shoda, Y., & Mischel, W. (1998). Personality as a stable cognitive-affective activation
network: Characteristic patterns of behavior variation emerge from a stable personality
structure. In S. J. Read & L. C. Miller (Eds.), Connectionist models of social reasoning and
social behavior (pp. 175–208). Mahwah, NJ: Lawrence Erlbaum.
Shultz, T., & Lepper, M. (1996). Cognitive dissonance reduction as constraint satisfaction.
Psychological Review, 103(2), 219–240.
Shultz, T., & Lepper, M. (1998). The consonance model of dissonance reduction. In
S. J. Read & L. C. Miller (Eds.), Connectionist models of social reasoning and social behavior.
Mahwah, NJ: Lawrence Erlbaum.
Smith, L. B., & Thelen, E. (1994). A dynamic systems approach to the development of cognition
and action. Cambridge, MA: MIT Press.
Thompson, M. M., & Zanna, M. P. (1995). The conflicted individual: Personality-based
and domain-specific antecedents of ambivalent social attitudes. Journal of Personality and
Social Psychology, 63(2), 259–288.
Valente, T. W. (2010). Social networks and health: Models, methods and applications. New
York: Oxford University Press.
Vallacher, R. R., Read, S. J., & Nowak, A. (2002). The dynamical perspective in personality
and social psychology. Personality and Social Psychological Review, 6, 264–273.
Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of “small-world” networks.
Nature, 393(6684), 440–442.
Wilensky, U., & Rand, B. (2015). An introduction to agent-based modeling: Modeling natural,
social and engineered complex systems with NetLogo. Cambridge, MA: MIT Press.
Witkiewitz, K., & Masyn, K. E. (2008). Drinking trajectories following an initial lapse.
Psychology of Addictive Behaviors, 22(2), 157–167. doi:10.1037/0893-164x.22.2.157
PART II

Interpersonal Dynamics
6
INTERACTION-DOMINANT
DYNAMICS, TIMESCALE
ENSLAVEMENT, AND THE
EMERGENCE OF SOCIAL BEHAVIOR
Brian A. Eiler, Rachel W. Kallen,
and Michael J. Richardson

Introduction
It is perhaps self-evident to say that the dynamics of human behavior are
complex. This is particularly true for social behavior, in that social structures
provide the prototypical example of a complex system. Yet it is this simple asser-
tion that best defines the ways in which we may uncover stable patterns of social
behavior, as well as investigate and understand why individuals behave the way
they do—namely, using the theoretical principles and methodological tools of
nonlinear dynamics and complexity science. It is troubling, therefore, that these
tools and principles remain relatively marginalized within the field of social
psychology, a field that instead concedes explanatory power to a more tradi-
tional, information processing, and non-dynamical social cognitive perspective.
Accordingly, in this chapter we suggest how researchers may investigate social
behavior from a complex systems perspective, and discuss the implications of
adopting such a framework for social psychology more broadly.
A central theme throughout this chapter will be the degree to which the
behavioral regularities (or irregularities) that characterize human social behavior
and cognition are the result of interaction-dominant dynamics and, as such, natu-
rally emerge from the reciprocally nested, multi-timescale interactions that exist
between mind, body, and environment (Eiler, Kallen, Harrison & Richardson,
2013; Richardson, Dale & Marsh, 2014). Thus, we first briefly define what
is meant by interaction-dominant dynamics and how stable patterns of behavior
emerge in interaction-dominant dynamical systems (IDDS). We then describe a
defining feature of IDDS, slow-to-fast timescale enslavement, which can be
leveraged to identify theoretical arguments and test hypotheses about the emer-
gence and self-organization of social behavior. In doing so, we review relevant
106  Eiler, Kallen, and Richardson

and noteworthy examples from the psychological sciences that demonstrate the
value of this approach, as well as providing a brief discussion of some of the
methodological tools that can be employed to study and understand interaction-
dominant, dynamical social systems.

Interaction-Dominant Dynamical Systems

The easiest way to understand what an IDDS is, is to contrast it with a com-
ponent dominant dynamical system, or CDDS. For a CDDS, stable behavior is
defined by the linear addition or combination of the intrinsic dynamics of com-
ponent processes (i.e., fixed, a priori defined modules). As such, the macroscopic
behavior of a CDDS is rigidly defined (hard-molded), context-independent, and
functionally closed. The canonical example of a CDDS is a pendulum clock,
whereby individual components perform specifically defined functions in order
to keep time. The functionality of each component and the system as a whole
is fixed in that it can only behave in one way. Thus, a mechanical clock cannot
spontaneously cook dinner, just as an oven cannot suddenly change its func-
tionality and become a timekeeper. Of course, the opposite is true for most
biological, non-mechanical, human perception–action systems. This is to say that
a human adult can be both a timekeeper and a chef, even at the same time. It is
for this very reason that human systems (including social systems), as well as other
complex biological systems, are best conceptualized as interaction-dominant
dynamical systems, in that the macroscopic behavior of an IDDS is soft-molded,1
highly context-dependent, and multi-functional.
The soft-molded and context-dependent nature of an IDDS is due to the
fact that the behavioral dynamics and functional organization exhibited by such
systems are typically self-organized and result from the reciprocal interactions
and couplings that exist between the components of a system, rather than the
intrinsic dynamics of the components themselves (Van Orden, Hollis, & Wallot,
2012). In other words, the time-evolving, macroscopic, observable behavior of
an IDDS is not determined by any single component process or by means of
a single centralized (executive) control function, but rather emerges from the
free interplay of system elements and component processes (Bak, 1996; Van
Orden et al., 2012). In this way, emergence is the result of nonlinear interactions
amongst components in which these nonlinearities give rise to non-obvious, or
emergent behavioral phenomena (e.g., Bak, 1996; Chalmers, 2006; Kelso, 1995;
Masataka, 1993). It is worth stating, however, that to say that the behavior of
an IDDS is self-organized is not to say that the behavior of an IDDS is “uncon-
trolled” or “uncontrollable.” On the contrary, the behavior of an IDDS is often
highly controlled and well organized in relation to a specific set of environmen-
tal constraints or functional goals. What makes it different from the behavioral
control inherent to a CDDS is that for an IDDS, behavioral control emerges in a
circularly causal manner, whereby lower-level component interactions produce
 The Emergence of Social Behavior  107

higher-order structural patterns of organization that then feed back to enslave,

and often maintain, the behavioral order of those very same lower-level com-
ponent relationships. In other words, fast-timescale processes lead to slow-timescale
processes that feed back to constrain the fast-timescale processes that gave rise to them.
With the latter point in mind, it is important to appreciate that the interactive
processes and spatial-temporal couplings that define the macroscopic behavior
of an IDDS are not restricted to local spatial-temporal scales or component con-
nections. Rather, these relationships exist across multiple spatial-temporal scales.
Accordingly, for an IDDS, stable or persistent patterns of behavioral order and
control arise from the structural asymmetries that emerge between processes at
differing rates of change. The idea that control operates from slow to fast in an
IDDS2 is by no means new, and has been well articulated elsewhere (Anderson,
Richardson, & Chemero, 2012; Kello, Beltz, Holden, & Van Orden, 2007;
Bruineberg & Rietveld, 2014; Eiler et  al., 2013; Friston, 2010; Karl, 2012;
Haken, 1983, 2004; Haken & Tschacher, 2010; Van Orden, Kloos, & Wallot,
2011; Van Orden, 2010; Van Orden et al., 2012). One of the most profound and
controversial implications of this idea is that the fast-timescale, neural process of
the brain cannot be responsible for controlling the slower-timescale processes of
perceptual-motor behavior alone (Van Orden et al., 2012). On the contrary, it
is more likely that the slower timescale processes of perceptual-motor behavior
(and the structured environment within which this behavior is situated) are what
define and control the emergent neural activity and organizational structures
inherent to central nervous system (CNS) functionality (Van Orden et al., 2012).
As Jordan and colleagues have so eloquently stated, the structure of organisms,
including the structures of the CNS and brain, are embodiments of the envi-
ronmental and social context that they are thoroughly, relationally, consistently,
and continuously embedded within (e.g., Jordan, 2008; Jordan, Bai, Cialdella, &
Schloesser, 2015; Jordan & Heidenreich, 2010).
With regard to social behavior, the implication of control operating from
slower to faster behavioral timescales is that although the fast-timescale behavior
of individuals or small groups within a society can modulate or perturb slower
societal dynamics, it is the slower-timescale dynamics of the society as a whole
(i.e., culture) that constrain and control the dynamics of individuals and small
groups. In a circularly causal manner, the dynamics of higher-order social and
cultural processes both emerge from and enslave the dynamics of lower-order
individual or small group processes (cf. Kitayama & Uskul, 2011). Accordingly,
the social context that emerges from ongoing human social interaction provides
the structure that controls and determines the dynamics of future human social
behavior.
Unlike the arguments mentioned above with regard to CNS control, these lat-
ter statements are by no means controversial, and are intuitively accepted within
the literature on social and cultural systems. As stated at the beginning of the
chapter, classifying a social system as a complex IDDS is self-evident. Indeed, it
108  Eiler, Kallen, and Richardson

would seem hard to argue against the notion that social and cultural behavior results
from interaction-dominant, time-evolving processes, or that such behavior does
not emerge from the reciprocal interactions that entail interpersonal, intragroup,
and intergroup processes. Even if one accepts these points to be true, however, it
is often difficult to see the ways in which researchers may experimentally exam-
ine and identify the complex dynamical processes that give rise to specific social
processes or social cognitive phenomena, or how adopting an IDDS perspective
might lead to a more concrete or pragmatic understanding (and explanation) of
why individuals behave as they do given a social context. The barrier to answer-
ing these questions and investigating social psychological phenomenon from a
self-organized, IDDS perspective more generally has been a lack of well-defined
theoretical and methodological principles for determining when social phenom-
enon and behavioral stabilities are the result of interaction-dominant dynamics and
for investigating how these interaction-dominant dynamical processes operate to
shape the social phenomenon in question. However, much of these theoretical
and methodological principles have been developed within the interdisciplinary
field of complexity science. Thus, what follows is a brief overview of key theoreti-
cal and methodological principles and how they may be employed to investigate
and understand social psychological phenomena. As already noted, we wish to
place particular emphasis on the principle of slow-to-fast timescale enslavement, thus
we begin with a more detailed explanation of this principle.

Slow-to-Fast Timescale Enslavement

Timescale enslavement refers to the fact that more slowly changing processes
emerge from, and yet disproportionately constrain, faster-timescale processes in
an IDDS. This is not to say that fast-timescale processes cannot influence slower-
timescale processes. For example, in a multistable system (i.e., when there is
more than one possible stable state or mode of behavior the system can exhibit)
small changes or perturbations associated with fast-timescale processes can some-
times lead to a qualitative change, or “switch,” between macroscopic modes of
behavioral organization. Such fluctuations can also amplify small asymmetries,
leading to a preferential stable state (Kondepudi & Gao, 1987). Persistent pertur-
bations can also operate to destabilize a slower-timescale dynamic by pushing a
system towards a tipping point and a new stable state. In this case, the activity of
a faster-timescale process works to modulate a system’s control parameter that,
when moved beyond a critical point, results in a bifurcation, or sudden change in
the qualitative state of the system. As an example, take the recent legalization of
same-sex marriage and the rapid increase in the positive national attitude towards
gay marriage in the United States. This sudden shift in the attitudes of individual
Americans could be conceptualized as the result of a bifurcation in US culture
that resulted from the persistent activity (social perturbations) of the LGBTQA+
community and their allies (circular causality in operation again).
 The Emergence of Social Behavior  109

Additionally, soft-modeled, self-organized IDDS often sit near the precipice

of instability, in that they exhibit stable behavior, but only critically. Synonymous
with the phenomena of self-organized criticality (Bak, 1996) and meta-stable sys-
tems (Kelso, 2012), this means that the stable order exhibited by an IDDS at
any point in time is not perpetual or immutable, but intermittent. In this way,
an IDDS is a critical system in which multiple behavioral patterns can exist, but
behavior manifests adaptively in response to changing contexts. As a relevant
example, consider the multiple social roles that individuals must switch between
during the course of the day. An individual might be a parent, a spouse, a friend, a
coach, a mentor, an employee, and a boss at different times and in different social
contexts. Here, each role corresponds to a critical or meta-stable state of behavior
that an individual must be able to quickly and flexibly switch between in real
time. These meta-stable states intermittently organize and control how individu-
als behave and interact with those around them. Yet this occurs only critically,
such that fast-timescale fluctuations or sudden changes in social context can eas-
ily bring about adaptive changes to the slower identity dynamics, by switching
between these meta-stable roles of behavioral control.
Note, however, that in all of the cases we just mentioned, the potentiality of
the macroscopic behavioral organization that a system exhibits must already be
structured in a meta-stable or self-critical way in order for faster-timescale pro-
cesses to modulate those slower-order dynamics. Thus, even in instances in which
faster-timescale processes functionally influence the higher-order behavioral
organization of an IDDS, this is only possible if the structure of the higher-order
states, collective-order parameters, or meta-state manifolds (i.e., set of meta-
stable roles) have already emerged. Again, slower-timescale dynamics emerge
from and then dominate the faster-timescale dynamics that gave rise to them.
Returning to our discussion of social behavior more generally, the implica-
tion of timescale enslavement is that more slowly changing social processes (e.g.,
social norms, fundamental belongingness needs, pervasive stereotypes, cultural
differences) exert a disproportionate influence on the fast-timescale, moment-to-
moment social behavior of individuals within groups and groups within societies
(see Figure 6.1). In this way, long-timescale processes provide a dynamic structure
that shapes and constrains the possible trajectory of moment-to-moment behavior.
Take, for example, when co-actors are asked to cooperatively move wooden
planks together from one location to another. Here, cooperation emerges as a
function of the slow-timescale dynamics of interaction history among co-actors.
In this way, participants who had previously been working together continued to
do so even when the task could be accomplished successfully alone (Isenhower,
Richardson, Carello, Baron, & Marsh, 2010). In this example, the stable mac-
roscopic state of “co-action” that emerges between cooperating individuals over
time is due to the extended action capabilities that such cooperative co-action
is perceived to provide. This functional extension of the individuals’ solo action
capabilities then comes to dominate the moment-to-moment actions of the
110  Eiler, Kallen, and Richardson

individuals’ current and future behavior (for more details about this specific
example, see Marsh, Chapter 9 in this volume). Similar effects define the per-
ception of momentum during sport, whereby competitive goals and confidence
during play are dependent on the history of momentum changes during a match
(Gernigon, Briki, & Eykens, 2010).
It is important to note that the aforementioned slower-timescale effects on the
trajectory of an individual’s or group of individuals’ moment-to-moment behav-
ior are not simply an effect of history per se, but rather appear as history effects
because changes in the higher-order stabilities that give rise to them operate at a
much slower timescale This self-sustaining (i.e., auto-catalytic) persistence can,
of course, have negative social consequences. For instance, the history of long-
timescale discrimination in labor markets has been shown to have a negative

FIGURE 6.1  (a) An illustration of how different levels of organization or behavioral

order are nested within a complex behavioral signal. (b) An illustration
of the reciprocal and mutually defined system of interactional constraint
that characterizes interaction-dominant dynamical systems. Note how
the causal basis of behavioral order is circular, with the slower/larger-
amplitude processes operating to enslave the micro-order of faster/
smaller-amplitude processes, while at the same time the faster/smaller-
amplitude processes operate to modulate or perturb the macro-order of
the slower/larger-amplitude processes. Adapted from Eiler et al. (2013),
with permission from the authors.
 The Emergence of Social Behavior  111

effect on learning ability and leads to situations of learned helplessness, which, in

turn, perpetuates discrimination in the labor market (Elmslie & Sedo, 1996). In
this way, such timescale enslavement effects reflect how prior social interactions
and experience influence collective belief systems or stable social or interpersonal
relationships that then constrain future social behavior in such a way that those
collective belief systems or social relationships persist. In sum, the slow-timescale
dynamics of social systems emerges from and enslaves the faster-timescale dynam-
ics of those processes and individuals that gave rise to it.

Examples of Interaction Dominant Dynamical

Social Systems
In the sections that follow, we further unpack how timescale enslavement can
be used to understand specific social phenomena. We situate this discussion
with regard to three interaction-dominant phenomena: synergistic social motor
coordination, social networks, and the fractal nature of human cognitive and behavioral
performance.

Synergistic Social Motor Coordination

Social motor coordination is defined as a stable, collective mode of dyadic or
group-level motor behavior that emerges and persists due to informational or
mechanical coupling. Common examples include the synchronized limb or body
movements of individuals walking side-by-side (Harrison & Richardson, 2009;
Van Ulzen, Lamoth, Daffertshofer, Semin, & Beek, 2008), rocking side-by-side
in rocking chairs (e.g., Richardson, Marsh, Isenhower, Goodman, & Schmidt,
2007; Richardson, Garcia, Frank, Gregor, & Marsh, 2012), the coordinated pos-
tural and gestural movements of two or more individuals during a conversational
interaction (e.g., Schmidt, Morr, Fitzpatrick & Richardson, 2012; Schmidt, Nie,
Franco, & Richardson, 2014; Shockley, Santana, & Fowler, 2003), or while
moving a large piece of furniture together (Richardson, Marsh & Baron, 2007).
Such interpersonal coordination is argued to underlie many types of interpersonal
and social activities and, moreover, is associated with behavior in pro-social,
affiliative, and positive ways (e.g., Marsh, Richardson, & Schmidt, 2009; Miles,
Griffiths, Richardson, & Macrae, 2010; Miles, Lumsden, Richardson, & Macrae,
2011; see also Marsh, Chapter 9 in this volume, for an excellent discussion). Of
particular relevance here is that the stable patterns of social motor coordina-
tion that characterize everyday social interaction are by definition the result of
interaction-dominant dynamical processes. That is, the stable patterns of social
motor coordination emerge from the visual, auditory, haptic, and verbal interac-
tions (couplings) that exist between two or more co-acting individuals. Moreover,
these stable states of social motor coordination correspond to higher-order, col-
lective organizations that enslave the behavioral order of the component limb or
112  Eiler, Kallen, and Richardson

body movements of individuals involved in the interaction, including the neural

activity inherent to an individual’s limb and body movements (Anderson et al.,
2012; Kelso, 1995).
Founded on interaction-dominant dynamical processes, stable patterns of
social motor coordination are therefore soft-molded, context-dependent states
of social organization. It is for these reasons that social motor coordination is
also argued to be synergistic (Dale, Fusaroli, Duran, & Richardson, 2013; Marsh,
Richardson, Baron, & Schmidt, 2006; Richardson, Marsh, & Schmidt, 2010).
With regard to human motor coordination more generally, a synergy is a coali-
tion of two or more degrees of freedom that are temporarily bound together to
operate as a single functional system. This temporary assemblage of biomechani-
cal elements are linked in such a way that (i) the number of component variables
that need to be controlled during task performance is significantly reduced, so-
called dimensional compression, and (ii) effective behavioral control is maintained
in the presence of unforeseen environmental perturbations or system noise by
the reciprocal compensation (mutual adaptation) of linked biomechanical elements
(Riley, Richardson, Shockley, & Ramenzoni, 2011; Romero, Kallen, Riley, &
Richardson, 2015). A social or interpersonal synergy is therefore a coalition
of motor degrees of freedom across two or more individuals that result in
(i) dimensional compression and (ii) reciprocal compensation (Riley et al., 2011).
Note, of course, that the synergistic processes of dimensional compression and
reciprocal compensation provide direct evidence of timescale enslavement, in
that synergistic coordinative structures by definition represent slower-timescale
stabilities that constrain and control the faster, moment-to-moment motor
actions of individuals engaged in a particular social motor task.
Although there is now a growing body of literature to support the hypothesis
that social motor coordination is synergistic and the result of IDDS processes (e.g.,
Black, Riley, & McCord, 2007; Dale et al., 2013; Riley et al., 2011, Ramenzoni,
2008), the most compelling evidence has come from a recent study by Romero
and colleagues (2015), in which they investigated the limb coordination that
occurred between two individuals performing an interpersonal targeting task (i.e.,
qualitatively similar to hand shaking). Utilizing an advanced mathematical mod-
eling technique (the uncontrolled manifold technique) to identify the magnitude of
dimensional compression and reciprocal compensation at both the intrapersonal
and interpersonal levels, results demonstrated that behavioral control occurred at
both levels, but that successful task performance was most strongly defined at the
interpersonal level (Romero et al., 2015). Moreover, the synergistic control that
occurred between co-actors emerged rapidly (within the first few trials), ensuring
robust task performance across several hundred trial sessions.
Richardson and colleagues (2015) have also provided strong evidence that
social motor coordination is synergistic. The authors investigated and mod-
eled the task dynamics of an interpersonal collision avoidance targeting task. Of
particular significance was the finding that the movement roles that emerged
 The Emergence of Social Behavior  113

between co-actors were complementary—each actor moved in a different yet

reciprocal manner between targets in order to ensure that task performance was
both robust and energetically efficient. Moreover, once these task roles were
established, they persisted across future task performance. Importantly, these roles
were a self-organized consequence of the shared task goal and trial-by-trial per-
formance outcomes (long-timescale dynamics), rather than a result of the faster
dynamics of individuals’ perceptual-motor organization. Indeed, the persistence
with which participants adopted a specific task role within and across trials was
a consequence of the real-time movement dynamics of each participant quickly
becoming enslaved and collectively controlled by the slow-timescale dynamics
of the complementary state of the interpersonal motor coordination that defined
task success (Eiler et al., 2013).
Beyond human movement, Fusaroli and Tylén (2015) have demonstrated that
synergetic properties of dialog are the best predictors of collective performance
for joint action tasks performed by dyads (see also Fusaroli, Rączaszek-Leonardi, &
Tylén, 2013). There is evidence to suggest that interpersonal synergies also exist
at the level of brain-to-brain coupling, in which brains are reciprocally influ-
ential via motor coordination in a shared environment (Hasson, Ghazanfar,
Galantucci, Garrod, & Keysers, 2012). It therefore seems likely that such syner-
gistic coordination processes can be extended both more locally (cf. Di Paolo &
De Jaegher, 2012) and more globally (cf. Eiler et al., 2013), in that synergistic
control processes exist at all levels of human behavior (neural, perceptual-motor,
interpersonal, social, etc.).
Key to demonstrating that a particular social coordination phenomenon is
synergistic and has emerged from interaction-dominant dynamical processes and
timescale enslavement, is to use methods that quantify the collective state of
the interpersonal or group dynamics that defined the phenomena in question.
For movement tasks that invite highly precise measurements of limb and body
positions over time, measures of relative phase (Kelso, 1995), phase space meth-
ods (Richardson et al., 2015), and cross wavelet spectral analysis (Issartel, Marin,
Gaillot, Bardainne, & Cadopi, 2006; Walton, Richardson, Langland-Hassan &
Chemero 2015) are particularly useful to compare coordination across multiple
timescales. For groups larger than two, cluster phase measures have been devel-
oped for measuring group synchrony (Frank & Richardson, 2010; Richardson
et  al., 2012). This allows for comparisons of short- and long-timescale coor-
dination, in which an IDDS perspective would predict that stable patterns of
coordination emerge over longer timescales to constrain the local movements of
the individuals who make up the group. Recurrence and cross-recurrence quan-
tification analyses (Webber & Zbilut, 2007) are also powerful tools for identifying
the underlying dynamics of interpersonal interaction, and can be used to identify
leader–follower relationships or other asymmetries (i.e., coupling strength), with
regard to both categorical and continuous movement, and behavioral data (Coco &
Dale, 2014; Richardson et al., 2015), and have been successfully employed for
114  Eiler, Kallen, and Richardson

the analysis of other synergistic behavioral streams such as eye movement coor-
dination (Richardson & Dale, 2005; Dale, Warlaumont, & Richardson, 2011),
postural coordination and control (Shockley et al., 2003; Riley et al., 2011), and
joint attention (Richardson, Dale, & Kirkham, 2007).
Finally, it is worth noting that synergistic social coordination is not modal-­
specific, but rather involves synergistic processes at the neural, physiological,
motor, perceptual, and linguistic levels simultaneously. To date, however, very lit-
tle research has been directed toward this cross-modal exploration (for exceptions
see Dale et al., 2013; Gorman et al., 2016). Consistent with an IDDS perspective,
therefore, researchers should be encouraged to devote more attention to uncov-
ering how social synergies are established, maintained, and organized across time.
It is likely that such research will not only lead to a better understanding of the
interactive, self-organized processes that shape and constrain social interaction,
but how perceptual, motor and cognitive aspects of human behavior are mutu-
ally and intrinsically related (i.e., provide greater insights into the embodied and
embedded nature of human and social perception, action and cognition).

Social Networks
Networks are, by definition, relational structures. Boccaletti and colleagues
(Boccaletti, Latora, Moreno, Chavez, & Hwang, 2006) argue that the central
issue for understanding any complex network is to graphically denote global
system properties by modeling components, or nodes, as the dynamical units,
and the links between them as representative of their interactions. Understood
within the IDDS framework, nodes are therefore made interdependent by vari-
ous types of connections, or ties (edges), that emerge over time to define the
structural order of the network (and the behavior of the nodes themselves) in a
circularly causal manner. Thus, it is the structural topology of the network that
emerges from node linkages, which then feeds back to enslave the links between
components in ways that give rise to emergent social phenomena (cf. Westaby,
Pfaff, & Redding, 2014).
Although there are many examples of networks of differing composition
(cells, the Internet, neurons in the brain, etc.), it is becoming increasingly clear
that similar interaction-dominant dynamical principles underlie the formation of
the complex structures that emerge in these different contexts (Albert & Barabási,
2002). It is also clear that networks exist at multiple temporal and physical scales
in which they are both embedded within, and interact with, other networks.
Accordingly, two complementary approaches have emerged with regard to
research on human networks, which typically relate network structure either
across disparate networks or across disparate processes. First, researchers have
sought to relate network-level properties across distinct networks (e.g., human
and orca whale societal structure; Hill, Bentley, & Dunbar, 2008) or seemingly
unrelated phenomena (e.g., geographic movement patterns and friendship ties;
 The Emergence of Social Behavior  115

Cho, Myers, & Leskovec, 2011). This approach underscores the universality
of timescale enslavement, in that each network system entails slower-timescale
structures that emerge to constrain faster-timescale observable behavior. For
example, with regard to social behavior, the network principle of homophily
(that similarity underlies social connection at the network level) structures
interactions as diverse as marriage, friendship, work, support, and information
exchange, and seems to constrain social connections in ways that have drastic
impact on information propagation, attitudes, and intergroup relations (e.g.,
McPherson, Smith-Lovin, & Cook, 2001). Indeed, this underlying dynamical
principle generates social structures with geographic, socioeconomic, racial/
ethnic, and attitudinal similarity that are irreducible to individual agents.
The underlying dynamical process of timescale enslavement operates to struc-
ture social networks in ways that demonstrate this principle across networks of
differing composition. Consider healthcare networks, where slow-timescale
emergent social infrastructure (the network structure of referral contacts for
physical and mental health) has been related to accessibility for Latino immigrants
in a non-traditional migration area and used to make targeted recommenda-
tions for reducing health disparities (Eiler, Bologna, Vaughn, & Jacquez, 2017).
Here, referrals are passed between similar healthcare providers such that changes
in policy or service provision (e.g., the absence of bilingual healthcare profes-
sionals) at individual locations constrains the entire network and subsequently
leads to inadequate healthcare service delivery for Latinos. Similar enslavement
effects have been observed with regard to smoking cessation, such that smokers
tend to quit when their immediate social group quits, and this quitting behav-
ior tends to spread through close individual ties initially and sparse connections
between groups of smokers over time (Christakis & Fowler, 2008). Some social
networks even tend to co-evolve in the case of organizational culture and indi-
vidual knowledge transfer (Lin & Desouza, 2010), demonstrating how networks
are both irreducible to individual dynamics and mutually structured by asym-
metries across temporal scales.
The second approach to understanding networks considers network prop-
erties from an IDDS perspective more directly. In this way, researchers tend
to explore how network structure at slow timescales relates to highly desirable
properties at faster timescales across many social domains: how slower-timescale
network behavior generates robust and highly effective faster-timescale behavior.
In particular, how many networks, such as the nervous system of the C. elegans,
collaborations between film actors, or the power grid in the United States (Watts &
Strogatz, 1998), are topographically structured in a way that is neither com-
pletely random (random ties between components), nor completely regular (ties
between components that follow a deterministic rule). In short, these and many
other networks are what are called small world networks, which are character-
istic of self-organized and self-sustaining systems that recruit or remove nodes
or connections between nodes during development (Watts & Strogatz, 1998).
116  Eiler, Kallen, and Richardson

The significance of these small world networks is that they display efficient infor-
mation propagation (i.e., attitudes, social norms), enhanced computational power
(i.e., wisdom of crowds), and synchronization (i.e., propensity to coordinate)
compared to other network types (cf. Watts & Strogatz, 1998). Small world
networks comprised of humans are also self-regulating, whereby agents draw
on various information sources distributed across the network such that a com-
mon and flexible understanding of governance emerges (Folke, Hahn, Olsson, &
Norberg, 2005). The self-regulating and information propagation properties
of small world networks have also been identified in models that capture the
dynamics of disease spread. For instance, Eubank and colleagues (2004) dem-
onstrated that targeted vaccination in specific network hub locations, combined
with early detection, can lead to herd immunity and containment in the case of
an outbreak. Here, the local interactions among infected individuals give rise to
a pattern of disease spread that then constrains the network in such a way as to
provide an opportunity for targeted intervention in a circularly causal manner.
Overall, network research in psychological science is consistent with an IDDS
approach, and can be used to demonstrate timescale enslavement and the self-
organized nature of social behavior. Furthermore, Westaby and colleagues (2014)
have developed a dynamic network perspective in which social networks influ-
ence processes at interpersonal, faster timescales (i.e., goals). While some research
exists that has modeled social network evolution and outcome (Snijders, Van de
Bunt, & Steglich, 2010; Westaby & Shon, Chapter 11 in this volume), another
interesting avenue for future research is the measurement of social networks
over time. Such research would provide key insights about how to model and
understand the dynamics of social network formation and change, and the role
that such formation and change plays in defining social engagement or inter-
action potentials. Teasing apart how asymmetries between interacting change
processes (e.g., network topology and individual dynamics) permit stable behav-
ior to emerge would also provide an opportunity for demonstrating how the
behavior of individuals both emerges from and imparts structure on social and
cultural processes (cf. Kitayama & Uskul, 2011). For example, women’s under-
representation in STEM (science, technology, engineering, and mathematics) at
the faculty level might be conceptualized as emerging from networks in which
cultural factors (i.e., departmental network structure) lead to isolation (i.e., fewer
network ties for women faculty), which, in turn, perpetuate isolation and may
limit opportunities for advancement or persistence.

Fractal Scaling of Human Performance

Researchers in psychology typically collapse multiple measurements into sum-
mary variables (i.e., mean, standard deviation) and assume that the variance
observed across multiple measurements is random and normally distributed.
For dynamic behavior, however, this is not typically the case, with behavioral
 The Emergence of Social Behavior  117

patterns over time often structured in a self-similar, fractal or scale-free man-

ner (e.g., Van Orden, Holden, & Turvey, 2003; Van Orden et  al., 2011, see
Figure 6.2). In simple terms, this means that the temporal variability observed in
human behavior and performance is correlated over time, with the structure of
smaller-scale fluctuations being statistically similar to the structure of larger-scale
fluctuations. Analogous to geometric fractals, this nested statistical structure is
characterized by an inverse relationship between the power and the frequency
of the observed variability in performance data over time. More specifically, the
relationship between the power of a specific change or fluctuation in a behavioral
measurement and how often that specific change in measurement is observed
(the frequency, f, with which a behavioral measurement is observed) follows an
inverse power-law function. Often referred to as 1/f or “pink” noise due to this
power–frequency relationship, such behavioral variation is the result of recip-
rocally defined inter-component interactions that propagate perturbations that
occur at one level of a system across all other system levels (Wang, Kádár, Jung, &
Showalter, 1999; Bak, Tang, & Wiesenfeld, 1987, 1988); hence the self-similarity
of behavioral variability across small and large scales of analysis.
With respect to our discussion of the IDDS approach to social behavior, what
is important here is that fractal behavior is a clear sign of nonlinearity, meta-­
stability and self-organized criticality (Van Orden, Holden, & Turvey, 2003).
Thus, 1/f scaling has become a benchmark signature of IDDS (Eiler et al., 2013;

FIGURE 6.2   xamples of simulated time series data (left column) and detrended
E
fluctuation analysis plots and Hurst (H) values (right column) for
brown, pink, and white signals.
118  Eiler, Kallen, and Richardson

Van Orden et al., 2003; Ihlen & Vereijken, 2010; Bak, 1996). It is for this reason
that the presence of fractal or pink noise (1/f scaling with an exponent ≈ 1) has
been argued to provide evidence that mind and body are intrinsically linked, and
that human behavior emerges from self-organized, interaction-dominant dynam-
ical processes (Van Orden et  al., 2003). Indeed, the fractal nature of human
behavior and performance variability over time has been studied quite exten-
sively over the last few decades (e.g., Kello et al., 2007; Kello et al., 2010; West,
2010; Dixon, Holden, Mirman, & Stephen, 2012), with this research consist-
ently demonstrating how stable human neural, perceptual-motor, and cognitive
behavior result from interaction-dominant dynamical processes.
Of particular significance here is the research that has demonstrated that time-
scale enslavement typically manifests, fractal, 1/f scaling. For example, functional
magnetic resonance imaging studies have demonstrated endogenous activity
(characterized by a fractal, 1/f structure) to be dominated by slow-timescale
dynamics, even in the absence of other experimental effects (e.g., Bullmore et al.,
2001). Perhaps the most compelling evidence comes from a wealth of research
that has consistently demonstrated how human cognitive performance is charac-
terized by fractal performance variability (e.g., Gilden, 2001; Ihlen & Vereijken,
2010; Kello et al., 2007; Van Orden et al., 2005), with this type of long-range
correlation being due to distributed neural processes that interact along multiple
scales, rather than due to simple feedback loops or other mechanisms typically
associated with modular approaches to cognition (Chen, Ding, & Kelso, 2001).
Beyond intrapersonal behavior and cognition, interpersonal and group behav-
iors have also been demonstrated to display fractal structure. For instance, the
collective reaction time variability of co-actors performing a Simon task is fractal
rather than random in nature (Malone, Castillo, Kloos, Holden, & Richardson,
2014). The fractal scaling of co-actors’ behavior also tends to become matched
over the course of an interaction (Marmelat & Delignières, 2012). Such complexity
matching appears to optimize information flow between coupled systems, thereby
enhancing the overall quality and flexibility of the social exchange. This type of
complex entrainment between co-acting individuals has also been observed in ver-
bal and conversational interaction (Abney, Paxton, Dale, & Kello, 2014; Kello,
Anderson, Holden, & Van Orden, 2008), interpersonal movement coordination
(Marmelat & Delignières, 2012; Fine, Likens, Amazeen, & Amazeen, 2015), neural
networks (Mafahim, Lambert, Zare, & Grigolini, 2015), and during anticipatory
control (Washburn, Kallen, Coey, Shockley, & Richardson, 2015). Furthermore,
Rigoli and colleagues (Rigoli, Holman, Spivey, & Kello, 2014) have argued that
complexity matching formalizes multi-scale coordination as fundamental to both
individual behavior and interaction more generally. This idea is consistent with
the general IDDS approach to social behavior, in that group processes enslave
individual processes in order to enact socially and functionally relevant behavior.
Moreover, complexity matching may increase informational coupling across nervous
systems that support shared goal states (cf. Mafahim et al., 2015).
 The Emergence of Social Behavior  119

Fundamental social processes also display fractal structure. For instance, indi-
vidual social processes such as self-esteem exhibit fractal scaling (Delignières,
Fortes, & Ninot, 2004) indicating that our global view of ourselves is dynamic
and only critically stable, self-sustained via interaction-dominant dynamical pro-
cesses. Changes in collective and synergistic emotional arousal and valence are
also associated with differential fractal structure (Tadić, Gligorijević, Mitrović, &
Šuvakov, 2013). Humans also tend to form social groups in discrete scale-invariant
ways, with group sizes that mimic herding behavior in financial markets (Zhou,
Sornette, Hill, & Dunbar, 2005). Moreover, robust (small world) social networks
in general possess a fractal topology and demonstrate scale-invariant statistical
structure over time (e.g., Hamilton, Milne, Walker, Burger, & Brown, 2007).
Indeed, fractal self-similarity pervades social processes at all levels of individual,
small group and cultural behavior, leading some to argue that fractal social systems
are not only a fundamental class of social organization (De Florio, Bakhouya,
Coronato, & Di Marzo, 2013), but typify the self-organized and interaction-
dominant nature of social behavior (Bradbury & Vehrencamp, 2014).

Conclusion
Here we have detailed a theory of social behavioral organization founded on a
key tenet of interaction-dominant dynamical systems, namely timescale enslave-
ment. Our goal was to explicate how IDDS theory and timescale enslavement
fit within the field of social psychology more broadly. As stated at the beginning
of this chapter, we do not view these principles of behavioral organization as
controversial or contrary to an intuitive understanding of how social behavior
emerges over time. Similar dynamical ideas, such as conceptualizing life spaces
as vector fields of attracting and repelling social forces (Lewin, Lippitt, & White,
1939) or coordinated action as a means for social inclusion (Asch, 1952), have
been theorized since the middle of the last century. Moreover, stating that inter-
actions between individuals are fundamental to the structure of human social
behavior is theoretically inherent to the field of social psychology.
Yet, if the principles of interaction dominance and timescale enslavement
are taken seriously, then how we understand and investigate social behavior
becomes somewhat different from that traditionally hypothesized within the field
of social psychology. Indeed, to take interaction dominance seriously means that
one should be committed to the idea of circular causality and self-referential,
self-sustainable processes of emergence—that the faster-timescale interactions of
individuals give rise to the historical constraints and macroscopic stabilities of the
social systems that those same individuals then find themselves embedded within.
This implies that the goal for social psychology should be to uncover how societal
and cultural processes are self-organized, emerge from, and enslave the behavioral
dynamics of the individuals embedded within those slower-timescale societal and
cultural processes.
120  Eiler, Kallen, and Richardson

To some extent, this requires a complete reconsideration of how we define

socially situated behavior. This is to say that social behavior is a contextually con-
strained property of a complex and dynamic agent–environment system, rather
than a property inherent to or possessed by individuals. That is, to be social is not
to be in the mere presence of another, but in fact to be defined by others and
to define others through ourselves. Moreover, social context is not simply the
addition of individuals into an objective reality, but an objective reality defined
by the social interactions that individuals are generatively embedded within.
Importantly, this grounds human behavior in principles that pervade science at
the broadest levels. For social psychologists, this means sociality is fundamental
to behavior—an idea that has long eluded traditional cognitive scientists that
typically subjugate our behavior as the mere product of brain signaling. Most
importantly, this approach allows us to address deep questions in social psychol-
ogy from a strong empirical position in which the answers that emerge will
demonstrate how the stable micro- and macroscopic patterns of human social
behavior emerge from the same IDDS principles that shape the stable micro- and
macroscopic patterns of behavior of other complex biological systems.

Notes
1 The term “soft-molded” refers to stable yet temporary behavioral patterns or functional
organization. Such behavioral patterns or functional organization are non-fixed, highly
flexible and can be created, re-organized, and destroyed in response to changes in envi-
ronmental context.
2 In this chapter, we have chosen to use the term “timescale enslavement” to refer to the
circular causality principle in which slower-timescale processes or dynamic stabilities
enslave faster-timescale processes or dynamic stabilities. It is worth nothing that this
principle is referred to as the slaving principle in synergetics (e.g., Haken, 1983, 2004).

References
Abney, D. H., Paxton, A., Dale, R., & Kello, C. T. (2014). Complexity matching in
dyadic conversation. Journal of Experimental Psychology: General, 143(6), 2304–2315.
doi:10.1037/xge0000021
Albert, R., & Barabási, A.-L. (2002). Statistical mechanics of complex networks. Reviews
of Modern Physics, 74(1), 47–97. doi:10.1103/RevModPhys.74.47
Anderson, M. L., Richardson, M. J., & Chemero, A. (2012). Eroding the boundaries
of cognition: Implications of embodiment. Topics in Cognitive Science, 4(4), 717–730.
doi:10.1111/j.1756-8765.2012.01211.x
Asch, S. (1952). Social psychology. New York: Prentice Hall.
Bak, P. (1996). How nature works: The science of self-organized criticality. New York: Springer-
Verlag.
Bak, P., Tang, C., & Wiesenfeld, K. (1987). Self-organized criticality: An explanation of
1/f noise. Physical Review Letters, 59(4), 381–384. doi:10.1103/PhysRevLett.59.381
Bak, P., Tang, C., & Wiesenfeld, K. (1988). Self-organized criticality. Physical Review A,
38(1), 364–374. doi:10.1103/PhysRevA.38.364
 The Emergence of Social Behavior  121

Black, D. P., Riley, M. A., & McCord, C. K. (2007). Synergies in intra- and interpersonal
interlimb rhythmic coordination. Motor Control, 11(4), 348–373.
Boccaletti, S., Latora, V., Moreno, Y., Chavez, M., & Hwang, D.-U. (2006). Complex
networks: Structure and dynamics. Physics Reports, 424(4–5), 175–308. doi:10.1016/j.
physrep.2005.10.009
Bradbury, J. W., & Vehrencamp, S. L. (2014). Complexity and behavioral ecology.
Behavioral Ecology, 25(3), 435–442. doi:10.1093/beheco/aru014
Bruineberg, J., & Rietveld, E. (2014). Self-organization, free energy minimization,
and optimal grip on a field of affordances. Frontiers in Human Neuroscience, 8, 599.
doi:10.3389/fnhum.2014.00599
Bullmore, E., Long, C., Suckling, J., Fadili, J., Calvert, G., Zelaya, F., Carpenter, T. A., &
Brammer, M. (2001). Colored noise and computational inference in neurophysiologi-
cal (fMRI) time series analysis: Resampling methods in time and wavelet domains.
Human Brain Mapping, 12(2), 61–78. doi:10.1002/1097-0193(200102)12:2<61::AID-
HBM1004>3.0.CO;2-W
Chalmers, D. J. (2006). Strong and weak emergence. In P. Clayton & P. Davies (Eds.),
The re-emergence of emergence. Oxford: Oxford University Press.
Chen, Y, Ding, M., & Kelso, J. A. S. (2001). Origins of timing errors in human sensorimotor
coordination. Journal of Motor Behavior, 33(1), 3–8. doi:10.1080/00222890109601897
Cho, E., Myers, S. A., & Leskovec, J. (2011). Friendship and mobility: User movement
in location-based social networks. In Proceedings of the ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, San Diego, CA, August 21–24,
2011 (pp. 1082–1090). New York: ACM. doi:10.1145/2020408.2020579
Christakis, N. A., & Fowler, J. H. (2008). The collective dynamics of smoking in a large
social network. New England Journal of Medicine, 358(21), 2249–2258. doi:10.1056/
NEJMsa0706154
Coco, M. I., & Dale, R. (2014). Cross-recurrence quantification analysis of categorical and
continuous time series: An R package. Frontiers in Psychology, 5, 1–14. doi:10.3389/
fpsyg.2014.00510
Dale, R., Fusaroli, R., Duran, N. D., & Richardson, D. C. (2013). The self-organization
of human interaction. In B. H. Ross (Ed.), The psychology of learning and motivation.
New York: Elsevier.
Dale, R., Warlaumont, A. S., & Richardson, D. C. (2011). Nominal cross recurrence
as a generalized lag sequential analysis for behavioral streams. International Journal of
Bifurcation and Chaos, 21, 1153–1161. doi:10.1142/S0218127411028970
De Florio, V., Bakhouya, M., Coronato, A., & Di Marzo, G. (2013). Models and
concepts for socio-technical complex systems: Towards fractal social organizations.
Systems Research and Behavioral Science, 30(6), 750–772. doi:10.1002/sres.2242
Delignières, D., Fortes, M., & Ninot, G. (2004). The fractal dynamics of self-esteem and
the physical self. Nonlinear Dynamics, Psychology, and Life Sciences, 8(4), 479–510.
Di Paolo, E., & De Jaegher, H. (2012). The interactive brain hypothesis. Frontiers in
Human Neuroscience, 6(1), 1–16. doi:10.3389/fnhum.2012.00163
Dixon, J. A., Holden, J. G., Mirman, D., & Stephen, D. G. (2012). Multifractal dynamics
in the emergence of cognitive structure. Topics in Cognitive Science, 4(1), 51–62. doi:10.
1111/j.1756-8765.2011.01162.x
Eiler, B.A., Bologna, D., Jacquez, F., & Vaughn, L. (2017). A social network approach
to understanding community partnerships in a non-traditional destination for Latinos.
Journal of Community Psychology, 45(2), 178–192. doi:10.1002/jcop.21841
122  Eiler, Kallen, and Richardson

Eiler, B. A., Kallen, R. W., Harrison, S. J., & Richardson, M. J. (2013). Origins of order
in joint activity and social behavior. Ecological Psychology, 25(3), 316–326. doi:10.1080/
10407413.2013.810107
Elmslie, B., & Sedo, S. (1996). Discrimination, social psychology, and hysteresis in
labor markets. Journal of Economic Psychology, 17(4), 465–478. doi:10.1016/0167-
4870(96)00021-9
Eubank, S., Guclu, H., Kumar, V. S. A., Marathe, M. V., Srinivasan, A., Toroczkai, Z., &
Wang, N. (2004). Modelling disease outbreaks in realistic urban social networks.
Nature, 429(6988), 180–184. doi:10.1038/nature02541
Fine, J. M., Likens, A. D., Amazeen, E. L., & Amazeen, P. G. (2015). Emergent com-
plexity matching in interpersonal coordination: Local dynamics and global variability.
Journal of Experimental Psychology: Human Perception and Performance, 41(3), 723–737.
doi:10.1037/xhp0000046
Folke, C., Hahn, T., Olsson, P., & Norberg, J. (2005). Adaptive governance of social-­
ecological systems. Annual Review of Environment and Resources, 30, 441–473. doi:10.
1146/annurev.energy.30.050504.144511
Frank, T. D., & Richardson, M. J. (2010). On a test statistic for the Kuramoto order
parameter of synchronization: An illustration for group synchronization during rocking
chairs. Physica D, 239(23–24), 2084–2092. doi:10.1016/j.physd.2010.07.015
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews
Neuroscience, 11(2), 127–138. doi:10.1038/nrn2787
Fusaroli, R., Rączaszek-Leonardi, J., & Tylén, K. (2013). Dialog as interpersonal synergy.
New Ideas in Psychology, 32(1), 147–157. doi:10.1016/j.newideapsych.2013.03.005
Fusaroli, R., & Tylén, K. (2015). Investigating conversational dynamics: Interactive align-
ment, interpersonal synergy, and collective task performance. Cognitive Science, 40(1),
145–171. doi:10.1111/cogs.12251
Gernigon, C., Briki, W., & Eykens, K. (2010). The dynamics of psychological momen-
tum in sport: The role of ongoing history of performance patterns. Journal of Sport and
Exercise Psychology, 32(3), 377–400.
Gilden, D. L. (2001). Cognitive emissions of 1/f noise. Psychological Review, 108(1), 33–56.
doi:10.1037//0033-295X.108.1.33
Gorman, J. C., Martin, M. J., Dunbar, T. A., Stevens, R. H., Galloway, T. L.,
Amazeen, P. G., & Likens, A. D. (2016). Cross-level effects between neurophysi-
ology and communication during team training. Human Factors, 58(1), 181–199.
doi:10.1177/0018720815602575
Haken, H. (1983). Synergetics, an introduction: Nonequilibrium phase transitions and self-­
organization in physics, chemistry, and biology. New York: Academic Press.

Haken, H. (2004). Synergetics: Introduction and advanced topics. Berlin: Springer-Verlag.
doi:10.1007/978-3-662-10184-1
Haken, H., & Tschacher, W. (2010). A theoretical model of intentionality with an appli-
cation to neural dynamics. Mind and Matter, 8(1), 7–18.
Hamilton, M. J., Milne, B. T., Walker, R. S., Burger, O., & Brown, J. H. (2007). The
complex structure of hunter-gatherer social networks. Proceedings of the Royal Society B:
Biological Sciences, 274(1622), 2195–2202. doi:10.1098/rspb.2007.0564
Harrison, S. J., & Richardson, M. J. (2009). Horsing around: Spontaneous four-legged
coordination. Journal of Motor Behavior, 41(6), 519–524. doi:10.3200/35-08-014
Hasson, U., Ghazanfar, A. A., Galantucci, B., Garrod, S., & Keysers, C. (2012). Brain-
to-brain coupling: A mechanism for creating and sharing a social world. Trends in
Cognitive Sciences, 16(2), 114–121. doi:10.1016/j.tics.2011.12.007
 The Emergence of Social Behavior  123

Hill, R. A., Bentley, R. A., & Dunbar, R. I. M. (2008). Network scaling reveals consist-
ent fractal pattern in hierarchical mammalian societies. Biology Letters, 4(6), 748–751.
doi:10.1098/rsbl.2008.0393
Ihlen, E. A., & Vereijken, B. (2010). Interaction-dominant dynamics in human cogni-
tion: Beyond 1/f α scaling. Journal of Experimental Psychology: General, 139(3), 436–463.
doi:10.1037/a0019098
Isenhower, R. W., Richardson, M. J., Carello, C., Baron, R. M., & Marsh, K. L. (2010).
Affording cooperation: Embodied constraints, dynamics, and action-scaled invari-
ance in joint lifting. Psychonomic Bulletin & Review, 17(3), 342–347. doi:10.3758/
PBR.17.3.342
Issartel, J., Marin, L., Gaillot, P., Bardainne, T., & Cadopi, M. (2006). A practical guide
to time-frequency analysis in the study of human motor behavior: The contribu-
tion of wavelet transform. Journal of Motor Behavior, 38(2), 139–159. doi:10.3200/
JMBR.38.2.139-159
Jordan, J. S. (2008). Wild agency: Nested intentionalities in cognitive neuroscience
and archaeology. Philosophical Transactions of the Royal Society B: Biological Sciences,
363(1499), 1981–1991. doi:10.1098/rstb.2008.0009
Jordan, J. S., Bai, J., Cialdella, V., & Schloesser, D. (2015). Foregrounding the back-
ground. In E. Dzhafarov, J. S. Jordan, R. Zhang, & V. Cervantes (Eds.), Contextuality
from quantum physics to psychology. Hackensack, NJ: World Scientific Publishing.
Jordan, J. S., & Heidenreich, B. A. (2010). The intentional nature of self-sustaining
systems. Mind and Matter, 8(1), 45–62.
Karl, F. (2012). A free energy principle for biological systems. Entropy, 14(11), 2100–2121.
doi:10.3390/e14112100
Kello, C. T., Anderson, G. G., Holden, J. G., & Van Orden, G. C. (2008). The pervasive-
ness of 1/f scaling in speech reflects the metastable basis of cognition. Cognitive Science,
32(7), 1217–1231. doi:10.1080/03640210801944898
Kello, C. T., Beltz, B. C., Holden, J. G., & Van Orden, G. C. (2007). The emergent
coordination of cognitive function. Journal of Experimental Psychology: General, 136(4),
551–568. doi:10.1037/0096-3445.136.4.551
Kello, C. T., Brown, G. D. A., Ferrer-i-Cancho, R., Holden, J. G., Linkenkaer-Hansen,
K., Rhodes, T., & Van Orden, G. C. (2010). Scaling laws in cognitive sciences. Trends
in Cognitive Sciences, 14(5), 223–232. doi:10.1016/j.tics.2010.02.005
Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge,
MA: MIT Press.
Kelso, J. A. S. (2012). Multistability and metastability: Understanding dynamic coordi-
nation in the brain. Philosophical Transactions of the Royal Society B: Biological Sciences,
367(1591), 906–918. doi:10.1098/rstb.2011.0351
Kitayama, S., & Uskul, A. K. (2011). Culture, mind, and the brain: Current evidence and
future directions. Annual Review of Psychology, 62(10), 419–449. doi:10.1146/annurev-
psych-120709-145357
Kondepudi, D. K., & Gao, M.-J. (1987). Passages through the critical point and the
process of state selection in symmetry-breaking transitions. Physical Review A, 35(1),
340–348. doi:10.1103/PhysRevA.35.340
Lewin, K., Lippitt, R., & White, R. K. (1939). Patterns of aggressive behavior in experi-
mentally created “social climates.” Journal of Social Psychology, 10, 271–299.
Lin, Y. & Desouza, K. C. (2010). Co-evolution of organizational network and individual
behavior: An agent-based model of interpersonal knowledge transfer. ICIS Proceedings,
153. Retrieved from http://aisel.aisnet.org/icis2010_submissions/153
124  Eiler, Kallen, and Richardson

Mafahim, J. U., Lambert, D., Zare, M., & Grigolini, P. (2015). Complexity matching
in neural networks. New Journal of Physics, 17(015003), 1–17. doi:10.1088/1367-
2630/17/1/015003
Malone, M., Castillo, R. D., Kloos, H., Holden, J. G., & Richardson, M. J. (2014).
Dynamic structure of joint-action stimulus–response activity. PLoS One, 9(2), e89032.
doi:10.1371/journal.pone.0089032
Marmelat, V., & Delignières, D. (2012). Strong anticipation: Complexity matching
in interpersonal coordination. Experimental Brain Research, 222(1–2), 137–148. doi:10.
1007/s00221-012-3202-9
Marsh, K. L., Richardson, M. J., Baron, R. M., & Schmidt, R. C. (2006). Contrasting
approaches to perceiving and acting with others. Ecological Psychology, 18(1), 1–38.
doi:10.1207/s15326969eco1801_1
Marsh, K. L., Richardson, M. J., & Schmidt, R. C. (2009). Social connection through
joint action and interpersonal coordination. Topics in Cognitive Science, 1, 320–339.
doi:10.1111/j.1756-8765.2009.01022.x
Masataka, N. (1993). Effects of experience with live insects on the development of
fear of snakes in squirrel monkeys, Saimiri sciureus. Animal Behavior, 46(4), 741–746.
doi:10.1006/anbe.1993.1251
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily
in social networks. Annual Review of Sociology, 27, 415–444. doi:10.1146/annurev.
soc.27.1.415
Miles, L. K., Griffiths, J. L., Richardson, M. J., & Macrae, C. N. (2010). Too late to
coordinate: Contextual influences on behavioral synchrony. European Journal of Social
Psychology, 40(1), 52–60. doi:10.1002/ejsp.721
Miles, L. K., Lumsden, J., Richardson, M. J., & Macrae, C. N. (2011). Do birds of a
feather move together? Group membership and behavioral synchrony. Experimental
Brain Research, 211(3–4), 495–503. doi:10.1007/s00221-011-2641-z
Ramenzoni, V. C. (2008). Effects of joint task performance on interpersonal postural coordination
Doctoral dissertation. Retrieved from ProQuest Information & Learning (AAI3323881).
Richardson, D. C., & Dale, R. (2005). Looking to understand: The coupling between
speakers’ and listeners’ eye movements and its relationship to discourse comprehen-
sion. Cognitive Science, 29(6), 1045–1060. doi:10.1207/s15516709cog0000_29
Richardson, D. C., Dale, R., & Kirkham, N. Z. (2007). The art of conversation is coor-
dination: Common ground and the coupling of eye movements during dialogue.
Psychological Science, 18(5), 407–413. doi:10.1111/j.1467-9280.2007.01914.x
Richardson, M. J., Dale, R., & Marsh, K. L. (2014). Complex dynamical systems in
social and personality psychology: Theory, modeling and analysis. In H. T. Reis &
C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (2nd ed.).
Cambridge: Cambridge University Press.
Richardson, M. J., Garcia, R. L., Frank, T. D., Gergor, M., & Marsh, K. L. (2012).
Measuring group synchrony: A cluster-phase method for analyzing multivariate
movement time-series. Frontiers in Physiology, 3, 1–10. doi:10.3389/fphys.2012.00405
Richardson, M. J., Harrison, S. J., Kallen, R. W., Walton, A., Eiler, B. A., Saltzman, E., &
Schmidt, R. C. (2015). Self-organized complementary joint action: Behavioral
dynamics of an interpersonal collision-avoidance task. Journal of Experimental Psychology:
Human Perception and Performance, 41(3), 665–679. doi:10.1037/xhp0000041
Richardson, M. J., Marsh, K. L., & Baron, R. M. (2007). Judging and actualizing intraper-
sonal and interpersonal affordances. Journal of Experimental Psychology: Human Perception
and Performance, 33(4), 845–859. doi:10.1037/0096-1523.33.4.845
 The Emergence of Social Behavior  125

Richardson, M. J., Marsh, K. L., Isenhower, R. W., Goodman, J. R. L., & Schmidt, R. C.
(2007). Rocking together: Dynamics of intentional and unintentional interpersonal
coordination. Human Movement Science, 26, 867–891, doi:10.1016/j.humov.2007.07.002
Richardson, M. J., Marsh, K. L., & Schmidt, R. C. (2010). Challenging egocentric
notions of perceiving, acting, and knowing. In L. F. Barrett, B. Mesquita, & E. Smith
(Eds.), The mind in context (pp. 307–333). New York: Guilford Press.
Rigoli, L. M., Holman, D., Spivey, M. J., & Kello, C. T. (2014). Spectral convergence
in tapping and physiological fluctuations: Coupling and independence of 1/f noise in
the central and autonomic nervous systems. Frontiers in Human Neuroscience, 8, 1–10.
doi:10.3389/fnhum.2014.00713
Riley, M. A., Richardson, M. J., Shockley, K., & Ramenzoni, V. C. (2011). Interpersonal
synergies. Frontiers in Psychology, 2(38), 1–7. doi:10.3389/fpsyg.2011.00038
Romero, V., Kallen, R. W., Riley, M. A., & Richardson, M. J. (2015). Can discrete joint
action be synergistic? Studying the stabilization of interpersonal hand coordination.
Journal of Experimental Psychology: Human Perception and Performance, 41(5), 1223–1235.
doi:10.1037/xhp0000083
Schmidt, R. C., Morr, S., Fitzpatrick, P., & Richardson, M. J. (2012). Measuring the
dynamics of interactional synchrony. Journal of Nonverbal Behavior, 36(4), 263–279.
doi:10.1007/s10919-012-0138-5
Schmidt, R. C., Nie, L., Franco, A., & Richardson, M. J. (2014). Bodily synchroni-
zation underlying joke telling. Frontiers in Human Neuroscience, 8, 633. doi:10.3389/
fnhum.2014.00633
Shockley, K., Santana, M., & Fowler, C. A. (2003). Mutual interpersonal postural con-
straints are involved in cooperative conversation. Journal of Experimental Psychology:
Human Perception and Performance, 29(2), 326–332. doi:10.1037//0096-1523.29.2.326
Snijders, T. A. B., Van de Bunt, G. G., & Steglich, C. E. G. (2010). Introduction to
stochastic actor-based models for network dynamics. Social Networks, 32(1), 44–60.
doi:10.1016/j.socnet.2009.02.004
Tadić, B., Gligorijević, V., Mitrović, M., & Šuvakov, M. (2013). Co-evolutionary mech-
anisms of emotional bursts in online social dynamics and networks. Entropy, 15(12),
5084–5120. doi:10.3390/e15125084
Van Orden, G. C. (2010). Voluntary performance. Medicina (Kaunas), 46(9), 581–594.
Van Orden, G. C., Holden, J. G., & Turvey, M. T. (2003). Self-organization of cog-
nitive performance. Journal of Experimental Psychology: General, 132(3), 331–350.
doi:10.1037/0096-3445.132.3.331
Van Orden, G. C., Holden, J. G., & Turvey, M. T. (2005). Human cognition and 1/f
scaling. Journal of Experimental Psychology: General, 134(1), 117–123. doi:10.1037/0096-
3445.134.1.117
Van Orden, G. C., Hollis, G., & Wallot, S. (2012). The blue-collar brain. Frontiers in
Psychology, 3, 1–12. doi:10.3389/fphys.2012.00207
Van Orden, G. C., Kloos, H., & Wallot, S. (2011). Living in the pink. Intentionality, well-
being, and complexity. In C. Hooker (Ed.), Philosophy of complex systems (pp. 629–672).
Oxford: Elsevier B.V.
Van Ulzen, N. R., Lamoth, C. J. C., Daffertshofer, A., Semin, G. R., & Beek, P. J. (2008).
Characteristics of instructed and uninstructed interpersonal coordination while walk-
ing side-by-side. Neuroscience Letters, 432(2), 88–93. doi:10.1016/j.neulet.2007.11.070
Walton, A. E., Richardson, M. J., Langland-Hassan, P., & Chemero, A. (2015).
Improvisation and the self-organization of multiple musical bodies. Frontiers in
Psychology, 6, 313. doi:10.3389/fpsyg.2015.00313
126  Eiler, Kallen, and Richardson

Wang, J., Kádár, S., Jung, P., & Showalter, K. (1999). Noise driven avalanche behavior in
subexcitable media. Physical Review Letters, 82(4), 855–858. doi:10.1103/PhysRev
Lett.82.855
Washburn, A., Kallen, R. W., Coey, C. A., Shockley, K., & Richardson, M. J. (2015).
Harmony from chaos? Perceptual-motor delays enhance behavioral anticipation in
social interaction. Journal of Experimental Psychology: Human Perception and Performance,
41(4), 1166–1177. doi:10.1037/xhp0000080
Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of “small-world” networks.
Nature, 393(6684), 440–442. doi:10.1038/30918
Webber, Jr., C. L., & Zbilut, J. P. (2007). Recurrence quantifications: Feature extractions
from recurrence plots. International Journal of Bifurcation and Chaos, 17(10), 3467–3475.
doi:10.1142/S0218127407019226
West, B. J. (2010). Fractal physiology and the fractional calculus: A perspective. Frontiers
in Physiology, 1, 12. doi:10.3389/fphys.2010.00012
Westaby, J. D., Pfaff, D. L., & Redding, N. (2014). Psychology and social networks:
A dynamic network theory perspective. American Psychologist, 69(3), 269–284.
doi:10.1037/a0036106
Zhou, W.-X., Sornette, D., Hill, R. A., & Dunbar, R. I. M. (2005). Discrete hierarchical
organization of social group sizes. Proceedings of the Royal Society B: Biological Sciences,
272(1561), 439–444. doi:10.1098/rspb.2004.2970
7
COMPUTATIONAL TEMPORAL
INTERPERSONAL EMOTION
SYSTEMS
Emily A. Butler, Jinyan Guan, Andrew Predoehl,
Ernesto Brau, Kyle Simek, and Kobus Barnard1

Emotions motivate and guide most complex human behavior (Frijda, 1986; Gross,
1999; Izard, 1977). If we could mathematically model emotions with enough
accuracy to predict emotional responding, we could intervene in a broad array
of domains, ranging across (but not limited to) close relationships, education,
parenting, business management, consumer behavior, work performance, health
behaviors, conflict resolution, and political negotiations. Most contemporary mod-
els frame emotion as an intrapersonal dynamic system comprised of subcomponents
such as appraisals, experience, expressive behaviors, and autonomic physiology that
interact over time to give rise to emotional states (Boker & Nesselroade, 2002;
Butner, Amazeen, & Mulvey, 2005; Hoeksma, Oosterlaan, Schipper, & Koot, 2007;
Kuppens, Allen, & Sheeber, 2010; Lewis, 2005; Lodewyckx, Tuerlinckx, Kuppens,
Allen, & Sheeber, 2010). An important extension of these models acknowledges
that many emotional episodes, if not most, occur in the context of social interac-
tions or ongoing relationships. When this occurs, an interpersonal emotion system
is formed, in which the subcomponents of emotion interact not only within the
individual, but across the partners as well (Boker & Laurenceau, 2007; Butler,
2011; Butner, Diamond, & Hicks, 2007; Chow, Ram, Boker, Fujita, & Clore, 2005;
Ferrer & Nesselroade, 2003; Granic & Hollenstein, 2003; Hsieh, Ferrer, Chen, &
Chow, 2010; Lodewyckx et al., 2010). Predicting and intervening in these tempo-
ral interpersonal emotion systems (TIES) requires statistical models that represent
complex, dynamic interdependencies across emotional subcomponents over time,
both within individuals and between partners in an emotional transaction.

Modeling TIES
Several aspects of TIES need to be considered when choosing a mathemati-
cal model to represent them, all of which can be seen in Figure 7.1, which
128  Emily A. Butler et al.

depicts typical emotional data appropriate for studying TIES (e.g., time series
observations of emotional experience, behavior or physiology from interact-
ing pairs of people). First, emotions are oscillatory (Boker & Nesselroade,
2002; Butner et  al., 2007; Chow et  al., 2005; Pettersson, Boker, Watson,
Clark, & Tellegen, 2013). Like a thermostat, people have their own individual
emotional set points that they tend to return to after being perturbed (Chow
et al., 2005; Pettersson et al., 2013). External stimulation provokes emotional
responses, but well-functioning people adapt to these experiences, both auto-
matically and via purposeful emotion regulation, and stabilize back to their
homeostatic set point. This combination of perturbations and self-regulation
results in an oscillating pattern of emotions across a range of time spans (e.g.,
from minutes to days) (Chow et  al., 2005; Helm, Sbarra, & Ferrer, 2012;
Pettersson et al., 2013).

FIGURE 7.1   rototypical emotional data appropriate for studying TIES (e.g., time
P
series observations of emotional experience, behavior, or physiology
from interacting pairs of people). Some measure of emotion is on the
y-axis, and time is on the x-axis. Dashed lines represent one partner, and
solid lines represent the other. Several features of interpersonal emotions
can be seen, including oscillations at different frequencies, damping/
amplification, and both in-phase and anti-phase between-partner
coupling.
 Computational Temporal Interpersonal Emotions  129

A second feature of emotions is that they may dampen or amplify over time.
Regulating emotions involves both negative feedback loops (e.g., A produces B,
which in turn inhibits A), and positive feedback loops (e.g., A produces B, which
produces more of A). Negative feedback dampens emotional oscillations, while
positive feedback amplifies them. Damping and amplification are most relevant
when examined in the context of a single regulatory process evolving over time.
For example, we may be more likely to observe damping or amplification in
laboratory-based studies that focus on specific regulatory tasks (e.g., resolving a
conflict with a partner), as opposed to daily diary studies that likely assess numer-
ous regulatory events over time. Thus, a dynamic model of emotions designed to
capture regulatory processes should allow for damping back to a homeostatic set
point, or amplification away from it.
A third feature is that partners’ emotions typically become interconnected,
due to both automatic and conscious mechanisms, including shared stimuli, con-
tagion, or being a stimulus for each other (Butler, 2011). Additionally, partners
may explicitly attempt to regulate each other’s emotions, whereby one partner
is motivated to change the other’s affective state (Diamond & Aspinwall, 2003;
Zaki & Williams, 2013). Therefore, partners’ oscillating emotions can become
coupled, meaning that some aspect of one partner’s emotional dynamics is influ-
encing some aspect of the other partner’s dynamics. For example, partners may
be pulled into or out of synchrony with each other, or have mutually damping
or amplifying effects on each other (Boker & Laurenceau, 2006; Butner et al.,
2005; Butner et al., 2007; Ferrer & Helm, 2012; Steele & Ferrer, 2011). Thus a
dynamic model of emotions in interpersonal contexts should allow for coupling
to exist between social partners.
A growing body of research guided by a dynamic systems perspective suggests
that conceptualizing relationships as dynamic interpersonal emotion systems pro-
vides a unifying theoretical framework (Butler, 2011). Although metaphors taken
from a dynamic systems framework are often invoked in research on emotion
and relationships, access to analytic tools to turn those metaphors into empiri-
cal outcomes has lagged behind (Boker & Laurenceau, 2007; Chow, Ferrer, &
Nesselroade, 2007; Granic & Hollenstein, 2006). As a result, social scientists are
typically limited to what is available in standard software packages, producing the
risk that the software package ends up dictating the models, rather than theo-
retical considerations about the phenomenon of interest. We are at risk of TIES
researchers looking where the light is, rather than where TIES processes are actu-
ally occurring. For example, if emotional oscillations are relevant to the research
question, then classic approaches such as repeated measures regression, temporal
multilevel models, growth modeling, latent change models, or sequential analyses
are inadequate (Boker & Nesselroade, 2002; Butner et al., 2005; Hessler, Finan, &
Amazeen, 2013). Such methods can assess the general emotional tone of an
interaction, trajectories of change across time, or within- and between-person
concurrent associations (synchrony) and time-lagged associations (transmission)
130  Emily A. Butler et al.

(Butler, 2011; Randall, Post, Reed, & Butler, 2013; Reed, Randall, Post, &
Butler, 2013). However, such methods do not allow for an oscillatory pattern
or damping/amplification, both of which may arise due to regulatory dynamics.
Furthermore, virtually all methods available in off-the-shelf software are depend-
ent upon distributional assumptions, the most important for TIES research being
the assumption of stationarity, meaning that the statistical properties of a time
series (e.g., mean and variance) do not change over time (Hsieh et al., 2010). This
is untenable when considering emotional interaction data, given that emotions
rise and fall, as well as fluctuate more or less, across time.

Mindful Modeling
Cross-disciplinary collaboration is our proposed solution for making appropriate
models available for TIES research. Our team includes social and computational
scientists working together to represent TIES theory in the form of testable
mathematical models, a process we call “mindful modeling.”2 The social scien-
tists provide a theory about some aspect of TIES, along with relevant data, and
the computational scientists provide a Bayesian generative model that embodies
the key aspects of that theory, along with the ability to estimate the parameters
of the model. A Bayesian generative model is a specification for the combined
(joint) probability distribution of: (1) observed data, (2) parameters of a math-
ematical model, and (3) latent variables (Gelman, Carlin, Stern, & Rubin, 2004;
Jordan, 2004; Koller & Friedman, 2009; Murphy, 2012). Prior knowledge, such
as established estimates for some of the model parameters based on prior research,
is encoded by distributions over parameters. One strength of generative models is
that they are conducive to having the parameters represent theoretical constructs.
This supports a top-down approach, whereby theory dictates the model. A key
philosophical point in applying our mindful modeling approach is separating
theoretical/mathematical modeling from inference. Inference refers to the process
of estimating model parameters given data, which is what statistics packages do
behind the scenes in order to return estimates of parameters such as the intercept
and slopes for a regression model. So, in other words, we do not allow our mod-
eling choices to be constrained by the inference options that are readily available,
but rather focus on the theory of interest to guide our decisions. This can lead to
challenging inference problems, but this is precisely the research domain of many
computational scientists, and they have developed extensive methods and tools
for addressing such problems.
A second key aspect of our approach is that we develop and test our models
based on their ability to predict held-out data (e.g., cross-validation) (Geisser,
1975; Mosier, 1951). In other words, some portion of the data is used to develop
the model and estimate the parameters; then we test the ability of that model
with those parameter values to predict new observations. We prefer models
that predict better in a new data set. In contrast, typical model selection and
 Computational Temporal Interpersonal Emotions  131

development procedures used in the social sciences, such as model fit indices
and significance testing, are known to over-fit the data, which means they capi-
talize on chance idiosyncrasies present in any given data set. Thus every time
we include a predictor in a regression model because it increases R-squared,
or free a path in a structural equation model because it improves the fit indices,
we are likely reducing the chance that the model will predict new data well.
The problem becomes increasingly serious as we move away from strict a pri-
ori hypothesis testing towards more exploratory or post hoc analyses. This is
a critical limitation, and is likely contributing to the replication crisis in social
psychology. These facts were noted in social science publications as early as the
1950s, but appear to have been largely forgotten (Cureton, 1950; Kurtz, 1948;
Mosier, 1951). Further, ultimately we would like our models to be able to
predict outcomes either in the future or in different circumstances for specific
people in particular relationships, not simply provide an estimate of a group
mean or average effect in one data set. Without the ideographic prediction
provided by cross-validation, our theories have little hope of achieving wide
ranging applied relevance.

An Example: Moderated and Drifting Linear

Dynamical Systems
As an example we present an investigation of TIES processes using self-reported
emotional data from interacting romantic partners (for details, see Guan et al.,
2015; Reed, Barnard, & Butler, 2015). Committed heterosexual couples engaged
in a conversation about their health-relevant lifestyle choices in the laboratory
(39 couples; average conversation length = 10.8 minutes). Following the con-
versation, the participants watched the videotape of their own interaction and
used a rating dial which turned through 180 degrees, with anchors of “negative,”
“neutral,” and “positive,” to indicate how they remembered feeling during the
conversation. This produced second-by-second continuous self-reports of emo-
tional experience ranging from -2 to +2 (see Figure 7.1 for typical data). The study
was focused on health processes in couples, and participants had been recruited
based on their dyadic weight status (e.g., both partners healthy weight, both
partners overweight, or mixed-weight). We hypothesized that in this context,
the partners’ relative body mass index (BMI) would contribute to the emotional
dynamics of the conversations.
Linear dynamical systems (LDS) provide a particularly useful class of models for
studying TIES because they can represent complex oscillatory patterns of inter-
connected time series (e.g., the dynamics of two persons’ emotions over time).
LDS refer to models in which the subsequent states of a set of variables are linear
functions of their current states. For this example, we focus on a particular sub-
set of linear dynamical systems often called coupled oscillator models (see Guan
et  al., 2015 for additional mathematical and technical details for all aspects of
132  Emily A. Butler et al.

this section). Coupled linear oscillators (CLO) are a particular version of LDS in
which the accelerations of a pair of oscillators are a linear function of their veloci-
ties and positions. CLO models have been fairly extensively used to represent
TIES processes because they model both intrapersonal and interpersonal char-
acteristics, including frequency, damping, and coupling (Boker & Laurenceau,
2006, 2007; Butner et al., 2007; Ferrer & Steele, 2014; Helm et al., 2012; Reed
et al., 2015; Steele & Ferrer, 2011; Steele, Ferrer, & Nesselroade, 2014). They are
also useful for distinguishing between qualitatively different TIES processes such
as coregulation and codysregulation (Butler & Randall, 2013; Reed et al., 2015).
Specifically, if we let xtw and xtM be the positions of two oscillators at time t,
with the first oscillator representing a woman’s emotional state and the second
being her male partner’s, then their joint emotional dynamics are defined by the
second-order differential equation

t = f
xW xt + dW xW
W W
t +c
W
(
xtM − xW
t  ) (7.1)

(
xtM = f M xtM + d M xtM + c M xW
t − xt
M
) (7.2)

Here, xt denotes the second derivative of the emotional observations (e.g.,
emotional acceleration) and xt denotes their first derivative (e.g., emotional
velocity), f w and f M can be translated into emotional oscillation frequencies for
the woman and man respectively, d w and d M represent damping parameters for
each partner, and c w and c M represent the coupling between the two partners.
One way to fit a CLO model to data is to use multilevel regression (for an
example and references, see Reed et  al., 2015). To do so, one estimates first
and second derivatives from the observed emotional time series data and then
uses the original observed data ( xtw and xtM ), along with the estimated first and
second derivatives, as predictor and outcome variables in a multilevel regression
model. This approach has the advantage that most social scientists with training
in graduate statistics will be able to implement it without needing to collabo-
rate with computational scientists. It also has many limitations, however, and
although ignoring any particular item in the following list is probably not prob-
lematic, ignoring many or all of them will likely result in spurious results. First,
estimating derivatives directly from data is inherently noisy and introduces addi-
tional error into the process. Second, we are usually not interested in emotional
acceleration at each time point per se, but rather the entire dynamic pattern of
interconnected partner’s emotions. Despite this, in the regression approach it is
specifically acceleration that is predicted. As a result, the prediction errors that
are minimized in the fitting process are not directly related to what we actually
want to predict. Third, the regression approach does not include a representation
of initial states, but dynamic systems are highly sensitive to start values, meaning
that very different dynamic patterns could arise from the same CLO model, given
different emotional starting points for the partners. Fourth, extending the model
 Computational Temporal Interpersonal Emotions  133

to include multivariate emotional observations (e.g., observed emotional experi-

ence and physiology from both partners, not just experience) is unwieldy, yet in
theory emotion is a multivariate system comprised of several components, and
ideally this would be represented by the model. Finally, the regression approach
assumes stationarity, which in this case means that the CLO parameters do not
change over time—an assumption which is likely violated in many interper-
sonal emotional interactions. Further, in theory we expect partners’ emotional
dynamics to change over time, not remain stationary, so an optimal model should
include that possibility for substantive reasons, not simply to avoid violating sta-
tistical assumptions.
To address all these limitations, we take a Bayesian generative modeling
approach. A generative model posits that the observed data is a random sample
from a (typically) complex probability distribution that includes probabilities for
both model parameters and observable data. In other words, the model pre-
scribes a combined distribution of model parameters and observables, potentially
informed by knowledge gained from prior research such as likely estimates for
some parameters. Thus, given the model parameters, we can generate samples
that should have similar characteristics to our observed data. Likewise, given a
set of parameter estimates or an observed data point, we can calculate the likeli-
hood that it came from the prescribed joint distribution. One thing this allows us
to do is search for likely parameter sets for complex models (e.g., the inference
process). Although the details of the procedure are often challenging, developing
such inference algorithms using sampling methods is an area of expertise for our
computational team. Another thing this allows us to do is evaluate models based
on how likely the data is given a set of parameters, or even more important, how
likely a new sample of data is, thus avoiding the over-fitting problem inherent in
traditional statistical procedures that was discussed above.
Generative models can include a lot of complexity. For example, at one
extreme we could specify that all the model parameters and observed data are
completely inter-correlated (e.g., everything is correlated with everything), but
such a model would be both impractical and scientifically useless, since it does
not simplify the world. We need to set some constraints, ideally driven by theory,
which is done by specifying conditional independencies. What this means in the
specific case of the observed data is that all the variables become independent
(e.g., there would be no remaining correlations among any of the variables), con-
ditioned on the model. In other words, if the model is accurate, then it accounts
for the observed correlation patterns in the data. In our example, we model a
latent dyadic emotional state that evolves following the dynamics of a coupled
oscillator. The observed emotional self-reports at each time point for each person
are posited to be independent, conditioned on their latent dyadic emotional state.
Specifically, if we knew each dyad’s true latent emotional states at every time
point, we theorize that it would account for all observed correlations within and
between partners’ emotions over time.
134  Emily A. Butler et al.

Our model can be broken into two parts for the purpose of description. First,
we use a state-space model to represent a CLO. The latent states in this model can
account for any number of observed variables, thus easily allowing for multivariate
extensions, although in this example we limit ourselves to the univariate ver-
sion where we consider only emotional self-reports. In addition, the parameters
of this model (e.g., the frequency, damping, and coupling parameters for each
partner) can either be stable across time or allowed to drift, thus accommodating
non-stationarity. The parameters can be different for each couple, but allowing
them to be completely free leads to over-fitting. Instead, in the second part of
the model we constrain the parameters for each couple to come from a common
distribution, with the means for each parameter potentially being a function of
moderators (here, BMI). This allows variability of the CLO parameters across
couples, and hence variability in their emotional dynamics, but also provides a
test of theoretical factors that may lead to similarities between couples (e.g., BMI).

State-Space Representation
Any system that is defined by a set of differential equations, such as a CLO, can be
represented in state space, which refers to all the possible configurations of a multi­
variate system. Concretely, each variable in a model provides one dimension, so
a model with five variables has five dimensions. Each possible combination of
those five variables represents a state of the whole five-dimensional system. The
general equations for a continuous-time LDS state-space model are

x = Axt + But + δt (7.3)

t

yt = Cxt + Dut + εt (7.4)

Here xt is a vector containing the system’s latent state variables at time t (note
that these are latent states, not observed ones). In our example, we are modeling
the joint emotional state of each male–female dyad, and so conceptually,3 xt contains
a two-dimensional state representing the true emotional state at time t for each
pair of people (e.g., one latent score for the male and one for the female). These
latent emotional states evolve over time and generate the emotional observations.
In our example, we focus on self-reported experience, but the latent states could
also generate both partners’ physiology or behavior, thus easily making it a
multivariate model if desired.
In Equation 7.3, x is the first derivative of the state variables, so this equa-
tion represents the process by which the states transition from one time point to
the next. The matrix A defines the dynamics by which the states transition, so in
our example A contains equations defining a CLO (e.g., similar to Equations 7.1
and 7.2 above). Obviously, this matrix could contain different equations defining
a different dynamic process, making the model very versatile. In addition, the
 Computational Temporal Interpersonal Emotions  135

components of A can be allowed to vary over time, thus modeling a non-stationary

process. In other words, the quality of the dynamics can change across time. The
optional vector u represents any measured external variables that influence the
dynamic process, and the matrix B determines how it affects the system. We do
not include these components in our example (e.g., Bu is set to zero), but they
could be used to represent something like noise in the room if we thought that
emotional dynamics might be perturbed by noise levels. For example, partners’
emotions may follow a CLO pattern quite well when they can hear each other, but
noise may interrupt that pattern if it disrupts their communication. The optional
δt represents stochastic fluctuations (e.g., unmeasured random errors) in the way
the system transitions between latent states at time t. We again set this component
to zero in our example, so that all random variance is accounted for either by
observation noise (e.g., measurement error) or smoothly changing parameter drift.
Equation 7.4 defines how the observed variables, yt, are related to the latent
state variables at time t. In the univariate case of our example, matrix C has a
particularly simple form that models the observed data as the latent state together
with observation noise, but more complex forms are possible. In the multivariate
case, for example, it could encode different measurement units for the observed
variables (e.g., different physiological variables might be recorded with differ-
ent scales and have different offsets—intercepts). Finally, the optional matrix D
specifies how the vector u affects the observations. For example, noise in the
room may impact the measurement quality of some variables (voice intonation)
and not others (physiology). This component is again set to zero in our example.
Finally, εt represents observation noise at time t, which we specify to be normally
distributed with mean zero and unknown variance.

Bayesian Hierarchical Model

Recall that a generative model prescribes a combined probability distribution of
model parameters and observables, potentially informed by prior knowledge such
as likely estimates for some parameters. Graphical models are a tool to visualize
the conditional dependence structure among a set of random variables, and are
useful for representing the set of independence relationships specified by a gen-
erative model. Figure 7.2 shows the graphical model for our example, omitting
parameter drift (e.g., the CLO parameters for the model depicted in the figure
stay the same across time). Nodes represent all the components of the model,
including observed variables (yit), latent states (xit), and model parameters (all
other nodes). Arrows represent dependencies, and a lack of arrows represents
independence. For example, the observed variables are dependent on the latent
states (e.g., the arrows go from the latent state to the observed variable) and
are conditionally independent of each other conditioned on those latent states
(e.g., the observed variables are not linked to each other by arrows because their
inter-correlations are accounted for by the latent states).
FIGURE 7.2   graphical model of the state-space model with BMI-dependent
A
Gaussian prior. Open circles represent the random variables of the
model, including observed variables (y) (shaded by convention), latent
states (x), and other model parameters. Small circles denote constants
which might be known, set by hand, or learned from data. Arrows
depict dependencies, and lack of arrows represents independence. By
convention, we describe how the model generates the data top-down
(this is called ancestral sampling). To begin, the shared BMI-coefficients
and the standard deviation, σ, come from (are generated by) a joint
Normal-Inverse-Gamma prior distribution. These, together with BMI,
ωi, determine the means, μ, of a distribution over CLO parameters.
This generates the CLO parameter set for a couple i, denoted by θi. To
use a CLO to specify latent state values, we need a starting point, and
these initial values, xt1, are generated from a normal distribution with
mean zero and variance one (depicted by Φx1). Given these initial values
and the CLO parameters, the rest of the values of the latent states are
determined in sequence. The latent states then generate the observed
values, yit. In the particular case of this example, this amounts to the
latent state values being used as the means of normal distributions with
shared standard deviation σ0. These distributions then generate the
observed yit.
 Computational Temporal Interpersonal Emotions  137

As shown in Figure 7.2, the pair of observed variables (e.g., both partners’ self-
reported emotional experience) for couple i at time t, yit, depend on the latent
state for couple i at time t, xit. Specifically, xit provides the means of a normal dis-
tribution for each partner’s observations (e.g., one distribution for men and one
for women) with shared standard deviations σ0. The first latent state for couple i
in turn comes from a prior distribution of emotional start values, Φx1, specified to
be normal with mean zero and variance one. Each subsequent state depends on
the prior state and a set of six CLO parameters for that couple, θi (e.g., both partners’
frequency, damping, and coupling). Those parameters are generated from normal
distributions with means and standard deviations for each CLO parameter speci-
fied by vectors μ and σμ respectively.
Up until here, the model represents a CLO in which each dyad could have
its own CLO parameters and hence unique emotional dynamics. Such a model is
very powerful, and will fit the data extremely well (see “Results” on pp. 139), but
would do so by over-fitting random noise and would have poor ability to predict
new data or speak to theoretical factors expected to influence emotional processes.
In particular, in the context of discussing health and lifestyle choices, we expected
body mass index to impact emotional dynamics in systematic ways. For example,
we expected that healthy-weight couples would be relatively unemotional com-
pared to overweight couples where health was a source of concern. Similarly, we
expected health to be a particularly contentious issue for mixed-weight couples
where one partner was overweight but the other was healthy. Thus we let the
means of the distribution of CLO parameters depend upon a linear function of
both partners’ BMI. Specifically, in Figure 7.2, ωi is a vector with couple i’s average
BMI (male BMI plus female BMI/2) and the difference between the partners’ BMI
(male BMI minus female BMI), α represents the offset (intercept) for each of the six
CLO parameters, β is the linear coefficient for BMI-average predicting each CLO
parameter, and γ is the linear coefficient for BMI-difference. Thus each couple’s
parameter set, θi (e.g., the six CLO parameters for couple i), depends on their joint
BMI status in a way that is similar across couples with similar BMI values. Finally,
the couple-shared BMI-coefficients, α, β, and γ, and shared standard deviation
σ are random variables generated from a joint Normal-Inverse-Gamma prior
distribution (e.g., represented in Figure 7.2 by μ0, Σ0, a0, b0).

Inference and Evaluation

Typically, inference proceeds by maximizing a function representing the prob-
ability of the model parameters given the observed data. For simple problems,
there might be a closed form solution, such as least squares for multiple regression.
For complex problems, however, this is not the case, and typically the approach
becomes a search in the parameter space for values that make the probability of
the parameters conditioned on the data most likely. This can be accomplished by
various forms of Markov-Chain Monte Carlo (MCMC) sampling. Inference in
138  Emily A. Butler et al.

the case of our model includes: (1) using sampling methods to find good estimates
for the parameters that are shared across couples, such as the BMI coefficients
(a process called learning), and (2) given good estimates of the shared parameters,
using those to get good estimates for couple-specific parameters (a process called
fitting). In our example, we use the couple-shared parameters to estimate couple-
specific emotional trajectories into the future.
We combine learning and fitting in a multi-stage model evaluation process
whereby we learn the shared model parameters in one subset of the data and
then use those learned estimates to predict the observed data in a different subset.
Specifically, we use ninefold cross-validation, which means we start by dividing
the couples into nine groups. We get estimates of the shared parameters using the
data from the couples in eight of the groups (learning groups). We then use those
shared parameters to estimate couple-specific parameters for the first 80% of the
conversation time for the held-out ninth group of couples (testing group). Finally,
we predict the remaining 20% of the time points for each held-out testing couple
by taking their estimated CLO parameters and evolving the process into the future
(e.g., predicting values for the future based on the CLO parameters obtained from
the past). We then repeat these steps eight more times, using each group in turn as
the testing group. We record the root mean squared error (RMSE) of prediction
for each testing couple for the first 80% of the conversation (fitting error based on
the same data used to estimate that couple’s parameters) and for the last 20% of the
conversation (prediction error based on new data that was not used in any stage
of parameter estimation). Models with low prediction error are less likely to be
capitalizing on chance and are more likely to be representing robust processes that
gave rise to the data, or in other words, real-world processes of substantive interest
such as emotional dynamics taking the form of coupled oscillators.
Our multi-stage cross-validation approach can be contrasted with the use of
fit statistics such as the AIC, DIC, or WAIC. Such fit measures will only work if
the assumptions used to derive them are met. For models as complex as the ones
we are implementing, it is hard to know what those assumptions should be, and
we have generally found that fit statistics do not concur with cross-validation
for our applications. Further, the gold standard for developing and validating
model fit statistics is to compare their results against cross-validation and the only
argument for using them is to avoid the computational cost of conducting actual
cross-validation (Gelman, Hwang, & Vehtari, 2014; Vehtari, Gelman, & Gabry,
2016; Vehtari, Mononen, Tolvanen, Sivula, & Winther, 2014).

Results
Table 7.1 presents mean RMSEs and their 95% confidence intervals across cou-
ples for both fitting and prediction for a sequence of models (since these are
error estimates, smaller values imply a better-fitting model). Non-overlapping
confidence intervals for any pair of RMSEs indicate that the estimates are
 Computational Temporal Interpersonal Emotions  139

significantly different, with p < .05. In general, one would prefer the model(s)
with the lowest prediction errors. The first model (Average), estimates each
person’s mean emotional ratings for the first 80% of the conversation and simply
predicts that mean will continue into the future 20%. The second model (Linear),
estimates each person’s intercept and slope for the first 80%, then predicts it will
continue on that trajectory into the future 20%. Notice that although the fitting
error is smaller for the Linear model than the Average, the reverse is true for the
prediction error, suggesting that any apparent benefit for the Linear model is
being driven by over-fitting. The third model (Couple-Specific CLO) estimates
a unique CLO model for each couple based on the first 80% of their conversa-
tion, then predicts that dynamic pattern will continue into the future 20%. Here
the presence of over-fitting is highly apparent, with very low fitting errors, but
very high prediction error. The fourth model (Couple-Shared CLO) estimates
a CLO model for the first 80% of the conversation in which all couple’s param-
eters come from a shared distribution (e.g., all couples share the same mean of
the CLO parameters). This shared dynamic pattern is then predicted to con-
tinue into the future 20%. Although this model does not fit quite as well as the
Couple-Specific model, it has the lowest prediction error so far, suggesting that
CLO dynamics are a fairly good characterization for the emotional self-reports
obtained in this context. Finally, the fifth model (BMI-CLO) estimates a CLO
model for the first 80% of the conversation in which the distribution of CLO
parameters depends on the couple’s BMI average and difference. Thus couples
who are similar to each other in their joint BMI (e.g., both healthy weight, both
overweight, mixed-weight) are predicted to show similar dynamics. This model
has the lowest prediction error, suggesting that a couple’s joint BMI has an influ-
ence on emotional dynamics in this context (although the confidence interval
overlaps slightly with the Couple-Shared CLO, suggesting a statistically marginal
improvement). To visualize this influence, Figure 7.3 shows the model-predicted
emotional trajectories for a couple in which the woman is heavier than the man,
as compared to the reverse. We can see that couples in which the woman is
heavier than the man may be emotionally disconnected and the woman may
show increasingly volatile emotional responses. This could contribute to poorly
modulated emotion responses when discussing health and lifestyle choice.

TABLE 7.1  Mean root mean squared errors and their 95% confidence intervals across
couples for fitting and prediction for a sequence of models.

Model Fitting Predicting

Average 0.52 (0.49–0.55) 0.70 (0.65–0.80)

Linear 0.47 (0.44–0.50) 0.76 (0.68–0.84)
Couple-Specific CLO 0.39 (0.36–0.42) 1.10 (0.91–1.29)
Couple-Shared CLO 0.44 (0.41–0.47) 0.68 (0.62–0.74)
BMI-CLO 0.41 (0.39–0.43) 0.59 (0.54–0.64)
140  Emily A. Butler et al.

FIGURE 7.3   odel-predicted emotional states over time, with the same initial
M
emotional start values but differing BMI moderator values. Black lines
are the emotional state for the male partner, and gray lines are for the
female partner. (a) The predicted emotional state for couples in which
the woman is heavier than the man. (b) The predicted emotional state
for couples in which the man is heavier than the woman.

Future Directions
One trivial extension of our model, similar to making it multivariate, is to rep-
resent social units including more than two members. Doing so simply involves
including variables for as many people as desired (e.g., three-person families, six-
person work groups, etc.) as indicative of the latent states (as values in the vector
xi of the state-space model). Of course, this implies that there is some meaningful
emotional pattern at the level of the social unit, such as hypothesized by family
systems (e.g., polarization or escalation) or theories about team functioning (e.g.,
group-think or synergy), but these theories could be tested using such a model.
A second extension we are working on is to allow qualitative shifts in emo-
tional dynamics across time using regime switching, which allows distinct state
changes across time. In our model so far, we have allowed the CLO parameters
to drift smoothly across time, which accommodates non-stationary emotional
processes and improves prediction. It does not, however, lend itself to substantive
interpretation. In contrast, regime switching can be used to model distinct shifts
in emotional dynamic states, which can be visualized and related to substantive
theory. Such states can either be specified a priori, or learned from the data. In
the latter exploratory case, cross-validation is particularly critical to avoid over-
fitting, and ideally the learned states would then be tested in an a priori fashion
on new data. For example, it may be common for couples to go through phases
of highly oscillatory emotional functioning, well characterized by a CLO pat-
tern, but spend time in relatively quiescent and disconnected phases in between,
 Computational Temporal Interpersonal Emotions  141

when the couple are less emotional or less involved with each other. Qualitative
shifts may be particularly relevant when studying longer periods of time, so this
extension is likely to be especially helpful for modeling emotional dynamics over
days, weeks, or even longer periods. Finally, we are working to make our meth-
ods available for other researchers via Web services (Predoehl, Guan, Butler, &
Barnard, 2015). Soon, you will be able to use our interface to upload your data,
choose model settings, and receive back a table of results similar to those reported
in this chapter. We encourage you to try it out (as of writing, we are in the test-
ing phase) and hope this service will support extensive, exciting, and cumulative
future research on TIES.

Notes
1 This research was supported in part by Grant 1R21HL109746-01A1 from the National
Heart, Lung, and Blood Institute awarded to the first author, and by Grant BCS-1322940
from the National Science Foundation awarded to the first and last authors.
2 We are also working to make all our modeling tools available to other researchers via
Web services. At the time of writing, we are in the testing phase and hope to have the
service fully functional soon at www.compties.org/software.html.
3 We say “conceptually” because, in practice, the CLO model includes second derivatives
as well as first, which leads to a slightly different set of equations (see Guan et al., 2015).
In particular, xt includes both the emotional state for each partner and their first deriva-
tives, but for the purpose of intuitive explanation, we ignore that detail here.

References
Boker, S. M., & Laurenceau, J. P. (2006). Dynamical systems modeling: An application
to the regulation of intimacy and disclosure in marriage. In T. A. Walls & J. L. Schafer
(Eds.), Models for intensive longitudinal data (pp. 195–218). New York: Oxford University
Press.
Boker, S. M., & Laurenceau, J. P. (2007). Coupled dynamics and mutually adaptive context.
In T. D. Little, J. A. Bovaird, & N. A. Card (Eds.), Modeling contextual effects in longitudinal
studies (pp. 299–324). Mahwah, NJ: Lawrence Erlbaum.
Boker, S. M., & Nesselroade, J. R. (2002). A method for modeling the intrinsic dynamics
of intraindividual variability: Recovering parameters of simulated oscillators in multi-
wave panel data. Multivariate Behavioral Research, 37(1), 127–160.
Butler, E. A. (2011). Temporal interpersonal emotion systems: The “TIES” that form relation-
ships. Personality and Social Psychology Review, 15, 367–393. doi:10.1177/1088868311411164
Butler, E. A., & Randall, A. K. (2013). Emotional coregulation in close relationships.
Emotion Review, 5, 202–210.
Butner, J., Amazeen, P. G., & Mulvey, G. M. (2005). Multilevel modeling of two cyclical
processes: Extending differential structural equation modeling to nonlinear coupled
systems. Psychological Methods, 10(2), 159–177.
Butner, J., Diamond, L. M., & Hicks, A. M. (2007). Attachment style and two forms
of affect coregulation between romantic partners. Personal Relationships, 14, 431–455.
Chow, S., Ferrer, E., & Nesselroade, J. R. (2007). An unscented Kalman filter approach to
the estimation of nonlinear dynamical systems models. Multivariate Behavioral Research,
42(2), 283–321.
142  Emily A. Butler et al.

Chow, S., Ram, N., Boker, S. M., Fujita, F., & Clore, G. (2005). Emotion as a thermostat:
Representing emotion regulation using a damped oscillator model. Emotion, 5(2),
208–225.
Cureton, E. E. (1950). Validity, reliability and baloney. Educational and Psychological
Measurement, 10(1), 94–96.
Diamond, L. M., & Aspinwall, L. G. (2003). Emotion regulation across the life-span:
An integrative perspective emphasizing self-regulation, positive affect, and dyadic
processes. Motivation and Emotion, 27(2), 125–156.
Ferrer, E., & Helm, J. L. (2012). Dynamical systems modeling of physiological coregulation
in dyadic interactions. International Journal of Psychophysiology, 88(3), 296–308.
Ferrer, E., & Nesselroade, J. R. (2003). Modeling affective processes in dyadic relations via
dynamic factor analysis. Emotion, 3(4), 344–360.
Ferrer, E., & Steele, J. S. (2014). Differential equations for evaluating theoretical models
of dyadic interactions. In P. C. M. Molenaar, K. M. Newell, & R. M. Lerner (Eds.),
Handbook of developmental systems theory and methodology (pp. 345–368). New York:
Guilford Press.
Frijda, N. H. (1986). The emotions. Cambridge: Cambridge University Press.
Geisser, S. (1975). The predictive sample reuse method with applications. Journal of the
American Statistical Association, 70(350), 320–328.
Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, D. B. (2004). Bayesian data analysis
(2nd ed.). London: Chapman & Hall.
Gelman, A., Hwang, J., & Vehtari, A. (2014). Understanding predictive information
criteria for Bayesian models. Statistics and Computing, 24(6), 997–1016.
Granic, I., & Hollenstein, T. (2003). Dynamic systems methods for models of develop-
mental psychopathology. Development and Psychopathology, 15, 641–669.
Granic, I., & Hollenstein, T. (2006). A survey of dynamic systems methods for devel-
opmental psychopathology. In D. Cicchetti & D. J. Cohen (Eds.), Developmental
psychopathology. Vol. 1: Theory and method (2nd ed.). New York: John Wiley & Sons.
Gross, J. J. (1999). Emotion and emotion regulation. In L. A. Pervin & O. P. John (Eds.),
Handbook of personality: Theory and research (2nd ed., pp. 525–552). New York: Guilford
Press.
Guan, J., Simek, K., Brau, E., Morrison, C. T., Butler, E. A., & Barnard, K. (2015).
Moderated and drifting linear dynamical systems. In Proceedings of the 32nd International
Conference on Machine Learning (Vol. 3, pp. 2463–2472). Stroudsburg, PA: International
Machine Learning Society.
Helm, J. L., Sbarra, D., & Ferrer, U. (2012). Assessing cross-partner associations in physi-
ological responses via coupled oscillator models. Emotion, 12(4), 748–762.
Hessler, E. E., Finan, P. H., & Amazeen, P. G. (2013). Psychological rhythmicities. In
J. P. Sturmberg & C. M. Martin (Eds.), Handbook of systems and complexity in health.
New York: Springer Science+Business Media.
Hoeksma, J. B., Oosterlaan, J., Schipper, E. M., & Koot, H. (2007). Finding the attractor
of anger: Bridging the gap between dynamic concepts and empirical data. Emotion,
7(3), 638–648.
Hsieh, F., Ferrer, E., Chen, S., & Chow, S. (2010). Exploring the dynamics of dyadic
interactions via hierarchical segmentation. Psychometrika, 75(2), 351–372.
Izard, C. E. (1977). Human emotions. New York: Plenum.
Jordan, M. I. (2004). Graphical models. Statistical Science (special issue on Bayesian
statistics), 19, 140–155.
 Computational Temporal Interpersonal Emotions  143

Koller, D., & Friedman, N. (2009). Probabilistic graphical models. Cambridge, MA: MIT
Press.
Kuppens, P., Allen, N. B., & Sheeber, L. (2010). Emotional inertia and psychological
adjustment. Psychological Science, 21, 984–991.
Kurtz, A. K. (1948). A research test of the Rorschach test. Personnel Psychology, 1(1),
41–51. doi:10.1111/j.1744-6570.1948.tb01292.x
Lewis, M. D. (2005). Bridging emotion theory and neurobiology through dynamic
systems modeling. Behavioral and Brain Sciences, 28, 169–245.
Lodewyckx, T., Tuerlinckx, F., Kuppens, P., Allen, N. B., & Sheeber, L. (2010). A
hierarchical state space approach to affective dynamics. Journal of Mathematical Psychology,
55, 68–83.
Mosier, C. I. (1951). The need and means of cross validation. I: Problems and designs of
cross-validation. Educational and Psychological Measurement, 11, 5–11.
Murphy, K. (2012). Machine learning: A probabilistic perspective. Cambridge, MA: MIT Press.
Pettersson, E., Boker, S. M., Watson, D., Clark, L. A., & Tellegen, A. (2013). Modeling
daily variation in affective circumplex: A dynamical systems approach. Journal of
Research in Personality, 47, 57–69. doi:10.1016/j.jrp.2012.10.003
Predoehl, A., Guan, J., Butler, E. A., & Barnard, K. (2015). CompTIES Web App.
Retrieved from www.compties.org/COM.html
Randall, A. K., Post, J. H., Reed, R. G., & Butler, E. A. (2013). Cooperating with
your romantic partner: Associations with interpersonal emotion coordination. Journal
of Social and Personal Relationships, 30(8), 1072–1095.
Reed, R. G., Barnard, K., & Butler, E. A. (2015). Distinguishing emotional co-regulation
from co-dysregulation: An investigation of emotional dynamics and body-weight in
romantic couples. Emotion, 15, 45–60. http://dx.doi.org/10.1037/a0038561
Reed, R. G., Randall, A. K., Post, J. H., & Butler, E. A. (2013). Partner influence and
in-phase versus anti-phase physiological linkage in romantic couples. International
Journal of Psychophysiology, 88, 309–316.
Steele, J. S., & Ferrer, E. (2011). Latent differential equation modeling of self-regulatory
and coregulatory affective processes. Multivariate Behavioral Research, 46(6), 956–984.
Steele, J. S., Ferrer, E., & Nesselroade, J. R. (2014). An idiographic approach to
estimating models of dyadic interactions with differential equations. Psychometrika,
79(4), 675–700. https://doi.org/10.1007/s11336-013-9366-9
Vehtari, A., Gelman, A., & Gabry, J. (2016). Practical Bayesian model evaluation using
leave-one-out cross-validation and WAIC. Journal of Statistics and Computing, 1–20.
Vehtari, A., Mononen, T., Tolvanen, V., Sivula, T. & Winther, O. (2014). Bayesian
leave-one-out cross-validation approximations for Gaussian latent variable models.
Journal of Machine Learning Research, 17(103), 1–38.
Zaki, J., & Williams, W. C. (2013). Interpersonal emotion regulation. Emotion, 13(5),
803–810. doi:10.1037/a0033839
8
MAPPING CO-REGULATION IN
SOCIAL RELATIONS THROUGH
EXPLORATORY TOPOLOGY
ANALYSIS
Jonathan E. Butner, Arwen A. Behrends,
and Brian R. Baucom

The family. We were a strange little band of characters trudging through life
sharing diseases and toothpaste, coveting one another’s desserts, hiding shampoo,
borrowing money, locking each other out of our rooms, inflicting pain and kiss-
ing to heal it in the same instant, loving, laughing, defending, and trying to figure
out the common thread that bound us all together.
(Bombeck, 1987, p. 11)

Introduction
Social relationships are inherently complex, with that complexity arising from
many sources and taking many forms. For example, individuals in relationships
are inherently interdependent on many levels (e.g., behavior, emotions, cog-
nition; Lansing & Berg, 2014). These interdependent relationships have led to
a revolution in the study of dyads, with the conclusion that dyads should not
merely be studied as a pair of individuals (Gottman, Murray, Swanson, Tyson, &
Swanson, 2002). Their relationships are adaptive, changing through time, show-
ing both stable and varying qualities. Recent work that considers the full range of
these possibilities demonstrates the potential advantages of this perspective rela-
tive to traditional methods of analysis (e.g., Baucom et al., 2015). This suggests
that the complexity of social relationships is difficult to depict from a reductionist
point of view.
In this chapter, we advance a simple premise: social relationships are better
understood through flexible statistical methods that are capable of modeling com-
plex, interdependent processes that evolve and emerge over time rather than
reducing their inherent complexity. We believe modeling with methods designed
for complex, interdependent processes can actually describe social relationships
 Mapping Co-Regulation in Social Relations  145

in a more tractable fashion (Nowak, 2004; Nowak & Lewenstein, 1994). One
class of modeling techniques that is especially well suited to this task is dynamical
systems models (DSMs). Dynamical systems models are a family of time series
techniques that aim to depict data through differential equations or difference
equations. As a whole, DSMs are very flexible modeling techniques that can
be used to capture patterns of change and, in this case, we will capitalize on the
link between two or more variables over time to represent the social relation-
ship context. In this chapter, we focus on the application of DSMs for modeling
co-regulation in romantic relationships. We begin by defining the concept of
co-regulation and describing how theoretical constructs in co-regulation can
be conceptualized from a dynamical systems perspective. We then provide an
example application of a DSM to modeling co-regulation in ambulatory heart
rate in a married couple. We close with a discussion of important issues to con-
sider when applying DSMs to modeling co-regulation and other complex forms
of interdependence in social relationships.

Co-Regulation: What It Is in Romantic Relationships

Co-regulation is a concept that is receiving rapidly growing attention in relation-
ship science. Though it has been variously defined and operationalized (for a
review, see Butler & Randall, 2013), the term “co-regulation” most commonly
refers to a linkage between romantic partners’ affective states that promotes
affective homeostasis within each partner (e.g., Sbarra & Hazan, 2008). That is,
co-regulation is the extent to which partners assist one another in maintaining
affective states. This assistance may be active or passive, but is in direct contrast
to the affective qualities one would observe if the partner were not in the roman-
tic relationship. Healthy, well-functioning relationships where partners feel safe
and are securely attached to one another are thought to promote co-regulation,
and recent research provides initial empirical support for this supposition (e.g.,
Butner, Diamond, & Hicks, 2007; Helm, Sbarra, & Ferrer, 2012; Helm, Sbarra, &
Ferrer, 2014). However, co-regulation can also be adverse, in that partners can
assist one another in maintaining poor affective states.
This definition suggests that there are several processes involved in co-regulation
and that a comprehensive statistical model of co-regulation should be able to
capture, describe, and distinguish each of these processes. First, this definition
makes a distinction between homeostatic processes that occur within an indi-
vidual partner and those that occur between partners. One way to illustrate this
key idea is to consider the emotional reactions of two romantic partners during
an argument. For example, suppose that Sam and Jordan disagree about whether
or not to let their son go to a party. Both Sam and Jordan are upset by the argu-
ment, and their different responses to one another’s emotional reactions illustrate
the differences between intrapersonal homeostatic mechanisms and interpersonal
homeostatic mechanisms. When Sam expresses frustration, Jordan is responsive
146  Butner, Behrends, and Baucom

to what Sam is saying, but Sam’s frustration does not impact Jordan’s distress
above and beyond the distress that Jordan was already feeling. In contrast, Sam
is very sensitive to Jordan’s distress such that the higher Jordan’s distress, the
higher Sam’s frustration in addition to the impact of the frustration that Sam was
already feeling. In this example, Jordan’s reactions are governed by intrapersonal
homeostatic mechanisms while Sam’s are governed by both intrapersonal and
interpersonal (i.e., co-regulatory) mechanisms.
A second implication of this definition of co-regulation is that co-regulation
occurs over time and can only be observed when there is temporal variability
in both partners’ affective states. This statement may seem obvious, but it raises
subtle yet crucial conceptual issues that are difficult to translate from theory to
statistical model. Because co-regulation is so strongly tied to the idea of homeo-
stasis, variability cannot be measured with regard to any particular point. Rather,
it should be measured with regard to a homeostatic set point, and many of the sit-
uations that provoke emotional responses in social relationships can also obscure
homeostatic set points. For example, consider Figure 8.1, which plots Jordan
and Sam’s emotional responses during their argument about whether or not to
let their son go to a party. Jordan’s responses clearly vary about a single point,
but Sam’s responses vary about a parabola that initially increases, but ultimately
decreases over time. While it would be possible to calculate both partners’ vari-
ability about his/her mean affective state or affective state at the beginning of the
argument, it is less clear that either approach would represent the same affective
construct (i.e., variability about a homeostatic set point) for both partners.

Conceptualizing Co-Regulation from

a Dynamical Systems Perspective
Numerous strategies could be used in considering the multitude of options for
translating these theoretical notions of what co-regulation is as a psychological
phenomenon into a statistical model. Consistent with our premise that social
relationships are better understood by examining the higher-order patterning,

FIGURE 8.1  Affective responses over time.

 Mapping Co-Regulation in Social Relations  147

we propose that one valuable approach to this task is to consider the degree to
which the theoretical assumptions about a psychological phenomenon, such as
co-regulation, are consistent with the theoretical assumptions underlying the sta-
tistical models. This proposition is not meant to suggest that processes in social
relationships cannot be accurately characterized using relatively simple statistical
models. Much to the contrary, we recommend using the simplest possible model
that is consistent with the theoretical underpinnings of the psychological phe-
nomenon. In the remainder of this section, we illustrate how DSMs satisfy these
criteria for modeling co-regulation.
DSMs are a class of modeling techniques that utilize differential equations or
difference equations to characterize change processes over time. DSMs are based
in dynamical systems theory, in that they inherently describe temporal patterns
in ways consistent with how systems theory assumes temporal processes function.
Dynamical systems theory is primarily concerned with understanding patterns
of change over time, under the assumption that these temporal patterns are a
byproduct of the interactions of many higher- and lower-order processes known
as emergence. Due to this assumption, systems theory makes several important
assumptions about the nature of change over time. First, dynamical system theory
assumes that the characteristics of a system can be observed by measuring the
behavior of the variables involved in the system. Applied to co-regulation, this
idea implies that it is possible to understand a relational process, such as co-regulation,
by observing the behavior of the individuals involved in the relationship.1 This
idea is commonly referred to as “emergence” in the dynamical systems literature,
and suggests that the “how” is a much more difficult question to assess, if not
impossible, and instead a researcher’s focus should be on what emerges. Second,
dynamical systems theory assumes that the ways a system changes over time are
best understood in terms of patterns (what emerges), which means that variables
tend to change over time in a predictable fashion or fashions (e.g., oscillating
about a single point, escalating over time, etc.). Applied to co-regulation, this
idea suggests that co-regulation can be described as partners’ typical responses
to each other’s behavior (e.g., when Jordan is more distressed, what is Sam’s
most likely response, and vice versa?). Third, dynamical system theory assumes
that some patterns of change over time are more likely to occur than others.
Applied to co-regulation, this idea implies that even though partners may have
typical ways of responding to one another, it is also possible, though less likely,
for them to deviate from their typical response. Fourth, these patterns differ in
their inherent stability, and many of the patterns are, in fact, defined by reactions
to outside forces generating temporary divergences from the pattern. Applied to
co-regulation, this idea implies that our description of a typical response is really
the degree to which the typical response is able to be maintained despite constant
forces that could lead to diverging from that response pattern.
Patterns that emerge under systems theory and that are applied to co-regulation
include both how changes in one partner’s affect are related to changes in the
148  Butner, Behrends, and Baucom

other partner’s affect as well as the point about which those changes occur. The
point about which changes occur is commonly called the set point in the dynami-
cal systems literature. In dynamical systems theory, a set point that a system is
drawn towards is called an attractor and a set point that a system is pushed away
from is called a repeller. Attractors are analogous to the idea of a homeostatic set
point—when the variables in a system move away from an attractor, they show a
pattern of returning to that set point. This pattern of change over time is similar
to the idea of homeostasis as representing a pattern of change where a person
returns to a preferred affective state after being perturbed from that state. In
contrast, repellers represent a set point that is unlikely to occur or that the system
avoids. A system can have one or more attractors, one or more repellers, or some
combination of the two. This quality of dynamical systems theory provides an
elegant solution to the ambiguity of defining homeostatic set points—rather than
defining them a priori, homeostatic set points (i.e., attractors) can be quantified
by observing the likelihood of patterns of change over time. The advantage of this
approach is that it allows for characterizing the system in terms of behavior as it
unfolds over time, which is what is of primary interest in studying co-regulation.

Modeling Social Relations

We illustrate how to construct a dynamical systems model of two romantic
partners’ data utilizing data from a single married couple where both the husband
and wife wore ambulatory heart monitors. These data are from a larger study
where couples completed seven days of ambulatory data collection after par-
ticipating in a thorough laboratory assessment. This particular couple was in the
clinically distressed range of relationship satisfaction, based on an average score
of 12 between the spouses on the four-item version of the Couple Satisfaction
Index (Funk & Rogge, 2007). We present results for a single day totaling
16 hours of heart rate data. Heart rate, measured in beats per minute (BPM),
was continuously measured during all waking hours using an Actiheart biosensor
(CamNTech); their data were recorded at 128 Hz sampling rate and aggregated
to one-minute epochs prior to analysis. Heart rate values greater than 130 BPM
were omitted from analyses as they were likely artifacts, resulting in a total of
973 minutes of data for the couple. Continuous GPS data were also recorded
during the seven days of ambulatory data collection; these data were used to
determine if the couple was together (within 250 feet of one another) or apart
(greater than 250 feet apart from one another) for any given reading.
Here we chose to restrict our analysis to a single dyad on a single day to keep
the example relatively simple. However, it is reasonable to analyze all couples
on all days simultaneously with the procedure we outline. Under this circum-
stance, the assumption would be that individuals all function within the same
dynamic system and that there is equivalency in the heart rate metrics across
couples. Centering techniques akin to multilevel modeling can be applied to help
 Mapping Co-Regulation in Social Relations  149

differentiate some of these issues, but would complicate our example.

Alternatively, analyzing a single couple and even on a single day has been pro-
moted as a way to help seek ergodic solutions (Molenaar & Campbell, 2009).
Ergodicity is a notion from physics that has been extrapolated to the social
sciences as having the same model both within and between individuals. As
is true of virtually all statistical approaches that can be used with nested data
(i.e., data sets with repeated observations nested within higher-order groups; in
the example in this chapter, repeated measurements of BPM are nested within
individuals and individuals are nested within couples), results from dynamical
systems models estimated on one group (i.e., one couple on one day) cannot be
assumed to apply to all couples on all days. This idea is consistent with the systems
theory concept of intrinsic dynamics—that the system for individuals (and in this
case, across days) could be quite different (Vallacher & Nowak, 1997). A seminal
example of intrinsic dynamics is seen in Molenaar’s illustration of how the Big
Five has a different factor structure within people in comparison to across people.
The implication of intrinsic dynamics for beginning dynamic systems modelers is
that ergodicity cannot be assumed to hold for results based on a case example and,
if one wishes to make interpretations about ergodicity, the consistency of results
within and between higher-order groups must be directly tested with approaches
that decompose within- and between-group variance and covariance, such as
multilevel modeling and multilevel structural equation modeling.

Covariation
Probably the hardest part of computational modeling under a dynamical systems
logic is knowing where to begin. Here we start with the singularly most common
examination, that of same-time covariation. To illustrate covariation, Figure 8.2
is a scatterplot of the husband BPM (on the y-axis) and wife BPM (on the x-axis),
with a loess smoother (a graphical line that is sensitive to data nonlinearities) and
best fitting straight line overlaid. The best fitting straight line directly represents
the covariation itself—the extent to which we observe highs and lows together
(positive relationship), or when one is high, the other being low (a negative rela-
tionship). As can be seen in Figure 8.2, there is some correspondence of points.
The correlation is 0.25 (95% confidence interval [CI]: 0.19 < 0.25 < 0.31), indi-
cating that when wives have higher heart rates, husbands also tend to have higher
heart rates. Interestingly, this correlation is substantially higher when together
(95% CI: 0.34 < 0.45 < 0.55) than when apart (95% CI: 0.13 < 0.20 < 0.27). This
result would commonly be interpreted as some simple evidence of there being
some sort of co-regulation.
The single biggest problem with a covariation approach to understanding sys-
tems is that covariation only tells us about the existence of value correspondence.
That is, when wives have higher heart rates, do husbands also have higher heart
rates (or lower, for a negative correlation)? There are many circumstances in
150  Butner, Behrends, and Baucom

FIGURE 8.2  Scatterplot of husband and wife BPM.

systems that can create correspondence of values and many that will not generate
correspondence. Some might call aspects from both “co-regulation.” As a hypo-
thetical example, consider husbands and wives hovering around a homeostatic
value alone coming just from intraindividual dynamics. Whenever the husband
or wife encounters various environmental situations, his or her heart rate changes
and, over time, moves back towards the homeostatic value—the attractor men-
tioned earlier. Notice that these events might only influence the husband or the
wife, and thus the return pattern over time can actually be counter to covariation.
However, the homeostatic pattern itself is one of covariation, in that husbands
and wives are converging towards the same value over time, and if they are
sitting at that value, it will generate consistency of that single point. If there exist
two such homeostatic values and the couple switches between these two stable
locales, one can then imagine a very strong degree of covariation due to the
relative placement of the homeostatic values to one another (e.g., do they fall on
a nice diagonal?). It would be difficult to argue that such regime switching was
co-regulation. We need a much more nuanced examination to know.
To better understand this issue of correspondence, Figure 8.3 is a two-­dimensional
kernel density plot of husband (on the y-axis) and wife (on the x-axis) BPM data.
This plot is a graphical way to estimate the probability density function (PDF), where
 Mapping Co-Regulation in Social Relations  151

the frequency of data points are represented topographically (a hill is many points) and
points that occur with greater probability are represented by taller hills. Notice that
there are two or three peaks (the third is at lower husband BPM). That is, our covari-
ation difference may well be this second scenario described above. Each peak could
be a homeostatic point in the combination of husband and wife BPM, where regime
switching between the two could be driving the existence of covariation.

Dynamical Systems as Maps

Topography is the drawing of maps, and topology is the math behind them. It
just so happens that topology is also the math behind dynamical systems (Butner,
Gagnon, Geuss, Lessard, & Story, 2014). So there is a direct correspondence
between the way maps function and our understanding of systems theory. In sys-
tems terms, the two-dimensional kernel density plot (Figure 8.3) also corresponds
to something known as the phase portrait—a map of where data converges over
time (Abraham and Shaw, 1992). The phase portrait is merely a scaled inverse of
the PDF (a hill on the PDF is a valley on a phase portrait; Guastello, 2011). From
our perspective, this map (the PDF or phase portrait) shows where BPM values

FIGURE 8.3  Two-dimensional kernel density plot of husband and wife BPM.
152  Butner, Behrends, and Baucom

are converging through time (i.e., the peaks). It allows us to start to understand
what covariation is capitalizing upon.
However, dynamical systems are about the time evolution of processes, thus
this also inherently highlights the ignorance of time in covariation. We can think
of this as one extreme treatment of our data—where time is ignored. Figure 8.3
just shows us where data coalesces. The other extreme is where time is explicit.
Common treatments of explicit time-based models are growth modeling and
forms of time series analyses like ARIMA models, in all their various forms.
Figure 8.4 shows a pair of time series plots for each heart rate—the graphical
equivalent to an explicit time representation. To generate a systems understand-
ing, we need to live between these two extremes. Time must be taken into
account, but we will do so implicitly rather than explicitly. That is, we will need
to link the evolution in Figure 8.4 with the map features in Figure 8.3.
It is helpful to consider the potential contributors to variation over time from
a systems perspective, and there are three potential contributors to the values
we observe (Riley & Turvey, 2002). In map terms, there are basins of attraction
where all activity in a topographical region is related (e.g., combinations of BPM
where husband and wife tend to reside as a function of constraints; e.g., stress
at work and home, common coping methods). In the case of an attractor, these

FIGURE 8.4  Time series plots of (a) wife BPM by time and (b) husband BPM by time.
 Mapping Co-Regulation in Social Relations  153

FIGURE 8.4  (continued)

tend to be regions that all function around a single set point. However, a basin
of attraction expands beyond the logic of just an attractor. For example cycles are
attractive looped trails (not unlike Gottman et. al.’s chain of negativity; Gottman
et al., 2002), and more complicated trails are possible.
The second contributors to the values we observe, from a systems perspective,
are perturbations. Modern systems theory uses mathematical techniques based on
the recognition that our examination is only part of the system, that the system is
open (Prigogine, 1977). The other parts of the system that are not being directly
examined still contribute to the variation observed in values through time as
constant changes to the values. Observing basins of attraction in the data (locations
or trails) is thus despite perturbations. Perturbations can then be thought of as a
stochastic element that helps define the stability of the attractive features—for
example, how strong the attractor is where stronger attractors are able to resist
larger perturbations.
The final contributor is measurement error, though this is much more limited
in scope once perturbations are taken into account in comparison to notions of
error in classical test theory. Specifically, error may just be a representation of
measurement imprecision. For example, under classical test theory, the impact of the
lighting in a room during a test would be measurement error on test performance.
154  Butner, Behrends, and Baucom

From a systems perspective, the lighting is part of the system itself, and thus
constitutes a perturbation instead. Techniques differ greatly on their ability to
distinguish perturbations from error. For example, a growth model makes no
distinction. However, systems-based techniques can be built to account for this
difference if they are able to distinguish the stability of a pattern. If we assume
perturbations always exist, then pattern stability inherently is in reaction to those
perturbations. The technique we will use will include notions of stability by esti-
mating perturbation resistance, but will not generate a specific perturbation term.
To begin to understand the role of time evolution, it helps to see how an
observed time series translates into trails instead of the map itself; a pair of time
series represent a series of vectors on the map. Figure 8.5 shows the logic.
Specifically, we can imagine a time series of northness and a time series of eastness.
The corresponding values represent a vector or series of vectors in northness/
eastness space. In our circumstance, the time series for the husband and wife from
Figure 8.4 correspond to a series of vectors in the husband/wife space, illustrated
in Figure 8.5d. Given the density of the data, it is difficult to see a clear trail (we
proportionally shortened the arrows to help distinguish all the vectors), and it is
possible that this represents multiple trails rather than just one (e.g., the couple is
switching between different trails).

FIGURE 8.5  Vector plot of husband and wife BPM.

FIGURE 8.5  (continued)
156  Butner, Behrends, and Baucom

FIGURE 8.5  (continued)

The Inferred Map

One common comment about dynamical systems is that it is a descriptive tech-
nique. So far, that is exactly what we have done—described the patterns of the
data through time. However, computational modeling approaches are inferential
ones. Within the context of our example, we want to generate the population-
level equations that depict the changes of heart rate through time. We want to
come to an understanding of the higher-order patterns within the data and how
these patterns connect to the ideas of co-regulation. Graphically, this corresponds
to generating an inferential version of Figure 8.5d. That is, we can extract an
inferred map from the data by using equations that conform to the types of topo-
graphical features.
From an inferential viewpoint, there are four common stable topographical
features we can observe from two first-order equations (more on this shortly)
with the addition of some interesting combinations (Abraham & Shaw, 1992;
Butner et  al., 2014; Kugler & Turvey, 1987). This includes the two already
depicted (attractor and repeller), but now in two-dimensional logic (husband
 Mapping Co-Regulation in Social Relations  157

and wife simultaneously). An attractor is the shared homeostatic combination of

BPMs the couple would congregate at in time. This is one possible description
for the peaks we observe in the PDF (or a valley in the phase portrait, since the
PDF and phase portraits are inverses of one another), with the steepness as an
indication of how quickly one moves towards the homeostatic point, but also
the stability of the homeostatic value as it is assumed to constantly be undergoing
perturbations (Figure 8.6a).

FIGURE 8.6   xample topographical features represented as vector plots and time
E
series.
158  Butner, Behrends, and Baucom

FIGURE 8.6  (continued)

A repeller is a case where there is a combination of heart rate values from

which the couple moves away in time (Figure 8.6b). It is a hill in a phase portrait
or the lack of data in a PDF, and thus would be represented by a lack of values
at or around it. Repellers are actually a case of instability rather than stability,
as they have a shared point where the couple can hypothetically sit. However,
the moment there is a perturbation, the couple will move away from that point
in time. The steepness of the repeller represents the speed at which one would
move away.
A saddle is a repeller in one direction and an attractor in another. In topo-
graphical terms, it is a ridgeline with the combinatory point of BPMs being the
location of the equivalent of the mountain pass (Figure 8.6c). Saddles often depict
the separation of different basins of attraction, where being on one side of the
saddle depicts a tendency to go to one attractive feature, while being on the other
side depicts going to a different one.
The last common topographical feature is known in systems theory as the limit
cycle (Figure 8.6d). It represents oscillations or, in map terminology, a looped
trail. Limit cycles capture the angular motion around a combination of BPM
values.
It is also possible to combine angular motion with the other various topo-
graphical features. For instance, one can have a cyclical attractor where there is
a homeostatic value the couple moves towards in time, but in a cyclical manner.
This angular motion is of particular interest for investigating social relations, as
it represents the coupling influence between partners—the notion of coupling
itself. Thus, cyclical properties suggest a push-pull relationship that emerges. This
push-pull relationship is the idea of interpersonal co-regulation.

Estimating the Map

A combination of several statistical techniques can be used to describe these
push-pull relationships. First and foremost, it is necessary to quantify change
 Mapping Co-Regulation in Social Relations  159

over time. There are several methods for achieving this aim (e.g., McArdle,
2009); here we propose the use of a simple difference score that is calculated
by subtracting the previous value of the time series from the current value of
the time series. If this quantity is positive, the time series has gone up in abso-
lute value, and conversely, if the quantity is negative, the time series has gone
down in absolute value. Second, we want to describe the associations between
change from one measurement point to the next (the difference score) and the
absolute value of the previous time point. As with quantifying change, there are
numerous methods for modeling this association (e.g., Butner, Berg, Baucom, &
Wiebe, 2014); here we propose using Structural Equation Modeling (SEM),
because SEM allows us to specify equations for both members of the couple
simultaneously and in a straightforward fashion. Finally, we want to allow for
the possibility that the system may be best described by one topographic feature
or a set of topographic features (e.g., two attractors and a repeller). We propose
that mixture modeling can be used to conduct an empirical test for the exist-
ence of qualitatively different features, as described by the equations for both
members of the couple that are parameterized using SEM. Qualitatively differ-
ent topographic features emerge as different latent classes in a mixture modeling
approach. Below we provide a detailed discussion of how to accomplish each of
these modeling steps.
All of the features described thus far require a pair of first-order coupled equa-
tions. By “first-order,” we mean that the equations capture some notion of the
first derivatives within them. In this particular case, we will build two first-order
difference equations where the differences are the change in BPM over time for
the husband and wife respectively. Husband and wife values of BPM are then
allowed to predict both differences. There are some distinctions between building
two first-order difference equations versus two first-order differential equations.
Furthermore, many approaches focus on second-order equations instead (where
the second derivatives are also included). These alternative cases are discussed
near the end of the chapter.
Attractors, repellers, saddles, limit cycles, and combinations of these can all be
directly extrapolated by a pair of first-order difference/differential equations.

(BPMHt+τ − BPMHt) = b0 + b1BPMHt + b2BPMWt + eHt

(BPMWt+ τ − BPMWt) = b3 + b4BPMHt + b5BPMWt + eWt (8.1)

Tau (τ) is some fixed distance in time. Let us focus on b1 and b5 respectively.
They represent the homeostatic nature of the topographical feature in Husband/
Wife space. To understand why this is the case, let us first consider a single equa-
tion and its graphical equivalent.

(BPMHt+τ − BPMHt) = b0 + b1BPMHt + eHt(8.2)

160  Butner, Behrends, and Baucom

Equation 8.2 represents a one-predictor linear regression where change is the out-
come and the value of what constitutes that change is the only predictor. Graphically,
we can think about this as a scatterplot with change on the y-axis and BPM on the
x-axis, as illustrated in Figure 8.7. Negative coefficients like the one in Figure 8.7
are indicative of attractors. To understand why this is, we must identify the set point
and then ask what changes we would expect in BPM if one were currently above
the set point (to the right of the set point on the x-axis) as opposed to below the set
point (to the left of the set point on the x-axis). The set point is the point at which
change would be zero, where our regression slope crosses the horizontal axis in
Figure 8.7.When BPM is higher than the set point (to the right), the regression pre-
dicts negative change—a decline in BPM.When BPM is lower than the set point (to
the left), the regression predicts positive change—an increase in BPM. Since time is
implicit here, we can start to imagine what would happen as our BPM is constantly
being updated.We converge on the set point.This is the behavior of an attractor.
Returning to Equation 8.1, this kind of relationship is occurring for both
husband and wife BPM simultaneously through b1 and b5. If both are negative,
then both represent attraction towards a shared set point. Two positive coef-
ficients indicate a repeller where, over time, one would diverge from the shared
set point. One positive and one negative indicate a saddle.

FIGURE 8.7  Scatterplot of change in husband BPM and change in wife BPM.
 Mapping Co-Regulation in Social Relations  161

B2 and b4 are the coupling coefficients. They represent how previous values
of the partner predict the other partner’s future changes. Thus, they capture the
swirling patterns within the couple. A true limit cycle will result in b1 and b5
being zero with substantial coupling relationships. That is, a true limit cycle never
moves closer to the set point which would be captured by b1 and b5. It merely
moves around it. Finding a limit cycle would be akin to saying that all detectable
co-regulation was interpersonal.
Latent Change Score (LCS) Modeling (McArdle, 2009) is a method in SEM
for estimating a pair of simultaneous difference score equations. LCS consists of
a series of dummy latent variables with constrained parameters so that all that is
estimated are the parameters that correspond to Equation 8.1. To see how this
works, consider a regression where earlier values in BPM are allowed to predict
future values of BPM with an intercept of zero and a regression slope of 1. The
residual from this regression ends up being the difference of the future value (the
observed value) minus the previous value (the predicted value). LCS constrains
these parameters and then uses a latent variable to capture the residual, repre-
senting a discrete change. Additional relationships are then estimated, treating
the discrete change latent variables as the outcomes from the earlier values, in
our case capturing Equation 8.1. LCS models can be expanded to include some
degree of data smoothing, and can be applied to panel data, but also used as a time
series approach, as we will do so here.
To convert LCS into a time series approach, one must structure the data into
what has been called a Toeplitz or time delay data structure. We illustrate this
in Table 8.1 such that the data are converted into a current and lead value for
husband and wife simultaneously (τ is fixed to 1 here). To smooth the data, one
need merely create additional lead and lag variables using the same value of τ and
extend the model equating like parameters over time.
The equations in Equation 8.1 are only capable of capturing a single top-
ographical feature. However, the PDF implies that there could be several
topographical features in the data, different locations of data density. Equation 8.1 is
known as a linear dynamic model, in that the predictors depict a linear association
with change (the outcomes). To capture more than one topographical feature or
changing topographical features, one must instead use nonlinear equation forms.
There are several ways to do this, including polynomials and interactions. Here
we utilize mixture modeling.

TABLE 8.1  Toeplitz data structure for husband and wife BPM.

Husband at T Husband at T+1 Wife at T Wife at T+1

BPMH1 BPMH2 BPMW1 BPMW2

BPMH2 BPMH3 BPMW2 BPMW3
BPMH3 BPMH4 BPMW3 BPMW4
BPMH4 BPMH5 BPMW4 BPMW5
162  Butner, Behrends, and Baucom

Mixture modeling is an analytic procedure where the PDF is estimated as the

cumulative function of a series of equations, rather than just one. That is, we will
estimate a series of equations, where the set of equations as a whole, weighted
by their commonality (i.e., the degree of consistency across the equations),
reproduce the entire PDF. Bauer (2007) criticized mixture modeling as only
really being able to reconstruct the PDF, and therefore being based on flawed
assumptions for common statistical goals. Given the relationship between the
PDF, the phase portrait, and the change equations we are utilizing, this criticism
is a strength here, in that the approach specifically focuses on the topographical
features of the data.
Mixture modeling can be used as an exploratory or confirmatory technique,
and has been incorporated into current versions of Mplus (Lanza, Flaherty, &
Collins, 2003). Fit indices, for example, can be used to determine the number
of latent classes (or groups of data vectors described by different equations, in
our case). Simulation work has shown that the Bayesian Information Criterion
(BIC) tends to minimize at the proper number of classes (Nylund, Asparouhov, &
Muthén, 2007). Furthermore, the Vuong Lo Mendell Rubin Likelihood ratio
(VLMR), Lo Mendell Rubin adjusted Likelihood ratio (aVLMR—better in
small samples), and the bootstrapped Likelihood ratio (BLRT) all test to see
if the current number of classes adds something above and beyond having one
less class (Henson, Reise, & Kim, 2007). Thus one can determine the number
of classes by iteratively increasing the number of classes until no gain in predic-
tion has been identified and settling on one less class (in the case of the fit index
approach).
Metaphorically, mixture modeling can be thought of as both a categori-
cal latent variable and a multiple group analysis where assignment to group is
unknown. Thus, anything that can be allowed to differ across groups or in rela-
tion to a latent construct in SEM can be used to distinguish the different classes.
In this case, we only allow parameters that would distinguish different types of
topographical features that differ across classes (i.e., different strengths of those
features and different locations of the features on the map). Figure 8.8 shows a
complete path diagram of the model, with all the parameters allowed to vary
across classes in black. Code for this procedure can be found in the appendix to
this chapter.

Results
The combination of fit indices all indicated that three classes was an improvement
over two classes (VLMR = 48.499, p = .02; aVLMR=172.588, p = .02), but four
classes did not improve data description over three classes (VLMR = 26.947,
p = .11; aVLMR = 103.898, p = .11). We therefore report the three-class solution.
Figures 8.9a, 8.9b, and 8.9c show three inferred vector plots generated in the
Apple Inc. program Grapher.
 BPM WT BPM WT+1

∆W

1
∆H

BPM HT BPM HT+1

FIGURE 8.8   ath model diagram of the bivariate LCS model of husband and wife
P
BPM. Dashed arrows indicate that a parameter is fixed to 1, and black
arrows indicate that a parameter was allowed to differ across groups.

(a)
y

125

100

75

50 x
50 60 70 80 90 100 110 120 130

FIGURE 8.9  Separate vector plots of the three class solution. Heat map scaling
is used to represent the rate of change. Darker shading (e.g., black)
indicate a faster rate of change, while lighter shading (e.g., grey)
indicate a slower rate of change.
 (b)

125

100

75

50 x
50 60 70 80 90 100 110 120 130

(c)
y

125

100

75

50 x
50 60 70 80 90 100 110 120 130

FIGURE 8.9  (continued)

 Mapping Co-Regulation in Social Relations  165

In addition, each figure includes trajectories inferred by the equations. Since

each class estimates a unique mean and standard deviation (variance) for husband
and wife BPM, we generated predicted trajectories extrapolating from simple
slopes logic from regression: one trajectory for each combination of means and
standard deviations above and below the means in BPM. We then used the
Runge-Kutta fourth-order algorithm for estimating the expected values several
steps into the future (all done within Grapher). This approach is particularly use-
ful in that it is able to also identify issues of nonstationarity. For example, if we
observe an attractor, what should normally occur under this procedure is that we
see the various starting points all converging on the attractor inside the cloud of
starting values. Under nonstationarity, the coefficients can indicate an attractor,
but when following this graphing procedure, the points will converge towards a
set point that is outside of the cloud of start values—essentially a series of growth
patterns. Figure 8.10 shows all three sets of inferred trajectories. Notice that we
have rebuilt the phase portrait, now with inferred trails.
Table 8.2 shows the table of coefficients (and 95% confidence intervals) for
the three different classes. Each of three topographical features are quite differ-
ent in terms of the coregulatory dynamics. The first feature accounts for 46%
of vectors. It is an attractor in that both of the coefficients of BPM predicting
its own changes were negative with 95% confidence intervals, excluding zero.

125

100

75

50 x
50 60 70 80 90 100 110 120 130

FIGURE 8.10  Combined vector plot of the three-class solution.

166  Butner, Behrends, and Baucom

Also notable, though, is that the coupling/crossover influences included zero

within the confidence intervals, suggesting that while the couple tended to con-
verge on a shared combination of heart rates, it was a pattern of intraindividual
regulation rather than interindividual. A better description was one of an emer-
gent dynamic of elevated heart rate for both individuals, such that both husband
and wife maintained their own elevated level.
The second topographical feature accounted for 35% of vectors. It represents
a lower combination of heart rates for the husband and wife, homeostatically.
It also includes non-zero coupling, in that the husband’s heart rate predicts
positive changes in the wife’s heart rate. This coupling relationship is consist-
ent with an interindividual relationship of co-regulation. Previous values of the
husband predict the changes we observe in the wife, and thus she seems sensi-
tive to both her own intraindividual pattern and the husband’s pattern. Notice
in Figure 8.10 that this coupling relationship is what generates the swirling
pattern in the topographical representation.
The third topographical feature accounted for 18% of vectors. Most inter-
esting is that only the wife’s own intraindividual effect excluded zero from the
confidence interval. The pattern of results suggested an attractor. But inferen-
tially, this is more a description of dysregulation where there did not tend to be
homeostatic responses, just a tendency for lower heart rates for the husband, but
not the wife.
We assessed the extent to which being together or apart was able to predict
class assignment. A dummy code was added to the mixture model to predict
class membership. However, adding this new variable can result in changing

TABLE 8.2  Parameter estimates and 95% CI for the three-class solution.

Class 1: High husband and 2: Mid heart rates 3: Low husband BPM
wife BPM

Husband 20.78 < 35.26* < 49.74 5.48 < 14.48* < 23.48 -26.42 < 9.19 < 44.81
intercept
Husband -0.48 < -0.38* < -0.28 -0.33 < -0.20* < -0.08 -0.37 < -0.17 < 0.03
own
Husband -0.12 < 0.007 < 0.14 -0.02 < 0.03 < 0.08 -0.25 < 0.03 < 0.32
< wife
Wife 23.54 < 33.01* < 42.48 -12.79 < -0.07 < 12.64 8.15 < 21.70* < 35.25
intercept
Wife -0.39 < -0.30* < -0.21 -0.27 < -0.18* < -0.10 -0.34 < -0.21* <
own -0.08
Wife < -0.09 < -0.04 < 0.00 0.021 < 0.20* < 0.37 -0.11 < -0.03 < 0.06
husband

Note: * p < 0.05.

 Mapping Co-Regulation in Social Relations  167

the observed results for the three topographical features. To ensure we were
predicting the three original classes, we included the posterior probabilities
of class assignment for each data point as training data in the new model that
then included the dummy code as a predictor of class assignment. This method
ensured replication of the original results while allowing for proper assessment of
the togetherness effect.
Togetherness only differentiated the third topographical feature from the
other two (1 from 2, Z = 1.566, p = .117; 1 from 3, Z = 2.655, p = .008; 2 from 3,
Z = 3.334, p = .001), in that being together decreased the likelihood of being in
the dysregulatory topographical feature. In essence, intrapersonal dysregulation
and interpersonal disconnection only occurred for the couple when they were
apart. When they were together, both the high-high attractor and the mid-mid
spiral attractor existed, possibly in a multistable arrangement where the couple
could switch between the two. Note that the couple were apart much more
frequently than being together, and thus all three states could be observed some-
times when they were apart.

Discussion
What does all this mean? Within a single couple in a single day, we see the full
range of co-regulatory patterns. When the spouses were together, their heart
rates tended to display one of two patterns. In this first pattern, change in each
spouse’s heart rate was governed by his or her own previous heart rate; this pat-
tern is consistent with intrapersonal regulation. In the second pattern, change in
each spouse’s heart rate was governed by his or her own previous heart rate, and
change in wife’s heart rate was also significantly associated with husband’s previ-
ous heart rate. This pattern is consistent with both intrapersonal regulation (both
spouses) and wife-to-husband co-regulation. Finally, when spouses were apart,
they were more likely to display a third pattern where change in wife’s heart
rate were associated with her own previous heart rate (intrapersonal regulation)
and change in husband’s heart rate was not significantly associated with his own
(dysregulation) or wife’s previous heart rate (disconnection).
While these findings must be interpreted with caution given that they are
exploratory and based on one day of heart rate data for one couple, several
aspects of the findings are consistent with existing research and substantive
theory. First, the differences in set points across the three patterns are consistent
with findings that spouses in distressed relationships show elevations in heart
rate when interacting with one another relative to when apart (e.g., Denton,
Burleson, Hobbs, Von Stein, & Rodriguez, 2001; Levenson & Gottman, 1985;
Nealey-Moore, Smith, Uchino, Hawkins, & Olson-Cerny, 2007; Smith et al.,
2009). More specifically, both spouses’ set point in the third class, which was
more likely to occur when spouses were separate, was lower than their set
168  Butner, Behrends, and Baucom

points in the other two classes. To the best of our knowledge, these find-
ings are the first evidence of this effect in daily life. Second, the variability
in within- and cross-partner associations between the two spouses and across
the three classes is consistent with a small but growing body of work dem-
onstrating that intrapersonal and co-regulatory patterns change depending on
circumstance (e.g., Helm et al., 2012; Helm et al., 2014). Lastly, the nature of
the co-regulatory effect in the second pattern is consistent with clinical theory.
The positive sign of the association between husband’s previous heart rate and
wife’s change in heart rate combined with the negative sign of the association
between wife’s previous heart rate and wife’s change in heart rate indicates
a compensatory influence of the two spouses’ previous heart rates on wife’s
change in heart rate. More specifically, wife’s heart rate returns to her set point
more slowly to the extent that husband’s heart rate is further away from his
own set point. Similar findings emerged in work examining treatment-seeking,
distressed spouses’ vocally encoded emotional arousal during relationship con-
flict. Prior to receiving couple therapy, to the extent that husbands were more
aroused, wives returned to their set points more slowly. However, after receiv-
ing couple therapy, wives’ rates of return to their set points was unrelated to
husbands’ levels of arousal (Baucom & Atkins, 2012). These findings are all
consistent with numerous theoretical models of romantic relationships, which
posit that relationship distress tends to co-occur with increased difficulty with
emotion regulation (for a review, see Baucom et al., 2015).

Issues to Consider within a Dynamical Systems Approach

We set out to illustrate an approach that embraced the complexity of co-regulation
while also aiming for as simple a description as we thought necessary for the
phenomena. This required several considerations along the way. Some of these
decisions may be different for other co-regulatory phenomena. We therefore
explicitly address these considerations in the following section.

Linear and Nonlinear Equations

In the broader literature of DSM, the distinction between linear and nonlinear
dynamics is paramount. One aspect of the distinction that warrants careful exami-
nation is that linear dynamic models are capable of describing nonlinear patterns
through time (Butner et al., 2014; Boker & Laurenceau, 2006). So we specifically
stipulate that a nonlinear dynamic model includes nonlinear terms such as poly-
nomials and interactions on the predictor side of the equation, or some approach
like the mixture modeling one presented here, which uses multiple linear terms
to reconstruct the nonlinear. The addition of these terms allows the equation to
capture multiple set points. This addition also allows for differences in change
rates across individuals—something that is not feasible in linear models.
 Mapping Co-Regulation in Social Relations  169

Linear dynamic models are able to capture just one set point, and assume that
the rate of change is constant. The benefit of using linear models is that they
can be easier to test and interpret. However, it is probably safe to assume that
most systems, including co-regulatory systems, are not linear. These models, for
example, are not sensitive to initial conditions (where a particular individual starts
out), and require that every person have the same set point. Thus, they lack the
descriptive and predictive power of nonlinear models.
There are multiple ways to create nonlinear models (Butner et al., 2014). Our
use of mixture modeling did exactly that by capitalizing on the notion that even
in nonlinear models, they are locally linear. That is, near the set points, system
dynamics can be captured by a linear dynamic model. The mixture modeling
approach then allowed for multiple linear dynamic models that then combine
together to reconstruct the nonlinear form. This approach has the added benefit
of the ease of interpretation from linear dynamic models while also allowing for
the additional descriptive precision and flexibility of nonlinear models.
Another method of using nonlinear terms is to add polynomials and/or other
interactions to the predictor side of the equations. This approach creates nonlin-
ear dynamic models that are capable of capturing multiple topographical features
(as stated above). The number of features that are captured in these models is a
function of the polynomial form included in the model. For example, a squared
predictor is capable of capturing two set points. A cubic predictor is capable of
capturing one to three set points. A quadratic predictor is capable of capturing
four set points.
Importantly, nonlinear models allow for multistability (the ability to have
multiple change patterns), differences in topographical feature stability (the likeli-
hood of being in one topographical features versus another one), and the ability
to model control parameters (variables that alter the topographical features).
Importantly, one of the losses from using this mixture modeling approach is in
our ability to distinguish when a system is multistable as opposed to being subject
to different dynamics under different levels of a control parameter. This issue is
discussed in more detail later in this chapter.

Orders of Equations
As mentioned earlier, DSMs can be constructed with either first- or second-order
equations. Our example case utilized two first-order coupled difference equa-
tions. A first-order difference equation models a form of velocity of a variable
(the change in position from one time point to the next; in this case, a difference
in consecutive values). In the example case, the regression equation outcome
was husband and wife’s change in heart rate (velocity). First-order equations can
typically only capture attractors, repellers, and saddles. Due to the coupling link
in Equation 8.1 (b2 and b4), the equation can also capture swirling patterns in the
state space (such as a spiral attractor or repeller).
170  Butner, Behrends, and Baucom

Second-order equations model acceleration, or the change in velocity. While

a pair of first-order models is capable of generating oscillations through limit
cycles (a function of the coupling relationships; the swirling pattern), second-
order equations assume oscillations exist. Coupling can still exist among two
second-order equations, but the coupling is about how two oscillatory patterns
push and pull one another rather than two values.
Consider the hypothetical couple Sam and Jordan from the beginning of the
chapter. It is possible that once a particular physiological threshold is reached,
Sam’s distress level increases much more rapidly than it was increasing before.
Similarly, the rate of change in Sam’s distress level could suddenly decrease once
a particular threshold is reached. Each of these changes in velocities could be a
function of intrapersonal dynamics. That is, Sam fluctuates irrespective of Jordan.
Jordan’s influence then may be in terms of where in the cycle Sam currently exists,
rather than whether Sam is currently higher or lower at a particular moment.
The decision of whether to use first- or second-order equations should
be guided by theory. Under an expectation of oscillations, the question is
whether you expect the interaction between Sam and Jordan to generate the
oscillations, or if Sam and Jordan would oscillate anyway and their relation-
ship is reflected in the linkages between the oscillations. In the first case,
first-order modeling is most appropriate because it constructs the oscillations
as a function of their interactions. In the second, each is expected to oscil-
late even if they never influence one another, and second-order models are
therefore most appropriate.

Time
Time is implicit within dynamical systems modeling (Butner et  al., 2014).
The focus in DSM is not as much on an individual’s linear movement from
point a to point b, but rather the dynamics of how the system as a whole
evolves over time. What DSM seeks to identify are the shifting patterns
and points of transition within a system. In other words, knowing where
an individual would end up if they were placed on a random location on a
topographical map. Rather than using time as a predictor of change, DSMs
use time to describe the change.
When utilizing DSMs, it is important to understand the difference between
discrete and continuous conceptualizations of time. A discrete conceptualization
requires that time be divisible, but really only in the division we observe. In this
sense, it is possible to break up time into meaningful individual segments while
allowing for the possibility that other temporal segments might generate different
dynamics. This is the approach used in our example analyses. Under the con-
tinuous conceptualization of time, any segment is plausible. That is, the timing
of the measurements themselves is essentially arbitrary, and can be meaningfully
thought of as representing a temporal sequence where the segments between
 Mapping Co-Regulation in Social Relations  171

measurements are infinitely small (the definition of a derivative is a difference in

the observed variable when the segment of time elapsed, in this case, approaches
zero). This distinction is important to understand in order to avoid confusion
regarding measurement periods and the actual unfolding of the phenomenon.
Issues related to time also crop up at several other points while constructing
a DSM. In the earlier stages, it is critical to identify the time scale at which the
phenomenon of interest is operating. This identification will inform your meas-
urement period. If a shorter time scale than the actual phenomenon is measured,
the local dynamics are likely to be overemphasized and the larger scaled pattern-
ing could be difficult to observe. For example, if a particular variable repeats a
particular pattern every five days and measurement is only collected for five days,
then all that could be observed is a stable value. It would be difficult to see that
it actually is repeating over time. In this case, the local dynamics are obscuring
the larger, emergent dynamics. When a phenomenon’s time scale is unknown,
it is preferable to oversample than to undersample. This dense scale of measure-
ment is unlikely to affect our results, and it would still be possible to observe the
larger-scale dynamics.
This logic can be extended to the co-regulation models. If measurement only
occurs while couples are together, the coupling link may be over- or underem-
phasized. It is necessary to identify where an individual’s homeostatic set point is
located when together and apart from their partner in order to properly observe
the coupling phenomenon.
Time is also important when constructing leads and lags (Butner et al., 2014).
This process is also referred to as creating a time delayed embedding space (Boker &
Laurenceau, 2006). The smallest plausible lag should be selected, because longer
lags could obscure shorter-term dynamics. The example case used a lag of 1,
which is typically used in DSMs. Longer lag values can be problematic with miss-
ing data, as each missing data point will generate several additional missing points.
However, longer lags can be beneficial, in that two consecutive measures can be
highly correlated (autocorrelated) because not enough time has passed to observe
change. This form of autocorrelation can mask the system dynamics and tends to
linearize system models overall. Thus, we want a small lag, but the smallest plausi-
ble one where change will be observed, rather than just the smallest.

Moderation and Multistability

Under DSMs, control parameters can be conceptualized as moderators (Butner
et al., 2014; Thelen & Smith, 2008). Control parameters are variables that change
the existing topology. In the example case, the together/apart dummy code was
a control parameter. This is simply a variable that moves individuals through a
phase transition (changing from one set point to another). Without this variable,
it would not be possible to determine when the trajectories were more strongly
associated with each of the three attractors.
172  Butner, Behrends, and Baucom

Consider the three attractors found in our example couple. The presence
of the high-high attractor and mid-mid spiral attractor indicate the possibility
that multistability is present. In this scenario, individuals can switch between
these two features. However, when the couple were together, we were more
likely to observe the third attractor (and decrease the probability of being in the
dsyregulatory feature).
It is important to note that control parameters are not considered to be causal
variables (Thelen & Smith, 2008). Instead, they are the points of transition from
one particular pattern to another. The goal of identifying a control parameter
is to explain how, when, and why people move between different states (such
as elevated heart rate). Essentially, you are looking for variables that predict the
system’s attractors. In our analysis, we would not claim that the couple being
together prevented the individuals from dysregulation, but rather this physical
proximity increases the likelihood of transitioning to another stable pattern (the
third attractor).
In co-regulation, DSMs also have to allow for differences in homeostatic set
points and strength of the coupling link. Each individual is a separate system.
Co-regulation arises from these systems being coupled together in some way.
However, this coupling link may not affect each partner the same way. It may
exert more force on one partner than the other. Recall the hypothetical couple,
Sam and Jordan. In their argument, Sam is more affected by Jordan’s distress than
Jordan is affected by Sam’s distress. Thus the link between the two systems is
stronger for Sam than for Jordan.

Conclusion
The need for ways to approach complexity that both embrace the complexity but
also keep the description relatively simple is paramount. As co-regulation contin-
ues to gain growing attention and social scientists increase their ability to collect
rich time-dependent data, approaches like the one described here are going to
be in demand. DSM models, under thoughtful application, can do just that. And
when that careful consideration allows for nonlinear systems approaches that
clearly map onto theory, they elevate the science. Ultimately, they provide a step
towards a new stage of social sciences.

Appendix
TITLE: Mixture Model of Bivariate Latent Change BPM

DATA:

! enter the name of the data set

FILE = couple6 day2.csv;
 Mapping Co-Regulation in Social Relations  173

VARIABLE:

! enter the names of the variables in the data set

NAMES = bpm_w1 bpm_w2 bpm_h1 bpm_h2 together;
usevariables = bpm_w1 bpm_w2 bpm_h1 bpm_h2;
missing is all.;
classes = node(3);
!Auxiliary with r3step specifies to use training data without
togetherness
! auxiliary = together(R3STEP);
! idvariable is time;

Analysis:

Type is Mixture;
!Random start values to help ensure solution is not just
local maxima
starts = 2500 500;
processors=4;

!optseed overrides random starts to only run the solution

with this seed
!value, recommend identify ideal seed value and then add
in the tech11
!and tech14 commands
!optseed = 907342;!for one node;
!optseed = 469158;!for two nodes;
optseed = 496016;!for three nodes;
!optseed = 346843;!for four nodes;

MODEL:

%Overall%
!Construct a latent representation of each variable
Lbpm_w2 by bpm_w2@1;
Lbpm_w1 by bpm_w1@1;
Lbpm_h2 by bpm_h2@1;
Lbpm_h1 by bpm_h1@1;

!Fix intercepts and variances of observed variables to zero

!Note that when building with more leads and lags & smooth-
ing, can estimate
!variances as error variances (usually equated)
bpm_w1@0 bpm_h1@0 bpm_w2@0 bpm_h2@0;
[bpm_w1@0 bpm_h1@0 bpm_w2@0 bpm_h2@0];
174  Butner, Behrends, and Baucom

!Perfect regression of Future = Past at a slope of 1 and

intercept of zero
Lbpm_w2 on Lbpm_w1@1;
Lbpm_h2 on Lbpm_h1@1;
Dbpm_w by Lbpm_w2@1;
Dbpm_h by Lbpm_h2@1;
lbpm_w2@0 lbpm_h2@0;
[lbpm_w2@0 lbpm_h2@0];

!Build our change variables grabbing what remains

Dbpm_w on lbpm_h1 lbpm_w1;
Dbpm_h on lbpm_h1 lbpm_w1;

Dbpm_w Dbpm_h;
[Dbpm_w Dbpm_h];
lbpm_w1 lbpm_h1;
[lbpm_w1 lbpm_h1];

%Node#1%
!Regression slopes
Dbpm_w on lbpm_h1 lbpm_w1;
Dbpm_h on lbpm_h1 lbpm_w1;

!Variances and means of BPM by class

Dbpm_w Dbpm_h;
[Dbpm_w Dbpm_h];
lbpm_w1 lbpm_h1;
[lbpm_w1 lbpm_h1];

%Node#2%
Dbpm_w on lbpm_h1 lbpm_w1;
Dbpm_h on lbpm_h1 lbpm_w1;

Dbpm_w Dbpm_h;
[Dbpm_w Dbpm_h];
lbpm_w1 lbpm_h1;
[lbpm_w1 lbpm_h1];

%Node#3%
Dbpm_w on lbpm_h1 lbpm_w1;
Dbpm_h on lbpm_h1 lbpm_w1;

Dbpm_w Dbpm_h;
[Dbpm_w Dbpm_h];
 Mapping Co-Regulation in Social Relations  175

lbpm_w1 lbpm_h1;
[lbpm_w1 lbpm_h1];

!%Node#4%
!Dbpm_w on lbpm_h1 lbpm_w1;
!Dbpm_h on lbpm_h1 lbpm_w1;

!Dbpm_w Dbpm_h;
![Dbpm_w Dbpm_h];
!lbpm_w1 lbpm_h1;
![lbpm_w1 lbpm_h1];

!tech11 for the vlmr and tech14 for the BLRT, both are com-
puter intensive and
!recommend identifying solution first
!Output:cinterval tech11; !tech14;

!Can output the posterior probabilities for each data vec-

tor for further
!examination

!Savedata:
!save=Cprob;
!file = ‘cprobc5d4.txt’;

Note
1 Dynamical systems theory provides ways in which a couple-level phenomenon such as
co-regulation could be examined from a single individual. This is a direct extrapolation
of Takens’ (1981) theorem into a measurement perspective.

References
Abraham, R. H., & Shaw, C. D. (1992). Dynamics: The geometry of behavior. Boston, MA:
Addison-Wesley.
Baucom, B. R., & Atkins, D. C. (2012). Polarization in marriage. In M. Fine and F. Fincham
(Eds.), Families theories: A content-based approach (pp. 145–166). New York: Routledge.
Baucom, B. R., Dickenson, J. A., Atkins, D. C., Baucom, D. H., Fischer, M.S., Weusthoff, S.,
Hahlweg, K., & Zimmermann, T. (2015). The interpersonal process model of demand/
withdraw behavior. Journal of Family Psychology, 29, 80–90.
Baucom, B. R., Sheng, E., Christensen, A., Georgiou, P. G., Narayanan, S. S., & Atkins, D. C.
(2015). Behaviorally-based couple therapies reduce emotional arousal during couple
conflict. Behaviour Research and Therapy, 72, 49–55.
Bauer, D. J. (2007). Observations on the use of growth mixture models in psychological
research. Multivariate Behavioral Research, 42, 757–786.
176  Butner, Behrends, and Baucom

Boker, S. M., & Laurenceau, J. P. (2006). Dynamical systems modeling: An appli-

cation to the regulation of intimacy and disclosure in marriage. In T. A. Walls &
J. L. Schafer (Eds.), Models for intensive longitudinal data (pp. 195–218). New York:
Oxford University Press.
Bombeck, E. (1987). Family:The ties that bind . . . and gag. New York: Random House.
Butler, E. A., & Randall, A. K. (2013). Emotional coregulation in close relationships.
Emotion Review, 5, 202–210.
Butner, J. E., Berg, C. A., Baucom, B. R., & Wiebe, D. J. (2014). Modeling coordina-
tion in multiple simultaneous latent difference scores. Multivariate Behavioral Research,
49(6), 554–570.
Butner, J., Diamond, L., & Hicks, A. (2007). Attachment style and two forms of emotion
co-regulation between romantic partners. Personal Relationships, 14, 431–455.
Butner, J., Gagnon, K. T., Geuss, M. N., Lessard, D. A., & Story, T. N. (2014). Using
topology to generate and test theories of change. Psychological Methods, 20, 1–25.
Denton, W. H., Burleson, B. R., Hobbs, B. V., Von Stein, M., & Rodriguez, C. P.
(2001). Cardiovascular reactivity and initiate/avoid patterns of marital communica-
tion: A test of Gottman’s psychophysiologic model of marital interaction. Journal of
Behavioral Medicine, 24, 401–421.
Funk, J. L., & Rogge, R. D. (2007). Testing the ruler with item response theory: Increasing
precision of measurement for relationship satisfaction with the Couples Satisfaction
Index. Journal of Family Psychology, 21, 572.
Gottman, J. M., Murray, J. D., Swanson, C. C., Tyson, R., & Swanson, K. R. (2002). The
mathematics of marriage. Cambridge, MA: MIT Press.
Guastello, S. J. (2011). Discontinuities: SETAR and catastrophe models with polynomial
regression. In S. J. Guastello & R. A. M. Gregson (Eds.), Nonlinear dynamical systems
analysis for the behavioral sciences using real data (pp. 251–280). Boca Raton, FL: CRC
Press.
Helm, J. L., Sbarra, D., & Ferrer, E. (2012). Assessing cross-partner associations in physi-
ological responses via coupled oscillator models. Emotion, 12, 748.
Helm, J. L., Sbarra, D. A., & Ferrer, E. (2014). Coregulation of respiratory sinus arrhyth-
mia in adult romantic partners. Emotion, 14, 522.
Henson, J. M, Reise, S. P., & Kim, K. H. (2007). Detecting mixtures from structural
model differences using latent variable mixture modeling: A comparison of relative
model fit statistics. Structural Equation Modeling, 14, 202–226.
Kugler, P. N., & Turvey, M. T. (1987). Information, natural law and the self-assembly of
rhythmic movement: Theoretical and experimental investigations. Hillsdale, NJ: Lawrence
Erlbaum.
Lansing, A. H., & Berg, C. A. (2014). Topical review: Adolescent self-regulation as a
foundation for chronic illness self-management. Journal of Pediatric Psychology, 39(10):
1091–1096.
Lanza, S. T., Flaherty, B. P., & Collins, L. M. (2003). Latent class and latent transition
analysis. In J. Schinka, W. Velicer, & I. Weiner (Eds.), Handbook of Psychology (Vol. 2,
pp. 319–325). Hoboken, NJ: John Wiley & Sons.
Levenson, R. W., & Gottman, J. M. (1985). Physiological and affective predictors
of change in relationship satisfaction. Journal of Personality and Social Psychology,
49, 85.
McArdle, J. J. (2009). Latent variable modeling of longitudinal data. Annual Review of
Psychology, 60, 577–605.
 Mapping Co-Regulation in Social Relations  177

Molenaar, P. C., & Campbell, C. G. (2009). The new person-specific paradigm in

psychology. Current Directions in Psychological Science, 18(2), 112–117.
Nealey-Moore, J. B., Smith, T. W., Uchino, B. N., Hawkins, M. W., & Olson-Cerny, C.
(2007). Cardiovascular reactivity during positive and negative marital interactions.
Journal of Behavioral Medicine, 30, 505–519. doi:10.1007/s10865-007-9124-5
Nowak, A. (2004). Dynamical minimalism: Why less is more in psychology. Personality
and Social Psychology Review, 8(2), 183–192.
Nowak, A., & Lewenstein, M. (1994). Dynamical systems: A tool for social psychology? In
R. R. Vallacher & A. Nowak, (Eds.), Dynamical systems in social psychology. San Diego,
CA: Academic Press.
Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of
classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation
study. Structural Equation Modeling, 14(4), 535–569.
Prigogine, I. N. G. (1977). Self-organization in nonequilibrium systems. Hoboken, NJ: John
Wiley & Sons.
Riley, M. A., & Turvey, M. T. (2002). Variability and determinism in motor behavior.
Journal of Motor Behavior, 34, 99–125.
Sbarra, D. A., & Hazan, C. (2008). Coregulation, dysregulation, self-regulation:
An integrative analysis and empirical agenda for understanding adult attach-
ment, separation, loss, and recovery. Personality and Social Psychology Review, 12,
141–167.
Smith, T. W., Uchino, B. N., Berg, C. A., Florsheim, P., Pearce, G., Hawkins, M.,
Henry, N. J. M., Beveridge, R. M., Skinner, M. A., Ko, K. J., & Olsen-Cerny,
C. (2009). Conflict and collaboration in middle-aged and older couples. II:
Cardiovascular reactivity during marital interaction. Psychology and Aging, 24(2),
274–286.
Takens, F. (1981). Detecting strange attractors in turbulence. In D. A. Rand and
L.-S. Young (Eds.), Dynamical systems and turbulence (lecture notes in mathematics)
(vol. 898, pp. 366–381). New York: Springer-Verlag.
Thelen, E., & Smith, L. B. (2008). Dynamic systems theories. In R. M. Lerner (Ed.),
Handbook of Child Psychology, Vol. 1: Theoretical models of human development (5th ed.,
pp. 563–634). Hoboken, NJ: John Wiley & Sons.
Vallacher, R. R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 8(2), 73–99.
9
DYNAMICS OF SYNCHRONY
AND JOINT ACTION
Kerry L. Marsh1

This chapter is concerned with social connection as it is manifested in unity states

that individuals form with others, as a consequence of each responding to the
other’s perceptuo-motoric behavior. How do coordinated social units rapidly
come into being, operate for some time—brief or relatively lasting—and dissolve
as rapidly? How does responsiveness to others who are nearby yield coordinated
states, even when there is no overt goal? When there is a goal, how do people
cooperate to complete an action? This chapter presents an embodied account
in which coordination and cooperation are defined not by between-person
harmony in verbal or cognitive processes—something that could play out via
asynchronous media communication, for instance. Rather, they are approached
as states that involve harmony at a physical level—something that could occur
on the football pitch, for instance. In this view, coordination and cooperation
involve congruencies in physical movement states: synchronous movement, or
other coordinated or complementary movements among people who have visual
systems and other senses, are in continual motion over time, and are co-present
in some shared physical setting. The assumption in this chapter is that progress
toward a computational and dynamical social psychology not only requires
that social psychologists substantially rethink data collection and data analytic
approaches, but make substantial theoretical advances as well.
The more traditional view of social connection phenomena views unity states
as operating in opposition to individual autonomy, as involving individual sacri-
fice in order to be a part of a whole. The challenge from that view is understanding
why/when individuals choose to cooperate rather than engage in competition or
self-interested action. Such strategic processes can be entirely virtual; they do not
necessarily involve physical movement or physical co-presence. They are also
defined by cognitive activity, and they require the heavy mental machinations
 Dynamics of Synchrony and Joint Action  179

of active “mind reading.” Researchers taking such a perspective are considerably

benefitted by the fact that social cognitive psychology has an excellent under-
standing of the science of mind reading. From the attribution era onward, we
have five decades of knowledge about how people make inferences about others’
intent and decide whether to trust that others will be cooperative or will instead
choose immediate self-interest, “defecting” from cooperation. We have well-
developed and nicely tested economic theoretic approaches to “playing such
games” that require sophisticated, high-level mentalizing processes so essential
for this kind of approach to cooperation and coordination.
I suggest in this chapter an alternative view to understanding cooperation
and the pull to connect and be a social unit with others: one that does not
require competition as the backdrop for cooperation, and thus does not posit
that mind reading is crucial. One danger of an account that is almost exclusively
grounded on mentalizing processes is that it leads to neglecting the surround, that
which is outside our brains (Mace, 1977; Richardson, Marsh, & Schmidt, 2010),
in determining sociality and connection processes. It neglects to consider the
influence of the social world we evolved in, the physical environmental context
that provided the embedding context for evolving humans. We are not merely
attribution-computing devices, but animate and intentional creatures who
respond to other animals within a shared grounding context. This primal context
includes the co-presence of other conspecifics in situations that unfold and
change over time and make pressing situational demands on us. We are equipped
by evolution to handle such demands—not solely because we can step back from
it and mentally analyze it. Rather, we are on the ground, actively engaged from a
first-person perspective in exploring that environment and those in it, by moving
our eyes, engaging other sensory systems, and moving our bodies. The premise of
this chapter is that understanding social connection in its most basic and diverse
forms requires a theoretical approach that is social-ecological2—grounded in the
social, physical and objectively real environments that surround behavior, and is
dynamic—involving ongoing regulation of behavior in response to the unfolding
information detected from our social world. Such a dynamic account will neces-
sarily require data that preserve rich temporal information, and analytic strategies
that can handle the more entangled, nonlinear nature of underlying causal forces
at operation (Richardson, Dale, & Marsh, 2014). This chapter, however, focuses
on advancing theory.
The first section of this chapter, “Minimalist Social Connection: The Emergence
of ‘Form’ or Synergies,” focuses on emergence of basic units, where the game’s
ultimate purpose is not the economic outcome, but the playing of the game
(the mere engagement with another individual), if you will. The underlying
theoretical issues about causality essential for understanding these processes are
described in the subsection “Theoretical Underpinnings: Self-Organization.”
The subsection “Empirical: Interpersonal Synchrony” discusses research that
illustrates these principles—namely, experiments on interpersonal synchrony.
180  Kerry L. Marsh

These subsections of the chapter examine sociality in terms of the minimal

conditions necessary for the emergence of the merest of social unity states.
The second section, “Connecting through Cooperation,” addresses social con-
nection emergence in the context of goal-directed action, specifically joint
action. Before discussing an empirical way to study joint action from an embod-
ied and ecological framework, here too, considerable theoretical groundwork is
needed because the principles are widely unfamiliar or the terms widely misun-
derstood.3 The subsection “Perceptual Processes” discusses perceptual principles
in depth, particularly the concept of affordances (i.e., what opportunities for
agency an environment provides). The subsection “Empirical Work on Affording
Cooperation” then discusses application of these principles in research on goal-
directed cooperative action.
Finally, the section “Future Directions: Affordances of Behavior Settings”
addresses the most neglected factor in understanding the emergence of social
connection states: the role of the physical setting.

Minimalist Social Connection: The Emergence

of “Form” or Synergies
A fundamental principle that underlies this approach is dynamic emergence of social
unity states. In the dynamic pickup of information from a rich environment, new
phenomena, basic sociality, emerges. In this first section, I focus on simple causal
principles that are essential for understanding interpersonal synchrony: dynami-
cal principles of self-organization. In this section, I suggest that by understanding
the dynamics of how lower-level physical movements spontaneously capture the
motions of others in our physical spaces, apparently trivial mutual influences of
incidental movement can create spontaneous interpersonal synchrony.

Theoretical Underpinnings: Self-Organization

This subsection and the next address how to look at the most fundamental means
of social interaction that allow for social connection, given evidence that syn-
chrony is a cause and a consequence of affiliative needs (e.g., Hove & Risen, 2009;
Lumsden, Miles, Richardson, Smith, & Macrae, 2012). Research on synchrony
illustrates how an approach that is not based on mind reading can proceed. To
explain how synchrony naturally emerges due to the dynamics of social perception-
action processes requires discussing three interrelated issues. First, causality under-
stood via self-organization principles is discussed. Second is a discussion of how
to conceptualize emergent structures when they come about as a consequence
of the first principle. This is essential for understanding emergence of social units
(dyad or group) created through self-organization—what can be called a social
“synergy.” Third is a discussion of what new predictions come about from such an
approach, describing the unique features of nonlinear dynamical systems.
 Dynamics of Synchrony and Joint Action  181

Causality
Self-organization means that coherent higher-order states of an overall system
emerge in a bottom-up fashion as a consequence of the individual elements
engaged in mutual adjustment to each other (Vallacher, Van Geert, & Nowak,
2015). Such a view suggests that states can evolve because of internal mechanics of
a system. No external operator drives the process, and there is no central internal
director that has particular responsibility for other processes unfolding.
When looking for a causal explanation, cognitive science commonly looks for
a controller that directs the action—DNA, intervention of others who train us,
or internal mental processes—that offer an explanation. Certain parts of the brain
light up during certain social cognitive processes, so that then becomes viewed
as the cause. This is oversimplified, of course; for example, modern evolutionary
views are epigenetic, and understand that mechanistic explanations require gene
X environment accounts. But it is not far wrong to say that an assumption that
the brain must re-present the world, simulate action of others, and make other
transformations of the outside world that bring it into the brain is viewed as the
primary causal principle underlying most areas of cognitive science.
It is easy to understand why that is so. When we see complicated behavior,
such as an individual’s highly coordinated movement of a large range of muscles
and joints in completing an action, we assume that such complexity must require
some plan—a motor program, for instance.4 We assume that complicated behav-
iors must involve skilled mental calculation or must have developed through
simple associative learning principles involving extraordinarily long chains of
associations. In this way, our brains have either built-in or developed through
experience a complex mental structure that directs sophisticated actions. Note
that an assumption is that contact with the external environment requires that it
be translated into our heads, re-presented, in order for us to plan our actions. The
assumption is that we rely on internal simulations of the outside world inside our
heads in order to be able to predict and respond to it.
An alternative view is to recognize phenomena such as brain activity during
a task, for instance, as an inextricable part of a system rather than the cause or
director of the activity. To call the brain the cause, Järvilehto (1998) believes,
is analogous to putting in the last puzzle piece to complete a jigsaw puzzle and
thinking that the piece is the cause of the depicted image (Järvilehto, 1998).
Even phenomena so readily viewed as brain-based, or at least localized in the
individual, such as emotion, he argues, cannot be localized in the brain, but
reside at the level of the dynamic living system, the organism environment system
(Järvilehto, 2001).
If one takes a radical system-oriented perspective, it yields the notion that the
causality principle is self-organization. Multiple components become collectively
organized, new states and new actions are achieved, without an executive system
directing the individual components (Turvey, 1990). Self-organization relies on
182  Kerry L. Marsh

certain assumptions about the underlying relations of the particular components

that make up the system. If components of a system have strong interdependen-
cies and particular connections, linkages among components can yield incredibly
complicated coordination patterns. Phenomena can emerge based merely on the
interaction of the components, and in response to constraints, press, or demands
on the system from without. A biological system is not like a mechanical one
that can be deconstructed into isolatable parts that still remain parts when decon-
structed, such as separating a bicycle into a wheel, chain, seat, and so forth.
Although the meaning of what a wheel does within a bicycle might not be
entirely clear without knowing the whole of which it is a part, it still has a
relatednessness to the other components that is highly different from the context
dependencies of components in biosocial-environmental systems (Juarerro, 2002).
A context dependent mechanical system might yield the astounding result that,
when pulling a wheel off a bicycle and inserting it into a different machine, it
morphs into an entirely different thing—unidentifiable any longer as a bicycle
wheel.

Structure
The kinds of structures that emerge from self-organization are dramatically
different than structures that an external controller would create. The human
skeletal system, for instance, entirely violates traditional principles of con-
struction that an architect would follow—it is neither heavy, rigid, nor sturdy
enough to pass muster as a load-bearing structure. But if we consider the entire
biomechanical system of the body, made up of components such as the muscles,
bones, fascia, and ligaments, it is clear that the system is unexpectedly strong,
efficient, and responsive because the body’s structure operates on a principle of
biotensegrity.
Tensegrity is the principle that underlies the (very strong) geodesic structures
that Buckminister Fuller was famous for promoting. Within the body, biotenseg-
rity means that muscle/tendons/bones are linked in such a way that there is a
continual balance of compression (by some elements) occurring within a context
of linkages that provide continual pull (tension) throughout the system (Ingber,
2008; Turvey, 2007; Turvey & Fonseca, 2014). The linkages between multiple
components allow them to function as a whole, a “synergy” or coordinative
structure, without something in charge of controlling the action of these com-
ponents. The very high-dimensional system of innumerable muscles and tendons
would make untenable computational demands for a motor program that must
compute current states and predict future states. Coordinated behavior is instead
more parsimoniously explained by the low-dimensional operation of coordina-
tive structures or synergies. In sum, this approach indicates that in the body,
qualitatively different states involving highly ordered, complex coordination can
come about as a result of demands of the environment.
 Dynamics of Synchrony and Joint Action  183

Once a synergy emerges, it has a reality that then constrains the components
comprising the whole. Forces exerted on the system, for example by situational
demands, yield one of two phenomena: resilience of the whole to temporary
perturbation (a tendency to bounce back to its current, stable state), but also
qualitatively new patterns of organization upon reaching some crucial amount of
press on the system. Causality in such a system resides at the level of interaction,
as a result of how the system is structured, and what forces are impinging on it
(Eiler, Kallen, & Richardson, Chapter 6 in this volume). Thus, putting pressure
on one part of the resilient system results in changes in other parts of the system.
Twisted in a person’s hand, a toy geodesic-type orb changes into a qualitatively
different shape, while maintaining its integrity in other ways.

Features of Nonlinear Dynamical Systems

Transient synergies, whether mechanically or biologically based states of coor-
dination, obey these two principles of resilience and change as well. Two
identically built pendulum clocks hanging near each other on a wall become
mechanically coupled through vibrations in the wall that influence each of them.
As a result, the clocks’ pendulums will come to move in a synchronized pattern.
A coupled oscillator dynamic (Haken, Kelso, & Bunz, 1985) similarly describes
biomechanical phenomena. The way components of the body are linked means
that one can hold identical weighted sticks (“pendulums”) in each hand and
easily swing the pendulums at a comfortable speed moving in precisely opposite
directions for a long time. Such a pattern of anti-phase synchrony is a fairly strong
attractor—in fact, it is impossible to swing each pendulum at a rhythm entirely
independent of the other. During anti-phase swinging, if pressure is exerted on
the system—for instance, if one must speed up—then at some critical speed,
an even stronger, more stable attractor pulls the system to a new state of order.
The nature of the interrelation of the limbs’ movements would change, with the
arms shifting to swing the pendulums in an in-phase synchrony state, in which
the pendulums are both swinging backward and both swinging forward at the
exact same time.
This last example also illustrates an important phenomenon in nonlinear
dynamical systems—namely, abrupt discontinuities. Nonlinear systems are ones
that are not well described by common regression equations. Rather, linear
increases in a variable across some range of values results in a non-commensurate
degree of change in another variable (Nowak & Vallacher, 1998; Richardson
et  al., 2014; Vallacher & Nowak, 1994; Vallacher, Read, & Nowak, 2002).
Increasing the pendulum speed or experiencing a subtle perturbation would have
minimal effect on anti-phase synchrony up to a certain increase in speed, but at
some critical point of increased speed, there is an abrupt change in the state of
the system. The system becomes briefly disordered (with temporarily increased
fluctuations), and then stabilizes into a new state: in-phase synchrony.
184  Kerry L. Marsh

Another crucial feature of such a system is that the present state cannot be
considered ahistorically—that is, it cannot be examined without considering the
immediate trajectory by which one got to that point; how one got “here” matters.
The quality of hysteresis means that there is a lack of symmetry in reaching one
state from different directions (e.g., reaching the current state from a state of
decreasing pressure versus increasing pressure on the system) in terms of its
consequences for future states. In other words, events could not be reversed in
sequence and get the outcome implied by their obverse. If one starts from a state
of in-phase behavior, a change in speed (e.g., decrease) does not shift one spon-
taneously to anti-phase behavior, despite the fact that speeding up in anti-phase
leads spontaneously to in-phase behavior.
The implications of a synergistic account for a single biological system may
seem easy to accept. Any child trying to pat their head and rub their tummy
learns the interdependence of apparently independent body components, as does
a novice drummer trying to simultaneously beat unrelated rhythms with two
drumsticks. It is a big conceptual leap, however, to imagine that two strangers
who lack any direct biomechanical connection could be coupled in a way that
would obey laws similar to those which coordinate the limbs of one’s body.

Empirical: Interpersonal Synchrony

Despite the implausibility that two strangers with no direct physical connection
are pulled to spontaneous states of coordination in their incidental movement,
research suggests that even fairly minimalistic interaction between two people—
e.g., happening to rock in a rocking chair alongside another person—can lead to
their becoming linked in a way such that they could be described as a social “synergy”
(Fusaroli, Rączaszek-Leonardi, & Tylén, 2013; Marsh, Richardson, Baron, &
Schmidt, 2006; Riley, Richardson, Shockley, & Ramenzoni, 2011). When
individuals are engaged in some kind of rhythmic movement such as swinging
their legs while seated or swinging a pendulum, or moving in a rocking chair
alongside another person doing the same, strangers can readily synchronize their
movements with each other when instructed to do so. Moreover, the conditions
under which they could readily do this, the types of synchrony patterns that were
possible, which patterns were more stable than others, and the deviations away
from perfect synchrony closely follow the predictions of bidirectional coupled
oscillator models (Richardson, Marsh, Isenhower, Goodman, & Schmidt, 2007;
Schmidt, Bienvenu, Fitzpatrick, & Amazeen, 1998; Schmidt, Carello, & Turvey,
1990a, 1990b; Schmidt, Christianson, Carello, & Baron, 1994).
We suggested that being mutually pulled into synchrony with another person
spontaneously might be the basis for a rudimentary social connection with others
(Marsh, 2010, 2013; Marsh et al., 2006; Marsh, Richardson, & Schmidt, 2009). If
the predictions of a dynamical model hold for unintentional synchrony, it would
lead to predictions that could not be anticipated by current perspectives that
 Dynamics of Synchrony and Joint Action  185

dominate explanations of mimicry and joint action, namely embodied simulation

accounts (e.g., Chartrand & Bargh, 1999; Sebanz, Knoblich, & Prinz, 2005). We
viewed interpersonal synchrony not as a cognitively mediated process such as prim-
ing (Bargh, Chen, & Burrows, 1996), but as the result of a field of forces, a view
more compatible with a field theory metaphor (Lewin, Lippitt, & White, 1939).
More concretely, we view synchrony as something that could be a consequence of
thermodynamic principles applied to the emergence of form or new “atomisms”—
new synergies (Iberall, 1987; Iberall & Soodak, 1987; Iberall, Wilkinson, & White,
1993; Soodak & Iberall, 1987).
Being pulled inadvertently into a synchrony state through one’s incidental,
idle, and non-goal-directed movements we likened to being swept into another
person’s orbit. The synergy of two people coupled in this way is like a social
eddy, analogous to a whirlpool of movement (Marsh et al., 2006). In the ocean,
a whirlpool has a real, forceful existence, and yet the entity only exists in the
relatedness that creates the force. It has a characteristic, visualizable form result-
ing from the interrelatedness of molecules flowing through it temporarily. The
particular molecules do not define the whirlpool, only the patterned force does.
The social eddy has no other real existence except in its temporal dynamics: its
harnessing of energy and creation of form.
From a social physics or thermodynamic perspective, social synergies or social
“eddies” can be understood as energy and dissipation processes occurring in a
near-equilibrium system (Iberall, 1987). Energy is either kinetic or potential energy
(i.e., conserved in bonds). When a system is at equilibrium, there is no motion,
no flow of energy. At near equilibrium, however, there are gradient differences
(differences in knowledge, social power, or for the case of concrete movement,
merely differences in energy of one individual’s movement) and as a result of
these gradient differences, flows of energies occur. Flows can be transient—the
mere capturing of another’s movement into a synchronous state, for instance.
Small fluctuations in arousal (presence of others) can be dissipated, or they can
be amplified and stabilized, leading to the emergence of form. Critical levels of
stress are achieved by reaching a certain level of physiological arousal, with touch,
close proximity, and language likely to induce such energy. Temporary eddies
of movement—emergent, temporary structured patterns in which two or more
people move as a unit—emerge when stress criticality is reached.
For more stable structures to form (a bond rather than transient sync), stronger
energy is required; this then becomes bound up in the relationship. Because indi-
viduals will always be buffeted by other sources of attractions—other potential
eddies that could form—the binding energy between two people must be stronger
than the energy that exists between the components and external agents. Iberall’s
social focus was on the implications of such an approach at a higher level of
analysis. Thus he discussed the development of civilizations and economic activity
in terms of the flow of variables between regions where there are gradient differ-
ences (e.g., economic activity of transporting wares and money between cities).
186  Kerry L. Marsh

But at the interpersonal scale that this chapter is concerned with, what would be
transported interpersonally might be informational products. Gradients would lead
to conversation, transmission of information, or orders given.
A final prediction of thermodynamic principles is with regard to dissolution
of bonds. When a given atomism, whether only temporary form (a social eddy)
or a stable structure (a relationship), is disrupted, energy is released. Energy flows
out of an interaction when synchrony dissipates, for instance. When more stable
structures are broken (relationships), substantially more energy is released.
Although most of these higher-order predictions extrapolated from the ther-
modynamics of energy flow and arousal have not yet been tested (e.g., using
physiological measures), the basic underlying dynamical model that predicts
how people become spontaneously synchronized when engaged in rhythmic
movement (e.g., the Haken-Kelso-Bunz model of bidirectional coupled oscil-
lators: Haken et al., 1985) has been supported. From an ecological perspective,
the means by which people are pulled into another’s orbit is informational.
Whereas mechanical and biological systems have physical direct linkages
between components, individuals are coupled by visual information about the
other person’s movements, or auditory information about the other person’s
movements. The key predictions of the model have been tested using a variety
of methods such as walking, swinging pendulums, moving in rocking chairs,
moving arms rhythmically, and shaking maracas, while using motion sensors
to track movement (Demos et al., 2012; Richardson, Garcia, Frank, Gregor, &
Marsh, 2012; Richardson, Marsh, & Baron, 2007; Richardson, Marsh, &
Isenhower et al., 2007; Richardson, Marsh, & Schmidt, 2005; Schmidt & O’Brien,
1997; Schmidt & Richardson, 2008; Van Ulzen, Lamoth, Daffertshofer, Semin, &
Beek, 2008). Motion sensors allow for continuously and precisely tracking
synchrony in a way that does not rely on more global and subjective assess-
ments by coders viewing videotapes. Coordination of even highly stochastic,
nonsinusoidal (wave-like) movements such as postural sway have been tracked,
using force plates or Wii technologies (Fowler, Richardson, Marsh, & Shockley,
2008; Shockley, Richardson, & Dale, 2009; Shockley, Santana, & Fowler,
2003). Moreover, new automated video analysis techniques allow for analysis
of synchrony in terms of ebbs and flows of total amounts of movement, or
coordination of behavior waves (Schmidt, Fitzpatrick, Caron, & Mergeche, 2011;
Schmidt, Morr, Fitzpatrick, & Richardson, 2012; cf. Newtson, 1994, 1998). As
predicted, seeing the movement of another person’s rocking chair or hearing
the movement is sufficient to lead individuals to coordinate their movements
without instruction (Demos et al., 2012; Richardson, Marsh, Isenhower et al.,
2007). (Indeed, coordination of movement has been found to occur even when
people actively attempt to not be affected by the others’ movements: Issartel,
Marin, & Cadopi, 2007.) Unlike intrapersonal coordination of limbs, sponta-
neous synchrony involves relative coordination. People shift into coordination
 Dynamics of Synchrony and Joint Action  187

for some period before falling out for a while, then becoming coordinated
again. Synchrony seems to be particularly important in early phases of forming a
connection, such as when being co-present with strangers or when interact-
ing with someone different in significant ways from oneself but with whom
a connection is desired (Miles, Griffiths, Richardson, & Macrae et  al., 2010;
Miles, Lumsden, Richardson, & Macrae, 2011). Moreover, a variety of predic-
tions based on dynamic models have been confirmed including how weaker
versus stronger coupling (by changing the amount of visual information avail-
able) affects synchrony, and the link between having similar natural frequencies
of movement and better coordination (see Marsh, Johnston, Richardson, &
Schmidt, 2009).
In particular, dynamical principles suggest that individuals with different
underlying perceptual-motor systems may be less likely to mutually influence
each other, leading to less synchrony. Some studies that focus on groups that have
perceptual-motor systems that are distinctively associated with dysfunction have
supported this prediction. One example is from a study of children who were on
the autism spectrum (and thus were diagnosed with significant perceptual-motor
deficits as well as social connection problems). They not only had more difficulty
doing basic drumming tasks alone (Isenhower et al., 2012), but they also showed
less spontaneous synchrony with their caregiver in the rocking chair paradigm
than did typically developing children (Marsh et al., 2013).
One prediction of the synergistic approach (Marsh et  al., 2009) that has
received minimal support is the mooring hypothesis. The hypothesis is focused on
individuals who are either overly variable in their movement, or in other ways
perceptually and motorically not well grounded in the rhythms of their physi-
cal world. This would mean they would be less resonant to the natural pacing
of temporal events, or even to the auditory information conveyed in speech or
music. Such individuals might benefit from the mutual influence of being pre-
sent with another individual in motion. The more stable or skilled individual
might provide an anchor onto whom less adaptively anchored individuals could
tether themselves. An alternative is that individuals who have more highly honed
senses to the rhythms of the world and are more responsive to those rhythms will
change their movement more to adjust to the other person in creating synchrony
rather than vice versa. To date, unpublished studies provide more evidence for
the latter hypothesis than for mooring effects, but forming clear conclusions
would be premature at this point.
A related hypothesis of anchoring also remains to be tested more thoroughly.
One assumption of a dynamical approach is that new modes of action (e.g., a social
synergy emerging from interactions of two initially independent actors) emerge
because they are lower energy states. Thus the gait patterns of a horse change
(between a trot, canter, or gallop) when the increased speed of movement makes a
different coordination of limbs easier and quantifiably more energy-efficient than
188  Kerry L. Marsh

the previous mode of locomotion. Based on this logic, a social synergy is predicted
to be more energy-conserving for the individual if they are operating as a part of
an easily coordinated ensemble. That teams of workers at hard labor chant or sing
together to coordinate their movement relies on this principle (McNeil, 1995). It
may be more efficient to hike a mountain by being able to perceptually entrain
to the rhythms of others’ footsteps in one’s visual field. Mere visual attention to a
rhythmic stimulus has been found to lead to unintentionally matching the external
rhythm (Lopresti-Goodman, Richardson, Silva, & Schmidt, 2009). Thus, it may
be less effortful to do difficult movement tasks (Nie, Caban, & Marsh, 2015) if
someone else is in sync with us. Intriguingly, rowers have a higher pain tolerance
after rowing with teammates than after rowing alone (Cohen, Ejsmond-Frey,
Knight, & Dunbar, 2010). Our belief is that the mechanism is not an explicit
motivational boost from shared experience. Instead, our assumption is that some
states (e.g., in-phase synchrony) are so readily achievable and promote cogni-
tive and perceptual fluency such that more perceptual and executive function
resources are available for other tasks. Such benefits should be apparent across a
wide range of outcomes, from pain tolerance to cognitive tasks (e.g., memory; for
indirect evidence, see Macrae, Duffy, Miles, & Lawrence, 2008 and Richardson
et al., 2005).
Understanding the most fundamental emergence of social connection with
others via quite trivial movements of individuals suggests the potential power
of a dynamical, ecological perception-based approach for understanding social
psychological phenomena. Research has strongly confirmed our suggestions that
synchrony is a fundamental basis for sociality. In the decade prior to the first inter-
personal synchrony studies that used dynamic, precise and objective measures of
synchrony, video coding observation studies suggested a link between rapport
and affiliation for dyads of interacting adults, and for infant–parent bonding pairs
as well (Bernieri, 1988; Bernieri, Reznick, & Rosenthal, 1988; Bernieri, Davis,
Rosenthal, & Knee, 1994; Condon & Sander, 1974; LaFrance, 1979). In the time
since Schmidt first studied interpersonal synchrony using motion sensors two
decades ago (for a review, see Schmidt & Richardson, 2008), research has pro-
liferated that uses interpersonal synchrony as an independent variable to induce
social connection (e.g., Wiltermuth & Health, 2009; Valdesolo, Ouyang, &
DeSteno, 2010), or as an outcome measure to assess sociality (e.g., Paxton &
Dale, 2013). As a whole,5 increased interpersonal synchrony is linked in the short
term as a cause or consequence of increased liking, stronger sense of team/group
cohesion, perceived smoothness, rapport, affiliation, and satisfaction with emo-
tional support (Bernieri, 1988; Bernieri et  al., 1994; LaFrance, 1979; Hove &
Risen, 2009; McNeil, 1995; Miles, Nind, & Macrae, 2009; Vacharkulksemsuk &
Fredrickson, 2012; for a review, see Vicaria & Dickens, 2016). Over the long
term, repeated experience with synchronization could have a defining effect on a
person’s psychological tendencies and have consequences for subsequent interac-
tion behavior (Novak, Vallacher, & Zochowski, 2005).
 Dynamics of Synchrony and Joint Action  189

Connecting through Cooperation

In the previous section, the scope of social connection phenomena was limited
to interpersonal synchrony. Understanding joint action, where two people must
coordinate to complete some action, might intuitively seem to require reading
the mind of the other in order to be able to coordinate. What would an ecologi-
cal, dynamical account require beyond the principles discussed in the previous
section to explain goal-directed action? Most other accounts rely on the indi-
vidual internally modeling or mentally simulating the other’s actions to do the
heavy lifting of crossing the gulf between other and self to engage in joint action
(Sebanz, Bekkering, & Knoblich, 2006). A social synergistic account requires
more in-depth understanding of the relatedness of the person to the environ-
ment, and the dynamic interdependence of action on perceptual flow. These
perceptual processes are discussed here in depth, focusing particularly on affor-
dance perception. Being able to cooperate with others means not only detecting
one’s own functional fit to the environment and what opportunities for solo
agency are available and when they are not, but also requires the ability to detect
what others have to offer, particularly when solo person–environment fit is defi-
cient (Marsh & Meagher, 2016). The next subsection, “Perceptual Processes,”
describes the underlying principles, and the subsection after that, “Empirical
Work on Affording Cooperation,” describes empirical work testing, such an
account of joint action.

Perceptual Processes
Traditional perspectives view perception in the following way. Sensation, as a
consequence of light hitting the retina, for instance, is viewed as occurring prior
to perception, which requires mental processes in order for the perceiver to
be able to piece together what those stimulations mean.6 The challenge, called
the binding problem, is understanding how static two-dimensional images on the
retina become bound together in order to construct a three-dimensional under-
standing of the objects we feel we are seeing. Within any individual static retinal
image, the entirety of a three-dimensional object is never depicted, only some
sides or parts of sides. Thus our brains have to fill in the missing parts and bind
them into coherent images. Each retinal image provides insufficient information
to specify where one object ends and another begins, and where in space things
are—what parts of the image are in the foreground or behind other objects, and
how far away things are. The mind must home in on particular cues that it has
learned through past experience in order to imperfectly infer these properties.
The fundamental problem is that the two-dimensional projection of the world
onto our retina is intractably nonspecific; any number of different situations
of the world could produce the same flat retinal image. The challenge for the
mind is deducing which state of the world has greater probability of producing
the image.
190  Kerry L. Marsh

However, Gibson’s theory of perception (Gibson, 1979) fundamentally

rejects such a perspective, suggesting that “the optical support for vision . . . is
not ambiguous, is not impoverished, and, therefore, not intractable” (Gibson,
1979, p. 103; Turvey & Shaw, 1999). Gibson believed that because animals
co-evolved with their environment, they would have evolved perceptual
systems that were capable of using the various forms of energy distributions of
their environment to gain veridical knowledge of the world. Humans and many
other animals, for instance, evolved in a world that is filled with ambient light
during the daytime. Light reflects off every surface that reaches our eyes. Gibson
posited that there may be information in the optic array that veridically specifies
what object is in front of another object, and the continuity versus the end-
ing (edges) of objects, for instance—provided that perceiver is in motion. Across the
transformations that occur in the optic array as a person merely moves their head,
there is structural persistence in aspects of the optic array, and we are sensitive
to these invariants. As we move, we correctly see where the edges of objects are
because texture accumulates at that leading edge. We see what is behind and in
front by the occlusion and revealing of textures in the world when we move
our head. Similar invariants veridically specify distance of objects. No inferential
process is required, only motion, so that information is specified via what persists
across the optic array over time.

Affordance Perception
Even more radically, Gibson (1977, 1979) hypothesized that if animals’ ecologi-
cal niches persisted over a sufficient numbers of generations, then there might be
information in the energy distributions that is specifically relevant to the actions
a given species needs to engage in for survival, beyond just information that
specifies the edges of objects, and how large, small, or distant they are. If visual
information can specify the edge of a chair, and haptic and other information
convey body awareness, attunement to information regarding more complex
constellations of features that specifies functionality of this special edge with
regard to one’s own body is not that far of a leap. Assume that information can
specify the edges of things. Assume as well that animals evolved to be highly fitted
to their world, and responsive with their bodies to the world they move through.
It is not a far leap from those assumptions that animals would have the perceptual
equipment to be able to immediately grasp the relatedness of things to them-
selves, and be able to discern, with proper attunement,7 what surfaces have the
proper array of qualities that allow the surface to be sit-on-able-for-me. Gibson
(1977, 1979) coined the term “affordances” to indicate the useful aspects of a
world that provide action opportunities for a given species. The term captures an
objective feature of the world, but one that is inherently relational—depending
on certain fittedness, certain capabilities, of a given species. Thus, whereas a level
ground affords easy transport of the body for a land animal via walking, it does
 Dynamics of Synchrony and Joint Action  191

not afford walking for a fish. The surface of water can afford walking-on for small
insects, but not for creatures that are of different size and weight.
Critically, researchers have defined perceiving affordances as being about agency
specifically, not the usefulness of objects more generally, nor about others’ agency.
Thus, saying an “apple affords nutrition” is incorrect (Michaels, 2003); rather, an
apple has the affordances of being holdable-in-one’s-hand, throwable, and bite-
into-able. Note, however, that we only talk about such affordances with respect to an
animal that has certain “ways of life,” reflecting the particular aspects in one’s envi-
ronment that are relevant—that is, our ecological niche (Rietveld & Kiverstein,
2014). Living things that occupy the same plot of land, whether bugs, earthworms,
or mammals, may occupy different ecological niches.
Evidence that animals are able to perceive affordances is provided by
experiments that systematically vary information provided to an animal that
specifies affordances, and showing “empirical evidence of differential behavior
tied to differential information availability” (Reed, 1996, p. 96). Reed (1996)
discussed experiments conducted with spiders and cats in which their behav-
ior reveals a sensitivity to distance, body-scaled to their capabilities. They
were highly successful at knowing whether a distance was leapable for them,
suggesting that some information covaried with their differential behavior.
In addition, Reed discussed studies Darwin conducted with earthworms that
demonstrated that earthworms had awareness of the usefulness of leaves for
plugging up their burrows and also demonstrated selectivity in the materials
they used. Behavioral discrimination was revealed, for instance, by pulling a
given leaf differently into the burrow depending on its features (e.g., shape).
In the research, trial and error could not explain behavior, and concepts such
as instincts, reflexes, or fixed action patterns were inadequate to the task of
explaining the rich flexibility and adaptability of the earthworms’ behavior
because they carried these tasks out even with materials not normally found
in that environment. The most parsimonious explanation for these animals’
behaviors is that something—namely, information that can be objectively quantified
and manipulated by a scientist—specifies affordances that animals use to regulate
their behaviors in an environment.
Gibson (1979) suggested that animals would also be sensitive to what other
conspecifics afford them, something that McArthur and Baron elaborated on
in an influential paper several years later (McArthur & Baron, 1983). Reviews
of the social perception and affordance literature suggest that people are highly
sensitive to dynamic perceptual information that specifies significant features of
individuals such as gender and age, and more complex features as well (Marsh &
Meagher, 2016; Marsh et al., 2006; McArthur & Baron, 1983; Johnson, 2013;
Johnston, 2013; Johnston, Hudson, Richardson, Gunns, & Garner, 2004).
Intriguing research suggests superficial cues such as hair, clothing, or wrinkles on
a face are not required to detect social features, but rather that there is something
in the temporal dynamics of moving form that specifies features such as age and
192  Kerry L. Marsh

gender (Berry, Kean, Misovich, & Baron, 1991; Runeson & Frykholm, 1983).
Researchers studying the kinematic specification of dynamics (KSD) put reflective
dots on key joints of an individual in motion, and film this under minimal light
such that the only information available to a perceiver is moving points of light
when the video plays. Not only can people detect what action someone is doing,
such as walking, dancing, or picking up a heavy box, they can also detect whether
a heavy or light object is being lifted, even when the actors are pretending to
lift something heavier. Static images of the point lights do not allow for such
detection. Most critically, perceivers also detect what opportunities for their own
actions are afforded by the presence of that individual—namely, social affor-
dances. Observers can detect when a dyad is cooperating or competing (Kean,
2000), or whether a person is carrying a child rather than rubbish (Hodges &
Lindhiem, 2006). Other KSD studies found that criminals can use the dynamics
of a person’s movement to determine the individual’s “muggability” (Johnston,
2013; Johnston et al., 2004).
In one sense, it is not that drastic a leap from the notion of direct perception
of an edge to the notion of direct perception of a unique constellation of features
(edges and the like) that are specifically relevant to how one’s action system
can make use of those features. But, in another sense, it is quite radical. With
the concept of affordances, the meaning of even quite abstract things, such as a
physical space, is not an arbitrary social construction. Instead, the environment
declares to us its meaning. This yields the hypothesis that our feelings and emo-
tional responses to places can be a consequence of our ability to rapidly see what
affordances are in a situation, even if the setting and features of the environment
are entirely new to us. Such responses should be detectable, albeit imperfectly,
by the words we use to describe environments. Somewhere between objective
fact (“What is the square footage of this room?”) and purely aesthetic evalua-
tions (“Do you like this room?”) are judgments that have both a functional and
valenced flavor, such as how spacious or cozy a room feels (Meagher & Marsh,
2015). These reactions to space, expressed in words and in deeds, should be
particularly apparent when the affordances of a space are most self-relevant—for
example, the dorm room or work space is our own territory rather than someone
else’s (Meagher, 2014).
Finally, an affordance-based perspective suggests that meaning is relational—it
is partly determined by what an individual is, in terms of their capabilities, as well
as what they could be, given the degree to which the physical and social environ-
ment completes us, by fitting an unmet need or untapped capability. Crucially,
meaning emerges out of dynamic interaction with the environment—new pos-
sibilities are encountered and created, with the new possibilities that come into
being with the presence of others being most significant of all.
In the following subsection, we first discuss affordances at the more immedi-
ate level, involving dyadic interaction. A decade ago, we began to explore the
question of how the presence of others holds out the promise of allowing us
 Dynamics of Synchrony and Joint Action  193

to become a more capable entity, by studying cooperation. The final section

of the chapter, “Future Directions: Affordances of Behavior Settings,” will
discuss the role of the broader situation in affording opportunities for social
connection.

Empirical Work on Affording Cooperation

Inspired by solo perception-action research, my colleagues and I hypothesized
that cooperative action, such as moving an object together, might be approached
from an affordance-based perspective rather than from an economic game theory
approach. One hallmark of an affordance-based approach is that there is infor-
mation in a situation about the task demands, and the other persons as well.
Cooperation would not be an abstract symbolic choice made with a person who
was not physically present, as is often the case in studies based on game theory,
but it would be examined as a behavior that involved physical actions. A social
synergistic perspective would view two individuals in cooperation as a new
entity, a social perception-action system in which the perceptions and actions of
each member of the ensemble were tightly responsive in real time to that of the
others.
Asch, for instance, talked about the example of two boys moving a log together
as creating a performance that was a wholly “new product, strictly unlike the sum
of [each boy’s] separate exertions” (Asch, 1952, pp. 173–174). A similar take on
embodied cooperation was offered by the philosopher Margaret Gilbert (1996,
2014). She viewed cooperative actions such as deciding to take a walk together
as being a situation in which actors simultaneously jointly commit to being as a
plural subject of action, with the attendant obligations and benefits to complete the
action as a social entity, much like Asch described.
If an affordance account is suitable for describing the emergence of coopera-
tive action, two key predictions from solo affordance research should hold. First,
cooperation should be predicted as a new “mode of action” that emerges from
solo action, with this mode of action precisely predictable by quantifying the
capabilities of individuals, and the demands of the task, taken with respect to each
other. This is the pi number prediction. A pi number is a body-scaled assessment
of the person–environment relation in which some quantity of an individual and
the environment are each assessed (e.g., in centimeters), and divided. It is called a
“pi number” because it is a dimensionless number, just as pi is. At crucial values
of a pi number, the system should shift from one action mode (e.g., solo action or
joint action) into the other mode. The second key prediction reflects dynamical
principles. Shifts between solo and joint action modes should have characteristic
dynamic features that previously have been found in the solo affordance research
upon shifting between solo action modes. For example, the system should display
features of a nonlinear dynamical system rather than a linear system, with abrupt
shifts in action modes.
194  Kerry L. Marsh

The task that we developed to examine cooperative action was similar to one
well studied in solo action: moving objects of different sizes. Previous research
had examined how children, for instance, will choose to pick up objects using
a one-handed grip versus a two-handed grip (Cesari & Newell, 1999; Van der
Kamp, Savelsbergh, & Davis, 1998). We thought the shifts between one actor
moving an object using a two-handed grip (a solo action mode) to a two-­person
action mode (joint action) should obey similar principles to that of the solo
­perception-action research. In our studies, we used narrow planks of wood of a
wide range of sizes, which were presented in a continuous sequence. Each plank
was painted red on the ends, and participants were told they could move the
planks however they chose, with the restriction that they could only touch the
ends of the planks, as if their middles had wet paint on them.

Pi Number Hypothesis
Research on a variety of actions such as passing through apertures (doorways),
climbing stairs, and moving objects supports the notion that these behaviors (and
perceptual judgments as well) are precisely predicted by pi numbers. For example,
Warren (1984) studied the affordance of stairs being climbable by normal
means—that is, upright, with only one foot resting on a step while the next leg
smoothly begins to move the other foot to the next step. He demonstrated that
when stairs cease to be comfortably climbable in this mode was not merely a
function of riser height, nor a function of how tall the different climbers were.
Rather, shifts between modes were entirely determined by a relational measure.
He calculated a body-scaled ratio of leg length to riser height, and found that a
consistent value of the pi number determined comfortable stair climbing across
individuals. By varying riser height as well, he could determine the pi num-
ber value at which normal stair climbing became consistently challenging for all
individuals. Similarly, children’s natural tendencies to pick up objects using one
hand or two are precisely predicted by a calculation that quantifies the relation
between hand size and object size: a pi number (Cesari & Newell, 1999; Van der
Kamp et al., 1998).
We first replicated the pi number prediction in an experiment in which solo
actors were asked to report how they would move planks of wood presented to
them—where their choices were using one hand, two hands, or an axe handle
with a claw-like end that would extend their arm span. In another experiment,
solo participants’ actual behaviors (using one hand, two hands, or the tool) were
assessed. These experiments supported the pi number prediction for solo actors,
where pi numbers were calculated as the relation between hand size and plank
size (for shifts between one- and two-handed grips) or as the relation between
arm span and plank size (for shifts between two hands and tool use).
In a joint action experiment (where pairs of participants took planks off a
conveyor belt, moving them either alone or together), we also found support for
 Dynamics of Synchrony and Joint Action  195

the pi number hypothesis. Key values of the pi number determined the bounda-
ries between engaging in solo action and joint action (Richardson, Marsh, &
Baron, 2007). A comparable experiment examined only perceptual judgments,
and replicated the results. In subsequent studies that more systematically var-
ied the height of participants, shifts between action modes were determined
by the value of the pi number for both tall dyads and short dyads (Isenhower,
Richardson, Carello, Baron, & Marsh, 2010).

Dynamical Hypotheses
Thus, emergence of a cooperative social unit occurred naturally in response to
the task becoming more challenging: cooperation was an emergent adaptation to
the situation. We also expected that the dynamics of cooperation’s emergence
and subsidence would display features of other nonlinear dynamical systems,
including abrupt shifts between action modes, and also the characteristic behavior
of hysteresis.
To allow us to look at dynamical features, another feature of our joint action
(or perception) experiment was that the sequencing of planks was varied, and
presented in a steady sequence with moderate pacing. Three planks of one length
were presented before the next length. In one condition, the planks were pre-
sented in order of ascending length; in another, they were in descending order.
For comparison, a random order of plank trios was presented. As predicted, when
the crucial pi number values occurred, the shift to a new mode of action was rela-
tively abrupt. That is, once the dyads shifted to a new mode of action, this action
mode was stable and the subsequent actions stayed in that new mode. Moreover,
the action experiments displayed hysteresis. The precise boundary between
joint action and solo action modes was shifted somewhat as a function of the
past trajectory of action. This pattern indicated that the system had multistability—
that is, regions where both solo action and joint action are stable attractors.
Hysteresis occurred in the ascending versus descending conditions, reflecting a
“stickiness” of past action. That is, the current mode persisted somewhat beyond
the pi value at which, in the random condition, behavior would have switched.
To clarify, in the descending lengths condition, where participants by necessity
start off in cooperative action, cooperation persisted into the multistable region.
Thus, slightly shorter planks were carried cooperatively in this condition, in con-
trast to the ascending condition, where planks of those lengths would have still
been carried alone. Interestingly, the opposite dynamical characteristic occurred
in the perception experiments; perception was anticipatory, or forward-looking.
As soon as the multistable region was reached, people’s judgments abruptly
shifted to the next action mode in that region.
The limits of these sets of experiments are that they focused solely on actions
in which another person extends one’s action possibilities in a quantitative man-
ner, where each lower action mode (one hand, two hand) is nested within a
196  Kerry L. Marsh

higher-order mode (two hands, two people). One assumption of a dynamical

approach, however, is that there will also be qualitatively new possibilities for
action that the presence of others provides for the individual. Emergence of new
“atomisms”—for example, a social synergy, should yield novel properties unique
to that atomism. In the plank moving experiments, we found the somewhat
surprising result that the precise pi numbers that determined action boundaries
between solo and joint action closely matched that of boundaries between one-
to two-hand action modes, suggesting, perhaps, close parallels in what constrains
the solo perception-action system and the social perception-action system.
However, as social ecological research moves into domains where there are
added complexities of affordances in the joint action domain, processes will differ
in important ways from results of solo perception-action experiments. A series of
recent perception studies examining how anticipated assistance carrying a heavy
weight affected perceived distance illustrates the challenge. Whereas some exper-
iments found results that mapped precisely onto what solo perception-action
experiments would predict—namely, distances would seem shorter because a
heavy weight will be easier to carry (Doerrfeld, Sebanz, & Shiffrar, 2012)—other
studies found the opposite (Meagher & Marsh, 2014). These differences were
presumed to occur because the presence of another person provides not only pos-
sibilities for action that lighten a load, but also negative interpersonal possibilities.
One might fear that one could let the other person down if one was unable to
do one’s part, for instance. Such interpersonal concerns change the equation in
significant ways.

“Perception-Based” Inaccuracies
The veridical perceptions involved in coordination and cooperation may seem
hard to reconcile with extensive evidence that judgments of social environments
are distorted by desire (Balcetis & Dunning, 2006; Balcetis & Lassiter, 2010).
Note, however, that perceptual judgmental processes will be expected to be
veridical only to the extent to which a scientist is able to objectively assess that
there is information available that specifies an affordance; only limited experimen-
tal research on cooperative action to date has taken an explicitly social affordance
perspective (for a review, see Marsh & Meagher, 2016).
In social judgment tasks, accuracy will increasingly decline to the degree
to which judgments are based not on perceiving affordances per se (detecting
action possibilities for oneself), but determining whether a situation provides
possibilities for another’s actions—whether a high shelf is reachable for them,
for instance. Although strong evidence is lacking, some recent research provides
indirect evidence for the hypotheses that (1) impoverished information as well
as impoverished experience with others’ capabilities is associated with greater
inaccuracies, and (2) perceptual attunements are constrained by social roles,
groups, recent experience, and repeated experiences inculcated through culture
 Dynamics of Synchrony and Joint Action  197

(Eiler et al., Chapter 6 in this volume; Marsh & Meagher, 2016; Meagher &
Kang, 2013; Ye, Cardwell, & Mark, 2009).
In many situations involving observing others’ interactions, where the situ-
ation involves multiple actors, the situation is more complex in that it involves
two or more agentic beings, and there are multiple possibilities for action present
for each agent. What response one has to such a situation will be highly affected
by attentional focus. Take, for instance, a situation of high potential conflict
and ambiguity, where one is making judgments as to whether a police officer’s
use of force is justified. Social processes bias the judgments by causing one to
focus on different aspects of the situation (Granot, Balcetis, & Schneider, 2014).
Group identification with one individual or the other will make some action
possibilities more salient via its redirection of visual attention to opportunities
for agency (or lack of agency) for one person rather than agentic possibilities
for the other (Granot et al., 2014). In general, the degree to which stimuli are
impoverished and the perceiver is focused on seeing a situation “for” another
person rather than with regard to one’s own capabilities, the more that motivated
inferential processes will dominate judgmental outcomes (Balcetis & Dunning,
2006; Balcetis & Lassiter, 2010) rather than veridical, affordance-based detection.
Even in motivation-distorting, highly inferential process-based situations (Granot
et  al., 2014), however, attentional measures (what part of a visual display one
focuses on) should be in support of inferential processes that follow.
Indeed, exploring the detection and enactment of affordances in rich and
complex social environments is a neglected area of research for social ecologi-
cal psychologists. However, there has been a flurry of interesting research on
social affordance detection and action in the sport psychology arena (e.g., Araújo,
Ramos, & Lopes, 2016; Silva, Garganta, Araújo, Davids, & Aguiar, 2013; Silva
et al., 2015)—albeit very little of it conducted by social psychologists. Research
from a social-ecological perspective has focused on emergent transient synergies
such as interpersonal synchrony, or on detection of social affordances, or emer-
gence of cooperative action. Little research has illuminated the environment side
of social affordances that is relevant to coordination and cooperation. In the final
section, we turn to this issue.

Future Directions: Affordances of Behavior Settings

The terms “environment,” “setting,” and “situation” have all been used in mul-
tiple and overlapping ways. Social psychology currently favors the term situation,
and sometimes uses it when assessing the kinds of activities people are engaged in,
in a particular location and time (Guillaume et al., 2016). Most often, the term is
currently used to categorize what is happening around a person in terms of high-
level interpersonal processes or the psychological reality of a setting, rather than
its spatiotemporal reality (e.g., Edwards & Templeton, 2005; Cantor, Mischel, &
Schwartz, 1982; Reiss, 2008; Ten Berge & De Raad, 2002). From an ecological
198  Kerry L. Marsh

and dynamical account, however, the only reasonable starting point for a taxon-
omy of situations is one that focuses on the degree to which a situation allows an
individual to fulfill relevant goals, as Yang, Read, and Miller (2009) proposed in
their comprehensive review of situation taxonomies. What would a research pro-
gram look like that takes the situation’s opportunities for various forms of agency
as its starting point? Unpacking an affordance-based account implies a wide range
of questions that social psychology has simply not pursued in any depth, with the
exception of some environmental psychology research, which did not have the
benefit of Gibson’s concepts to allow a principled (rather than purely descriptive)
approach to pursue those issues.
First, an affordance-based account implies certain assumptions about situa-
tions: that they should be conveyed by information available for pickup, not that
they are mere cultural constructions, or that because of habit, or experience, one
believes the situation allows certain opportunities for action. That a situation tells
us certain actions are possible here must be in some way conveyed by informa-
tion available to an active perceiver. But almost no research has pursued these
questions: What is it in a situation that allows it opportunities for certain mani-
festations of agency, and restrains others? And from the perspective of the goals
of this chapter in particular, what aspects of a situation provide opportunities or
close down opportunities for social connection? To what extent do the opportu-
nities of the environment universally evoke certain behaviors (e.g., with young
enough children, when encountering an unusually bouncy surface, nearly all will
jump up and down)? To what extent do the social affordances for connection in
a setting interact with individual traits (Yang et al., 2014) or need states to predict
environmental responses (Meagher & Marsh, 2017)?
The social ecological perspective cares about a physical environment, one
which has a “durable existence”—a spatial and temporal reality (Barker, 1968;
Schoggen, 1989), and is not merely the construal of an environment, or a sub-
jective representation of an environment. Roger Barker developed an area of
research focused on ecological environments, and more specifically, developed
the notion of behavior settings which provides direction for a way forward (Heft,
2001). A behavior setting is a part of the ecological environment that has defin-
able temporal and spatial boundaries—one can know that the setting will be in
existence at particular times (it might be advertised, for instance), and a perceiver,
entering a behavior setting unexpectedly would be able to detect where the
boundary is. Behavior settings cease existence during the times when they do
not occur (e.g., when a basketball game is not occurring, or when music class is
not being held), even though aspects of the milieu (e.g., basketballs, court, music
stands) may continue to exist. Important aspects of what defines a behavior set-
ting are objective, observable “standing patterns of behavior by people—that
is, specific sequences of people’s behavior that regularly occur within particular
settings” (Schoggen, 1989, pp. 2–3).
 Dynamics of Synchrony and Joint Action  199

Importantly, there is an interdependency of participants in the setting, but

the existence of the behavior setting is not dependent on a particular individual’s
behavior. Thus a standing pattern of behavior “has unique and stable character-
istics that persist even when current inhabitants of the setting are replaced by
others” (Schoggen, 1989, p. 31). If one player or referee is injured or unavailable,
another person fills that role, and the standing pattern of playing or refereeing
remains intact. An important aspect of the behavior setting is therefore the
“dynamic, on-going pattern of relationships among individuals” (Heft, 2007, p. 97).
Barker discovered behavior settings by starting out to focus on individual dif-
ferences in children’s behavior—dispositional differences— and finding instead
that children’s behaviors varied less across children than they varied across settings.
Knowing what behavior setting a child was in substantially predicted aspects of
their behavior, much more so than knowing the particular behavioral tendencies
of the child (Barker, 1978; Barker & Wright, 1954; Barker, Wright, Barker, &
Schoggen, 1961).
Barker’s rich and extremely detailed observations of the behavior of members
of behavior settings within a small Midwestern US town and a small British
town (Barker & Schoggen, 1973; Barker & Wright, 1951) illustrated how strong
the possibilities and constraints of our environments are for social behavior.
Barker’s perspective provides a fertile starting point for research determining the
informational and dynamical features that define situations, considered from an
affordance-based perspective (Yang et al., 2009). However, only a few hypotheses
Barker developed about behavior settings have been tested using rigorous experi-
mental (or even correlational) methodologies (e.g., Heft, Hoch, Edmunds, & Weeks,
2014). For a field that has been calling for a more situated, embodied, behavioral
and situation-focused approach (e.g., Agnew, Carlston, Graziano, & Kelly, 2009;
Baumeister, Vohs, & Funder, 2007; Rauthmann, Sherman, & Funder, 2015;
Rauthmann, Sherman, Nave, & Funder, 2015; Reiss, 2008; Smith & Semin,
2004), this offers an excellent way forward to understanding the environment-
based features that provide a grounding for social coordination and cooperation.

Conclusion
In this chapter, I have suggested that an approach to social psychology that reaches
into its evolutionary roots to focus on the local interactions between co-present
actors in a local environmental niche provides a substantially new approach to
understanding the nature and emergence of sociality phenomena such as interper-
sonal synchrony and joint action. Dynamical principles explain how self-organizing
principles can provide the causal mechanism for social connection’s emergence,
and active perceptual exploration processes are essential as well. The approach sug-
gests a way for the field to become unstuck from its all too common focus on the
individual (or at best, the dyad) as the unit of analysis for study.
200  Kerry L. Marsh

Notes
1 This work was completed while the author was serving at the National Science Foundation.
The views expressed in this chapter are solely those of the author and do not necessarily
represent those of the NSF.The author thanks Harry Heft, Stephen Read, Brian Eiler, and
Andrzej Nowak for comments on an earlier version of this chapter.
2 The term “ecological psychology” is inherently problematic, as this term has been claimed
by a wide-range of individuals such as population biologists, behavioral ecologists, and
cognitive, clinical, evolutionary, and developmental psychologists and anthropologists
who have a traditional cognitivist bent, but wish to signal a concern with situated or
evolutionary influences. Typical concerns for “ecology,” for instance, might include using
levels of explanations that are mindful of how the studied behavior would normally occur
in natural, real-world contexts, or would normally be situated in important community
contexts (families, cultures). Here I use the term more narrowly, to include ecological
psychologists who view humans as first and foremost animals, and who focus on environ-
ments in a substantial way in their research—specifically, the ecological niche in which
a species evolved, and not merely an animal’s general habitat or a human’s interpersonal
or cultural context. More specifically, my account focuses on the work of those whose
theoretical concepts derive from Gibson’s (1977, 1979) theory of perception, a theory that
provided an alternative to classical constructivist approaches to perception (Marr, 1982).
3 For example, the term “affordance” was widely popularized by publication of
Norman’s otherwise excellent book on the design of everyday objects (Norman, 1988).
Unfortunately, his inaccurate use of the term (as he acknowledges) was not corrected
until the book was revised (Norman, 2002).
4 This section liberally paraphrases what my collaborators and I have explained elsewhere
(Marsh, Johnston, Richardson, & Schmidt, 2009, p. 1220).
5 A caution to readers that the size of the relationships between synchrony and self-
reported liking, affiliation, and self-other connection appear overstated by the published
literature, not surprisingly (Open Science Collaboration, 2015). Correspondence with
other researchers and the author’s own experience suggest that at least in minimal
interaction paradigms, effects using any particular self-report measures are not always
replicable, much like analogous effects in the mimicry literature.
6 See Marsh (2015) for further elaboration about these issues surrounding perceptual
processes.
7 This is not to say that experience has no role. With new capabilities, for instance
(a child learning to crawl, or someone wearing shoes with high blocks attached; Mark,
1987), experience is needed in order to develop perceptual attunement to information
that specifies affordances. Perceptual cliff experiments with crawling infants indicate
that ability to detect a falling-off place is not instantaneously perceived by an infant; it
requires some crawling experience in order to learn what information to attend to in
the environment (Gibson & Walk, 1960).

References
Agnew, C. R., Carlston, D. E., Graziano, W. G., & Kelly, J. R. (Eds.) (2009). Then
a miracle occurs: Focusing on behavior in social psychology theory and research. New York:
Oxford University Press.
Araújo, D., Ramos, J. P., & Lopes, R. J. (2016). Shared affordances guide interpersonal
synergies in sports teams. In N. P. Passos, K. Davids, & J. Y. Chow (Eds.), Interpersonal
coordination and performance in social systems (pp. 165–178). London: Routledge.
Asch, S. (1952). Social psychology. New York: Prentice Hall.
Balcetis, E., & Dunning, D. (2006). See what you want to see: Motivational influences on
visual perception. Journal of Personality and Social Psychology, 91, 612–625.
 Dynamics of Synchrony and Joint Action  201

Balcetis, E., & Lassiter, G. D. (Eds.) (2010). The social psychology of visual perception. New
York: Psychology Press.
Bargh, J. A., Chen, M., & Burrows, L. (1996). Automaticity of social behavior: Direct
effects of trait construct and stereotype activation on action. Journal of Personality and
Social Psychology, 71, 230–244.
Barker, R. G. (1978). Need for an eco-behavioral science. In R. G. Barker et al., Habitats,
environments, and human behavior (pp. 36–48). San Francisco, CA: Jossey-Bass.
Barker, R. G., & Schoggen, P. (1973). Qualities of community life: Methods of measuring
environment and behavior applied to an American and an English town. San Francisco, CA:
Jossey-Bass.
Barker, R. G., & Wright, H. F. (1951). One boy’s day: A specimen record of behavior. New
York: Harper.
Barker, R. G., & Wright, H. F. (1954). Midwest and its children: The psychological ecology of
an American town. New York: Row, Peterson, & Co.
Barker, R. G., Wright, H. F., Barker, L. S., & Schoggen, M. (1961). Specimen records of
American and English children. Oxford: University of Kansas Publications.
Baumeister, R., Vohs, K. D., & Funder, D. C. (2007). Psychology as the science of self-
reports and finger movements: Whatever happened to actual behavior? Perspectives on
Psychological Science, 2, 396–403.
Bernieri, F. J. (1988). Coordinated movement and rapport in teacher–student interactions.
Journal of Nonverbal Behavior, 12, 120–138.
Bernieri, F. J., Davis, J. M., Rosenthal, R., & Knee, C. R. (1994). Interactional syn-
chrony and rapport: Measuring synchrony in displays devoid of sound and facial affect.
Personality and Social Psychology Bulletin, 20, 303–311.
Bernieri, F. J., Reznick, J. S., & Rosenthal, R. (1988). Synchrony, pseudosynchrony,
and dissynchrony: Measuring the entrainment process in mother–infant interactions.
Journal of Personality and Social Psychology, 54, 243–253.
Berry, D. S., Kean, K. J., Misovich, S. J., & Baron, R. M. (1991). Quantized displays of
human movement: A methodological alternative to the point-light display. Journal of
Nonverbal Behavior, 15, 81–97.
Cantor, N., Mischel, W., & Schwartz, J. C. (1982). A prototype analysis of psychological
situations. Cognitive Psychology, 14, 45–77.
Cesari, P., & Newell, K. M. (1999). The scaling of human grip configurations. Journal of
Experimental Psychology: Human Perception and Performance, 25, 927–935.
Chartrand, T. L., & Bargh, J. A. (1999). The chameleon effect: The perception–
behavior link and social interaction. Journal of Personality and Social Psychology, 76,
893–910.
Cohen, E. E., Ejsmond-Frey, R., Knight, N., & Dunbar, R. I. (2010). Rowers’ high:
Behavioural synchrony is correlated with elevated pain thresholds. Biology Letters, 6,
106–108.
Condon, W. S., & Sander, L. W. (1974). Synchrony demonstrated between movements
of the neonate and adult speech. Child Development, 45, 456–462.
Demos, A. P., Chaffin, R., Lewis, A., Begosh, K. T., Daniels, J. R., & Marsh, K. L.
(2012). Rocking to the beat: Effect of music and partner’s movements on spontane-
ous interpersonal coordination. Journal of Experimental Psychology: General, 141, 49–53.
Doerrfeld, A., Sebanz, N., & Shiffrar, M. (2012). Expecting to lift a box together makes
the load look lighter. Psychological Research, 76, 467–475.
Edwards, J. A., & Templeton, A. (2005). The structure of perceived qualities of situations.
European Journal of Social Psychology, 35, 705–723.
202  Kerry L. Marsh

Fowler, C. A., Richardson, M. J., Marsh, K. L., & Shockley, K. (2008). Language
use, coordination, and the emergence of cooperative action. In A. Fuchs &
V. K. Jirsa (Eds.), Coordination: Neural, behavioral and social dynamics (pp. 261–279).
Berlin: Springer-Verlag.
Fusaroli, R., Rączaszek-Leonardi, J., & Tylén, K. (2013). Dialog as interpersonal synergy.
New Ideas in Psychology, 32, 147–157.
Gibson, E. J., & Walk, R. D. (1960). The “visual cliff.” Scientific American, 202, 67–71.
Gibson, J. J. (1977). The theory of affordances. In R. Shaw & J. Bransford (Eds.), Perceiving,
acting, and knowing: Toward an ecological psychology (pp. 67–82). Hillsdale, NJ: Lawrence
Erlbaum.
Gibson, J. J. (1979). The ecological approach to visual perception. Boston, MA: Houghton
Mifflin.
Gilbert, M. (1996). Living together: Rationality, sociality, and obligation. New York: Rowman &
Littlefield.
Gilbert, M. (2014). Joint commitment: How we make the social world. New York: Oxford
University Press.
Granot, Y., Balcetis, E., & Schneider, K. E. (2014). Justice is not blind: Visual attention
exaggerates effects of group identification on legal punishment. Journal of Experimental
Psychology: General, 143, 2196–2208.
Guillaume, E., Baranski, E., Todd, E., Bastian, B., Bronin, I., Ivanova, C., et al. (2016).
The world at 7:00: Comparing the experience of situations across 20 countries. Journal
of Personality, 84, 493–509.
Haken, H., Kelso, J. A. S., & Bunz, H. (1985). A theoretical model of phase transitions in
human hand movements. Biological Cybernetics, 51, 347–356.
Heft, H. (2001). Ecological psychology in context: James Gibson, Roger Barker, and the legacy of
William James’s radical empiricism. Mahwah, NJ: Lawrence Erlbaum.
Heft, H. (2007). The social constitution of perceiver-environment reciprocity. Ecological
Psychology, 19, 85–105.
Heft, H., Hoch, J., Edmunds, T., & Weeks, J. (2014). Can the identity of a behavior
setting be perceived through patterns of joint action? An investigation of place
perception. Behavioral Sciences, 4, 371–393.
Hodges, B. H., & Lindhiem, O. (2006). Carrying babies and groceries: The effect of moral
and social weight on caring. Ecological Psychology, 18, 93–111.
Hove, M. J. & Risen, J. L. (2009). It’s all in the timing: Interpersonal synchrony increases
affiliation. Social Cognition, 27, 949–960.
Iberall, A. S. (1987). A physics for study of civilizations. In F. E. Yates (Ed.), Self-organizing
systems: The emergence of order (pp. 521–540). New York: Plenum.
Iberall, A. S., & Soodak, H. (1987). A physics for complex systems. In F. E. Yates
(Ed.), Self-organizing systems: The emergence of order (pp. 499–520). New York:
Plenum.
Iberall, A. S., Wilkinson, D., & White, D. (1993). Foundations for social and biological
evolution: Progress toward a physical theory of civilization and of speciation. Laguna Hills,
CA: Cri-de-Coeur Press.
Ingber, D. E. (2008). Tensegrity and mechanotransduction. Journal of Bodywork & Movement
Therapies, 12, 198–200.
Isenhower, R. W., Marsh, K. L., Richardson, M. J., Helt, M., Schmidt, R. C., & Fein, D.
(2012). Rhythmic bimanual coordination is impaired in young children with Autism
Spectrum Disorder. Research in Autism Spectrum Disorders, 6, 25–31.
 Dynamics of Synchrony and Joint Action  203

Isenhower, R. W., Richardson, M. J., Carello, C., Baron, R. M., & Marsh, K. L. (2010).
Affording cooperation: Embodied constraints, dynamics, and action-scaled invariance
in joint lifting. Psychonomic Bulletin and Review, 17, 342–347.
Issartel, J., Marin, L., & Cadopi, M. (2007). Unintended interpersonal co-ordination:
“Can we march to the beat of our own drum?” Neuroscience Letters, 411, 174–179.
Järvilehto, T. (1998). The theory of the organism–environment system. I: Description of
the theory. Integrative Physiological & Behavioral Science, 33, 321–334.
Järvilehto, T. (2001). Feeling as knowing, Part II: Emotion, consciousness and brain
activity. Consciousness & Emotion, 2, 75–102.
Johnson, K. L. (2013). Social constraints on perceiving biological motion. In K. L. Johnson &
M. Shiffrar (Eds.), Visual perception of the human body in motion: Findings, theory, and
practice. New York: Oxford University Press.
Johnston, L. (2013). Kinematics and vulnerability perception. In K. L. Johnson &
M. Shiffrar (Eds.), Visual perception of the human body in motion: Findings, theory, and
practice. New York: Oxford University Press.
Johnston, L., Hudson, S. M., Richardson, M. J., Gunns, R. E., & Garner, M. (2004).
Changing kinematics as a means of reducing vulnerability to attack. Journal of Applied
Social Psychology, 34, 514–537.
Juarerro, A. (2002). Dynamics in action: Intentional behavior as a complex system. Cambridge,
MA: MIT Press.
Kean, K. J. (2000). An investigation into the interpersonal kinematics of cooperation and
competition. Dissertation Abstracts International, Section B: The Sciences and Engineering,
60, 4965.
LaFrance, M. (1979). Nonverbal synchrony and rapport: Analysis by the cross-lag panel
technique. Social Psychology Quarterly, 42, 66–70.
Lewin, K., Lippitt, R., & White, R. K. (1939). Patterns of aggressive behavior in experi-
mentally created “social climates.” Journal of Social Psychology, 10, 271–299.
Lopresti-Goodman, S. M., Richardson, M. J., Silva, P. L., & Schmidt, R. C. (2009).
Period basin of entrainment for unintentional visual coordination. Journal of Motor
Behavior, 40, 3–10.
Lumsden, J., Miles, L. K., Richardson, M. J., Smith, C. A., & Macrae, C. (2012). Who
syncs? Social motives and interpersonal coordination. Journal of Experimental Social
Psychology, 48, 746–751.
Mace, W. M. (1977). James J. Gibson’s strategy for perceiving: Ask not what’s inside your
head, but what your head’s inside of. In R. Shaw & J. Bransford (Eds.), Perceiving,
acting, and knowing: Toward an ecological psychology (pp. 43–65). Hillsdale, NJ: Lawrence
Erlbaum.
Macrae, C. N., Duffy, O. K., Miles, L. K., & Lawrence, J. (2008). A case of hand waving:
Action synchrony and person perception. Cognition, 109(1), 152–156.
Marr, D. (1982). Vision. San Francisco, CA: W. N. Freeman.
Marsh, K. L. (2010). Sociality from an ecological, dynamical perspective. In G. R. Semin &
G. Echterhoff (Eds.), Grounding sociality: Neurons, minds, and culture (pp. 43–71).
London: Psychology Press.
Marsh, K. L. (2013). Coordinating social beings in motion. In K. L. Johnson & M. Shiffrar
(Eds.), People watching: Social, perceptual, and neurophysiological studies of body perception
(pp. 236–257). New York: Oxford University Press.
Marsh, K. L. (2015). Social ecological context of conversing. Ecological Psychology, 27(4),
310–334. doi:10.1080/10407413.2015.1086229
204  Kerry L. Marsh

Marsh, K. L., Isenhower, R. W., Richardson, M. J., Helt, M., Verbalis, A. D.,
Schmidt, R. C., & Fein, D. (2013). Autistic and social disconnection in interpersonal
rocking. Frontiers in Integrative Neuroscience, 7(4), 1–8. doi:10.3389/fnint.2013.00004
Marsh, K. L., Johnston, L., Richardson, M. J., & Schmidt, R. C. (2009). Toward a radi-
cally embodied, embedded social psychology. European Journal of Social Psychology, 39,
1217–1225.
Marsh, K. L., & Meagher, B. R. (2016). Affordances and interpersonal coordination. In
P. Passos, K. Davids, & J. Y. Chow (Eds.), Interpersonal coordination and performance in
social systems (pp. 245–258). London: Routledge.
Marsh, K. L., Richardson, M. J., Baron, R. M., & Schmidt, R. C. (2006). Contrasting
approaches to perceiving and acting with others. Ecological Psychology, 18, 1–37.
Marsh, K. L., Richardson, M. J., & Schmidt, R. C. (2009). Social connection through
joint action and interpersonal coordination. Topics in Cognitive Science, 1, 320–339.
Mark, L. S. (1987). Eyeheight-scaled information about affordances: A study of sitting and
stair climbing. Journal of Experimental Psychology: Human Perception and Performance, 13,
361–370.
McArthur, L. Z., & Baron, R. M. (1983). Toward an ecological theory of social percep-
tion. Psychological Review, 90, 215–238.
McNeill, W. H. (1995). Keeping together in time: Dance and drill in human history. Cambridge,
MA: Harvard University Press.
Meagher, B. M. (2014). The emergence of home advantage from differential perceptual
activity. Unpublished doctoral dissertation, University of Connecticut, Storrs, CT.
Meagher, B. R., & Kang, J. J. (2013). Gender identification as a constraint on detecting
multiple affordances. In T. Davis, P. Passos, M. Dicks, & J. A. Weast-Knapp (Eds.),
Studies in perception and action XII: Seventeenth International Conference on Perception and
Action (pp. 106–109). New York: Psychology Press.
Meagher, B. R., & Marsh, K. L. (2014). The costs of cooperation: Action-specific percep-
tion in the context of joint action. Journal of Experimental Psychology: Human Perception
and Performance, 40, 429–444.
Meagher, B. R., & Marsh, K. L. (2015). Testing an ecological account of spaciousness in
real and virtual environments. Environment and Behavior, 7, 782–815.
Meagher, B. R., & Marsh, K. L. (2017). Seeking the safety of sociofugal space:
Environmental design preferences following social ostracism. Journal of Experimental
Social Psychology, 68, 192–199.
Michaels, C. F. (2003). Affordances: Four points of debate. Ecological Psychology, 15,
135–148.
Miles, L. K., Griffiths, J. L., Richardson, M. J., & Macrae, C. (2010). Too late to coor-
dinate: Contextual influences on behavioral synchrony. European Journal of Social
Psychology, 40, 52–60.
Miles, L. K., Lumsden, J., Richardson, M. J., & Macrae, C. (2011). Do birds of a feather
move together? Group membership and behavioral synchrony. Experimental Brain
Research, 211, 495–503.
Miles, L. K., Nind, L. K., & Macrae, C. N. (2009). The rhythm of rapport: Interpersonal
synchrony and social perception. Journal of Experimental Social Psychology, 45,
585–589.
Newtson, D. (1994). The perception and coupling of behavior waves. In R. R. Vallacher &
A. Nowak (Eds.), Dynamical systems in social psychology (pp. 139–167). New York:
Academic Press.
 Dynamics of Synchrony and Joint Action  205

Newtson, D. (1998). Dynamical systems and the structure of behavior. In K. M. Newell

(Ed.), Applications of nonlinear dynamics to developmental process modeling (pp. 199–220).
Mahwah, NJ: Lawrence Erlbaum.
Nie, L., Caban, J., & Marsh, K. L. (2015). Comparing solo and joint syncopation shows
dyadic facilitation. In J. A. Weast-Knapp, M. L. Malone, & D. H. Abney (Eds.), Studies
in perception and action XIII: Eighteenth International Conference on Perception and Action
(pp. 69–72). New York: Psychology Press.
Norman, D. (1988). The psychology of everyday things. New York: Basic Books.
Norman, D. (2002). The design of everyday things (revised and expanded ed.). New York:
Basic Books.
Nowak, A., & Vallacher, R. R. (1998). Dynamical social psychology. New York: Guilford
Press.
Nowak, A., Vallacher, R. R., & Zochowski, M. (2005). The emergence of personality:
Dynamic foundations of individual variation. Developmental Review, 25, 351–385.
Open Science Collaboration (2015). Estimating the reproducibility of psychological
science. Science, 349(6251), aac4716.
Paxton, A. & Dale, R. (2013). Argument disrupts interpersonal synchrony. Quarterly
Journal of Experimental Psychology, 66, 2092–2102.
Rauthmann, J. F., Sherman, R. A., & Funder, D. C. (2015). Principles of situation
research: Towards a better understanding of psychological situations. European Journal
of Personality, 29, 363–381.
Rauthmann, J. F., Sherman, R. A., Nave, C. N., & Funder, D. C. (2015). Personality-
driven situation experience, contact, and construal: How people’s personality traits
predict characteristics of their situations in daily life. Journal of Research in Personality,
55, 98–111.
Reed, E. S. (1996). Encountering the world: Toward an ecological psychology. New York:
Oxford University Press.
Reiss, H. T. (2008). Reinvigorating the concept of the situation in social psychology.
Personality and Social Psychology Review, 12, 311–329.
Richardson, M. J., Dale, R., & Marsh, K. L. (2014). Complex dynamical systems in
social and personality psychology: Theory, modeling and analysis. In H. T. Reis &
C. M. Judd (Eds.), Handbook of research methods in social and personality psychology
(2nd ed., pp. 251–280). New York: Cambridge University Press.
Richardson, M. J., Garcia, R. L., Frank, T. D., Gregor, M., & Marsh, K. L. (2012).
Measuring group synchrony: A cluster-phase method for analyzing multivariate move-
ment time-series. Frontiers in Fractal Physiology, 3, 405. doi:10.3389/fphys.2012.00405
Richardson, M. J., Marsh, K. L., & Baron, R. M. (2007). Judging and actualizing intraper-
sonal and interpersonal affordances. Journal of Experimental Psychology: Human Perception
and Performance, 33, 845–859.
Richardson, M. J., Marsh, K. L., Isenhower, R. W., Goodman, J. R. L., & Schmidt, R. C.
(2007). Rocking together: Dynamics of unintentional and intentional interpersonal
coordination. Human Movement Science, 26, 867–891.
Richardson, M. J., Marsh, K. L., & Schmidt, R. C. (2005). Effects of visual and verbal
couplings on unintentional interpersonal coordination. Journal of Experimental
Psychology: Human Perception and Performance, 31, 62–79.
Richardson, M. J., Marsh, K. L., & Schmidt, R. C. (2010). Challenging the egocentric
view of coordinated perceiving, acting and knowing. In L. F. Barrett, B. Mesquita, &
E. R. Smith (Eds.), Mind in context (pp. 307–333). New York: Guilford Press.
206  Kerry L. Marsh

Rietveld, E., & Kiverstein, J. (2014). A rich landscape of affordances. Ecological Psychology,
26, 325–352.
Riley, M. A., Richardson, M. J., Shockley, K., & Ramenzoni, V. C. (2011). Interpersonal
synergies. Frontiers in Psychology, 2, 38. doi:10.3389/fpsyg.2011.00038
Runeson, S., & Frykholm, G. (1983). Kinematic specification of dynamics as an infor-
mational basis for person-and-action perception: Expectation, gender recognition, and
deceptive intention. Journal of Experimental Psychology: General, 112, 585–615.
Schmidt, R. C., Bienvenu, M., Fitzpatrick, P. A., & Amazeen, P. G. (1998). A compari-
son of intra- and interpersonal interlimb coordination: Coordination breakdowns and
coupling strength. Journal of Experimental Psychology: Human Perception and Performance,
24, 884–900.
Schmidt, R. C., Carello, C., & Turvey, M. T. (1990a). Critical fluctuations and non-
equilibrium phase transitions in the coordination between two people. Journal of
Experimental Psychology: Human Perception and Performance, 16, 227–247.
Schmidt, R. C., Carello, C., & Turvey, M. T. (1990b). Phase transitions and critical fluc-
tuations in the visual coordination of rhythmic movements between people. Journal of
Experimental Psychology: Human Perception and Performance, 16, 227–247.
Schmidt, R. C., Christianson, N., Carello, C., & Baron, R. (1994). Effects of social
and physical variables on between-person visual coordination. Ecological Psychology, 6,
159–183.
Schmidt, R. C., & O’Brien, B. (1997). Evaluating the dynamics of unintended inter-
personal coordination. Ecological Psychology, 9, 189–206.
Schmidt, R. C., Fitzpatrick, P., Caron, R., & Mergeche, J. (2011). Understanding social
motor coordination. Human Movement Science, 30, 834–845.
Schmidt, R. C., Morr, S., Fitzpatrick, P., & Richardson, M. J. (2012). Measuring the
dynamics of interpersonal synchrony. Journal of Nonverbal Behavior, 36, 263–279.
Schmidt, R. C., & Richardson, M. J. (2008). Dynamics of interpersonal coordination.
In A. Fuchs & V. K. Jirsa (Eds.), Coordination: Neural, behavioral and social dynamics
(pp. 281–307). Berlin: Springer-Verlag.
Schoggen, P. (1989). Behavioral settings: A revision and extension of Roger G. Barker’s Ecological
Psychology. Stanford, CA: Stanford University Press.
Sebanz, N., Bekkering, H., & Knoblich, G. (2006). Joint action: Bodies and minds
moving together. Trends in Cognitive Sciences, 10, 71–76.
Sebanz, N., Knoblich, G., & Prinz, W. (2005). How two share a task: Corepresenting
stimulus–response mappings. Journal of Experimental Psychology: Human Perception and
Performance, 31(6), 1234–1246.
Shockley, K., Richardson, D. C., & Dale, R. (2009). Conversation and coordinative
structures. Topics in Cognitive Science, 1, 305–319.
Shockley, K., Santana, M. V., & Fowler, C. A. (2003). Mutual interpersonal postural
constraints are involved in cooperative conversation. Journal of Experimental Psychology:
Human Perception and Performance, 29, 326–332.
Silva, P., Esteves, P., Correia, V., Davids, K., Araújo, D., & Garganta, J. (2015). Effects
of manipulating player numbers vs. field dimensions on inter-individual coordination
during youth football small-sided games. International Journal of Performance Analysis of
Sport, 15, 641–659.
Silva, P., Garganta, J., Araújo, D., Davids, K., & Aguiar, P. (2013). Shared knowledge or
shared affordances? Insights from an ecological dynamics approach to team coordina-
tion in sports. Sports Medicine, 43, 765–772.
 Dynamics of Synchrony and Joint Action  207

Smith, E. R., & Semin, G. R. (2004). Socially situated cognition: Cognition in its social
context. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 36,
pp. 53–117). San Diego, CA: Elsevier Academic Press.
Soodak, H., & Iberall, A. S. (1987). Thermodynamics and complex systems. In F. E. Yates
(Ed.), Self-organizing systems: The emergence of order (pp. 460–469). New York: Plenum.
Ten Berge, M. A., & De Raad, G. (2002). The structure of situations from a personality
perspective. European Journal of Personality, 16, 81–102.
Turvey, M. T. (1990). Coordination. American Psychologist, 45, 938–953.
Turvey, M. T. (2007). Action and perception at the level of synergies. Human Movement
Science, 26, 657–697.
Turvey, M. T., & Fonseca, S. (2014). The medium of haptic perception: A tensegrity
hypothesis. Journal of Motor Behavior, 46, 143–187.
Turvey, M. T., & Shaw, R. (1999). Ecological foundations of cognition. I: Symmetry and
specificity of animal–environment systems. Journal of Consciousness Studies, 6, 85–110.
Vacharkulksemsuk, T., & Frederickson, B. L. (2012). Strangers in sync: Achieving
embodied rapport through shared movements. Journal of Experimental Social Psychology,
48, 399–402.
Valdesolo, P., Ouyang, J., & DeSteno, D. (2010). The rhythm of joint action: Synchrony
promotes cooperative ability. Journal of Experimental Social Psychology, 46, 693–695.
Vallacher, R. R., & Nowak, A. (Eds.) (1994). Dynamical systems in social psychology. San
Diego, CA: Academic Press.
Vallacher, R. R., Read, S. J., & Nowak, A. (2002). The dynamical perspective in per-
sonality and social psychology. Personality and Social Psychology Review, 6, 264–273.
Vallacher, R. R., Van Geert, P., & Nowak, A. (2015). The intrinsic dynamics of psycho-
logical process. Current Directions in Psychological Science, 24, 58–64.
Van der Kamp, J., Savelsbergh, G. J. P., & Davis, W. E. (1998). Body-scaled ratio as a con-
trol parameter for prehension in 5- to 9-year-old children. Developmental Psychobiology,
33, 351–361.
Van Ulzen, N. R., Lamoth, C. J. C., Daffertshofer, A., Semin, G. R., & Beek, P. J.
(2008). Characteristics of instructed and uninstructed interpersonal coordination while
walking side-by-side. Neuroscience Letters, 432, 88–93.
Vicaria, I. M., & Dickens, L. (2016). Meta-analyses of the outcomes of interpersonal
coordination. Journal of Nonverbal Behavior, 40, 335–361.
Warren, W. H. (1984). Perceiving affordances: Visual guidance of stair climbing. Journal of
Experimental Psychology: Human Perception and Performance, 10, 683–703.
Wiltermuth, S. S. & Heath, C. (2009). Synchrony and cooperation. Psychological Science,
20, 1–5.
Yang, Y., Read, S. J., Denson, T. F., Xu, Y., Zhang, J., & Pedersen, W. C. (2014). The
key ingredients of personality traits: Situations, behaviors, and explanations. Personality
and Social Psychology Bulletin, 40, 79–91.
Yang, Y, Read, S. J., & Miller, L. C. (2009). The concept of situations. Social and Personality
Psychology Compass, 3, 1018–1037.
Ye, L., Cardwell, W., & Mark, L. S. (2009). Perceiving multiple affordances for objects.
Ecological Psychology, 21, 185–217.
10
FROM INTERACTION TO
SYNCHRONIZATION IN
SOCIAL RELATIONS
Robin R. Vallacher and Andrzej Nowak1

Interpersonal relations are established through social interactions. This connection

is obvious, of course, but the means by which a complex relationship can develop
from simple interaction is not as clear. One could argue that the exchange of
compliments and agreed-upon attitudes in social interaction is the key to estab-
lishing a bond between people (e.g., Byrne, 1971; Swann, 1990; Thibaut &
Kelley, 1959). But relationships are not limited to individuals who encounter
each other already inclined to think, feel, and act the same way, nor to individuals
who feel rapport at first sight. Relationships, like other social processes, evolve
in time, and do so in a manner that promotes change in both individuals. We
suggest that the development of interpersonal relationships can be understood by
framing social interaction in terms of synchronization, a process that characterizes
coupled dynamical systems (Kaneko, 1993; Shinbrot, 1994). The central idea is
that interacting individuals are not static entities, maintaining fixed thoughts and
action tendencies, but instead represent separate systems capable of displaying rich
dynamics, and that the synchronization of their respective dynamics produces a
higher-order system—the dyadic relationship—with its own dynamic properties.
In this chapter, we document the role of synchronization in close relationships
and present a formal model, implemented in computer simulations, that illustrates
how synchronization develops in interpersonal contexts.
People in their daily lives clearly recognize the importance of synchronization
in their social relationships. Thus, two individuals comprising a positive relation-
ship are said to “be on the same wavelength” or to “resonate with one another,”
suggesting that their internal states and overt behaviors are coordinated on some
timescale. As the relationship develops, their spontaneous reactions to events
become increasingly similar and their moment-to-moment thoughts and feelings
when in one another’s presence rise and fall in tandem. Individuals in problematic
 From Interaction to Synchronization  209

relationships, in contrast, are described as “being out of synch with each other,”
with their respective reactions and feelings following different temporal trajecto-
ries. They may display the same pattern of thought and behavior when exposed
to strong and unambiguous stimuli—a traumatic event or a celebratory occasion,
for example—but when the stimuli diminish, the individuals fall back into their
respective idiosyncratic patterns that evolve independently.
Despite the centrality of interpersonal synchronization in people’s intuitive
understanding of relationships, this construct has received sparse attention in social
psychology. For the most part, theorists and researchers have conceptualized both
short-term interactions and long-term relationships in terms of global variables
that characterize the interaction or relationship as a whole. The interactions
between individuals or among members of a group, for example, are commonly
investigated in terms of constructs such as cooperation versus competition, com-
patibility of motives and goals, and the structure of norms and roles (e.g., Biddle &
Thomas, 1966; Dawes, 1980; Levine & Moreland, 1998; Berkowitz & Walster,
1976; Wish, Deutsch, & Kaplan, 1976). In some approaches, the temporal aspects
of social interaction are emphasized, but the focus is usually restricted to the
development of strategies for achieving personal or shared goals (e.g., Axelrod,
1984; Messick & Liebrand, 1995; Thibaut & Kelley, 1959).
Synchronization between individuals takes on a different and more precise
meaning when viewed from the perspective of nonlinear dynamical systems. In
the dynamical account, two (or more) people are synchronized to the extent
that the actions, thoughts, and feelings of one person are related over time to
the actions, thoughts, and feelings of the other person or persons (cf. Marsh,
Richardson, & Schmidt, 2009; Nowak & Vallacher, 1998; Nowak, Vallacher, &
Zochowski, 2002). Because the temporal pattern of a person’s phenomenal expe-
rience and overt behavior can be quite complex and hard to predict (Vallacher &
Nowak, 1997), assessing the synchronization of two (or more) such patterns
poses a daunting task. Fortunately, methods and tools have been developed to
characterize temporal coordination in physical systems, including those capable
of highly complex and chaotic behavior (e.g., Shinbrot, 1994). We suggest that
this approach, based on principles of nonlinear dynamical systems, can be adapted
to the investigation of synchronization in interpersonal systems. In particular, we
employ coupled logistic equations—the simplest dynamical systems capable of
chaotic behavior—to investigate synchronization dynamics.

The Coordination of Behavior

The most basic and ubiquitous form of synchronization is the correlation over
time between individuals with respect to some parameter. This phenomenon has
been investigated extensively in the context of movement coordination (e.g.,
Beek & Hopkins, 1992; Bernieri, Reznick, & Rosenthal, 1988; Delaherche
et al., 2012; Hove & Risen, 2009; Macrae, Duffy, & Lawrence, 2008; Marsh,
210  Robin R. Vallacher and Andrzej Nowak

Richardson, & Schmidt, 2009; Miles, Griffiths, Richardson, & Macrae, 2010;
Miles, Nind, Henderson, & Macrae, 2010; Miles, Nind, & Macrae, 2009;
Newtson, 1994; Pelose, 1987; Schmidt, Beek, Treffner, & Turvey, 1991;
Schmidt & Richardson, 2008; Shockley, Santana, & Fowler, 2003; Turvey,
1990). For example, a person is asked to swing his or her legs in time to a met-
ronome and the other person is asked to match those movements. Two forms
of coordination are typically observed: in-phase synchronization (the individuals
swinging their legs in unison) and anti-phase synchronization (the individuals
swinging their legs with the same frequency, but in the opposite direction). In
a variation of this approach, the individuals are asked to synchronize out-of-
phase. Up to a certain rate of movement, people can follow these instructions
successfully. But beyond a certain tempo, the individuals cannot maintain this
form of synchronization, and switch spontaneously to in-phase synchronization.
Hysteresis—a signature feature of nonlinear dynamical systems—is also com-
monly observed in this research (cf. Kelso, 1995). Thus, when the tempo is
decreased to some value, the individuals can re-establish anti-phase coordina-
tion, but this tempo is significantly lower than the point at which they originally
started to coordinate in-phase.
In-phase coordination, then, is the easiest form to achieve and maintain, and
it may be the only form that can be sustained as coordination becomes more
difficult (e.g., as the tempo of behavior is increased). This constraint on coordi-
nation may find expression in more involving social situations. Under conditions
conducive to high stress, for example, people may find it impossible to coor-
dinate their behavior in any form other than in-phase. In a crowded club or
auditorium that suddenly bursts into flames, taking turns in exiting through the
doors is the only form of coordination that would make evacuation possible, yet
the occupants are typically unable to do so. In similar fashion, when the level of
emotionality reaches a critical point in a conversation, the individuals may find it
impossible to take turns speaking.
In recent years, psychologists have investigated the role of behavioral synchro-
nization in social relations, with particular emphasis on affect and liking. Much
of this research has focused on behavioral mimicry, commonly referred to as the
chameleon effect (e.g., Chartrand & Bargh, 1999; Dijksterhuis & Bargh, 2001).
Mimicry is ubiquitous in interpersonal life, occurring in about a third of ordinary
social interactions, and it typically unfolds with little or no conscious awareness
on the part of interacting individuals (e.g., Chartrand & Bargh, 1999; Lakin &
Chartrand, 2003). People’s unconscious tendency to mimic one another occurs
in multiple channels, including facial expressions, mannerisms, physical gestures,
and posture (e.g., Dijksterhuis, 2005). Mimicry appears essential to empathy,
rapport, and liking in social interaction. Even something as seemingly trivial as
two people rubbing their respective faces in unison functions as “social glue” that
binds people together (e.g., Bernieri, 1988; Chartrand & Bargh, 1999; Chartrand,
Maddux, & Lakin, 2005; Dijksterhuis, 2005; Gueguen, 2004; LaFrance, 1979;
 From Interaction to Synchronization  211

Lakin & Chartrand, 2003; Lakin, Jefferis, Cheng, & Chartrand, 2003; Maurer &
Tindall, 1983; Stel, Van Baaren, & Vonk, 2008; Stel & Vonk, 2010; Van Baaren,
Holland, Kawakami, & Van Knippenberg, 2004). But recent research suggests
that synchronization rather than mimicry may be responsible for this effect.
People who synchronize their respective finger movements to a metronome
rather than to one another’s movements, for example, tend to like one another
(Hove & Risen, 2009).

The Synchronization of Internal States

The synchronization of overt behavior is clearly an important component of
interpersonal relations, but the synchronization of people’s internal states—their
thoughts and feelings—has broader significance for social relations (e.g., Nowak
et al., 2002; Tickle-Degnen & Rosenthal, 1987). Internal states represent a broad
category, ranging from states that operate on short timescales, such as mood or
arousal, to features that reflect fairly enduring properties of a person that operate
on considerably longer timescales, such as personality traits, values, goals, and
temperament. A concern with synchronization of internal states is salient for
people even in simple social interaction. In preparing for an interaction with a
stranger, for example, people tend to change their mood to match the anticipated
affective state of the stranger, even if this means toning down a positive mood in
favor of a more subdued mood (e.g., Erber, Wegner, & Thierrault, 1996).
The quality of a relationship is signaled by the degree to which individuals
coordinate in-phase with respect to their respective internal states (e.g., Baron,
Amazeen, & Beek, 1994; Butler & Randall, 2013; McGrath & Kelly, 1986;
Nowak, Vallacher, & Zochowski, 2005; Guastello, Pincus, & Gunderson, 2006;
Sbarra & Hazan, 2008; Tickle-Degnen & Rosenthal, 1987). Indeed, the synchro-
nized ebb and flow of moods, emotional reactions, and judgments may convey
deeper insight into the nature of a relationship than might global properties such
as the average sentiment, the amount of information exchanged, or the summary
judgments reached. In essence, synchronization binds individuals in a relation-
ship into a higher-order functional unit that has its own dynamic properties. In
a close romantic relationship, for example, the synchronized flow of thoughts,
feelings, and actions cannot be reduced to the individuals’ personal dynamics. We
speak of the “couple” as if it, rather than the separate individuals, constituted the
relevant unit of analysis. The emergence of higher-order functional units through
synchronization can be used to characterize other enduring and meaningful rela-
tionships, including long-term friendships, tight-knit groups, and successful work
and sports teams.
To capture the essence of interpersonal synchronization and explore its rel-
evance and generality, we propose a formal model that depicts the emergence of
synchronization of both overt behavior and internal states in social relationships.
The key assumptions of the model are tested in computer simulations of the
212  Robin R. Vallacher and Andrzej Nowak

emergence of synchronization in a dyad as the individuals develop a progressively

closer relationship. Based on the simulation results, we discuss the implications
of synchronization dynamics for topics and issues of interest in the study of social
experience.

Humans as Nonlinear Dynamical Systems

The backdrop and rationale for the model are derived from the study of nonlinear
dynamical systems (e.g., Haken, 1978; Holland, 1995; Schuster, 1984; Strogatz,
2003; Weisbuch, 1992). In its most basic sense, a dynamical system is a set of
interconnected elements that evolve in time, with the elements influencing, or
adjusting, to each other to achieve a coordinated state that characterizes the sys-
tem as a whole. Because of the influence arising within the system, the system
may display a pattern of change in some system-level property in the absence
of external influence. Depending on the topic, elements can represent anything
from atoms to animals and planets. The task of dynamical systems theory is to
specify how the elements interact with each other over time to promote global
properties and behavior characterizing the system as a whole. In physics, the issue
is how subatomic particles interact to produce visible matter and forces; in ecol-
ogy, the focus is how animals interact to generate and maintain a balance between
predator and prey; in cosmology, a central concern is how planets and other
celestial bodies influence each other to produce stable orbits.
Nonlinear systems are noteworthy for their tendency to demonstrate emer-
gence. This simply means that the higher-order property or behavior that results
from the mutual influence among elements cannot be reduced to the properties
of the elements. The stable pattern of planetary orbits, for example, depends on
the mutual gravitational influences among the planets, not on the properties of
the planets in isolation. Rather than being imposed on the system by outside
forces, in other words, the higher-level properties and behaviors emerge from the
internal workings of the system itself. For this reason, the process responsible for
emergence is referred to as self-organization.
Once a higher-level property or behavior emerges in a dynamical system, it
constrains the behavior of the elements that give rise to it, and can capture the
behavior of new elements that enter the system. When a stable orbit emerges in
a planetary system, for example, it constrains the orbits of each planet and can
capture the orbit of passing asteroids and comets. Because it functions to “attract”
the behavior of both existing and new elements, the system-level property is
referred to as an attractor. A system’s attractor actively resists change due to outside
influences, and thus serves to stabilize the system. If system-level change occurs, it
is because the mutual influences among the elements are weakened, so that they
evolve independently of one another, which undermines the attracting power of
the system’s higher-order property or behavior. From this disassembled state of
affairs, however, the system is primed for self-organization and emergence to a
 From Interaction to Synchronization  213

new higher-order state that provides a different configuration of the lower-level

elements. A dynamical system, in other words, tends to display periods of stabil-
ity and resistance to disruption and change, punctuated by periods of disassembly
that set the stage for a new round of self-organization and emergence.
Because humans display change over time, and often do so in the absence of
external influence, they can be conceptualized—and presumably investigated—as
dynamical systems (Vallacher, Van Geert, & Nowak, 2015). Every domain of
human functioning reflects the complementary tendencies of stability and change
due to the mutual influence among the elements comprising the domain. The
mind is in constant motion, generating an endless flow of momentary thoughts
and feelings, but it also demonstrates strong attractor tendencies that provide
coherence and stability in social judgment. Social interactions, meanwhile, dem-
onstrate a continuous flow of words and gestures that is never replicated exactly,
yet they admit to regularities and patterns that qualify as attractors. And social
relationships are characterized by the evolution of roles and sentiments, with the
emergence of scripts and agendas that function as attractors for the actions and
thoughts of the partners. In recognition of the complex, dynamic, and emergent
nature of human experience, the approach of nonlinear dynamical systems has
emerged in recent years as an integrative model for the topical landscape of social
psychology (cf. Boker & Wenger, 2007; Butner, Gagnon, Geuss, Lessard, &
Story, 2014; Guastello, Koopmans, & Pincus, 2009; Nowak & Vallacher, 1998;
Vallacher, Read, & Nowak, 2002; Vallacher & Nowak, 1994, 1997, 2007).
A psychological system may have more than one attractor, each representing
a different global system-level property that provides stability and coherence for
the system. Which attractor governs a system’s dynamics at a given point in time
depends on the initial states or starting values of the system’s elements. The set
of initial states that converge on each attractor represents the basin of attraction for
that attractor. For a person or a group characterized by multiple attractors, then,
the process in question can display different equilibrium tendencies, each associ-
ated with a distinct basin of attraction. Within each basin, different initial states
will follow a trajectory that eventually converges on the same stable value. Even
a slight change in the system’s initial state, however, will promote a large change
in the system’s trajectory if this change represents a state that falls just outside
the original basin of attraction and within the basin for an alternative attractor.
The potential for multiple attractors in a psychological system captures the intui-
tion that people can have different or even mutually contradictory patterns of
thought, feeling, and action.
The attractor landscape (e.g., single versus multiple) that characterizes a
dynamical system depends in part on the value of one or more control param-
eters that influence the mutual interactions among the elements of the system.
In the research on synchronization of leg swinging, for example, the tempo of
leg swinging is the relevant control parameter. When the value of this parameter
(i.e., tempo) increases and reaches a threshold, the form of synchronization
214  Robin R. Vallacher and Andrzej Nowak

changes (e.g., from anti-phase to in-phase). The relevant control parameters

for various psychological systems have yet to be identified, but likely candi-
dates include stress, arousal, subjective importance, rate of information intake,
task or judgment difficulty, and perceived control (Vallacher & Nowak, 2007).
Changes in the values of these factors can shape the manner in which momentary
thoughts, feelings, and actions self-organize to promote the emergence of attrac-
tors for the system. Depending on the value of a system’s control parameter, the
system may evolve in very different ways, including convergence on a single
attractor, the generation of two (or more) attractors, and very complex patterns
of evolution resembling randomness (i.e., deterministic chaos). Mild stress or
arousal, for example, may promote a flow of thought and feeling that converges
on a single attractor for judgment and global affect, whereas more intense stress
or arousal may promote a flow of thought and feeling that results in conflicting
attractors for judgment and affect. A particularly intense level of stress or arousal,
meanwhile, might change the interactions of momentary thoughts and feelings,
such that the person’s global judgment or affect is characterized by chaos rather
than by attractors with fixed values.
Dynamical systems come in many forms, but the simplest system capable of
complex behavior is the logistic equation (Feigenbaum, 1978). The logistic equa-
tion involves repeated iteration, so that the output value representing the system’s
state (x) at one step (n) becomes the input value for the equation at the next step
(n+1). The current value of the system’s state (which varies between 0 and 1),
in other words, depends on the state’s previous value—that is, xn+1 = f(xn). This
dependency has two opposing components. First, the higher the previous value,
the higher the current value; specifically, xn+1 equals xn multiplied by the value of
the system’s control parameter, represented as r. Second, the higher the previous
value, the lower the current value; specifically, xn+1 equals (1-xn) multiplied by the
value of r. The combined effect of these competing tendencies is expressed as
xn+1 = rxn(1-xn). Depending on the value of r, repeated iteration of the logis-
tic equation can display qualitatively different patterns of behavior (pattern of
changes in x), including the convergence on a single value (i.e., attractor), oscil-
latory (periodic) changes between two or more values, and very complex patterns
of behavior resembling randomness (i.e., deterministic chaos).
We suggest that the logistic equation provides a useful way of conceptual-
izing human dynamics (cf. Nowak & Vallacher, 1998; Nowak et al., 2002). In
this approach, x represents a person’s behavior, and changes in x represent vari-
ations in the intensity of the behavior. The control parameter, r, corresponds to
internal states (e.g., moods, arousal levels, traits, values) that shape the person’s
pattern of behavior (changes in x over time). The opposing forces expressed
by the logistic equation capture the idea of conflict, which is a key concept in
many psychological theories. In classic research on motivational conflict (Miller,
1944), for example, movement toward a goal increases both approach and avoid
tendencies. Research on achievement motivation (e.g., Atkinson, 1964), in turn,
 From Interaction to Synchronization  215

has identified two concerns, the desire for success and the fear of failure, that
combine in different ways to produce resultant motivation. Research on thought
suppression, meanwhile, suggests that attempts at suppression activate an ironic
process that works at cross-purposes with the attempted suppression (cf. Wegner,
1994). More generally, the assumption that conflicting forces or tendencies are
central to human experience is common to most issues of psychological inter-
est (e.g., short-term versus long-term self-interest, impulse versus self-control,
autonomy versus social identity, egoism versus altruism).

Social Interaction as the Coupling of

Nonlinear Dynamical Systems
If an individual can be represented as a dynamical system, then social interaction can be
conceptualized and investigated as the coupling or synchronization of two (or more)
dynamical systems. Accordingly, we have employed coupled logistic equations, which
have a successful track record in modeling the synchronization of physical systems
(e.g., Shinbrot, 1994), to model the synchronization of interacting individuals. The
basic idea is that when the value of the behavior (x) for one person (i.e., equation)
depends not only on its previous value but also to some degree on the value of x for
the other equation, the two equations tend to become synchronized over time. This
idea is easy to appreciate in the context of human interaction: the behavior of each
person depends not only on his or her preceding behavior, but also on the preceding
behavior of the other person. In formal terms, this influence or coupling is operational-
ized according to the following equations:
r1x1 (t )(1 − x1 (t )) + α r2 x 2 (t )(1 − x 2 (t )) (10.1)
x1 (t + 1) = 
1+α
r2 x 2 (t )(1 − x 2 (t )) + α r1x1 (t )(1 − x1 (t ))  (10.2)
x 2 (t + 1) =
1+α
To the value of the dynamical variable representing one person’s behavior (x1),
one adds a fraction, denoted by (alpha), of the value of the dynamical variable
representing the behavior of the other person (x2). The size of alpha represents
the strength of coupling and reflects the mutual influence or interdependency of
the interacting individuals. When the fraction is 0, there is no coupling on the
behavior level, whereas a value of 1.0 signifies that each person’s behavior is deter-
mined equally by his or her preceding behavior and the preceding behavior of the
other person. Intermediate values of alpha represent intermediate values of coupling.

Modeling the Synchronization of Behavior

When the control parameters (r) of each person have the same value, the depend-
ence between their respective behaviors causes the systems to synchronize
216  Robin R. Vallacher and Andrzej Nowak

completely, with the temporal changes in x1 and x2 becoming identical (e.g.,

Kaneko, 1984). In the real world, of course, the respective control parameters
of two individuals are rarely (if ever) identical, nor does the same degree of
mutual influence or interdependence characterize all relationships. In an initial
set of computer simulations, therefore, we investigated how the coordination
of dynamical variables (corresponding to individuals’ behavior) depends on the
similarity of control parameters (corresponding to individuals’ internal states) and
the strength of coupling (corresponding to the strength of influence between
individuals).
We began each simulation with a random value of the dynamical variables
for each person, drawn from a uniform distribution that varied from 0 to 1. We
assume that individuals often tend to display complex temporal patterns that can
be difficult to identify and may seem unpredictable. Accordingly, the control
parameter for one system (one person) was held constant at a value of 3.67,
which corresponds to low levels of chaotic behavior. We systematically varied
the value of the control parameter for the other person between values of 3.6 and
4.0, which corresponds to the highest value of the chaotic regime. We let the
simulations run for 300 steps, so that each system had a chance to come close to
its pattern of intrinsic dynamics (i.e., pattern of changes in x) and both systems
had a chance to synchronize. For the next 500 simulation steps, we recorded the
values of the dynamical variables (behavior) for each system and measured the
degree of synchronization (i.e., the difference between the dynamical variables).
The primary results reflected the intuitions expressed earlier. The degree of
synchronization tended to increase both with increases in alpha (mutual influ-
ence) and with increasing similarity in r. This suggests that mutual influence
and similarity in internal states can compensate for one another in achieving a
particular level of synchronization in social interaction. Thus, despite relatively
weak mutual influence, two people can achieve a high degree of synchronization
if their respective internal states are similar. If the individuals have different inter-
nal states, however, high mutual influence—for example, constant monitoring,
communication, or mutual reinforcement—is necessary to maintain the same
level of synchronization.
The simulation results also revealed less straightforward, but interesting effects
for behavior coordination due to variation in alpha. For very high values of alpha,
the predominant mode of coordination was in-phase synchronization. But for
low values of alpha, the simulation results revealed different modes of coordi-
nation. In addition to in-phase synchronization, the coupled systems displayed
anti-phase synchronization (analogous to turn-taking in behavior), independence
in behavior, and yet more complex forms of coordination. This suggests that
when mutual influence is relatively weak, a richer repertoire of modes of coor-
dination is available. There was also a tendency for the systems to stabilize each
other under relatively weak coupling, so that each system behaved in a more
regular (e.g., less chaotic) manner than it would have without the weak influence
 From Interaction to Synchronization  217

(cf. Ott, Grebogi, & York, 1990). Considered together, these results suggest that
for high levels of mutual influence and control (e.g., constant monitoring and
control), the behavior of one person largely tends to mirror the behavior of the
other partner. For relatively weak mutual influence, in contrast, behavior coor-
dination can be manifest in more complex and less obvious forms. The complex
and nonlinear nature of coordination in this case suggests that observers may find
it difficult to note or understand the ways in which the behaviors of the individu-
als are synchronized.
These results imply that mutual influence in close relationships (cf. Thibaut &
Kelley, 1959) may be a mixed blessing. To be sure, if two individuals have very
dissimilar internal states, they may nonetheless achieve a fair degree of coordi-
nation by directly influencing one another’s behavior. But at the same time,
this scenario creates the potential for instability in the relationship because the
dynamics of the two people will immediately diverge as soon as the influence is
weakened. In contrast, a high level of similarity in the setting of control parame-
ters can maintain synchronization for a considerable period of time when mutual
influence is broken. Even if the individuals’ behaviors do not synchronize in
time, the overall form of their respective dynamics will remain similar, making
the reestablishment of coordination at a later time relatively easy. In couples
characterized by similarity in their respective internal states, then, relatively little
influence or communication is required to maintain coordination and thus main-
tain the relationship. Couples characterized by weak similarity in their internal
states, in contrast, may be able to maintain coordination in their behavior, but
only through strong and sustained attempts at mutual influence. Because such
couples thus are at heightened risk for a breakdown in coordination, they may
engage in events together that induce a common mood and also bring about
coordination on a behavioral level. Activities that are affectively positive (e.g.,
dancing, entertainment, sexual relations) can clearly have this effect, but so can
negatively toned events, such as witnessing a tragic event or having a heated
argument.

Modeling the Synchronization of Internal States

Modeling the synchronization of internal states (i.e., the similarity of control
parameters) is straightforward, at least in principle. One need only assume that
on each simulation step, the values of each person’s control parameter drifts in
the direction of the value of the other person’s control parameter. How quickly
the control parameters begin to converge is determined by the rate of this drift
and the size of the initial discrepancy between the values of the respective control
parameters. This mechanism works fine if one can assume that both individuals
can directly observe or estimate the settings of one another’s control parameters.
However, direct observation of the internal states of an interaction partner may
be difficult, or even impossible in some cases. People engage in considerable
218  Robin R. Vallacher and Andrzej Nowak

effort when communicating or attempting to infer one another’s dispositional

qualities and other internal states (cf. Jones & Davis, 1965; Kunda, 1999; Nisbett &
Ross, 1980; Wegner & Vallacher, 1977). Despite the multiplicity of cognitive
means available to people, the precise nature of a person’s momentary state or
relevant chronic disposition may be opaque to others—or wildly misperceived.
Interacting individuals may lack insight or clear inferences into one another’s
internal states, but they may nonetheless achieve similarity in these states by
means of behavioral coordination. This possibility follows from several lines of
social psychological research. Research on facial feedback (e.g., Strack, Martin, &
Stepper, 1988), for example, has demonstrated that when people are induced to
mechanically adopt a specific facial configuration linked to a particular affective
state (e.g., disgust), they tend to experience that affective state. The matching
of one’s internal states to one’s overt behavior has also been demonstrated for
behavior that is interpersonal in nature. Even role-playing, in which a person fol-
lows a scripted set of actions vis-à-vis another person, often produces noteworthy
changes in the role-player’s attitudes and values to match his or her overt actions
(e.g., Zimbardo, 1970).
To simulate this mechanism in our model, we allowed each system (i.e., per-
son) to modify the value of its own control parameter in order to match the other
system’s pattern of behavior. So although the exact value of the other person’s
control parameter is invisible to the person, he or she remembers the partner’s
most recent set of behaviors (i.e., the most recent values of x), as well as his or
her own most recent behaviors. The person compares the other person’s pattern
with his or her own, and adjusts his or her own control parameter until the two
behavior patterns match (Zochowski & Liebovitch, 1997). If the other person’s
observed behavior pattern is more complex (e.g., more chaotic) than the per-
son’s own behavior pattern, he or she adjusts (increases) the value of his or her
own control parameter (internal state) until similarity in their respective behavior
patterns is achieved. But if the other person’s behavior is less complex than the
person’s own behavior, the person decreases slightly the value of his or her own
control parameter. In effect, interacting individuals can “discover” the internal
state of one another by monitoring and matching the dynamics of one another’s
behavior.
Figure 10.1 displays the progressive similarity of internal states in accord-
ance with this scenario under relatively weak coupling (alpha = 0.25). Time in
simulation steps is represented on the x-axis, and the difference between the
two systems in their respective control parameters (thick gray line) and in their
dynamical variables (thin black line) is portrayed on the y-axis. The figure shows
that the two systems become progressively similar in the values of their control
parameters and perfectly synchronized in their behavior over time. These results
suggest that attempting behavioral synchronization under relatively weak levels
of mutual control over one another’s behavior facilitates the matching of one
another’s internal state.
 From Interaction to Synchronization  219

FIGURE 10.1   evelopment of synchronization under relatively weak mutual

D
influence. From A. Nowak, R. R. Vallacher, & M. Zochowski
(2002). The emergence of personality: Personal stability through
interpersonal synchronization. In D. Cervone & W. Mischel (Eds.),
Advances in personality science (Figure 1, p. 305). New York: Guilford
Press. Reprinted with permission of Guilford Press.

Figure 10.2 displays very different results when the simulations were per-
formed with a relatively high value of coupling (alpha = 0.7). Although virtually
perfect synchronization in behavior develops almost immediately, the control
parameters of the two systems fail to converge completely, even after many inter-
action opportunities (1000 simulation steps). Strong mutual influence causes full
synchronization of behavior, even for systems with very different values of their
respective control parameters. Even though full synchronization is achieved, the
two individuals do not have a clue that their control parameters are different. If
the coupling were suddenly removed, the dynamics of the two individuals would
immediately diverge. This suggests that although very strong influence promotes
behavioral coordination, it is likely to hinder synchronization at a deeper level.
The simulation results have interesting implications for interpersonal dynamics.
They suggest, first of all, that there is an optimal level of influence and control
over behavior in social relationships. If influence is too weak, synchronization
may fail to develop at all. On the other hand, very strong influence is likely
to prevent the development of a relationship based on mutual understanding,
emotional similarity, and empathy. Partners who monitor and control one
another may fully synchronize their behavior, but they are unlikely to adopt
the internal states necessary to maintain such behavior in the absence of such
monitoring and control. And as noted earlier, high values of coupling restrict the
220  Robin R. Vallacher and Andrzej Nowak

FIGURE 10.2   evelopment of synchronization under relatively strong mutual

D
influence. From A. Nowak A. Nowak, R. R. Vallacher, &
M. Zochowski (2002). The emergence of personality: Personal
stability through interpersonal synchronization. In D. Cervone &
W. Mischel (Eds.), Advances in personality science (Figure 2, p. 306). New
York: Guilford Press. Reprinted with permission of Guilford Press.

range of possible modes of coordination in a relationship. Mutual influence that

is too strong may therefore result in a relationship that is experienced as highly
predictable and potentially boring. On balance, the optimal degree of coupling is
one that enables effective coordination, but does so without directly influencing
the behavior of the relationship partners. In effect, moderate coupling allows for
the internalization of common internal states in a relationship. And under such
conditions, relationships may develop rich dynamics and the potential to switch
between different modes of coordination. The optimal degree of coupling, in
sum, is the minimal amount necessary to achieve synchronization.

Synchronization and Personality

It is widely assumed in psychology that personality plays an important role in
social interaction and the formation of interpersonal relationships. Indeed, no one
would quibble with the suggestion that people’s stable characteristics (e.g., traits,
values, temperament) bias the choice of interaction partners and influence the
likelihood of success in establishing meaningful and enduring relationships with
those who are chosen. An extraverted person, for example, seeks out a variety of
other people for social interaction and is likely to form a fairly extensive network
of friends and acquaintances. A shy person, on the other hand, approaches social
 From Interaction to Synchronization  221

interaction in a considerably more tentative fashion, interacts with far fewer indi-
viduals, and tends to have a relatively limited circle of friends and acquaintances.
One can look at the process in reverse, however, and consider how social
interactions shape personality. After all, a person’s personality comes from some-
where. The synchronization dynamics model implies that a person’s history of
social interactions plays a crucial role in personality development. In interacting
with someone, the individual develops internal states that facilitate synchroniza-
tion, provided the interactions proceed with moderate rather than strong mutual
influence. Particularly if interactions with someone are sustained over time, the
emergent internal states become engraved in the person’s cognitive-affective
system as attractors. The resultant attractors operate as ready-made parameters
for subsequent interactions, and thus promote efficiency in the person’s social
relations. A person who has repeated interactions with someone whose behav-
ior is characterized by high energy, for example, is likely to develop an internal
setting for energetic behavior that shapes how he or she interacts with others.
Synchronization dynamics also account for personality complexity and explain
why some people have a wider repertoire of stable internal states than do others.
As noted earlier, nonlinear dynamical systems are often characterized by multista-
bility, with more than one attractor shaping the system’s dynamics. The emergence
of multiple attractors characterizes human systems as well (e.g., Nowak &
Vallacher, 1998; Vallacher, Van Geert, & Nowak, 2015). It is easy to appreciate
the emergence of multistability in personality from this perspective.With increas-
ing diversity among the people with whom one interacts and forms relationships,
one is likely to develop a corresponding proliferation of stable internal states—
attractors for cognitive, affective, and behavioral tendencies—for interpersonal
encounters. Even a social recluse is bound to interact with a variety of people and
is certain to have more than one more or less enduring relationship. For those
who are more socially inclined—no doubt the majority of people, given humans’
strong belongingness concerns (Baumeister & Leary, 1995)—many social rela-
tionships are likely to develop over a lifetime, with each one holding potential
for establishing a different internal setting for behavior. People with extensive
and diverse social relations are especially capable of acting in diverse fashion, with
each mode representing control by a different internal state. In effect, each person
has as many personality tendencies as he or she has different types of relationships.
In principle, some people may lack any stable attractors with respect to any
internal state relevant to social relationships. This possibility follows from research
on individual variation in action identification, which has shown that people
differ in the extent to which they think about their behavior across numerous
domains in relatively low-level, mechanistic terms as opposed to higher-level,
comprehensive terms reflecting effects, consequences, and self-evaluative impli-
cations (Vallacher & Wegner, 1989). Because they focus on the granular aspects
of their behavior, “low-level agents” tend to be unclear about what they are
like with respect to broad personality traits, making them more open to social
222  Robin R. Vallacher and Andrzej Nowak

feedback regarding these personal qualities than are “high-level agents.” It is

noteworthy that low-level agents are more likely than their higher-level coun-
terparts to have experienced environmental and interpersonal instability during
childhood (Vallacher & Wegner, 1989). This finding is consistent with the syn-
chronization dynamics model since sustained interaction with specific individuals
is essential for engraving attractors for internal states.
The synchronization dynamics model also provides insight regarding a some-
what puzzling conclusion regarding the relative impact of parents and peers on
children’s personality development (e.g., Dunn & Plomin, 1990; Harris, 1995;
Scarr, 1992). Children spend far more time with their parents than with any sin-
gle peer when they are very young and thus highly impressionable, yet research
suggests that they develop personality traits that are more similar to those of their
peers than to those of their parents. Though controversial, this conclusion is con-
sistent with the synchronization model and the results of its associated computer
simulations. Because the parent–child relationship is characterized by strong cou-
pling (monitoring and control), children have little need to internalize the settings
of their parents’ control parameters in order to achieve and maintain behavioral
coordination with them. In monitoring a child’s behavior, praising it when it is
considered appropriate and providing discipline when it is less so, parents are exert-
ing fairly constant and strong influence over what the child does.To the extent that
the child is aware of the parents’ behavior control, it learns to act in accordance
with the relevant reinforcement contingencies, rules, expectations, and so forth.
Of course, children internalize important lessons from these experiences, and
thus may develop control parameters that resemble their parents’. But children’s
need to adopt their parents’ internal states pales in comparison to the strategic value
of matching the internal states of their peers. Unlike parents, peers are not in a
position to monitor or control a child’s behavior on a daily basis. Beyond that,
peers are considerably less faithful interaction partners, and the surface structure
of their behavior tends to be more erratic as well. Because of these characteristics
of peer relations—relatively weak coupling, relationship instability, and potential
for unpredictable behavior—there is practical value in learning the internal bases
for peers’ behavior. Then, too, it is simply easier to resonate with the interests,
moods, and thoughts of someone who is similar to oneself in age, power, experi-
ences, and competencies. Children certainly love, admire, and appreciate their
parents, but they are less likely to empathize and identify with them than they are
with their peers. In short, because synchronization with peers has all the ingre-
dients for convergence on common control parameters, it is more influential in
establishing children’s characteristic ways of thinking and acting.

Caveats and Conclusions

The dynamical systems perspective on close relations resonates with lay intui-
tion. In both cases, a close relationship is tantamount to the achievement and
 From Interaction to Synchronization  223

maintenance of behavioral and mental synchronization between two people.

So although coupled non-linear dynamical systems were originally developed
in mathematics and the physical sciences, they provide important insight into
the dynamics of close relationships. Computer simulations confirm that syn-
chronization occurs both with respect to dynamical variables, representing overt
behavior, and control parameters, representing internal states (e.g., moods, val-
ues, personality traits, attitudes). These forms of synchronization are intimately
related. In the simulations, a feedback loop exists such that each system changes
its own control parameter to match the level of complexity in the dynamical
variable of the other system. The model also acknowledges that the strength
of mutual influence is an important factor in close relationships. This factor,
represented by the strength of coupling (alpha), captures the essence of various
social processes, including direct communication, promises of reward or threats
of punishment, or other forms of control.
The computer simulations revealed that the synchronization of internal states
in a relationship facilitates the synchronization of behavior. Interestingly, dif-
ficulty experienced in achieving behavioral synchronization may be used as a
guide for achieving synchronization with respect to internal states. By the same
token, very strong behavioral synchronization resulting from strong mutual influ-
ence (i.e., very high values of alpha) tends to hinder rather than facilitate the
development of similarity in people’s internal states. So, for a close relationship to
develop and endure, the partners’ control over one another’s behavior must not
be too strong. Although intuitively appealing, the synchronization model—and
the dynamical systems perspective generally—is relatively new to psychology, so
caution should be exercised when generalizing the results we have described to
human relationships. There are important differences between physical systems
and psychological systems, and it is an open question whether certain unique
features of human thought and action (e.g., consciousness, intentionality) can be
meaningfully reframed in terms of formal properties that are common to dynami-
cal systems in other areas of science. This question, fortunately, has become an
empirical question because of the rigorous methods and tools that have been
developed in recent years (e.g., Boker & Wenger, 2007; Butner et  al., 2014;
Guastello et al., 2009; Nowak & Vallacher, 1998; Vallacher et al., 2015).
Although the logistic equation captures important features of human systems
(e.g., conflicting forces), it is not clear at this early stage which synchroniza-
tion phenomena are specific to the coupling of logistic equations and which
reflect coupled nonlinear systems generally. The use of the logistic equation
was motivated in part by the philosophy of dynamical minimalism (Nowak,
2004). The key point in this approach is that the simplest set of rules should be
employed when attempting to model a real-world issue. As noted at the outset,
the logistic equation is the simplest dynamical system capable of demonstrating
chaotic (highly complex and unpredictable) behavior, which presumably cap-
tures the complex dynamism of human thought and behavior. Nonetheless, the
224  Robin R. Vallacher and Andrzej Nowak

robustness of the results we have observed to this point will ultimately depend
on their generality across different models of dynamical systems.
At this stage in its development, the synchronization model cannot capture
the complexity of human behavior and people’s internal (psychological) states.
Overt behavior can take a variety of forms, not all of which are likely to syn-
chronize equally readily or in the same fashion. Internal states, for their part, are
multidimensional (e.g., emotions, values, plans, temperaments) and it is not clear
whether each manifestation functions in the same way as a control parameter for
interpersonal synchronization. Finally, certain synchronization phenomena may
be unique to humans, reflecting the influence of social and cultural norms, or
perhaps expressing the unique biological properties of humans (e.g., sophisticated
brains capable of self-awareness). So, although the synchronization model may
provide insight into the nature of generic phenomena, empirical research with
humans is critical for determining how these processes play out in the various
domains of human interpersonal experience.
These caveats notwithstanding, the synchronization model and its imple-
mentation with computer simulations point to striking parallels between the
dynamics of coupled logistic equations and the dynamics of close relationships.
The simulation results showing that strong influence can compensate for differ-
ences in people’s internal states, for example, resonates with intuitions regarding
control in close relationships. The results are also consistent with research dem-
onstrating that psychological changes obtained under relatively weak (and subtle)
influence tends to be more enduring than are the more rapid changes obtained
through excessive (rewarding and aversive) control (cf. Vallacher, Nowak, &
Miller, 2003). In sum, framing human interpersonal dynamics in terms of syn-
chronization dynamics prevalent in the natural world provides a conceptual
and methodological bridge between psychological intuitions and the precision
afforded by physical science.

Note
1 Andrzej Nowak acknowledges the support of a grant from the Polish Committee for
Scientific Research [DEC-2011/02/A/HS6/00231].

References
Atkinson, J. W. (1964). An introduction to motivation. Princeton, NJ: Van Nostrand.
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Baron, R. M., Amazeen, P. G., & Beek, P. J. (1994). Local and global dynamics of social
relations. In R. R. Vallacher & A. Nowak (Eds.), Dynamical systems in social psychology
(pp. 111–138). San Diego, CA: Academic Press.
Baumeister, R. F., & Leary, M. R. (1995). The need to belong: Desire for interper-
sonal attachments as a fundamental human motivation. Psychological Bulletin, 117,
497–529.
 From Interaction to Synchronization  225

Beek, P. J., & Hopkins, B. (1992). Four requirements for a dynamical systems approach to
the development of social coordination. Human Movement Science, 11, 425–442.
Berkowitz, L., & Walster, E. (Eds.) (1976). Advances in experimental social psychology
(Vol. 9). New York: Academic Press.
Bernieri, F. J. (1988). Coordinated movement and rapport in teacher–student interactions.
Journal of Nonverbal behavior, 12, 120–138.
Bernieri, F. J., Reznick, J. S., & Rosenthal, R. (1988). Synchrony, pseudosynchrony, and
dissynchrony: Measuring the entrainment process in mother-infant interactions. Journal
of Personality and Social Psychology, 54, 243.
Biddle, B. S., & Thomas, E. J. (Eds.) (1966). Role theory: Concepts and research. New York:
Wiley.
Boker, S. M., & Wenger, M. J. (2007). Data analytic techniques for dynamical systems in the
social and behavioral sciences. Mahwah, NJ: Lawrence Erlbaum.
Butler, E. A., & Randall, A. K. (2013). Emotional coregulation in close relationships.
Emotion Review, 5, 202–210.
Butner, J., Gagnon, K. T., Geuss, M. N., Lessard, D. A., & Story, T. N. (2014). Using
topology to generate and test theories of change. Psychological Methods, 20, 1–25.
Byrne, D. (1971). The attraction paradigm. New York: Academic Press.
Chartrand, T. L., & Bargh, J. A. (1999). The chameleon effect: The perception–behavior
link and social interaction. Journal of Personality and Social Psychology, 76, 6, 893–910.
Chartrand, T. L., Maddux, W., & Lakin, J. (2005). Beyond the perception–behavior
link: The ubiquitous utility and motivational moderators of nonconscious mimicry. In
R. Hassin, J. Uleman, & J. A. Bargh (Eds.), The new unconscious (pp. 334–361). New
York: Oxford University Press.
Dawes, R. M. (1980). Social dilemmas. Annual Review of Psychology, 31, 169–193.
Delaherche, E., Chetouani, M., Mahdhaoui, A., Saint-Georges, C., Viaux, S., & Cohen, D.
(2012). Interpersonal synchrony: A survey of evaluation methods across disciplines.
IEEE Transactions on Affective Computing, 3, 349–365.
Dijksterhuis, A. (2005). Why we are social animals: The high road to imitation as social
glue. In S. Hurley & N. Chater (Eds.), Perspectives of imitation: From cognitive neuroscience
to social science (Vol. 2, pp. 207–220). Cambridge, MA: MIT Press.
Dijksterhuis, A., & Bargh, J. A. (2001). The perception–behavior expressway: Automatic
effects of social perception on social behavior. In M. P. Zanna (Ed.), Advances in experi-
mental social psychology (Vol. 33, pp. 1–40). San Diego, CA: Academic Press.
Dunn, J., & Plomin, R. (1990). Separate lives: Why siblings are so different. New York: Basic
Books.
Erber, R., Wegner, D. M., & Thierrault, N. (1996). On being cool and collected: Mood
regulation in anticipation of social interaction. Journal of Personality and Social Psychology,
70, 757–766.
Feigenbaum, M. J. (1978). Quantitative universality for a class of nonlinear transforma-
tions. Journal of Statistical Physics, 19, 25–52.
Guastello, S. J., Koopmans, M., & Pincus, D. (Eds.) (2009). Chaos and complexity in psychol-
ogy:The theory of nonlinear dynamical systems. New York: Cambridge University Press.
Guastello, S. J., Pincus, D., & Gunderson, P. R. (2006). Electrodermal arousal between
participants in a conversation: Nonlinear dynamics and linkage effects. Nonlinear
Dynamics, Psychology, and Life Sciences, 10, 365–399.
Gueguen, N. (2004). Mimicry and seduction: An evaluation in a courtship context. Social
Influence, 4, 249–255.
226  Robin R. Vallacher and Andrzej Nowak

Haken, H. (1978). Synergetics. Berlin: Springer.

Holland, J. H. (1995). Emergence: From chaos to order. Reading, MA: Addison-Wesley.
Harris, J. R. (1995). Where is the child’s environment? A group socialization theory of
development. Psychological Review, 102, 458–589.
Hove, M. J., & Risen, J. L. (2009). It’s all in the timing: Interpersonal synchrony increases
affiliation. Social Cognition, 27(6), 949–960.
Jones, E. E., & Davis, K. E. (1965). From acts to dispositions: The attribution process
in person perception. In L. Berkowitz (Ed.), Advances in experimental social psychology
(Vol. 2, pp. 220–266). New York: Academic Press.
Kaneko, K. (1984). Like structures and spatiotemporal intermittency of coupled logistic
lattice: Toward a field theory of chaos. Progress in Theoretical Physics, 72, 480.
Kaneko, K. (Ed.) (1993). Theory and applications of coupled map lattices. Singapore: World
Scientific.
Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge,
MA: MIT Press.
Kunda, Z. (1999). Social cognition: Making sense of people. Cambridge, MA: MIT Press.
LaFrance, M. (1979). Nonverbal synchrony and rapport: Analysis by the cross-lag panel
technique. Social Psychology Quarterly, 42, 66–70.
Lakin, J. L., & Chartrand, T. L. (2003). Using nonconscious behavioral mimicry to create
affiliation and rapport. Psychological Science, 14, 334–339.
Lakin, J. L., Jefferis, V. E., Cheng, C. M., & Chartrand, T. L. (2003). The chame-
leon effect as social glue: Evidence for the evolutionary significance of nonconscious
mimicry. Journal of Nonverbal Behavior, 27, 145–162.
Levine, J. M., & Moreland, R. L. (1998). Small groups. In D. T. Gilbert, S. T. Fiske, &
G. Lindzey (Eds.), The handbook of social psychology (4th ed., Vol. 2, pp. 415–469).
New York: McGraw-Hill.
Macrae, C. N., Duffy, O. K., & Lawrence, J. (2008). A case of hand waving: Action
synchrony and person perception. Cognition, 109, 152–156.
Marsh, K. L., Richardson, M. J. & Schmidt, R. C. (2009). Social connection through
joint action and interpersonal coordination. Topics in Cognitive Science, 1, 320‒339.
Maurer, R. E., & Tindall, J. H. (1983). Effects of postural congruence on client’s
perception of counselor empathy. Journal of Counseling Psychology, 30, 158–163.
McGrath, J. E., & Kelley, J. R. (1986). Time and human interaction: Toward a psychology of
time. New York: Guilford Press.
Messick, D. M., & Liebrand, V. B. G. (1995). Individual heuristics and the dynamics of
cooperation in large groups. Psychological Review, 102, 131–145.
Miles, L. K., Griffiths, J. L., Richardson, M. J., & Macrae, C. N. (2010). Too late to
coordinate: Contextual influences on behavioral synchrony. European Journal of Social
Psychology, 40, 52–60.
Miles, L. K., Nind, L. K., Henderson, Z., & Macrae, C. N. (2010). Moving memories:
Behavioral synchrony and memory for self and others. Journal of Experimental Social
Psychology, 46, 457–460.
Miles, L. K., Nind, L. K., & Macrae, C. N. (2009). The rhythm of rapport: Interpersonal
synchrony and social perception. Journal of Experimental Social Psychology, 45, 585–589.
Miller, N. E. (1944). Experimental studies of conflict. In J. M. Hunt (Ed.), Personality and
the behavior disorders (pp. 431–465). New York: Ronald.
Nisbett, R., & Ross, L. (1980). Human inference: Strategies and shortcomings of social
judgment. Englewood Cliffs, NJ: Prentice Hall.
 From Interaction to Synchronization  227

Newtson, D. (1994). The perception and coupling of behavior waves. In R. R. Vallacher &
A. Nowak (Eds.), Dynamical systems in social psychology (pp. 139–167). San Diego, CA:
Academic Press.
Nowak, A. (2004). Dynamical minimalism: Why less is more in psychology. Personality
and Social Psychology Review, 8¸183–192.
Nowak, A., & Vallacher, R. R. (1998). Dynamical Social Psychology. New York: Guilford
Press.
Nowak, A., Vallacher, R. R., & Zochowski, M. (2002). The emergence of personality:
Personal stability through interpersonal synchronization. In D. Cervone & W. Mischel
(Eds.), Advances in personality science (pp. 292–331). New York: Guilford Press.
Nowak, A., Vallacher, R. R., & Zochowski, M. (2005). The emergence of personality:
Dynamic foundations of individual variation. Developmental Review, 25, 351–385.
Ott, E., Grebogi, C., & York, J. A. (1990). Controlling chaos. Physics Review Letters, 64,
1196–1199.
Pelose, G. C. (1987). The functions of behavioral synchrony and speech rhythm in
conversation. Research on Language and Social Interaction, 20, 171–220.
Sbarra, D. A., & Hazan, C. (2008). Coregulation, dysregulation, self-regulation: An
integrative analysis and empirical agenda for understanding adult attachment, separa-
tion, loss, and recovery. Personality and Social Psychology Review, 12, 141–167.
Scarr, S. (1992). Developmental theories for the 1990s: Development and individual
differences. Child Development, 63, 1–19.
Schmidt, R. C., Beek, P. J., Treffner, P. J., & Turvey, M. T. (1991). Dynamical sub-
structure of coordinated rhythmic movements. Journal of Experimental Psychology:
Human Perception and Performance, 17, 635–651.
Schmidt, R. C., & Richardson, M. J. (2008). Dynamics of interpersonal coordination.
In W. A. Fuchs & V. K. Jirsa (Eds.), Coordination: Neural, behavioral and social dynamics
(pp. 281–307). Berlin: Springer-Verlag.
Schuster, H. G. (1984). Deterministic chaos. Vienna: Physik Verlag.
Shinbrot, T. (1994). Synchronization of coupled maps and stable windows. Physics Review
E, 50, 3230–3233.
Shockley, K., Santana, M., & Fowler, C. A. (2003). Mutual interpersonal postural con-
straints are involved in cooperative conversation. Journal of Experimental Psychology:
Human Perception and Performance, 29, 326–332.
Stel, M., Van Baaren, R. B., & Vonk, R. (2008). Effects of mimicking: Acting prosocially
by being emotionally moved. European Journal of Social Psychology, 38, 965–976.
Stel, M., & Vonk, R. (2010). Mimicry in social interaction: Benefits for mimickers,
mimickees, and their interaction. British Journal of Psychology, 101, 311–323.
Strack, F., Martin, L. L., & Stepper, S. (1988). Inhibiting and facilitating conditions of the
human smile: A nonobtrusive test of the facial feedback hypothesis. Journal of Personality
and Social Psychology, 54, 768–777.
Strogatz, S. (2003). Sync:The emerging science of spontaneous order. New York: Hyperion Books.
Swann, W. B., Jr. (1990). To be adored or to be known? The interplay of self-enhancement
and self-verification. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation
and cognition (Vol. 2, pp. 404–448). New York: Guilford Press.
Thibaut, J. W., & Kelley, H. H. (1959). The social psychology of groups. New York:
Wiley.
Tickle-Degnen, L., & Rosenthal, R. (1987). Group rapport and nonverbal behavior.
Review of Personality and Social Psychology, 9, 113–136.
228  Robin R. Vallacher and Andrzej Nowak

Turvey, M. T. (1990). Coordination. American Psychologist, 4, 938–953.

Vallacher, R. R., & Nowak, A. (Eds.) (1994). Dynamical systems in social psychology. San
Diego, CA: Academic Press.
Vallacher, R. R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 8, 73–79.
Vallacher, R. R., & Nowak, A. (2007). Dynamical social psychology: Finding order
in the flow of human experience. In A. W. Kruglanski & E. T. Higgins (Eds.),
Social psychology: Handbook of basic principles (2nd ed., pp. 734–758). New York:
Guilford Press.
Vallacher, R. R., Nowak, A., & Miller, M. E. (2003). Social influence and group dynamics.
In I. Weiner (Series Ed.) & T. Millon & M. J. Lerner (Vol. Eds.), Handbook of psychology,
Vol. 5: Personality and social psychology (pp. 383–417). New York: Wiley.
Vallacher, R. R., Read, S. J., & Nowak, A. (2002). The dynamical perspective in person-
ality and social psychology. Personality and Social Psychology Review, 6, 264–273.
Vallacher, R. R., Van Geert, P., & Nowak, A. (2015). The intrinsic dynamics of psycho-
logical process. Current Directions in Psychological Science, 24, 58–64.
Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal agency: Individual vari-
ation in action identification. Journal of Personality and Social Psychology, 57, 660–671.
Van Baaren, R. B., Holland, R. W., Kawakami, K., & Van Knippenberg, A. (2004).
Mimicry and prosocial behavior. Psychological Science, 15, 71–74.
Wegner, D. M. (1994). Ironic processes of mental control. Psychological Review, 101,
34–52.
Wegner, D. M., & Vallacher, R. R. (1977). Implicit psychology. New York: Oxford
University Press.
Weisbuch, G. (1992). Complex system dynamics. Redwood City, CA: Addison-Wesley.
Wish, M., Deutsch, M., & Kaplan, S. J. (1976). Perceived dimensions of interpersonal
relations. Journal of Personality and Social Psychology, 33, 409–420.
Zimbardo, P. G. (1970). The human choice: Individuation, reason, and order versus dein-
dividuation. In W. J. Arnold & D. Levine (Eds.), Nebraska Symposium on Motivation
(Vol. 17, pp. 237–307). Lincoln, NB: University of Nebraska Press.
Zochowski, M., & Liebovitch, L. (1997). Synchronization of trajectory as a way to control
the dynamics of the coupled system. Physical Review E, 56, 3701.
PART III

Collective Dynamics
11
SIMULATING THE SOCIAL
NETWORKS IN HUMAN
GOAL STRIVING
James D. Westaby and DaHee Shon

Introduction
Describing, predicting, and explaining the complexities of human goal pursuit and
behavior over extended periods of time is an ambitious pursuit for psychologists and
social scientists. Relying upon conventional methods, unfortunately, is often limited
because they have considerable difficulty navigating the various contingencies that
humans face over time, the uncertain social environments into which they traverse,
and the various events that can change at any probabilistic moment. Fortunately,
computational modeling and computer simulations can allow researchers to apply
and extend parameters from contemporary theories of psychology to navigate these
more complex environments and make predictions about emergent system out-
comes, such as goal achievement, performance, and emotional contagion.
According to Harrison and colleagues (Harrison, Lin, Carroll, & Carley,
2007), a variety of simulations can be created, such as (1) agent-based models, which
examine how factors of interest can impact individuals simulated in a computer,
(2) systems dynamics models, which model “the behavior of the system as a whole”
(Harrison et al., p. 1238), and (3) cellular automata models, which examine how
actors are influenced by the characteristics of neighboring cells. Simulations can
not only address important linear and non-linear dynamics over time, they also
allow researchers to arbitrarily change parameters of interest to see their impact
on long-term system outcomes (Smith & Conrey, 2007). In contrast to the vol-
ume of computational research applied in the hard sciences, there is relatively less
simulation research in the social sciences, despite its relevance to a wide variety
of research topics examining dynamical issues (Richetin et al., 2010; Vallacher,
Read, & Nowak, 2002; Weinhardt & Vancouver, 2012). And even less work
has examined how social networks are motivationally involved in goal pursuits,
despite the exponential growth of social network research (Borgatti, Mehra,
232  James D. Westaby and DaHee Shon

Brass, & Labianca, 2009). In response, this chapter explores how simulations can
be used to model the complex ways in which social networks influence emer-
gent goal striving processes and system outcomes. To traverse this new domain, a
dynamic network theory perspective will be used to guide our thinking (Westaby,
Pfaff, & Redding, 2014). This line of theorizing adds value over traditional goal
striving research in psychology, because it comprehensively examines how social
networks influence goal striving processes. Our approach is tangentially related
to some of Castelfranchi, Cesta, Conte, and Miceli’s (1993) computational theo-
rizing on the interdependence of goals in social systems, although our approach
systematically connects goal striving to explicit social network roles. The the-
ory also adds value over traditional social network approaches in sociology and
organizational science, because it is the first to systematically infuse goal nodes
into social networks. Goal nodes represent potential stimuli that entities may be
wanting, desiring, or pursuing (or not). Traditional network research has focused
on structural links between entities alone (e.g., individuals, groups, or other collective
units), and not on how those entities may be connected to specific goal pursuit nodes
(e.g., how person X is supporting person Y who is striving for goal Z). Figure 11.1
illustrates these contrasts, as discussed below.

FIGURE 11.1  Chart comparisons. + indicates positive link. − indicates negative

link. Quotation marks illustrate links in the traditional social network
approach that have different motivational orientations in the dynamic
network chart (e.g., “+” link between Spouse and Relative is actually
a supportive resistance (V) link working against Person X’s weight
loss goal. Neutral path is not included in the positivity ratio, given
ambiguity about positive or negative direction. G = goal striving.
S = system supporting. P = goal preventing. V = supportive resisting.
N = system negating. System reacting (R), interacting (I), and
observing (O) are not illustrated, for simplicity. All manifest links are
scored 1 for strength in both approaches.
 Social Networks and Goal Striving  233

The structure of the chapter is as follows. First, we review some basic network
concepts. Second, we provide an overview of dynamic network theory and how
it is different from traditional network theories. Third, given the importance
of emergent phenomena (Nowak, Vallacher, & Zochowski, 2005; Vallacher &
Nowak, 1997), we demonstrate how a rudimentary computational model can
be derived from concepts in the theory to simulate emergent network power
in a person’s goal striving, which represents strong social networks mobilized
to assist in dynamic goal pursuit. We assume in this chapter that “dynamic”
means processes and outcomes changing in time. Third, we introduce network
goal graphing, which sets the stage to create computational simulations to forecast
and visualize the ways in which social networks dynamically impact emergent
outcomes in real-world goal striving. In all, such network goal analysis has
applications across social, organizational, and counseling psychology, as well as
in conflict resolution, coaching, and consulting.

Traditional Social Network Frameworks

Researchers have made enormous strides in describing network structures and
processes in numerous domains, such as physics, technology, and the social and
organizational sciences. There are several key concepts to keep in mind when
thinking about networks. The first is that a network often represents how vari-
ous entities (also known as vertices or nodes) are interconnected via links, edges,
or ties (Burt, Kilduff, & Tasselli, 2013). For example, in highway systems, one
may be interested in understanding how various cities (nodes) are connected
by highways or rail (links). Or a social network researcher may be interested
in how various individuals on a social networking platform are connected with
one another. Once a network structure is identified or plotted, researchers can
start examining a number of useful metrics to describe the system (Borgatti et al.,
2009; Wasserman & Faust, 1994). For example, network density represents the
degree to which the network is saturated with linkages, ranging from no connec-
tions to complete connections between all entities. Another powerful concept is
related to the level of centrality in the system. Generally speaking, this represents
the extent to which information or resources are flowing through key entities in
the network. When centrality is high, certain entities in the network are receiv-
ing (or mediating) a greater amount of information or resources than others,
which in turn may give them advantage by controlling the flow of information
or resources in the system (Burt et al., 2013).1 Other system-level concepts of
interest to research include network size and network diversity.
From a psychological perspective, one can also examine the overall level of
positivity or negativity in a system through linkages with positive or negative
valence, such as friendship (positive) and foe (negative) networks. In Figure 11.1a,
we illustrate such linkages in the context of a network involved in a person’s
goal pursuit of losing weight. Here, we see that four of the five individuals are
234  James D. Westaby and DaHee Shon

positively linked, which gives a positivity ratio of .90. This is derived by dividing
the total number of unambiguous positive paths by total number of positive and
negative paths. The strength of manifest links are scored 1 in this simple illustra-
tion. As demonstrated later, results from dynamic network theory will suggest
that this system has much more conflict than implied by the traditional approach.
Although traditional network research has added enormous capacities to analyze
complex social systems, it has not sufficiently drawn upon psychology to explain
how network entities may be dynamically striving for their specific goals, wishes,
and desires. Dynamic network theory attempts to fill this gap.

Dynamic Network Theory

In dynamic network theory (Westaby et al., 2014), implicit or explicit judgment
and decision-making processes are presumed to trigger the activation of one of
eight types of social network role behaviors (or their multiplex combinations).
The eight roles include goal striving, system supporting, goal preventing, sup-
portive resisting, system negating, system reacting, interacting, and observing.
Each role is summarized in Table 11.1, and then illustrated graphically.

TABLE 11.1  The eight social network roles in dynamic network theory.

Social network Label Meaning Examples

role

1. Goal striving G Direct motivation Wants, desires, intentions, wishes, or

toward a goal. motivated states directed toward
goal nodes of interest (e.g., I want
to get a new job).
2. System S Providing support Helping, supporting, assisting,
supporting to others in aiding, advising, or backing up
a network in another’s efforts to advance a goal
their pursuits. pursuit (e.g., my friend in my
network is trying to help me get
a new job).
3. Goal P Direct motivation Preventing, hindering, thwarting,
preventing against or attempting to stop a given goal
someone’s goal. (e.g., my boss is trying to prevent
my quitting).
4. Supportive V Providing support Helping, supporting, assisting,
resisting to network aiding, advising, promoting, and
entities that backing up another’s resistance
are showing or negativity toward another’s
resistance or goal (e.g., the company VP is
negativity supporting my boss’s efforts at
about a given trying to prevent my quitting by
goal pursuit. offering me a raise).
5. System N Having negative Negative attitude, feeling, affect, or
negating orientations disagreement directed at another’s
toward others goal striving (e.g., my coworker
in their goal is angry about my potential
pursuit. leaving, because there will be no
replacement).
6. System R Having negative Negative attitude, feeling, affect, or
reacting reactions disagreement directed at another’s
toward others resistance or negativity about a
that are goal pursuit (e.g., my spouse is
resisting upset by my coworker’s anger
another’s toward me).
pursuit.
7. Interacting I Entities in the Those near others in their pursuits,
vicinity of but not necessarily paying
another’s attention to what’s going on.
pursuit.
8. Observing O Entities only Bystanders, observers, and curious
observing entities that are attending to what’s
or simply going on in a goal or behavior, but
being aware not helping or hurting others in
of another’s the efforts (e.g., some customers
pursuits. are observing whether I will
quit the company, simply out of
curiosity).

Example system-level concepts

Goal pursuit links (also The amount of goal In Figure 11.1: 1 G link from Person
known as network striving and system X to the goal plus 1 S link from
motivation) supporting forces the Friend to Person X plus 1 S
that fuel a goal link from the Spouse to Person
pursuit. X plus 1 S link from Doctor
to Person X, and 1 S link from
Doctor to Spouse (i.e., 1 G +
4 S’s = 5).
Network resistance The amount of In Figure 11.1: 1 P link from Person
goal prevention X to goal plus 1 P link from
and supportive Spouse to goal plus 1 V link from
resistance in Spouse to Person X plus 1 V link
system. from Relative to Person X, and
1 V link from Relative to the
Spouse (i.e., 2 P’s + 3 V’s = 5).
Positive focus ratio The relative level of Goal pursuit links (G and S) divided
functional forces by all goal pursuit, network
in a social system resistance, and negativity links (i.e.,
around a given in Figure 11.1: 5/(5 + 6) = .45).
goal.
236  James D. Westaby and DaHee Shon

The Eight Social Network Roles

Because goal nodes are systematically inserted into network structures, the theory
can explain how networks are pivoting around goal striving processes (or not).
It does so through the eight social network roles. First, the goal striving (G) role
illustrates individuals in a network trying to pursue a given goal, aspiration, or
behavior under study. Goal striving subsumes a variety of important psychologi-
cal antecedents of behavior, such as the wants, dreams, wishes or desires toward
a goal node (Austin & Vancouver, 1996). It also accounts for the importance
of behavioral intention, one of the best predictors of human behavior (Ajzen,
1991; Fishbein & Ajzen, 2010; Westaby, 2005; Westaby, Probst, & Lee, 2010;
Westaby, Versenyi, & Hausmann, 2005). Second, system supporting (S) represents
those entities helping those in goal pursuit or encouraging those not pursuing
a goal to start doing so, thereby accounting for powerful support processes on
behavior (Rhoades & Eisenberger, 2002). These network forces are predicted
to facilitate emergent goal achievement and performance in complex systems,
especially when entities have system competency concerning the goal, such as
objective skill or high self-efficacy (Stajkovic & Luthans, 1998).2 Third, going
beyond past approaches that have addressed negative interpersonal links in net-
works (Labianca, Brass, & Gray, 1998), to account for direct network resistance
to a focal goal, goal preventing (P) represents entities more directly preventing,
obstructing, or hindering the goal pursuit node, explicitly or implicitly. Fourth,
supportive resistance (V) represents those indirectly supporting others in resistance
efforts. The above two forces are predicted to inhibit emergent performance of
those pursuing a target goal in ego-centric systems.
Fifth, as for negative interpersonal emotions, system negating (N) represents
negative orientations to those working toward a goal. Sixth, in contrast, system
reactance (R) represents those negatively reacting to those showing negativ-
ity or resistance to a given goal, such as when your friend gets upset with
another who has intentionally obstructed your goal pursuit. While the former is
negativity toward those pursuing the goal, the latter is negativity toward those
resisting the goal pursuit. Hence, system reactance represents a form of negativ-
ity or arousal that is affirming the goal pursuit. Ironically, it could therefore be
a positive force, often defending those in goal pursuit against those resisting it.
In contrast to the above factors that reflect strong motivational or emotional
linkages to goal nodes or to others, the last two roles represent more periph-
eral ones that can inadvertently impact goal pursuits. Seventh, the interacting role
(I), in its exclusive form, represents individuals simply interacting around oth-
ers involved in a goal pursuit without necessarily paying attention to another’s
pursuits. This can affect performance in some settings, such as an individual try-
ing to walk to work quickly, who becomes slowed by the density of people
on a crowded sidewalk (Westaby, 2012a); those interactants that emerge in the
network around a given goal pursuit are neither trying to help nor hinder the
 Social Networks and Goal Striving  237

person—they have simply emerged around the person’s own goal pursuit. Eighth,
and lastly, those engaged exclusively in observing role behavior (O) represent those
observing, watching, or learning about a person’s goal pursuit, but not intention-
ally helping or hurting the process. These roles can affect performance in some
cases as well, such as when experienced goal strivers increase performance when
being observed, hence accounting for classic social facilitation effects in social
psychology (Geen, 1991). More broadly, it is important to note that individuals
can express multiplex combinations of the eight roles, which allows the theory to
account for vastly different ways that individuals may demonstrate motivational,
emotional, and peripheral linkages to others and their goals.3

Charts, Metrics, and Conceptual Differences

In contrast to purely verbal theories, a distinct advantage of dynamic network
theory for computational simulation research is its utilization of dynamic net-
work charts to characterize and metricize complex human systems. Dynamic
network charts show how social network entities are helping, hurting, or just
peripherally involved in specific cases of goal pursuit. All paths in the charts are
based on the finite set of roles proposed in the theory. These charts take the form
of directed graphs (Snijders, Van de Bunt, & Steglich, 2010), but with the unique
inclusion of a goal node. Goals play a special role in these networks because they
allow researchers to examine how the entities in the network are involved in
the given goal pursuit, such as a friend and coworker being supportive of one’s
new career plans (i.e., two system supporter links to one’s career goal striving).
Hence, we can now visualize and quantify how social network entities and their
role linkages are pivoting around the goal node(s). This helps us theoretically
explain (not just describe) how social networks are interpersonally and moti-
vationally working for or against a given goal or behavior, instead of merely
examining how interpersonal linkages alone may be influential.
However, this does not mean to imply that focusing on interpersonal connec-
tions exclusively is without merit. This has tremendous value, and its intuitive
nature has been helpful in launching the network perspective to study a vast array
of social, organizational, biological, technological, and physical systems (Westaby
et al., 2014). However, to better predict and explain human behavior, we believe
that incorporating goal nodes and corresponding social network roles is a critical
next step to advance theory and scientific method.
Dynamic network charts can be focused in different ways. For example, an
ego-centric version can show how entities in a network are behaviorally involved
in (1) a focal person’s overall goal pursuit, or (2) the person’s goal pursuit at
specific snapshots in time. We first illustrate the overall format in the context of
Person X’s attempts at trying to lose weight. We will illustrate the latter format
in subsequent sections. In contrast to the traditional network approach shown in
Figure 11.1a, the goal node is directly inserted into the dynamic network chart as
238  James D. Westaby and DaHee Shon

shown in Figure 11.1b (i.e., “Person X’s goal of weight loss”). Here we can see
how entities and their motivated role linkages are connected to Person X’s goal
pursuit (or not). For simplicity, we focus on strong links here, although future
research can examine weak links as well, given their potential importance in
securing advantage (Granovetter, 1973). In the dynamic network charts below,
the strength of manifest paths are scored 1. See Westaby (2012a) for additional
ways to scale path strength, such as accounting for intensity levels.
In Figure 11.1b, we can see that Person X has often tried hard to lose weight
(G path), but has also knowingly binge-eaten at times (P path); hence, a multiplex
role combination path to the goal node exists in the chart (i.e., coming from the
G and P paths). Person X’s Friend and Spouse have also supported (S) Person X’s
weight loss attempt, although Person X’s Spouse has worked directly against it at
times by bringing home fattening food (P) and encouraging Person X to enjoy it
(V path). Hence, the Spouse can manifest a complex set of non-linear behaviors
that can differentially impact Person X over time. Further, Person X’s Relative,
who is extremely overweight, has shown supportive resistance to Person X by
encouraging overeating during their dinner parties (V path).
From an emotional linkage perspective, the Relative also becomes upset when
Person X does not eat a lot of food during dinner parties (N dashed path, where the
black dot on path symbolizes de-affirmation against goal and visually distinguishes
the path from V). This negativity is the same as the negativity seen in the traditional
network approach. However, Person X does not react with negative emotions to
this resistance, given that Person X realizes that the Relative is rather overweight
and likely biased.4 Lastly, a doctor has provided strong support and encouragement
(S) for Person X’s dieting goal as well as supportive discussions directed at the
Spouse to help encourage the Spouse’s support of Person X’s dieting (S).

Differences with Traditional Network Analysis

There are several differences between a traditional network approach and a
dynamic network theory approach. A first major difference, as mentioned earlier,
is that goal nodes are used to show how networks are connected to goal pursuits.
This helps us understand how social forces are motivated around goals (or not). A
second difference concerns the interpretation of positive and negative linkages. In
the above example, consider the link between the Spouse and Relative. For the
traditional approach, these two entities are linked positively, because they share
a similar motivational orientation. In contrast, the dynamic network chart illus-
trates that this similarity in motivation is being leveraged against Person X’s goals
and wishes to lose weight, and therefore represents a negative resistance force in
this goal system. A third difference can be seen when looking at overall system
dynamics. For example, consider the presumed overall positivity or negativity of
network forces in Person X’s weight loss network. In the traditional approach,
we see that there is high positivity among the interpersonal network connections
 Social Networks and Goal Striving  239

in Figure 11.1a. That is, the positivity ratio is .90, where four out of the five link-
ages are positive in nature. However, the dynamic network chart reveals far less
positivity in the system. Here, the positive focus ratio from recent work on dynamic
network theory (Westaby, Woods, & Pfaff, 2016) is used to generally show the
functional forces in a social system involved in a given goal (G and S) where no
resistance or conflict is manifest (P, V, N, and R). In this case, five out of the 11
social network role paths demonstrate purely functional orientations. The rest are
de-affirming or negative in nature. This results in a positive focus ratio of .45, and
suggests that the traditional network analysis of .90 may be over-estimating the
positive social forces involved in the goal pursuit. Future research needs to further
examine such comparisons (Westaby et al., 2014; Westaby & Redding, 2014).5
Although the above concepts and metrics are useful in helping describe and
explain complex goal striving systems, such representations are not enough to
address dynamical changes over time. In this regard, Snijders and his colleagues
have done important work examining how traditional interpersonal linkages in
networks may impact behavior, and have used novel methods and program-
ming to examine these effects, such as longitudinal analyses, Markov modeling,
and SIENA computer modeling (e.g., Snijders, Van de Bunt, & Steglich, 2010).
Dynamic network theory adds to this important line of work by directly inserting
goal nodes into social network structures along with social network roles to those
nodes. A promising line of research would be to use concepts, methods, and pro-
gramming developed by Snijders and others interested in explaining behavior in
order to examine how including such goal nodes and their linkages may help us
better explain emergent goal achievement, behavior, and other system outcomes
of interest. Goals play a critical role here by allowing us to see how various enti-
ties are emerging in a social network to help, such as providing system support
to a goal striver, or hinder, such as enacting goal prevention (e.g., trying to talk
a person out of a goal pursuit or behavior). In the next sections, we attempt to
further extend our explanation of dynamic network processes through the use of
computational modeling in more complex systems over time.

Simulating the Development of Powerful Networks

Because the framework is grounded in computational metrics, dynamic network
theory has the potential to provide new insight into how networks can impact
emergent variables and system outcomes over time. Although the theory could
be applied in various computational ways, we start with a basic simulation that
attempts to examine how individuals derive powerful social networks to help
achieve their goals. Computationally, we use a rudimentary agent-based model
to simulate how individuals build such networks, given environments with vary-
ing degrees of difficulty, for example. This simulation examines how agents, who
are the simulated individuals in the computer program, stochastically encounter
others in their social networks that may help or hinder them in their pursuits.
240  James D. Westaby and DaHee Shon

These stochastic environments are ones in which various social encounters are
randomly and probabilistically determined.
The simulation uses select parameters in dynamic network theory to calcu-
late the emergent power of the networks helping or inhibiting a person’s goal
pursuit. For example, it is assumed in the computational model that agents who
encounter others in a social network that have motivational qualities (i.e., partner
goal striving and system supporting roles), strong system competency (another
key factor in the theory), and facilitative environments are presumed to be able
to build stronger networks to help achieve their goals. In contrast, agents that
encounter others that show network resistance (i.e., goal preventing and support-
ive resistance), negativity (i.e., system negating), and/or difficult environments
are computationally assumed to have less potential of achieving goals and per-
forming well overall. When agents probabilistically encounter such situations
frequently, they will likely have problems materializing their goals into success.
Throughout history, there are countless stories of individuals, groups, and even
societies that cannot break such difficult cycles, which can lead to frustration,
aggression, and conflict in various cases.

The Computational Environment

The following illustrates key features of the basic computational model. A sto-
chastic computer environment was first created for the agent in the program.
That is, on each round of the computer simulation, the agent was programmed
to encounter an entity in the network that had a random configuration of social
network role attributes that could impact his or her goal pursuit in conjunction
with an environment that ranged from somewhat easy to extremely difficult in
achieving the goal. The configuration and impact of the role attributes was based
on a theoretical interpretation of effects proposed in dynamic network theory
(Westaby, 2012b). Future research could also base computations upon results
from empirical studies on real participants, whenever possible (e.g., using results
from regression or path coefficients).
Computationally, the more functional behavior the other network entities
provide in less harsh environments, the greater emergent network power com-
putationally accrued by the agent in his or her goal pursuit over time. Functional
behaviors typically include more goal striving and system supporting and less
goal prevention, supportive resistance, and system negation, while harsh environ-
ments generally represent those situations where resources are harder to obtain to
accomplish the goal.6 Network power can be generally conceptualized as strong
social networks mobilized to assist in goal pursuit. We wanted to simulate how
a focal agent builds such power over time. In this study, the rolling total for
network power increases on each round when the agent encounters others that
provide functional behavior and competence without network de-affirmation,
and in the context of facilitative environments. For example, in a positive context
where the agent encountered another person in his or her network that showed
 Social Networks and Goal Striving  241

no network resistance, no dysfunctional negativity, high network motivation

(i.e., G and S roles), high system competence, and a facilitative environment, the
agent would have the greatest computed increase in emergent network power
(e.g., a computation increment of 2). In contrast, if network resistance was encoun-
tered without any network motivation, it would stifle an increment in emergent
power (e.g., a computation increment of 0). The encounter was not helpful to
build the person’s network, and would be reflected as such in the computations.7
Hence, a high score for network power represents an agent’s capacity to have a
strong network that is willing and able to help the person achieve his or her goal.
Computing and plotting an agent’s network power over iterations allows us to
see how well agents are able to build network power in their encountered social
systems. In Figure 11.2, we plot the network power functions observed across
the computer simulations for five different individuals. Each of these plots show
the degree to which each specific agent was able to acquire a strong network
that has the potential to facilitate his or her goal pursuit. We ran our simulations
over 1000 rounds per agent, given that people can encounter a large number of
individuals across the span of long-term goals. This fits with other agent-based
models that run several hundred iterations to simulate behavior (e.g., 600 for
Kaufmann, Stagl, & Franks, 2009).8

FIGURE 11.2   mergent network power among five simulated agents. The y-axis
E
reflects cumulative network power.
242  James D. Westaby and DaHee Shon

As one can see, and fitting our expectations, we saw great variation in people’s
capacities to build powerful networks in their goal pursuits. Some agents started
to have success early, and were able to build more powerful networks with
much social capital, such as Agent 1. In contrast, Agents 2, 3, and 4 required
more time to gain traction, but were then able to start building more powerful
networks, although at somewhat different rates and slopes. In contrast, Agent 5
had great difficulty ever gaining traction in building a powerful network, either
because the environment was not facilitative, or because when it was facilita-
tive, other people created serious network resistance or there was a lack of goal
motivation in the system. In line with nonlinear perspectives (Vallacher, Van
Geert, & Nowak, 2015), these results show strong nonlinear effects both within
and between individuals, which multilevel statistics or hierarchical linear mod-
eling (HLM) could formally examine for significance (Curran & Bauer, 2011;
Hofmann, 1997). In this example, different individuals can end up in very differ-
ent places, simply as a result of random outcomes and how they may accrue over
time. These results do not take into account initial individual differences, which
future models should address. In contrast, as seen in the last graph in Figure 11.2,
which aggregated the data across all agents, a more linear trend was found when
integrating, thereby washing away the various nonlinear complexities that were
occurring for different agents in the system. There are several things we learn
from these results. First, individuals experienced great variation in their abilities
to develop strong networks to help them in their pursuits, which mimics what
can be seen in the vast discrepancies of achievement around the world. Second,
the results are revealing because they show how nonlinear trends impact the abil-
ity to build strong networks. Instead of everyone simply accruing more network
power over time in a linear manner, some individuals did not gain traction until
after some time passed. Then, they were able to finally start acquiring a help-
ful network. The implication is that individuals may need to persist through
considerable resistance before enjoying some success, akin to the concept of
perseverance and Grit (Duckworth et al., 2007). However, there were also some
simulated individuals, despite their efforts, who were never able to gain traction.
In such contexts, a profound implication is that the individuals may need to
consider finding an entirely new social network system that may increase their
probability of finding support and minimizing resistance. At a macro level, one
can see this when individuals in impoverished countries decide to take huge risks
by attempting to migrate to other places that are perceived as more resourceful,
social, and less constraining.

Adding Complexity
Although the above simulation utilized select concepts from dynamic network
theory, it did not include other dynamics that could be important during goal
pursuit. Hence, in the spirit of a “building block approach” (Harrison et al., 2007)
 Social Networks and Goal Striving  243

where simple models are expanded, we illustrate several potential extensions.

First, various sub-processes in Vancouver and colleagues’ multiple-goal pursuit
model (Vancouver, Weinhardt, & Schmidt, 2010) could be examined, such
as those concerning negative feedback loops in control theory, resource limits
(e.g., deadlines), expected performance lags, and learning rules. As of now, some
of these sub-processes are more generally accounted for in dynamic network
theory’s system competency and feedback and change constructs.9 Joining these
theoretical perspectives could provide greater computational specification and
precision.
Second, computational work could examine how habit (or past behavior)
interacts with “remembering” to impact network roles and behavioral change.
Integrating aspects of Tobias’ (2009) model of habit, habit decay, and forgetting
with constructs from dynamic network theory may be helpful. Third, it would be
important for research to model how observers in a network decide whether or
not to provide system support (e.g., behavioral, financial, moral, etc.) or partner
goal striving—a form of goal contagion. This also has the potential to add breadth
to research on bystander interventions, helping, and prosocial behavior (Dovidio,
Piliavin, Schroeder, & Penner, 2006).
Another promising area of computational modeling needs to address the
network rippling of emotions—a form of emotional contagion. According to
dynamic network theory (Westaby et al., 2014), upon goal accomplishment or
salient goal progress, goal strivers and system supporters are expected to experi-
ence a rippling of positive emotion across the network, while those in resistance
roles would feel negative emotion. For example, if Pat achieves a goal, Pat will
feel good, as well as Pat’s immediate supporters—this positive feeling can then
ripple to others in the network as they hear of the success, assuming that they
support Pat’s goal pursuit. Inverse outcomes are predicted for those in network
resistance roles (P and V) in this same system. That is, they are expected to
become upset by Pat’s success as they learn of it. Those in peripheral roles (I and O)
are not expected to experience large changes in emotions, unless they suddenly
transform into a supporter or supportive resistor, for example, because of what
they have seen in the network.
Future research needs to also examine how the network rippling of emotions,
as feedback loops, may further motivate social network role behaviors (or not).
To stimulate theorizing, we posit here that the networking rippling of emotions
will systematically impact future social network role behaviors. For example, on
one hand, we anticipate that the network rippling of positive emotions will act
as a form of feedback that elevates or maintains goal striving and system sup-
porting intensity in networks over time. On the other hand, those in network
de-affirmation roles are expected to enact more complex dynamics based on the
emotional feedback, ranging from complete motivational shutdown to motiva-
tional inspiration, if future resistance is deemed to be inefficacious or efficacious,
respectively.
244  James D. Westaby and DaHee Shon

The Network Goal Graphing of Real-World Systems

We believe that one of the most promising and perhaps most ambitious areas
for future computational research in the social sciences is to examine how past,
present, and future anticipated variables in people’s real-world pursuits can
predict long-term emergent goal achievement, performance, and other rel-
evant outcomes. Computer simulations are extremely valuable for this level of
complexity, because they allow us to integrate numerous social, psychological,
organizational, and environmental variables that can help explain the rich context
of goal striving over time. Although important efforts have been made to simulate
behavioral change and social influence using computational models in specific
content domains (Kaufmann, Stagl, & Franks, 2009; Orr & Plaut, 2014; Hu &
Puddy, 2011; Schwarz & Ernst, 2009; Wang, Huang, & Sun, 2012; Wang &
Hu, 2012), this work has not fully examined how complex motivational struc-
tures in networks can impact changes in emergent goal pursuit outcomes over
time. Moreover, although work on scenario planning has looked generally at
how to make plans for the future, this work has not examined how interpersonal
and goal connections relate to past, present, and potential future pursuits. Nor has
previous work provided frameworks that can parsimoniously visualize the most
relevant complexity, including earlier work on network scripts (Westaby, 2012a).
To stimulate and facilitate modeling in this arena, we introduce a new network
goal graphing approach to help conceptualize and visualize how complex social
networks impact varying levels of emergent outcomes over time in long-term
goal pursuits. We hope this framework can provide a foundation from which
researchers can then use computational modeling to make projections and fore-
casts about emergent outcomes into the distal future on real human behavior and
system outcomes. The future projections could then be analyzed and compared
to empirical results, to further test and rigorously calibrate theoretical advances
(Kaufmann et al., 2009; Tobias, 2009). We are very interested in examining how
a focal individual (group, organization, or even society) is interacting with other
important entities entering or leaving the network or changing their motivation
levels over time, while being mindful of past, present, and potential future attrac-
tor and stability dynamics that could impact the system (Warren, 2006).
Generally speaking, we theorize that motivational influences from network
players can generate profound nonlinear effects on emergent outcomes. Some
may be good, such as a new wealthy customer providing major support for a com-
pany’s goal of selling products (e.g., system support for a company’s goal striving),
while others can be harmful (e.g., a drug dealer getting an adolescent addicted to
hard drugs, where the dealer’s goal prevention and supportive resistance is work-
ing against the adolescent’s normal goal striving for healthy school and family
goals). We know of no research to date that has attempted to computationally
model or even fully conceptualize the complex influences from such moti-
vated networks on human goal and behavioral striving and resulting outcomes.
 Social Networks and Goal Striving  245

Because dynamic network theory proposes that only eight social network roles
are needed to explain critical elements of goal pursuit, it is hoped to provide
a parsimonious way to explain high levels of complexity in elaborate human
systems.
To illustrate, a network goal graph is presented in Figure 11.3, which elabo-
rates on Person X’s weight loss goal striving from Figure 11.1. The top portion
of the graph shows the network entities directly or indirectly involved in the goal
pursuit or target behavior, while the bottom portion shows the emergent levels of
goal achievement, performance, or relevant outcome(s) over time. Going beyond
previous dynamic network charts, the goal node is allowed to vary on the y-axis
to show these outcomes over time. The social network roles are visualized by the
paths in the graph (also known as connections, lines, or arcs). These paths show
how the network entities on the top portion of the chart are attempting to influ-
ence the emergent goal achievement, performance, or system outcome(s) that
are varying on the goal node shown in the lower part of the chart. In this way,
we can now more easily see how the various network dynamics are differentially
impacting emergent criteria over the system’s lifespan, as shown by variability
in the goal node. Previous charting in dynamic network theory did not provide
these dynamic visualizations.
Theoretically, we hypothesize that changes in goal or performance node levels
result from the different social network roles being activated. We also assume that
the implemented network role behaviors maintain their force on the goal node
over time until other events are encountered that trigger new or different social
network role behaviors. Thus, one can think of any change to social network

FIGURE 11.3  Network goal graph of Person X’s weight loss pursuit (G = goal
striving; S = system supporting; P = goal preventing; V = supportive
resisting; N = system negating). Superscript denotes when interpersonal
links are activated on timeline.
246  James D. Westaby and DaHee Shon

role behaviors as also changing the trajectory or direction of the goal node level;
this trajectory is then maintained on emergent goal node levels until other events
occur that again change network role behaviors.10 However, as illustrated later,
we predict that attractor dynamics can have independent, (re-)stabilizing effects
on goal node levels as well.

Example
The full network goal graph in Figure 11.3 illustrates the network dynamics
influencing Person X’s weight loss goal, but now over precise points in time.
These portrayals provide much more dynamical detail about Person X’s struggles
and successes with weight loss as compared to either a traditional social network
analysis or static dynamic network chart as demonstrated in Figure 11.1. The
story is clearer because time and context are taken into account. To illustrate,
we can now see that at major event 1, Person X started intense weight loss goal
striving after being inspired by the supportive friend (S). Major events from the
timeline are connected to the social network roles with superscript indicators
(e.g., S1 to denote system support from this supportive friend at event 1).11 This
behavior (G) contributed to an emergent period of linear weight loss until a
stressful event in Person X’s life (event 2) presumably led to intentional over-
eating (P) and a nonlinear change in weight gain, although Person X did try to
maintain some dieting behavior (G) during this time frame. Fortunately, Person
X got re-inspired (G) around event 3 by the Friend again (S) and stopped all
overeating (no P) and maintained a healthy diet routine (G). This led to nonlin-
ear change in weight loss and steady success over time. However, upon getting
married, a host of new family dynamics emerged in the network and impacted the
system during event 4. For example, Person X’s new Spouse provided support for
the weight loss goal on one hand (S), but also supported Person X’s overeating
at other times (V), especially because the Spouse loved to overeat. The Spouse
also brought home a lot of fattening food that prevented Person X from serving
healthier choices at dinner (Spouse’s direct P path to the goal).
To add, Person X’s overweight Relative loved family gatherings where every-
one ate a lot of questionable food choices, and pressured Person X to join in the
festivities at those times (V). Given the new family dynamics, Person X entirely
stopped dieting (no G) and started overeating again (P). This led to fast and non-
linear weight gain. The pattern finally stopped when Person X developed cardiac
problems, which led to serious discussions with a cardiologist during event 5 who
said Person X must engage in a healthy lifestyle or it could lead to a catastrophic
cardiac event. Such supportive encouragement to Person X for dieting and to the
Spouse (S) to help Person X diet was enough to recommit goal striving (G) and
stop overeating (no P).
Researchers can also focus their analyses on discrete points in time. This is
helpful because the network goal graphs can become visually congested in large
 Social Networks and Goal Striving  247

systems over time.12 To illustrate, Figure 11.4 shows the major motivational
changes in Person X’s network at important times, such as critical turning or
tipping points (Gladwell, 2000). Computer technology can allow researchers
to easily visualize motivational dynamics at different time periods or view the
entire system moving and changing over time. Researchers can also create and
use system-level metrics, such as shown on the bottom of Table 11.1, to pre-
dict emergent system outcomes, preferably using multilevel statistical methods
(Curran & Bauer, 2011; Hofmann, 1997).13
In the network goal graph in Figure 11.3, we linked social network role
behaviors to a goal achievement/performance node that varies along the y-axis.
Alternatively, researchers could use other aggregated summary variables to for-
malize influence effects over time at the system level. For example, the positive
focus ratio could also be plotted over time as a varying node. Furthermore,
both the predicted positive focus ratio node and goal performance node could
be jointly displayed, ideally in standardized form, to show the relationship
between network-level predictions using the positive focus ratio and over-
all goal performance. Please see the appendix to this chapter for additional
details about link strength, metrics, and data aggregation in the ongoing
example.

Attractor Mechanisms
An attractor is an important concept from physics that has relevance to human
behavior (Warren, 2006). A fixed-point attractor, to illustrate, represents “the
state to which a system evolves over time and to which it returns after being

FIGURE 11.4   etwork influence on the goal at major events in time (G = goal striving;
N
S = system supporting; P = goal preventing; V = supportive resisting;
N = system negating).
248  James D. Westaby and DaHee Shon

perturbed” (Vallacher et al., 2015, p. 59). Goal achievement/performance node

data can provide critical information about attractors, which we theorize can
serve as important independent variables to further explain goal pursuit, perfor-
mance, and relevant outcomes over time. For example, the greater the stability
of a goal performance node level over time, the stronger an attractor becomes for
that level into the future, thereby serving as a re-stabilizing force when the system
is perturbed.14 Such propositions have particular relevance for simulating how
network linkages may interface with previously accrued goal or performance
levels over time. To illustrate, if a person’s goal achievement has been on a steady
decline for an extended period of time, one would predict that this would inde-
pendently impact future performance (beyond network role activations) and thus
need to be accounted for in simulation algorithms. This is akin to how attractors
can have strong effects on stabilizing system outcomes, but now contextualized as
interfacing with social network role changes over time. Future research will need
to examine these joint effects in efforts to better simulate and predict long-term
system outcomes.

Future Forecasting
One of the most ambitious applications of network goal graphing is its use in
computational and simulation science to make future forecasts, predictions, and
visualizations of system outcomes over time. Such forecasts could be based upon
relevant past, present, and future anticipated variables in a system. This is difficult
because of the challenges in making long-term projections about outcomes in
complex systems (Tetlock & Gardner, 2015; Tetlock & Mellers, 2011), but one
in which we think computational and simulation approaches become important
because of the host of variables that likely need to be considered. Hence, a critical
task in computational work is to accurately select, vet, and weigh valid predictors,
such as in the context of data underlying network goal graphs. (See the appendix
to this chapter for additional details.)
There are a number of potential variables that researchers could consider
from a dynamic network theory perspective. First, researchers could assess the
variation, change, intensity, and duration of past, present, and future anticipated
network role linkages across the various entities in a network as well as emer-
gent goal achievement, performance, or relevant outcome data. Such data could
come from a number of sources with the easiest being self-report. For example,
imagine that for the initial data shown in Figure 11.3, Person X reported his or
her perceptions of the past and present network roles and monthly weight levels.
(We discuss forecast data below.)
Second, given that multiple actors are often involved in our goal pursuits,
another powerful source of data could come from these other known individu-
als in the network (Kilduff, Crossland, Tsai, & Krackhardt, 2008) (e.g., Person
X’s friend, etc.). Research has also shown that some individuals may have more
 Social Networks and Goal Striving  249

accuracy when they are more intimately involved (Connelly & Ones, 2010) or
when they have characteristics of superforecasters (Tetlock & Gardner, 2015). In
our context, participating entities could be asked to judge the degree to which
various people are enacting different network role behaviors in a system, includ-
ing their own role. In such cases, agreement indices could also be created, which
provides insights into dynamic network intelligence (DNI) (Westaby et al., 2014).
For example, if the Friend indicates support for Person X’s dieting, which is
the same as Person X’s perception, Person X would have DNI for that linkage.
Researchers could then calculate the degree to which individuals in a network
have system-wide DNI, and perhaps then use or weight this information in com-
putational projections.15
Third, more objective information could be assessed in other lines of work.
For instance, as a critical dependent variable, Person X’s weight could be col-
lected weekly by a research team. Researchers could also make assumptions about
objective indicators of role behaviors, instead of relying on self-report. To illus-
trate, Person X’s goal striving could be inferred by the number of times Person X
went to a gym or spent money on weight loss programs or healthy food choices
(where receipts could be objectively examined).16 This is akin to other researchers
making assumptions about how financial spending connects to underlying
motives (Aknin et al., 2013).
Fourth, relevant demographic, individual difference, or environmental data
can become useful in computational forecasts. For example, one could theorize
that target individuals under study will observe the behavior from similar others
in the future and adopt similar network roles (i.e., using the observer role under
a homophily motive; Kossinets & Watts, 2009; Powell, White, Koput, & Owen-
Smith, 2005), when they lack confidence in their anticipated behavior and past
behavior is highly variable.17
Last, to provide contextual information for the goal or behavioral pursuits,
researchers could assess important events or issues occurring during the pursuit
as well as future anticipated events and issues. For example, Person X in our
ongoing dieting example may be able to better visualize and anticipate that his
or her own goal prevention likelihoods will increase somewhat during an antici-
pated retirement event that Person X anticipates will be stressful. Psychologically,
information and experiences from these events would feedback to the judgment
and decision-making process that triggers the selection of the eight network roles
(and/or their combinations) which are the immediate and direct antecedents of
important system outcomes.

Forecasting Example
To stimulate ways to think about such computational modeling in future
research, we hypothetically illustrate what a computational model’s forecast could
look like in the far right portion of Figure 11.3. In this case, such projections
250  James D. Westaby and DaHee Shon

could be based not only on past trajectories and fluctuations of Person X’s
weight gains and losses (around potential attractor states), but also on anticipated
network role behaviors and anticipated weight loss and gain in the future. In
our example, given that Person X may anticipate moderately high goal striv-
ing and a small amount of goal prevention during an expected stressful move
at event 6 in conjunction with somewhat predictable cyclical trends in past
weight loss and gain, one simulation model may project that Person X’s cur-
rent weight levels will be maintained until the stressful event happens, at which
point greater goal prevention will occur and generate a slight increase in weight
gain. However, this would be temporary, since the event may be perceived to
be short-term and past tendencies to re-establish relatively stronger goal striving
(over goal prevention) would result in a stabilization of moderately successful
weight loss.18
Akin to how meteorologists make different trajectories of hurricane paths
based upon assumptions in competing models (i.e., often referred to as “spaghetti
noodles” in the media), computational researchers should estimate in an a priori
fashion different trajectories of emergent outcomes based on different theoretical
parameter estimates and different stochastic assumptions about future environ-
ments. Then they could test and calibrate the different trajectories with observed
data over time—a form of “model grounding” (Harrison et  al., 2007). In this
case, the rival models would be calibrated to the empirical data that become avail-
able over time (Kaufmann et al., 2009; Tobias, 2009). In such work, researchers
could refine their computational algorithms in ways that best predict future net-
work forces and emergent states, which would then hopefully have applicability
across agents and contexts, and therefore generate greater external validity. It also
has implications in artificial intelligence applications, providing computers with
important parameters to search for in the social network, along with strategies
for maximizing functional social network dynamics and minimizing or avoiding
dysfunctional ones.

Summary
This chapter has used dynamic network theory to computationally explore how
social networks are involved in human goal striving processes over time and in
complex environments. In contrast to traditional network analysis, such net-
work goal analysis inserts goal nodes into network structures. This is a major
development because motivational links to those goals can then be directly
assessed, such as goal striving and goal preventing, and then their impact on
system outcomes examined, such as goal achievement, performance, and the
network rippling of emotions. The approach also models interpersonal role link-
ages that serve goal striving, such as system support, or potentially hinder it,
such as system negation. In all, the framework is expected to improve on the
prediction of important system outcomes over traditional approaches. We used
 Social Networks and Goal Striving  251

the theory’s proposed set of network role linkages to illustrate how computa-
tional models can be derived to simulate emergent network power in a person’s
goal striving. For more complex modeling, we introduced network goal graphing
to conceptualize, visualize, and computationally forecast how social networks
impact goal pursuit outcomes over longer time spans in upward, downward, or
stabilized patterns or trajectories. The importance of attractor sub-mechanisms
and dynamic network intelligence were illustrated to further explain important
complexities in human pursuits.19

Appendix: Underlying Data and Concepts

in Network Goal Graphs
This appendix provides further details about network goal graphs and how they
contain information about path strength, aggregation, and corresponding goal
performance. We illustrate this in the context of Person X’s weight loss goal,
where the data in the table below corresponds to the paths shown in Figures 11.3
and 11.4. Column 1 indicates major events occurring over time, such as Person
X being inspired by a friend to lose weight at event 1. This event theoretically
translates into (or is mediated by) Person X deciding to goal strive (G) with critical
support from a friend (S). The strength of these two role paths, each scored 1, can
be seen in the next two columns in the table.20 The other social network role
behaviors manifest in the example are shown in subsequent columns (i.e., P, V,
and N). As for aggregating to the system-level, the next column illustrates the
positive focus ratio (Westaby et al., 2016), which generally represents the functional
forces in a social system around a given goal. This metric is computed by divid-
ing the strength of G and S goal pursuit paths by all social network role paths in
the system (peripheral I and O roles excluded). For example, for event 1, there is
one goal striving path (scored 1) and one system supporting path (scored 1), with-
out any other resistance or negativity paths. Thus, the ratio is 1.0 (i.e., 2/2). At
event 2, the positive focus ratio decreases to .5, given that Person X shows both
goal striving (G) and goal preventing (P)—a common struggle for people trying
to lose weight (i.e., 1/2). The last column shows goal performance for Person
X in terms of subsequent pounds lost until a new change occurs in the system
(e.g., Person X lost 72 pounds after being inspired through G and S mechanisms
until a new change occurred during event 2). One would predict that the goal
striving and system supporting activated at event 1 at the path level (and/or
positive focus ratio at the system level) would impact subsequent performance.
Theoretically, the implementation of the social network roles is presumed to
result in a stable state of behavior in the system that impacts the trajectory of
goal node performance until new changes or events occur in the system, either
from the environment or from internal decision-making processes that result in
the activation of new social network roles. Statistically, if enough data points
are gathered, within-person regressions could be calculated on data in the table
252  James D. Westaby and DaHee Shon

to see which roles are having the strongest impact on performance variation (or
multilevel modeling could be used, including between-person data). Practically,
when longitudinal data are unavailable or untenable, individuals could be sur-
veyed about system paths over different time periods, such as reporting major
past, present, and expected future events or changes, along with data on social
network role paths and performance. Objective indicators of performance are
preferred whenever possible (Westaby, 2012a). Coefficient weighting from the
resulting statistical analyses, and/or results from additional policy capturing meth-
ods that manipulate future contingencies using survey methods, could be used, in
part, in simulating how a person’s performance would change over time under
different conditions. In addition, the stability of past goal performance could be
used when simulating future behavior, which we expect would independently
impact system outcomes over time, as discussed in the “Attractor Mechanisms”
section of the chapter. Although some exploratory ideas are presented here, more
research is needed to examine other possibilities.

Events over Social network roles in Person X’s network Positive Subsequent
time focus ratio1 performance
Goal System Goal Supportive System
striving supporting preventing resisting negating
(G) (S) (P) (V) (N)

1. Inspired 1 1 0 0 0 2/2 = 1.0 72 pounds

by friend lost
2. Stress and 1 0 1 0 0 1/2 = .5 33 pounds
overeating gained
3. Re-inspired 1 1 0 0 0 2/2 = 1.0 17 pounds
lost
4. New family 0 1 2 3 1 1/7 = .14 47 pounds
gained
5. Doctor 1 3 0 0 0 4/4 = 1.0 43 pounds
discussion lost (last
data
point)
6. Potential 1 1 1 0 0 2/3 = .67 20 pounds
future stress gained
(expected)

Note: 1 Other aggregated summary variables could be calculated, plotted, and tested as well, such as
the network affirmation ratio.

Notes
  1 There are different types of centrality as well, such as betweenness centrality and degree
centrality, representing different ways to analyze the flow of information through a
social network.
 Social Networks and Goal Striving  253

  2 Future research should also examine potential competency levels on various social net-
work role linkages, extending previous theorizing that focused on entity-level system
competency alone (Westaby, 2012a).
  3 For example, an SIO multiplex role linkage illustrates a person supporting (S) another
while also interacting a lot around the person (I) and watching the person’s behavior (O).
In contrast, a pure S role behavior could represent someone who morally supports
someone (S), but does not interact closely in the goal striving (no I), and nor does the
person have much time to observe the person’s goal striving (no O). Multiplex linkages
have been important in network frameworks (Snijders et al., 2010). However, dynamic
network theory bounds the number of possible multiplex linkages by the finite set of
roles in the theory.
  4 Otherwise, a system reactance path (R) would be shown back to the Relative, which
can be represented as a solid path with a black dot to symbolize the negative emotion
that affirms/protects the goal—otherwise, the path would not be visually distinguished
from system support linkages. See Westaby et al. (2014) or Westaby (2012a) for more
charting specifics. As a less complicated technique, role symbols can be simply placed
on all-black paths.
  5 The network affirmation ratio (Westaby, 2012a) represents the degree to which network
roles are affirming the goal, including motivational properties of defensive system reac-
tance: (ΣG + ΣS ΣR)/(ΣG + ΣS + ΣR + ΣP + ΣV + ΣN). Because no R was illustrated
in this example, this ratio is the same as the positive focus ratio.
  6 Constructive resistance can also occur from goal prevention and supportive resistance
when goal modifications help goal strivers better achieve their goals (Westaby et al.,
2014), although this was not examined in this exploratory simulation.
  7 For negative affect and emotions in networks, the effects were presumed to be more
complex in this model (Westaby, 2012b). For example, system negation can help peo-
ple learn and adapt in some cases, especially when others showing the emotion are
competent in the goal pursuit and are supportive (i.e., increasing computed network
power). In contrast, if a person becomes distracted from system negation or if the
other person has little competence in their suggested changes, it could reduce goal
striving motivation or result in a poor new strategy (i.e., thus not increasing com-
puted network power).
  8 Although we only simulated five individuals, researchers could also generate an entire
distribution of possible behavioral outcomes for a large set of simulated individuals
(Smaldino, Chapter 4 in this volume), given the ease of running simulations. Then,
deviations from normal expectations could be generated. We focused on five networks
for simplicity and to make the simple point that a wide dispersion of emergent out-
comes can occur in small samples.
  9 Research should also account for the concept of authorization levels in dynamic net-
work theory, such as computing an additional interaction term for authorized entities
who have power to dictate system direction (e.g., a company executive has the power to
prevent others from engaging in many behaviors, which would squelch those activities,
perhaps wisely or unwisely). Such effects may be more relevant in certain environments,
such as in organizational, medical, and governmental domains.
10 Otherwise, we would also need to show all network role paths for each emergent data
point (e.g., time 1 G, time 2 G, time 3 G), which would add complexity to the charts.
Research needs to examine these issues.
11 Superscripts are used because subscripts are already used in original dynamic network
charts to indicate intensity levels and performance, such as S2(3), indicating high sys-
tem support (1 = significant, 2 = intense) with high role performance on the linkage
(-3 = very poor performance, 3 = very strong performance) (Westaby, 2012a).
12 Such congestion also happens in traditional network analyses of large systems, although
a dynamic network theory approach is fortunately bounded by the given goal, and may
be less prone to the “boundary problem” (Westaby, 2012a).
254  James D. Westaby and DaHee Shon

13 Data on the separate social network role behaviors would be available at each snapshot
in time within each system as well, and could be used as lower-level predictors in
multilevel testing.
14 Researchers should also examine the utility in eliminating previous restrictions in
ego-centric dynamic network charts, such as allowing multiple roles (instead of just G
and P) to proceed directly to goal nodes.
15 Forecasts from individuals with high DNI could be weighted more strongly in projec-
tions than individuals who demonstrate low DNI on past and current roles. If these
assumptions result in greater calibration with future outcomes, such insights should be
weighted and accounted for in finalized algorithms for future research, which could
then be further tested and calibrated. When collecting data, researchers may also want
to consider the benefits of placing individuals into even-handed implemental mindsets
(Gollwitzer, 2012) when trying to improve long-term predictions.
16 Simulations should also examine group and social interactions in real time based upon
Westaby et al.’s (2016) new observational framework, which examines how levels of
productive and satisfying interactions may result from various combinations of goal
striving behavior during interactions (e.g., pure talking without affect), system support-
ing (e.g., agreement statements, nodding), goal preventing (e.g., disagreeing without
affect), and system negating (e.g., condescending bullying behavior), and their multi-
plex combinations, such as GS (e.g., talking while warmly smiling toward another).
17 For example, if external research shows that the demographic makeup of Person X in
our ongoing example is correlated with weight gain at time Y, this data could be used
in the forecasting algorithm, if Person X has low confidence about his or her network
role behavior and performance at Time Y and past behavior is too variable to discern
stable attractor points or lacks identifiable cycles.
18 Other predictor data could be integrated into computer algorithms as well, including
the aggregation of past, present, and/or future anticipated data from other people in
the network as well as weighting by dynamic network intelligence. Computer algo-
rithms could also be based upon results from empirical studies of real participants
using structural equation models or relevant regressions, which can show the relative
weight of the various factors in dynamic network theory, akin to how research has
used parameters from SEM and regression to set simulation and change parameters
(Schwarz & Ernst, 2009).
19 The authors thank the editors,Yoshi Kashima, and Jennifer Talevich for their helpful
comments, and Genie Song for her help programming the agent-based computer
model.
20 See Westaby (2012a) for further details about scoring path intensity, such as 1 meaning
significant strength of a path and 2 meaning intense strength of a path. In this chapter,
all paths are scored 1, for simplicity.

References
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human
Decision Processes, 50, 179–211.
Aknin, L. B., Barrington-Leigh, C. P., Dunn, E. W., Helliwell, J. F., Burns, J., Biswas-
Diener, R., Kemeza, I., Nyende, P., Ashton-James, C. E., & Norton, M. I. (2013).
Prosocial spending and well-being: Cross-cultural evidence for a psychological universal.
Journal of Personality and Social Psychology, 104, 653–652.
Austin, J. T., & Vancouver, J. B. (1996). Goal constructs in psychology: Structure,
process, and content. Psychological Bulletin, 120, 338–375.
Borgatti, S. P., Mehra, A., Brass, D. J., & Labianca, G. (2009). Network analysis in the
social sciences. Science, 323, 892–895.
 Social Networks and Goal Striving  255

Burt, R. S., Kilduff, M., & Tasselli, S. (2013). Social network analysis: Foundations and
frontiers on advantage. Annual Review of Psychology, 64, 527–547.
Castelfranchi, C., Cesta, A., Conte, R., & Miceli, M. (1993). Foundations for interac-
tion: The dependence theory. In P. Torasso (Ed.), Advances in artificial intelligence
(pp. 59–64). Berlin: Springer-Verlag.
Connelly, B. S., & Ones, D. S. (2010). Another perspective on personality: Meta-analytic
integration of observers’ accuracy and predictive validity. Psychological Bulletin, 136,
1092–1122.
Curran, P. J., & Bauer, D. J. (2011). The disaggregation of within-person and between-
person effects in longitudinal models of change. Annual Review of Psychology, 62,
583–619.
Dovidio, J. F., Piliavin, J. A., Schroeder, D. A., & Penner, L. (2006). The social psychology
of prosocial behavior. Mahwah, NJ: Lawrence Erlbaum.
Duckworth, A. L., Peterson, C., Matthews, M. D., & Kelly, D. R. (2007). Grit:
Perseverance and passion for long-term goals. Journal of Personality and Social Psychology,
92, 1087–1101.
Fishbein, M., & Ajzen, I. (2010). Predicting and changing behavior: The reasoned action approach.
New York: Psychology Press.
Geen, R. G. (1991). Social motivation. Annual Review of Psychology, 42, 377–399.
Gladwell, M. (2000). The tipping point: How little things can make a big difference. New York:
Back Bay Books/Little, Brown.
Gollwitzer, P. M. (2012). Mindset theory of action phases. In P. Van Lange,
A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of social psychology
(Vol. 1, pp. 526–545). London: SAGE Publications.
Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78,
1360–1380.
Harrison, J. R., Lin, Z., Carroll, G. R., & Carley, K. M. (2007). Simulation mod-
eling in organizational and management research. Academy of Management Review,
32, 1229–1245.
Hofmann, D. A. (1997). An overview of the logic and rationale of hierarchical linear
models. Journal of Management, 23, 723–744.
Hu, X. L., & Puddy, R. (2011). Cognitive modeling for agent-based simulation of
child maltreatment. In J. Salerno, S. J. Yang, D. Nau, & S. K. Chai (Eds.), Social
computing, behavioral-cultural modeling and prediction (pp. 138–146). Berlin: Springer-
Verlag.
Kaufmann, P., Stagl, S., & Franks, D. W. (2009). Simulating the diffusion of organic farming
practices in two new EU member states. Ecological Economics, 68, 2580–2593.
Kilduff, M., Crossland, C., Tsai, W., & Krackhardt, D. (2008). Organization network
perceptions versus reality: A small world after all? Organizational Behavior and Human
Decision Processes, 107, 15–28.
Kossinets, G., & Watts, D. J. (2009). Origins of homophily in an evolving social network.
American Journal of Sociology, 115, 405–450.
Labianca, G., Brass, D. J., & Gray, B. (1998). Social networks and perceptions of intergroup
conflict: The role of negative relationships and third parties. Academy of Management
Journal, 41, 55–67.
Nowak, A., Vallacher, R. R., & Zochowski, M. (2005). The emergence of personality:
Dynamic foundations of individual variation. Developmental Review, 25, 351–385.
Orr, M. G., & Plaut, D. C. (2014). Complex systems and health behavior change: Insights
from cognitive science. American Journal of Health Behavior, 38, 404–413.
256  James D. Westaby and DaHee Shon

Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network
dynamics and field evolution: The growth of interorganizational collaboration in the
life sciences. American Journal of Sociology, 110, 1132–1205.
Rhoades, L., & Eisenberger, R. (2002). Perceived organizational support: A review of the
literature. Journal of Applied Psychology, 87, 698–714.
Richetin, J., Sengupta, A., Perugini, M., Adjali, I., Hurling, R., Greetham, D, & Spence, M.
(2010). A micro-level simulation for the prediction of intention and behavior. Cognitive
Systems Research, 11, 181–193.
Schwarz, N., & Ernst, A. (2009). Agent-based modeling of the diffusion of environmen-
tal innovations: An empirical approach. Technological Forecasting and Social Change, 76,
497–511.
Smith, E. R., & Conrey, F. R. (2007). Agent-based modeling: A new approach for
theory building in social psychology. Personality and Social Psychology Review, 11,
87–104.
Snijders, T. A. B., Van de Bunt, G. G., & Steglich, C. E. G. (2010). Introduction to actor-
based models for network dynamics. Social Networks, 32, 44–60.
Stajkovic, A. D., & Luthans, F. (1998). Self-efficacy and work-related performance: A
meta-analysis. Psychological Bulletin, 124, 240–261.
Tetlock, P. E., & Gardner, D. (2015). Superforecasting: The art and science of prediction. New
York: Crown.
Tetlock, P. E., & Mellers, B. A. (2011). Intelligent management of intelligence agencies:
Beyond accountability ping-pong. American Psychologist, 66, 542–554.
Tobias, R. (2009). Changing behavior by memory aids: A social psychological model of
prospective memory and habit development tested with dynamic field data. Psychological
Review, 116, 408–438.
Vallacher, R. R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 8, 73–99.
Vallacher, R. R., Read, S. J., & Nowak, A. (2002). The dynamical perspective in per-
sonality and social psychology. Personality and Social Psychology Review, 6, 264–273.
Vallacher, R. R., Van Geert, P., & Nowak, A. (2015). The intrinsic dynamics of psycho-
logical process. Current Directions in Psychological Science, 24, 58–64.
Vancouver, J. B., Weinhardt, J. M., & Schmidt, A. M. (2010). A formal, computational
theory of multiple-goal pursuit: Integrating goal-choice and goal-striving processes.
Journal of Applied Psychology, 95, 985–1008.
Wang, M. H., & Hu, X. L. (2012). Agent-based modeling and simulation of com-
munity collective efficacy. Computational and Mathematical Organization Theory, 18,
463–487.
Wang, S. W., Huang, C. Y., & Sun, C. T. (2014). Modeling self-perception agents in an
opinion dynamics propagation society. Simulation: Transactions of the Society for Modeling
and Simulation International, 90, 238–248.
Warren, W. H. (2006). The dynamics of perception and action. Psychological Review, 113,
358–389.
Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications.
Cambridge: Cambridge University Press.
Weinhardt, J. M., & Vancouver, J. B. (2012). Computational models and organizational
psychology: Opportunities abound. Organizational Psychology Review, 2, 267–292.
Westaby, J. D. (2005). Behavioral reasoning theory: Identifying new linkages underly-
ing intentions and behavior. Organizational Behavior and Human Decision Processes, 98,
97–120.
 Social Networks and Goal Striving  257

Westaby, J. D. (2012a). Dynamic network theory: How social networks influence goal pursuit.
Washington, DC: American Psychological Association.
Westaby, J. D. (2012b). Emergence and contagion in dynamic network systems. Paper pre-
sented at Dynamical Systems and Computational Modeling in Social Psychology Preconference,
San Diego, CA.
Westaby, J. D., Pfaff, D. L., & Redding, N. (2014). Psychology and social networks: A
dynamic network theory perspective. American Psychologist, 69, 269–284.
Westaby, J. D., Probst, T. M., & Lee, B. C. (2010). Leadership decision-making: A
behavioral reasoning theory analysis. Leadership Quarterly, 21, 481–495.
Westaby, J. D., & Redding, N. (2014). Social networks, social media, and conflict
resolution. In P. T. Coleman, M. Deutsch, & E. C. Marcus (Eds.), The handbook
of conflict resolution: Theory and practice (3rd ed., pp. 998–1022). San Francisco, CA:
Jossey-Bass.
Westaby, J. D., Versenyi, A., & Hausmann, R. C. (2005). Intentions to work during
terminal illness: An exploratory study of antecedent conditions. Journal of Applied
Psychology, 90, 1297–1305.
Westaby, J. D., Woods, N., & Pfaff, D. L. (2016). Extending dynamic network theory to
group and social interaction analysis: Uncovering key behavioral elements, cycles, and
emergent states. Organizational Psychology Review, 6, 34–62.
12
COMPUTATIONAL MODELS
OF SOCIAL INFLUENCE AND
COLLECTIVE BEHAVIOR
Robert J. MacCoun

A photo of the blackboard in Richard Feynman’s office features his assertion

that “What I cannot create, I do not understand.”1 This chapter is an encourage-
ment to social psychologists to solidify and advance their understanding of social
influence and collective behavior by creating and testing computational models—
models that can be formally stated with enough specificity to allow them to be
run as computer simulations.
Almost a century ago, Floyd Allport (1920) suggested that there are two kinds
of data in social psychology:

(1) the behavior of an individual in direct response to social stimulus, that is

in response to some form of behavior in others, and, (2) behavior which is
the response to a non-social stimulus, e.g., a column of figures to be added,
or a meal to be eaten, when such response is modified by the presence and
actions of other persons.

Much of contemporary social psychology involves the second category; direct

study of the first category is actually less common than an outsider to the field
might suppose. But Allport suggested that it was the first category that posed “the
master problem” for social psychology—“the problem of the relation of a single
individual’s action to the collective result of many individuals acting together”
(quoted in Brooks & Johnson, 1978). This chapter takes as its focus efforts to make
progress on that problem—specifically, the nature of the relationship between the
opinions and behaviors of individuals and those of others around them.
To make this topic manageable, I ignore individual differences in suscepti-
bility and influence, for example those produced by differences in power and
status (see Abrams, 1980). Also, some economists and political scientists may
be surprised by my treatment of models as mechanisms or algorithms rather
 Models of Influence and Collective Behavior  259

than equilibrium solutions deduced from axioms, but that reflects my training—
psychologists are wary of both axioms and equilibria, for a variety of reasons
appropriate to our discipline. At the same time, I will blur some distinctions that
are vitally important for a psychology of social influence, including descriptive
versus injunctive norms (Cialdini, Reno, & Kallgren, 1990), French and Raven’s
(1960) taxonomy of forms of social power (e.g., rewards, legitimacy, expertise),
and Kelman’s (1958) distinction between compliance, identification, and inter-
nalization, all of which I discuss elsewhere (MacCoun, 2012, 2015). Perhaps the
most surprising omission from a chapter on social influence is the almost com-
plete lack of attention given to the content of argumentation, but that’s because
a key goal of the chapter is to show how well social influence can be described
by strength in numbers rather than strength in arguments.
Throughout this chapter, I’ll attempt to maintain a consistent notation across
the different models I discuss, which is not necessarily the notation in which each
model is described elsewhere. Following Latané (1981), I’ll generally refer to
S = number of Sources who are advocating one position (opinion, trait, behavior)
and T = number of Targets who currently favor an alternative, with s = S/N repre-
senting the proportion of the local population currently favoring the first option.
(This language comes from studies of conformity and persuasive appeals; in stud-
ies of mutual deliberation, it is arbitrary which faction is considered the sources
and which is considered the targets.) I will use p as a dependent variable for the
probability that an individual chooses to convert from Option 2 to Option 1, and
P to refer to the probability that the group chooses Option 1.

Five Key Ideas

Before delving into the details of specific models, it is worth reviewing some
verbal propositions about this domain. These are not “laws”—at least not in the
deterministic sense; rather, they are empirical regularities.

People Influence Each Other

A corollary of this first point is that people monitor each other; we are exquisitely
attuned to those around us, and those of us who aren’t have difficulty surviving.
(The reader will already be familiar with many classic discussions of this point,
but for insightful new perspectives, see Boyer, Firat, & Van Leeuwen, 2015;
Leary, 2005.) This is so obvious that it may not seem worth mentioning; it would
seem to be the very bedrock of the social sciences. But the reader can verify from
any recent journal issue that most social science analyses omit others’ responses
from the right-hand side of any tested model. In the laboratory, one reason for
this omission is that experiments rarely involve interacting individuals or multiple
observation periods in which people can respond at time t + 1 to behaviors at time t.
In the field, social influence is often ignored for a different reason; the relation-
ship between the group and individual is usually endogenous, making causal
260  Robert J. MacCoun

identification very difficult (MacCoun, Cook, Muschkin, & Vigdor, 2008). But
the arrival of big data from social media is solving the first problem while it is
demanding that we grapple with the second.

Social Influence Rises with the Number of Sources

As a very rough first approximation, the probability that an actor changes opin-
ions rises with the proportion of others who hold a position the actor doesn’t
currently hold. Indeed, Mullen (1983) showed that an extremely simple model
(the “Other-Total Ratio”) can account for a lot of variance in social behavior:

S  (12.1)
p = m 
N 
where m is a scaling constant that serves as a “ceiling” parameter.2 However, we
will see that other models provide a much better approximation. There are even
cases where social influence appears to be non-monotonic (MacCoun, 2012): for
example, when people don’t want to be “on the bandwagon” (Brewer, 1991), or
when a lone dissenter seems to have negative social influence (Cialdini et al., 1990).

Social Influence Is (Often) Marginally Diminishing

A somewhat better approximation comes from positing that influence is margin-
ally decreasing (concave) in influence sources. This is a feature of many models,
and it has multiple causes. In social impact theory (Latané, 1981; Nowak, Szamrej, &
Latané, 1990), this follows from the interpretation of social impact as a psycho-
physical phenomenon (Stevens, 1957), with a power function
k
S  (12.2)
p = m 
T 
where k is an exponent that varies from situation to situation. Note that the ceil-
ing parameter (m) in each of the above models (and some that follow) provides
a second mechanism for producing decreasing marginal influence. A third con-
tributing factor, not explicitly modeled in any of the models presented here, is
information-theoretic: Wilder (1977) has demonstrated that marginal influence
drops off more rapidly as a function of how similar (and hence partially redun-
dant) sources are to each other. A similar result is implied by a Bayesian analysis
of consensus judgments (Dawes, 1989).

Social Influence Is Disproportionate (Majority Amplification)

In Figure 12.1, the 45-degree diagonal line shows the baseline of strictly pro-
portional influence: that is, if the individual probability of voting for Option 1
is p = .75, the group probability of choosing Option 1 is P = .75.
 Models of Influence and Collective Behavior  261

The thick grey line in Figure 12.1 shows the predictions of Condorcet’s
(1785) famous “jury theorem.” The sigmoidal shape predicted by Condorcet’s
model exhibits disproportionality—a simple majority of, say, 6/10 actually has a
greater than .6 probability of choosing Option 1. Note that the Condorcet model
does not predict simple diminishing marginal influence; rather, the Condorcet
function is concave for a large number of sources, but it is convex for a smaller
number of sources, producing an S-shaped function.
Condorcet’s model follows from elementary probability theory, under two
assumptions: (1) group decisions can be described by a majority rule, and (2) the
probability that the group chooses Option 1 is given by
N
P = ∑pn (1 − p )N −n , n1 > .5N 
1 1 (12.3)
n1

where P is the probability of the group choosing Option 1, p is the probability

that an individual votes for Option 1, and n1 is the size of a majority voting for
Option1. Thus, for a nine-person group, the summation is across the five initial
splits that would produce a majority for Option 1: 5:4, 6:3, 7:2, 8:1, and 9:0.
On the basis of a third assumption—that members are more likely to be cor-
rect than not (i.e., that p > .5)—Condorcet made a strong normative claim that

FIGURE 12.1  Comparing the Condorcet and Lorge-Solomon models.

262  Robert J. MacCoun

the group choice is therefore superior to individual choice. There is ample evi-
dence that this assumption is often wrong. As seen in Figure 12.1, the formal
model itself predicts that if p > .5, then P >> .5. But it also predicts that if p < .5,
then P << .5. Thus, Condorcet’s real contribution is not his dubious normative
argument, but his insight that group processes can produce what Rosenwein and
Campbell (1992) aptly labeled majority amplification. Note that the inflection point
in Figure 12.1 can be considered an implicit threshold or “tipping point.” The
most famous conformity experiments—those of Asch (1956)—show this kind of
sigmoidal pattern because the second source had more marginal impact than the
first. In sigmoidal patterns, marginal influence is increasing (convex) before the
inflection point, but decreasing (concave) beyond it.

Social Influence Is Often Asymmetrical

The models considered thus far are symmetrical in the following sense: The influ-
ence of a given faction is a function of its relative size, but not of the position it
advocates, and hence a majority’s drawing power is the same in a 60:40 split or
a 40:60 split. But social influence is often asymmetrical (see MacCoun & Kerr,
1988; Kerr & MacCoun, 2012). One reason is that one side may be arguing the
“correct” position, where “correctness” is defined by some shared conceptual
system (arithmetic, symbolic logic, a rule book, an encyclopedia) that allows
advocates to demonstrate that they are right (see Laughlin, 2011). Lorge and
Solomon (1955) offered the following normative model of optimal group per-
formance on such tasks:

P = 1 − (1 − p ) 
N
(12.4)

As a normative model, this assumes that “truth wins”—that the group will find
the correct answer so long as one member finds it. In other words, a faction of a
given relative size (say, 25% of the group) will have more drawing power when
it is arguing for the position most demonstrably “correct” relative to the shared
conceptual scheme. Empirically, even in tasks with a clearly demonstrable cor-
rect answer, it often takes more than two initial advocates for the truth to prevail
(“truth-supported wins”; see Laughlin, 2011; see also Asch, 1956). Thus, social
influence involves both strength in numbers and strength in arguments (Deutsch &
Gerard, 1955). But the asymmetry need not be rooted in either factual or logi-
cal arguments; it might instead reflect a shared cultural attitude; for example, in
criminal juries, advocates of acquittal have greater drawing power than compa-
rably sized factors for conviction, reflecting the asymmetric “reasonable doubt
standard” (MacCoun & Kerr, 1988; Kerr & MacCoun, 2012).

The Social Decision Scheme Framework

Davis (1973) proposed a more general modeling framework for social-psychological
research on group decision processes, one that has been quite productive for
 Models of Influence and Collective Behavior  263

many decades (Davis & Witte, 1996; Kerr, MacCoun, & Kramer, 1996; Parks &
Kerr, 1999; Stasser, Kerr, & Davis, 1989). In the social decision scheme (SDS) frame-
work, the group decision process is parsed into two components—the sampling
process and the decision process.
The sampling process describes the probability πi that the group will start with
a particular initial split—e.g., 10 votes for Guilty versus two votes for Not Guilty.
For the sampling function in a dichotomous choice situation, Davis proposed the
binomial expansion

πi = pn (1 − p )N −n , ni > .5N 
i i (12.5)

where ni is a given number of votes for option 1 in initial split i. The major
theoretical contribution of this component of the SDS approach is to remind
us that how situations evolve is heavily influenced by the processes determining
which participants comes to find themselves in the situation. But in reality, most
SDS papers ignore this component because in social psychology experiments,
the composition of groups is either experimentally controlled (rarely), or else
the sampling process is a haphazard matter of convenience. And in any case, the
binomial expansion is often dubious because it assumes (a) that the sampling of
one participant is independent of the sampling of other participants, and (b) once
sampled, groups will actually persist long enough to reach a decision. In reality,
neither assumption is tenable in the real world.
The real meat of the SDS approach comes from the notion of a social deci-
sion scheme. A particular social decision scheme D gives the probability dij that
the group will adopt option j given initial split i. (For finer-grained analyses
of moment-by-moment individual and group transition probabilities, see Stasser
et al., 1989.) The expected outcomes are given by πidij. The power of the SDS
framework is that it encourages researchers to consider a variety of theoretical
decision schemes. For example, in the Proportionality scheme—a useful baseline
model—di = i/N. A “Simple Majority” scheme is given by

i
1.0 if > .5 
N
i
di = 0.5 if = .5
N
i (12.6)
0.0 if < .5
N

Note that for this scheme, πD is equivalent to Condorcet’s model (Equation 12.3).
A “Truth Wins” scheme (assuming the first option is “correct”) is given by

1 if i ≥ 1
di = 
0 if i < 1 (12.7)
264  Robert J. MacCoun

Since the SDS approach was first articulated, scholars have occasionally con-
fused D matrices with formal voting rules. This is not correct; D is a summary
statement of all the combined individual and group factors that lead the group
to behave a particular way. Thus, D can readily be estimated or tested in situ-
ations where there is no formal voting rule (and indeed, when groups may not
even think of themselves as forming a group), and the D that best characterizes
a group’s behavior is often different than the group’s formal voting rule—for
example, groups operating under a formal unanimity requirement tend to behave
as if they were using a Two-Thirds Majority rule.
A large variety of such schemes have been proposed and examined. SDS
researchers can either test the goodness of fit between each D and some data
set, or use the data to describe an empirically derived D (Kerr, Stasser, & Davis,
1979). For example, MacCoun and Kerr (1988) meta-analyzed 11 experiments
to estimate the typical asymmetric scheme that best characterizes juries operating
under a reasonable-doubt standard.
Arguably, an important strength of the SDS approach is that it shifts the
focus from a search for “What is the model account of social influence?” to
“Which models are best for describing which situations?” Thus, Simple and
Two-Thirds Majority schemes fare best in judgmental tasks with no clear cri-
teria for conclusively demonstrating a correct answer; in tasks that do have
such criteria, groups are better characterized by a “Truth-Supported Wins” D
(in which at least two members need to initially endorse the solution) than a
Truth-Wins D.

The BOP Framework

MacCoun (2012) demonstrated that, despite their strengths, existing influence
models like the Other-Total Ratio and Social Impact Theory have important
limitations. In particular, models that fare well in a conformity paradigm (where
multiple sources unidirectionally influence a target) perform more poorly in a
deliberation paradigm (where factions bidirectionally influence each other), and
vice versa. And despite its many strengths, the SDS approach has limitations of its
own. It is more suitable for characterizing group decisions than the effects of oth-
ers on individual decisions, and the machinery of multiple matrices doesn’t always
facilitate descriptive statements and generalizations. Finally, existing models aren’t
always well integrated with more general theories of choice and methods of data
analysis.
MacCoun (2012, 2015) proposed a family of models for identifying and charac-
terizing the apparent “burden of (social) proof” (BOP) in social influence settings.
The BOP framework is an attempt to be integrative in two different senses.
First, it attempts to provide a common set for formalisms that are equally
applicable across distinct social influence paradigms: the conformity paradigm,
the deliberation paradigm, the diffusion-of-innovations paradigm, etc. And
second, the BOP models can be directly, explicitly, and formally linked to
 Models of Influence and Collective Behavior  265

many other bodies of theory in psychology and the social sciences, including
the Luce strict utility model, the McFadden random utility model, item-
response theory, and the Schelling tipping point model (see MacCoun, 2012),
as well as signal detection theory and prospect theory (see MacCoun, 2015).
The core model in this approach is the bBOP (originally “bidirectional burden
of proof”) model
m (12.8)
p= 
 S 
1 + exp c  − b  
 N 

in which b is a threshold parameter denoting the inflection point at which the

target is more likely than not to adopt the source’s advocated position, and c is
a clarity parameter which reflects the steepness of the threshold. A low level of c
approaches random responding, while a high level of c approaches an abrupt step
function. When the parameters are estimated from aggregate data, c is inversely
related to the standard deviation of individual thresholds. Thus, c indexes the
degree to which there is a socially shared threshold or perceived burden of proof
in one direction or the other. When c is very high for a population, behavioral
changes are correlated across actors and it is meaningful to talk of tipping points;
e.g., Schelling’s (1978) demonstration of tipping in housing segregation used a
step function. When c is lower, individual changes are only weakly correlated and
social change is more gradual (or, at the lower limit, a random walk). In empiri-
cal applications so far, I’ve needed to include the m parameter to fit data in the
conformity paradigm, where multiple sources impinge on a single target; in other
paradigms, the c and b parameters seem to suffice (m is set to 1) (MacCoun, 2012).
One variant, uBOP (unidirectional burden of proof), like social impact theory,
uses the S/T ratio (in effect, an odds format) rather than the S/N ratio.
m (12.8)
p= 
 S 
1 + exp c  − b  
  T 

For complicated reasons that are now moot, MacCoun (2012) conjectured that
uBOP would better describe behavior in the conformity paradigm and bBOP
would better describe behavior in the deliberation paradigm, but in fact, the
models both performed quite well. Figure 12.2 shows two examples. The top
row shows that the Other-Total Ratio and Social Impact Theory each predict a
pattern of monotonically increasing but diminishing marginal social influence.
Thus both do well at fitting Milgram and colleagues’ famous study of whether
urban passersby would look up in the sky (Milgram, Bickman, & Berkowitz,
1969), as a function of how many sources (Milgram’s students) were already
gazing up. But neither predicts Asch’s 1956 finding that a second apparently
wrong source had greater impact than the first one. As seen in the bottom row
of Figure 12.2, uBOP fits both patterns quite well.
 Milgram et al. (1969) Asch (1951) Kerr & MacCoun (1985)

0.8 0.30
0.8
0.20
0.4 0.4
0.10

p (change)
p (change)
0.0 0.00 0.0

p (guilty verdict)
0 5 10 15 0 5 10 15 2 4 6 8 10
Sources Sources Number favoring guilt (of 12)
Data OTR SIT

Milgram et al. (1969) Asch (1951) Kerr & MacCoun (1985)

0.8 0.30
0.8
0.20
0.4 0.4
0.10

p (change)
p (change)
0.0 0.00 0.0
p (guilty verdict)

0 5 10 15 0 5 10 15 2 4 6 8 10
Sources Sources Number favoring guilt (of 12)
Data uBOP bBOP

FIGURE 12.2   omparing fit to three data sets. Top row: OTR and SIT. Bottom row: uBOP and bBOP. See MacCoun (2012)
C
for a detailed presentation.
 Models of Influence and Collective Behavior  267

Even though bBOP is more broadly useful, uBOP is helpful for character-
izing Asch-type conformity situations involving small numbers of participants
and a single target, because it puts the b threshold on an absolute “person” metric
rather than a relative “percentage” metric. For example, in the Asch (1956) con-
formity study, the estimated thresholds were .67 for bBOP and 2.10 for uBOP.
The former value is awkward to interpret because both the numerator and the
denominator change as more confederates are added to the experiment, but the
latter directly tells us that Asch’s respondents began to “tip” toward conformity
(to a clearly incorrect option) once there were at least two people disagreeing
with them. Note that the formulae also predict Asch’s finding that adding a sec-
ond target to the denominator significantly reduced conformity.
The Milgram et al. (1969) and Asch (1956) studies exemplify a conformity
paradigm, where multiple sources impinge on a single target. In a deliberation
paradigm, there is a fixed group size, so any increase in sources produces a reduc-
tion in targets. As seen in the third column of Figure 12.2, OTR and SIT do a
poor job of characterizing data from criminal mock juries (Kerr & MacCoun,
1985). The jury data show a sigmoidal pattern, but one that is asymmetric—
largely due to the reasonable-doubt standard which gives advocates of acquittal a
deliberative advantage (MacCoun & Kerr, 1988; Kerr & MacCoun, 2012).

Modeling Continuous Judgments

While there are large literatures examining the predictors that influence con-
tinuous group judgments (e.g., juries’ compensatory damage awards in personal
injury trials), there has been relatively little work on modeling the process by
which groups combine their opinions into continuous quantitative judgments.
One reason is that, to a first approximation, one can often predict these judg-
ments fairly well with the median or mean of the individual member’s personal
recommendations (see Graesser, 1978). Hinsz (1999) describes a large number of
possible models falling into six basic categories: central-tendency, consensus-based,
faction-attraction, coalition-based, distance-influence, and dictator schemes. But
far too few studies have examined the relative fit of these models across tasks and
settings, and it is premature to offer strong views about model validity (Hinsz,
1999; Davis et al., 1997).
Nevertheless, there are some important empirical regularities that any viable
model should predict (or more accurately, postdict). The first is the wisdom of crowds
(WOC) effect (Surowiecki, 2005), first documented by Francis Galton (1907), in
which an aggregate (median or mean) of a large number of individual judgments
tends to be far more accurate than that of most or even all of the individuals. The
WOC effect is really a statistical phenomenon—a demonstration of the cancel-
lation of random errors predicted by the law of large numbers. Indeed, a WOC
effect is more likely to be obtained if the individuals in the crowd are completely
unaware of each others’ judgments (Lorenz, Rahut, Schweiter, & Helbing, 2011).
268  Robert J. MacCoun

A second is group polarization, the tendency for post-deliberation mean judg-

ments to be more extreme in the direction of the initial mean judgment (for a
review, see Isenberg, 1986). Note that group polarization is most commonly doc-
umented in cases involving bipolar judgments with a meaningful midpoint—for
example, pro-to-anti attitudes, bad-to-good evaluations, etc. Indeed, the phenom-
enon is partly explainable by the majority amplification effect discussed above—a
boost in the relative drawing power of whichever pole has a majority of initial
advocates (see Stasser et al., 1989). Group polarization provides an important cor-
rective to the rosy view of groups implied by the WOC effect. While groups (or
actually, aggregation) provide an effective way of attenuating noise (random error),
group deliberation can often amplify bias (systematic error; see Kerr et al., 1996).
Something akin to group polarization has been observed for groups making
judgments on a monetary scale—for example, juries deciding how much to award
in punitive damages. Schkade, Sunstein, and Kahneman (2000) have documented
a severity shift such that the group’s recommended award is often considerably
larger than the mean of their initial opinions. Though there may be some element
of “emotional contagion” involved, the authors argue that the effect is partly
attributable to the difficulty of making judgments on a scale that is bounded at
zero on one end but unbounded at the other (see MacCoun, 2005).

Modeling Growth over Time

It is useful to distinguish two classes of models that are easily confused with
each other. The first class, social influence modeling, characterizes the probability
of change in the prevalence of a trait, behavior, or opinion as a function of the
relative popularity of that characteristic. The second class, social growth modeling,
characterizes the change in prevalence of a trait, behavior, or opinion as a func-
tion of time. The approaches are confusable because each produces plots showing
monotonic increases in influence—often sigmoidal—but the horizontal axes are
different (prevalence of an opinion versus elapsed time). In principle, any social
influence model can be recast as a social growth model (e.g., through simple
iteration), but the converse is not necessarily true because some models use time
as an independent variable (rather than as an index).
If Condorcet (1785) is the grandfather of social influence modeling, Verhulst
(1838) is the grandfather of social growth modeling.Verhulst’s model begins with
the proposal that the rate of growth of a condition (a disease, an opinion, a behavior)
in a population is a function of the product of those who already have the condi-
tion (in our terms, S) and those that do not (in our terms, T), weighted by some
constant r. This basic assumption was so influential in epidemiological modeling
that Wilson and Worcester (1944) named it “the law of mass action.” As seen on the
left side of Figure 12.3, this rate reaches its peak when half the population is already
“infected.” (Statistically, this is the point of maximum variance for a dichotomous
variable.) If we iterate this rate from an initial small S, we find something like the
 Models of Influence and Collective Behavior  269

FIGURE 12.3  The “law of mass action,” with r = 0.01 (left) and the pattern of
growth over time it produces (right).

sigmoid curve on the right panel of Figure 12.3. Note that, as Verhulst himself rec-
ognized, this “law” is based on an assumption of uniform mixing of the two groups
of people that is convenient as an approximation but unlikely to be strictly true.
Verhulst’s logistic model is the integral of this rate; put differently, the “law
of mass action” is the derivative of Verhulst’s logistic model. Strictly speaking,
Verhulst originally modeled the growth of a single population, but his model is
readily adapted to persuasion (or infection) processes in which the increase in
sources (or infected people) brings about a reduction in targets (or susceptible
people). His full model incorporates a ceiling-type parameter, K, often called the
“carrying capacity” or the “saturation rate.” It is the maximum possible number
of people (in a population growth analysis), sources (in a social influence analysis),
or infected (in an epidemic analysis). Using our notation, it is equivalent to mN,
where, as above, m is a ceiling parameter between 0 and 1.

K (12.10)
St =
 K  − rt
1+  e
 S0 − 1 

The basic logic of Verhulst’s model has been extended in various ways in fields
such as epidemiology and sociology (Daley & Gani, 1999; Newman, Barabási, &
Watts, 2006; Smaldino, Janssen, Hillis, & Bednar, 2015), where state-of-the-
art models are considerably more complex, making far more sophisticated
270  Robert J. MacCoun

assumptions about contact, susceptibility, network structure, and population

heterogeneity. Another large literature in growth modeling examines the diffu-
sion of innovations (see Mahajan, Muller, & Bass, 1995). MacCoun (2012) shows
that the following iterative version of bBOP, called iBOP, does a good job of
fitting data on technological innovations, political participation, and the spread of
drug use among college students:
m − st −1 (12.11)
st = st −1 + 
1 + exp  −c ( s − b ) 

This kind of growth modeling is rare in social psychology, partly because

longitudinal data are rarely collected in experiments. And there are good
reasons to be wary of growth modeling as a tool for social psychology. First,
trend data are difficult to interpret causally; for example, what can look like
social momentum due to imitation or conformity might actually be a lagged
response to common information (for examples, see MacCoun, et al., 2008),
or simply autocorrelated error. Second, the Verhulst tradition favors deduc-
tive models from calculus that provide closed-form equilibrium solutions.
But such solutions come at a price; for the models to be tractable, they have
to make many simplifying assumptions about population structure and mixing
(see Rahmandad & Sterman, 2008).

Simulating Communities of Interacting Agents

Simulation modeling provides a more flexible (and arguably more realistic)
approach, exploring growth processes numerically (i.e., through iterative compu-
tation) rather than analytically (i.e., through closed-form equilibrium solutions).
Monte Carlo simulation is one approach; a computer draws simulated samples
and/or sampled mixes of parameter settings in order to examine the implications
of a model under a variety of plausible settings (for examples, see MacCoun,
2012). Monte Carlo simulation has become incredibly fast and easy; it is a tool
every researcher should use regularly, if only as an “intuition pump” when think-
ing through research problems.
Agent-based modeling (ABM; see Nowak, Szamrej, & Latané, 1990; Railsback, &
Grimm, 2011; Smaldino, Calanchini, & Pickett, 2015)3 is a powerful tool for
exploring questions that would be nearly impossible to examine experimen-
tally. ABM has its origins in mathematician John Conway’s computer game Life
(Gardner, 1970) and the simple but highly insightful checkerboard simulations
conducted by Thomas Schelling (1978). In the cellular automata approach, a grid
of locations (cells) is randomly seeded with “agents,” who then interact iteratively
with their neighbors and their environment. The agents are simply lines of com-
puter code consisting of rules for exploration, interaction, and locomotion. In the
newer network approach (Newman et al., 2006), agent relationships can take on
a much broader array of structures.
 Models of Influence and Collective Behavior  271

The left side of Figure 12.4 shows a “Moore neighborhood” in which each
agent is exposed to up to eight neighbors (using compass directions: NW, N, NE,
E, SE, S, SW, and S). In the example, the 16 cells have been randomly seeded
with 10 agents. In a torus formation, the neighboring locations wrap past the
borders; thus in a torus, Agent 10’s neighbors would include Agents 6 and 9, but
also Agents 2 and 4 and 7. In a grid formation, there are border effects, which
some modelers prefer to avoid. But of course, the real world has many borders.
Allowing empty cells and borders enables the simulation to encompass greater
heterogeneity of exposure or connectedness. For example, Agent 1 is far more
isolated in the grid formation than the torus. Agents with more neighbors are
exposed to more information, and if the simulation allows it, they have more
potential exchange partners. Another factor is whether agents are allowed to
“see” beyond their immediate Moore neighborhood to some radius greater than
one cell; MacCoun (2012) showed that this vision parameter had a profound
effect on social influence patterns, enabling minority opinions to persist by reduc-
ing minority isolation.
The right side of Figure 12.4 shows the same set of relationships as a net-
work, with various common indices of network structure. Network models are
a more general formalism than two-dimensional cellular automata (which can

FIGURE 12.4  Left: a randomly seeded Moore neighborhood of 10 individuals

(in 16 locations). Right: the equivalent social network.
272  Robert J. MacCoun

be considered the subclass of “lattice networks”), though an n-dimensional

cellular automata model can in principle represent any network (Bonacich,
2001). But real-world social networks have some properties that are better
represented by lattice structures than by some of the more esoteric network
structures that have received recent attention (e.g., “small-world networks”;
Newman et al., 2006).
One such property is clustering (see Nowak et al., 1990). Figure 12.5 shows
the 50th iteration of each of four different runs of an agent-based simulation
using the bBOP model. In each case, there is a 51 × 51 grid initially seeded
with 910 red and 910 blue agents; white cells are empty. In all four simulations,
agents were assigned a clarity value of c = 20 and a threshold value of b = .5.
In the City 1 simulations (top row), the bBOP model determines the probabil-
ity that an agent will find its neighbors too dissimilar and move to the nearest
randomly chosen open cell. In the City 2 simulations, however, agents cannot
move, but they can change colors (opinions, behaviors, political parties), and the
same bBOP model gives the probability of change. In all four simulations, 50
iterations was sufficient to produce a considerable amount of clustering—reds
with reds and blues with blues. But the simulations suggest that mobility produces
different clustering than mutability. In the City 1 configuration, the empty cells are
redistributed in a manner than provides buffers or “moats” separating the groups.
As a result, the City 1 simulations produce greater segregation and less between-
group contact and exposure.

Are Model Results “Results”?

Is this differential pattern of clustering a “finding” in the same sense as, say, Asch’s
(1956) finding that conformity pressures on a lone target tend to level off after
three sources? No, of course not. Simulations are an exercise in theory build-
ing, and they enable theories of considerable complexity—too much complexity
for us to simulate by intuition—using fairly simple rules. We can use empiri-
cal evidence to validate features and predictions of our models, but the models
themselves don’t produce empirical results. But this is not a profound shortcoming.
First, theory is important for science, and social psychological “theories” are often
little more than a vague set of verbal propositions that offer simple ordinal or
directional predictions and little clear linkage to theories of other phenomena in
the field. Second—and here the author can speak from three decades wearing a
second hat as a professional public policy analyst—it is wrong to think that we
generalize from our data when we offer predictions or suggestions for coping
with real-world problems. Rather, we generalize from our theories, as Lewin’s
dictum “nothing so useful as a good theory” reminds us. Empirical data are essen-
tial for theory building and theory testing, but to paraphrase Keynes, those who
think they prefer hard facts to academic theories are usually unaware that they are
in the thrall of some dead theorist.
 (a) Move t1 d70 p50 v1 c10 b50 5404
Period 50

(b) Move t1 d70 p50 v1 c10 b50 1623

Period 50

FIGURE 12.5  S imilar patterns of emergent clustering from bBOP in models where
agents can relocate but not change opinions (“Move”, plots a and b)
or can change opinions but not relocate (“Change”, plots c and d).
Note the “buffer zones” that form in the “Move Cities.”
 (c) Change t1 d70 p50 v1 c10 b50 3853
Period 50

(d) Change t1 d70 p50 v1 c10 b50 3852

Period 50

FIGURE 12.5  (Continued)

 Models of Influence and Collective Behavior  275

The Excessive Distrust of Free Parameters

This raises the issue of free parameters, with which many psychologists have at least
a passing familiarity. For many academic psychologists, a model’s free parameters
are at best an embarrassment, if not an object of scorn (see Roberts & Pashler,
2000). In brief, free parameters are model components whose values are fitted to
a data set by using an iterative algorithm. For example, spreadsheet programs and
most statistical packages have algorithms that will seek the values of free param-
eters that minimize the discrepancy (e.g., mean squared error) between a model’s
predictions and the observations in the data set. Clearly, some caution is war-
ranted. First, ceteris paribus, models with more free parameters will fit data better
than models with fewer parameters, and a model that closely fits data with no free
parameters is impressive indeed. Second, many reasonable models will provide an
impressive correlation to data that are monotonically increasing (or decreasing)
as a function of an independent variable. According to Occam’s razor, we should
always prefer the simplest model, and fit indices like the Aikake information cri-
terion steeply penalize free parameters. But according to what’s sometimes called
“Einstein’s razor,” we want our models to be as simple as possible, but no simpler.
MacCoun (2012) provides a good illustration of the choice between these
criteria. Looking at a dozen data sets, some models (e.g., social impact theory)
sometimes outperformed the BOP models with respect to the Aikake criterion.
Yet across all 12 data sets (involving multiple research paradigms), the BOP mod-
els consistently performed well (and performed best on average), whereas other
models were much less consistent, sometimes failing badly. Arguably, the BOP
models are superior with respect to what philosophers of science call the “best
system” criterion for scientific theories, which balances simplicity (fewer axioms
for deduction, or free parameters for induction) with strength (greater informa-
tiveness or explanatory power; see Lewis, 1973). And it is not clear why we even
need to be parsimonious. Computation is now extremely cheap and relatively
easy. Rather than using a handful of empirical tests to “eliminate” models, we
can share our modeling code and routinely examine a whole set of plausible
models, discovering how they perform in different situations—and in the process,
familiarizing our junior collaborators with a range of ways of conceptualizing
the domain. Following this approach, models may remain viable contenders for
much longer, but when we finally abandon them, we will have a firm sense of
why we are doing so and why they were wrong or incomplete.
Psychologists have an unduly narrow view of parameters and the role that
they can play in scientific progress. Free parameters needn’t be a gimmick to
bolster a theory; they can be a source of valuable information leading to better
understanding of phenomena. Oddly, psychologists often balk at a “mathematical
model” with two to four free parameters, yet few blink when encountering, say,
10 free parameters in a structural equation model (which is, of course, a math-
ematical model) involving four variables or a logistic regression equation with
nine explanatory variables.
276  Robert J. MacCoun

Fitting free parameters can and should be seen as a tool for investigation and
estimation rather than a definitive test for validation. This requires us to seek
parameters that can develop a meaningful psychological interpretation. For exam-
ple, the best-fitting uBOP model of the Asch (1956) data, with 4 (S/T – 2.1),
can be re-parameterized as an ordinary logit model, with -8 + 4S/T. But uBOP
interprets Asch’s data as a threshold-matching process with the threshold at just
over two sources, whereas the logit model has little apparent social-psychological
content.
MacCoun (2012, p. 371) demonstrated that, indeed, the BOP models provide
a fairly good fit to even randomly generated monotone series. But this would
be true for all the influence models; the fit to real data is notably better than to
these “decoy” data sets, and most importantly, the estimated BOP coefficients
have values that imply we are fitting data generated by a random non-threshold
process.

Parameter Space: Our Next Frontier?

A great strength of parameterization is that it allows us to begin mapping out
where different research studies and paradigms fit in “parameter space.” For
example, in physics, basic mechanical laws tell us how much force is required to
move an object in contact with other objects, but only if we know the friction
coefficients of each of the objects. Originally, these were free parameters. But
with repeated application, these coefficients became known properties of con-
crete, wood, glass, water, ice, etc. And in time, our developing maps of parameter
space can inform us where to collect data and what we should expect to see if
our theories are right—as with the Table of Elements in chemistry, or the way
cosmologists used a mapping of physical parameters to translate supernova obser-
vations into an inference that the universe is expanding at an accelerating rate.
After such lofty examples, Figure 12.6 offers a much humbler example, plotting
bBOP coefficients from various data sets involving either criminal jury delibera-
tion or else “brainteaser”-type intellectual problems (MacCoun, 2014). Because
the direction of an asymmetric threshold is a matter of convention (a high thresh-
old for guilt is a low threshold for innocence), we can convert the threshold into
an absolute index of asymmetry, forming the horizontal axis. Because the clarity
parameter is quite variable, we can plot it in base-10 logs, forming a vertical axis.
It turns out (MacCoun, 2012) that many of the most influential Social Decision
Scheme D matrices are actually points in this parameter space: Proportionality,
Simple Majority, Two-Thirds Majority, Truth Wins, and Truth-Supported
Wins. Three of these are plotted as “landmarks” on Figure 12.3’s map. The map
makes readily apparent that the criminal jury studies occur on a different “con-
tinent” than the intellective task studies. The intellective tasks produce greater
asymmetry, as we would expect when there is a shared conceptual scheme that
facilitates the demonstration of a correct solution. But the criminal jury process is
 Models of Influence and Collective Behavior  277

FIGURE 12.6  “ Map” of the bBOP parameter space, showing two types of group
decision tasks as distinct “continents” Adapted from MacCoun (2014).

also asymmetric, and it also has greater clarity—the groups all seem to recognize
that a sufficiently sizeable amount of disagreement implies that they should acquit
the defendant (MacCoun & Kerr, 1988).
Developing maps of parameter space can also aid in the design of new experi-
ments, identifying strong tests, but also helping us distinguish a theory’s “terra
firma” from regions outside its boundaries of validity or utility. For example:
What factors promote threshold clarity? Do homogenous groups show greater
clarity than mixed groups? Do socialization and training in a conceptual system
promote threshold clarity and correlated behavior, and if so are those effects lim-
ited to the shared domain?

Concluding Thought
Having opened the chapter with Feynman’s construction metaphor, I apologize
for mixing it with a cartography metaphor. But my fantasy is that the social
psychology textbooks a decade or two from now might include both blueprints
for constructing models and an atlas of colorful maps of parameter spaces for
various phenomena. The blueprints and maps would be so familiar that any two
social psychologists would nod in agreement when a third says, “Well, to a first
approximation. . . .” In such a discipline, social psychologists would be cautious
about offering worldly expert advice for models that haven’t been built and areas
that are not well mapped out. We would have a clearer sense of the frontier of
our knowledge, of uncharted but potentially promising territories, and the kinds
of vehicles needed to “get there from here.”
278  Robert J. MacCoun

Notes
1 https://en.wikiquote.org/wiki/Talk:Richard_Feynman.
2 Bass (1969, p. 216) made this assumption in formulating his influential model of the
diffusion of innovations, stating that the probability of an initial product purchase “is a
linear function of the number of previous buyers.” But Bass also included an intercept
term to allow for the probability that the purchaser is an innovator who makes the pur-
chase independently of the number of other purchasers. On this latter point, see Boyd
and Richerson’s (1985) model of cultural evolution and discussion in MacCoun (2015).
3 For example, see U. Wilensky’s NetLogo, available from http://ccl.northwestern.edu/
netlogo/.

References
Abrams, R. (1980). Foundations of political analysis: An introduction to the theory of collective
choice. New York: Columbia University Press.
Allport, F. H. (1920). The influence of the group upon association and thought. Journal of
Experimental Psychology, 3, 159–182.
Asch, S. E. (1956). Studies of independence and conformity: A minority of one against a
unanimous majority. Psychological Monographs, 70(whole no. 416).
Bass, F. (1969). A new product growth model for consumer durables. Management Science,
15, 215–227.
Bonacich, P. (2001). The evolution of exchange networks: A simulation study. Journal of
Social Structure, 2(5). Retrieved from www.cmu.edu/joss/content/articles/volume2/
Bonacich.html
Boyd, R., & Richerson, P. J. (1985). Culture and the evolutionary process. Chicago, IL:
University of Chicago Press.
Boyer, P., Firat, R., & Van Leeuwen, F. (2015). Safety, threat, and stress in intergroup
relations: A coalitional index model. Perspectives on Psychological Science, 10, 434–450.
Brewer, M. B. (1991). The social self: On being the same and different at the same time.
Personality and Social Psychology Bulletin, 17, 475–482.
Brooks, G. P., & Johnson, R. W. (1978). Floyd Allport and the master problem in social
psychology. Psychological Reports, 42, 295–308.
Cialdini, R. B., Reno, R. R., & Kallgren, C. A. (1990). A focus theory of normative
conduct: Recycling the concept of norms to reduce littering in public places. Journal of
Personality & Social Psychology, 58, 1015–1026.
Condorcet, M. (1785). Essay on the application of analysis to the probability of plurality deci-
sions. In K. M. Baker (Ed. and Trans.), (1994). Condorcet: Selected writings (pp. 131–138).
Brookfield, VT: Edward Elgar.
Daley, D. J., & Gani, J. (1999). Epidemic modelling: An introduction. Cambridge: Cambridge
University Press.
Davis, J. H. (1973). Group decision and social interaction: A theory of social decision
schemes. Psychological Review, 80, 97–125.
Davis, J. H., & Witte, E. H. (Eds.) (1996). Understanding group behavior: Consensual action by
small groups (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum.
Davis, J. H., Zarnoth, P., Hulbert, L., Chen, X. P., Parks, C., & Nam, K. (1997). The
committee charge, framing interpersonal agreement, and consensus models of group
quantitative judgment. Organizational Behavior and Human Decision Processes, 72, 137–157.
Dawes, R. A. (1989). Statistical criteria for establishing a truly false consensus effect. Journal
of Experimental Social Psychology, 25, 1–17.
 Models of Influence and Collective Behavior  279

Deutsch, M., & Gerard, H. B. (1955). A study of normative and informational social influ-
ences upon individual judgment. Journal of Abnormal and Social Psychology, 51, 629–636.
French, J. P. R., Jr., & Raven, B. (1960). The bases of social power. In D. Cartwright &
A. Zander (Eds.), Group dynamics (pp. 607–623). New York: Harper & Row.
Galton, F. (1907). Vox populi. Nature, 75, 450–451.
Gardner, M. (1970). Mathematical games: The fantastic combinations of John Conway’s
new solitaire game “Life.” Scientific American, 223, 120–123.
Graesser, C. C. (1978). A social averaging theorem for group decision making. Dissertation
Abstracts International, 38(11-B), 5647.
Hinsz, V. B. (1999). Group decision making with responses of a quantitative nature: The
theory of social decision schemes for quantities. Organizational Behavior and Human
Decision Processes, 80, 28–49.
Isenberg, D. J. (1986). Group polarization: A critical review and meta-analysis. Journal of
Personality and Social Psychology, 50, 1141–1151.
Kelman, H. C. (1958). Compliance, identification and internalization: Three processes of
attitude change. Journal of Conflict Resolution, 2, 51–60.
Kerr, N. L., & MacCoun, R. J. (1985). The effects of jury size and polling method on
the process and product of jury deliberation. Journal of Personality and Social Psychology,
48, 349–363.
Kerr, N. L., & MacCoun, R. J. (2012). Is the leniency asymmetry really dead?
Misinterpreting asymmetry effects in criminal jury deliberation. Group Processes and
Intergroup Relations, 15, 585–602.
Kerr, N. L., MacCoun, R. J., & Kramer, G. (1996). Bias in judgment: Comparing
individuals and groups. Psychological Review, 103, 687–719.
Kerr, N. L., Stasser, G., & Davis, J. H. (1979). Model testing, model fitting, and social
decision schemes. Organizational Behavior and Human Decision Processes, 23, 399–410.
Latané, B. (1981). The psychology of social impact. American Psychologist, 36, 343–356.
Laughlin, P. R. (2011). Group problem solving. Princeton, NJ: Princeton University Press.
Leary, M. R. (2005). Sociometer theory and the pursuit of relational value: Getting to the
root of self-esteem. European Review of Social Psychology, 16, 75–111.
Lewis, D. (1973). Counterfactuals. Oxford: Blackwell.
Lorenz, J., Rahut, H., Schweitzer, F., & Helbing, D. (2011). How social influence can
undermine the wisdom of crowd effect. Proceedings of the National Academy of Science,
108, 9020–9025.
Lorge, I., & Solomon, H. (1955). Two models of group behavior in the solution of
Eureka-type problems. Psychometrika, 20, 139–148.
MacCoun, R. J. (2005). Media reporting of jury verdicts: Is the tail (of the distribution)
wagging the dog? DePaul Law Review, 55, 539–562.
MacCoun, R. J. (2012). The burden of social proof: Shared thresholds and social
influence. Psychological Review, 119, 345–372.
MacCoun, R. J. (2014). Alternative maps for the world of collective behavior. Invited
comment. Behavioral and Brain Sciences, 37, 88–90.
MacCoun, R. J. (2015). Balancing evidence and norms in cultural evolution. Organizational
Behavior and Human Decision Processes, 129, 93–104.
MacCoun, R. J., Cook, P., Muschkin, C., & Vigdor, J. (2008). Distinguishing spurious
and real norm effects: Evidence from artificial societies, small-group experiments, and
real schoolyards. Review of Law & Economics, 4(3), 695–714.
MacCoun, R. J., & Kerr, N. L. (1988). Asymmetric influence in mock jury deliberation:
Jurors’ bias for leniency. Journal of Personality and Social Psychology, 54, 21–33.
280  Robert J. MacCoun

Mahajan, V., Muller, E., & Bass, F. (1995). Diffusion of new products: Empirical gener-
alizations and managerial uses. Management Science, 14, G79–G88.
Milgram, S., Bickman, L., & Berkowitz, L. (1969). Note on the drawing power of crowds
of different size. Journal of Personality and Social Psychology, 13, 79–82.
Mullen, B. (1983). Operationalizing the effect of the group on the individual: A self-
attention perspective. Journal of Experimental Social Psychology, 19, 295–322.
Newman, M., Barabási, A. L., & Watts, D. J. (2006). The structure and dynamics of networks.
Princeton University Press.
Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitudes to public opinion: A
dynamic theory of social opinion. Psychological Review, 97, 362–376.
Parks, C. D., & Kerr, N. L. (1999). Twenty-five years of social decision scheme theory.
Organizational Behavior and Human Decision Processes, 80, 1–2.
Rahmandad, H., & Sterman, J. (2008). Heterogeneity and network structure in the
dynamics of diffusion: Comparing agent-based and differential equation models.
Management Science, 54, 998–1014.
Railsback, S. F., & Grimm, V. (2011). Agent-based and individual-based modeling: A practical
introduction. Princeton, NJ: Princeton University Press.
Roberts, S., & Pashler, H. (2000). How persuasive is a good fit? A comment on theory
testing. Psychological Review, 107, 358–367.
Rosenwein, R. E., & Campbell, D. T. (1992). Mobilization to achieve collective action
and democratic majority/plurality amplification. Journal of Social Issues, 48, 125–138.
Schelling, T. C. (1978). Micromotives and macrobehavior. New York: W. W. Norton.
Schkade, D. A., Sunstein, C. R., & Kahneman, D. (2000). Deliberating about dollars: The
severity shift. Columbia Law Review, 100, 1139–1176.
Smaldino, P. E., Janssen, M. A., Hillis, V., & Bednar, J. (2015). Adoption as a social marker:
The diffusion of products in a multigroup environment. Retrieved from http://arxiv.org/
abs/1507.04775
Smaldino, P. E., Calanchini, J., & Pickett, C. L. (2015). Theory development with agent-
based models. Organizational Psychology Review, 5, 300–317.
Stasser, G., Kerr, N. L., & Davis, J. H. (1989). Influence processes and consensus models
in decision-making groups. In P. Paulus (Ed.), Psychology of group influence (2nd ed.,
pp. 279–326). Hillsdale, NJ: Lawrence Erlbaum.
Stevens, S. S. (1957). On the psychophysical law. Psychological Review, 64, 153–181.
Surowiecki, J. (2005). The wisdom of crowds. New York: Anchor Books.
Verhulst, P.-F. (1838). Notice sur la loi que la population suit dans son accroissement.
Correspondance Mathématique et Physique, 10, 113–121.
Wilder, D. A. (1977). Perception of groups, size of opposition, and social influence.
Journal of Experimental Social Psychology, 13, 253–268.
Wilson, E. B., & Worcester, J. (1944). The law of mass action in epidemiology. Proceedings
of the National Academy of Science, 31, 24–34.
13
MODELING CULTURAL
DYNAMICS
Yoshihisa Kashima, Michael Kirley,
Alexander Stivala, and Garry Robins1

Culture is an enigma in contemporary social psychology. Despite its brief prominence

in the 1960s (e.g., Triandis, 1964), social psychological curiosity about culture
waned over the heyday of social cognition in the 1970s and 1980s, only to regain
its prominence in the 1990s with the publication of Triandis (1989) and Markus
and Kitayama (1991) in Psychological Review. Over the last two and a half decades,
cultural comparative approaches to culture and psychology have accumulated
a large body of literature documenting cultural differences around the globe,
though primarily comparing East Asian- and Western European-based cultural
groups, and have brought about a number of insights into cultural diversity across
humanity. However, this literature has been largely silent on the question of the
dynamics of culture—namely, the stability and change of culture, and the mecha-
nisms that drive the trajectories of cultural change over time (see, e.g., Kashima &
Gelfand, 2012). Cultural dynamics is concerned with these questions—how do
individuals’ thoughts, feelings, and actions in interaction with others in particular
contexts generate the movements of culture—its formation, maintenance, and
transformation over time (e.g., Kashima, 2000).
At the heart of this lies the fundamental question for social psychology. How
do human individuals interact with the actual and imagined others? In so doing,
how do we influence each other? Ever since Sherif and Asch, social influence
has been a core concern of social psychology; culture is at one level “what social
influence influences” (Axelrod, 1997, p. 207). From the concept of human rights
to the landing on the Moon, and from genocides in intractable intergroup con-
flicts to anthropogenic climate change, culture enables the astonishing human
adaptiveness and achievements in society and in nature, while at times exhibiting
surprising maladaptiveness with pathological and tragic consequences in both.
With all its glories and failings, the dynamic complexity of human sociality is
282  Kashima, Kirley, Stivala, and Robins

fundamental to human culture. Culture enables human sociality; human sociality

constitutes culture. Given their interdependency, culture must be an integral part
of human social psychology.
Computational modeling—especially agent-based modeling (e.g., Railsback &
Grimm, 2012)—is a useful methodological tool in research on cultural dynamics
for the development of a coherent theory, derivation of testable hypotheses,
understanding of the past and present, as well as forecasting of future possibili-
ties. Because of the number and heterogeneity of actors, the complexity of social
interactions, and different timescales involved (social interaction in situ may
change in a short timespan, but institutions remain stable over a longer period of
time), some formal representations (e.g., difference and differential equations) are
often useful in theorizing and describing cultural dynamics. However, these very
characteristics often make the type of formal analytical methods necessary for the
modeling of cultural dynamics outside the training of social psychologists, and
difficult, if not impossible, to use (e.g., A. Nowak & Vallacher, 1998). Agent-
based modeling approaches, however, enable social psychologists to explicate
their assumptions about culture, construct explicit models of cultural dynamics,
and explore their implications in a principled fashion. This chapter is designed to
provide a broad and selective introduction to diverse literatures on computational
approaches to cultural dynamics.

Culture and Its Dynamics

Culture is an essentially contested concept—depending on one’s theoretical per-
spective, it can be defined in a multitude of ways. In the current approach, we
define culture as a set of non-genetically transmissible information that is commonly avail-
able, accessible, and applicable in a human group. Cultural information typically takes
the form of ideas (e.g., liberty, equality, and fraternity) or practices (e.g., how to
deliberate, vote, and determine an outcome). Availability means information is
there to be learned if an individual wants to; accessibility implies that informa-
tion has been learned and can be brought to mind; and applicability suggests that
information can inform the individual’s action. Several aspects of this definition
are worth highlighting:

1. Culture is information. Some theorists (e.g., Herskovits, 1948; Triandis,

1994) have defined culture as the human-made part of the environment.
We acknowledge that the human-made part of the environment (largely
consisting of artefacts) carries or embodies cultural information, and cultural
information enables humans to construct it; however, we distinguish what
represents information from information itself.
2. Culture is not a group. Culture is often equated with a group or a collection
of human individuals that continue to exist over a period of time where
a group can vary in size from relatively small communities to the whole
 Modeling Cultural Dynamics  283

of humanity. We acknowledge that a culture is definable when a group is

defined; however, we regard culture as information carried by members of a
group or embodied in the artefacts constructed by them, and as such, cultural
information can continue to exist and inform human action even after the
group ceases to exist (e.g., Classic Egyptian, Mesopotamian, Greek, Roman,
Indian, Chinese cultures).
3. Cultural information differs from genetic information in mode of transmission. Culture
consists of information that is socially transmitted, rather than genetically
transmitted, from one person to another. Social transmission of cultural infor-
mation can occur in various forms: formal schooling and explicit instruction,
co-participation in joint activities, with or without language or other symbolic
means. Although genetic transmission occurs only from a biological parent to
a biological child, cultural transmission occurs within and between genera-
tions, from older to younger, but also from younger to older, generations.

In this perspective, cultural dynamics is concerned with the trajectory of persistence

and change of a set of cultural information and associated characteristics (e.g., struc-
ture, organization, and frequency and spatial distributions) in a group over time, and
the social psychological mechanisms that drive these temporal dynamics.

Sources of Cultural Dynamics

Critical aspects of cultural dynamics are the generation and retention of information
in the culture of a group. That is, when new cultural information is generated,
it is added to the set; when it is retained over time (e.g., across generations), it
is kept in the group’s culture and potentially used in the future. The mechanism
that drives the generation and retention of cultural information is the social trans-
mission of cultural information between people (for recent research on this topic, see
Kashima, 2016a). This perspective is broadly in line with a host of theories of
cultural evolution (e.g., Boyd & Richerson, 1985; Campbell, 1975; Cavalli-Sforza &
Feldman, 1981; Dawkins, 1976; Richerson & Boyd, 2005; Sperber, 1996).
Kashima, Peters, and Whelan (2008) called this metatheory neo-diffusionism.
In this perspective, there are at least four sources of cultural dynamics
(Kashima, 2014). New information may be generated by invention within the
group (e.g., Campbell, 1960; Simonton, 2010) or by importation from another
group (e.g., Bartlett, 1923, 1932); once cultural information is present, it may
be retained by selection—selecting it in for its benefit, or selecting it out for its
cost (e.g., Boyd & Richerson, 1985; Cavalli-Sforza & Feldman, 1981)—or by
drift, non-selective stochastic transmission and retention (e.g., Bentley, Hahn, &
Shennan, 2004; Hahn & Bentley, 2003). The four sources of cultural dynamics
differ in whether new information is added to a given culture, and whether adap-
tation is involved (i.e., benefit enhancement or cost reduction; see Table 13.1).
Presumably, information is imported or selected during the process of adaptation;
284  Kashima, Kirley, Stivala, and Robins

TABLE 13.1  Sources of cultural dynamics.

Random Adaptive

Addition Invention Importation

Retention Drift Selection

Note: Each source differs in terms of whether it pertains to the addition of new cultural information
or the retention (or removal) of cultural information within a population, and whether it occurs ran-
domly or due to adaptive processes. We chose the term “random” here, but it may be better to say
it is neutral (i.e., non-adaptive) or blind (i.e., without foresight). The characterization of invention as
blind is due to Campbell (1960).

it may be invented or may drift due to stochastic processes that are neutral to
adaptation (i.e., do not necessarily result in adaptation or maladaptation).
The sources of cultural dynamics presented here provide a framework for
our discussions about diverse computational approaches to cultural dynamics. In
this chapter, we begin our coverage with social psychological models of cultural
dynamics (Abelson & Bernstein, 1963; A. Nowak, Szamrej, & Latané, 1990), and
then move to two prominent approaches to cultural dynamics—Axelrod’s (1997)
model of cultural dissemination, and evolutionary game theoretic approaches
to evolution of cooperation. We show that these approaches focus on com-
plementary aspects of cultural dynamics, and that each has unique strengths in
dealing with some aspects, but not others. On one hand, the Axelrod model
has been used to explore the dynamics deriving from transmissions of cultural
information and the role of drift and to some extent of importation; however, it
does not address invention, or most importantly, selection. On the other hand,
evolutionary game theoretic approaches have a unique strength in examining
the importance of the selection process in cultural (and genetic) evolution. In
this chapter, we will thus discuss how the existing approaches complement each
other, and also point to the gap in the existing theory—neither has addressed the
process of invention (other than as a random process of mutation).
We hasten to add that there are many analytical (for a systematic exposition,
see McElreath & Boyd, 2007) and simulation (e.g., Carley, 1991; Hutchins, 1995)
models of culture and cultural dynamics; however, they are too extensive and
varied to be covered in the present chapter. It suffices to say that these attempts
model some of the above processes at the micro-level of cognition and commu-
nication, and explore their consequences at macro-levels of organization, society,
and culture. Nevertheless, there is a tradition in social psychology to attempt to
describe the trajectory of cultural change with computer simulations of individu-
als’ micro-level interactions. We start with this social psychological literature.

Social Psychology of Cultural Dynamics

The social psychological tradition of cultural dynamics research began with pub-
lic opinion dynamics. Note that in the current view of culture, a cultural element
 Modeling Cultural Dynamics  285

may be a public opinion. Given a certain proposition (i.e., cultural information)

such as “Climate change is occurring, and it is largely human-caused,” individu-
als in a group (e.g., Americans) can have a variety of opinions, and the frequency
distribution of opinions on a binary variable such as pro versus con, or a continu-
ous bipolar dimension that indicates a degree of agreement or disagreement with
the proposition can be regarded as an aspect of this group’s culture. Obviously,
the distribution of opinions about public issues such as climate change dynami-
cally changes over time (e.g., Brulle, Carmichael, & Jenkins, 2012). Describing
the temporal trajectory of public opinions and examining the social psychological
mechanisms underlying such opinion changes are both integral aspects of research
on cultural dynamics (e.g., Kashima, 2014).
Abelson and Bernstein (1963) were the first to examine such public opinion
dynamics in computer simulations. In their model, agents had opinions about
fluoridation of water supplies, interacted with other agents as a function of their
relationships with them, conversed with them about fluoridation as a function of
their interest in the issue, and updated their opinions and their relationships with
their interaction partners; these steps occurred iteratively over time. Although most
of the then existing models of attitude formation and change predicted a uniformly
pro or con distribution of opinions in the community (Abelson, 1964), Abelson and
Bernstein’s simulation model showed a polarization of opinions—opinions would
become more extreme—so that those agents that held initially pro opinions are
likely to hold more pro opinions, and those that held con opinions become more
con in opposition as the controversy runs its course. In other words, public opin-
ions could polarize, and the polarization might persist and exacerbate—a finding
that accorded much better with the empirical reality of controversies in realpolitik.
Although Abelson and Bernstein (1963) showed that the emergence of consen-
sus is not a universal consequence of simulation models, thus laying a foundation
for simulating cultural dynamics, their model did not have an explicit spatial
arrangement. It was A. Nowak and colleagues (1990) who explicitly defined a
space (depending on its interpretation, it can be construed as a physical or social
space). They suggested that agents can be located in a grid-like structure (i.e., lattice),
where they would interact with other agents in their spatial neighborhoods and
exert social influences on each other as specified by Latané’s (1981) social impact
theory. More importantly, because their simulation model had an explicit defini-
tion of space, they could also model the effect of spatial distance—it is known
that opinions of others who are psychologically close (Latané’s immediacy; for
empirical evidence, see also Latané, Liu, Nowak, Bonevento, & Zheng, 1995)
have a greater persuasiveness. Together with the nonlinearity in attitude change
processes assumed in the model (see Lewenstein, Nowak, & Latané, 1992), their
model produced spatial clustering of public opinions—even if there is a strong
majority (i.e., a large proportion of agents have pro or con opinions), an opinion
minority could survive if its members are spatially close to each other and cluster
together in space (for empirical evidence, see Latané & L’Herrou, 1996).
286  Kashima, Kirley, Stivala, and Robins

These developments have enabled modeling of a single cultural element (e.g.,

opinion), and examination of the consequences of micro-level communication
dynamics among agents on the macro-level trajectory of the opinion distribution
in the group. This line of investigation can be extended in two different ways:
multidimensionality and adaptiveness of culture. We will examine them in turn.

Multidimensionality of Culture: The Axelrod

Model of Cultural Dissemination
One of the characteristics of the foregoing model is that it models only a single cul-
tural element. However, culture is obviously not a single element, but typically
consists of multiple elements. Furthermore, cultural elements are not inde-
pendent of each other, but are considered to be “patterned” (e.g., Kroeber &
Kluckhohn, 1952; Triandis, 1996) or to form a configuration. How can a pattern
of cultural elements emerge in computer simulations? Latané (1996) suggested
one potential mechanism for the emergence of a cultural pattern. That is, when
multiple cultural elements exist, but when each cultural element is influenced by
the same mechanism of social influence such as dynamic social impact theory,
even in the absence of any inherent relations between them, there will be a spatial
clustering of each cultural element. However, by virtue of the spatial clustering
of these cultural elements, the elements that happen to occur within a given
spatial cluster become correlated just by happenstance. A. Nowak, de Raad, and
Borkowski (2012) showed this to be the case in their computer simulation (see
also DellaPosta, Shi, & Macy, 2015; for a review, see Harton & Bullock, 2007).
This line of investigation was further extended by the Axelrod model of
cultural dissemination (Axelrod, 1997). In Axelrod’s original model, cultural
information is assumed to be represented as a set of multiple attributes (e.g.,
religion, language, taste) which each agent possesses. It is represented as a culture
vector of F features, each of which can take q possible values (traits). For instance,
one feature can be opinion about climate change, and then each agent can have
one of several different types of opinions (e.g., “climate change doesn’t exist,”
“climate change is happening, but not anthropogenic”). Another feature may
be opinion about economic development (e.g., “market economy is the best
form of resource distribution,” “government intervention is sometimes impor-
tant to regulate the economic process”). This way, there may be a number of
cultural elements. When two agents interact, one agent’s cultural information is
transmitted to the other, and as a result changes the latter’s culture vector. This
transmission process, however, is constrained, so that the model assumes that
cultural transmission occurs only when the agents already share some cultural
information. In other words, if two agents have no cultural information in com-
mon, they cannot interact or transmit cultural information. This assumption can
be interpreted as suggesting that (1) similar people tend to interact with each
other (i.e., homophily), (2) people use their shared culture to communicate new
 Modeling Cultural Dynamics  287

cultural information, or both (1) and (2). Note that this process implies that cul-
tural similarity begets cultural similarity—that is, culturally similar agents become
more culturally similar to each other. Does this mean that cultural differences
eventually disappear? Even with multidimensional cultures, Axelrod (1997)
showed that cultural diversity can persist under some circumstances.

Basic Model
The dynamics of the model are as follows. As in A. Nowak and colleagues
(1990), each agent is placed on a lattice site (every site on the lattice is occu-
pied) with four neighbors (i.e., above, below, right, and left; Von Neumann
neighborhood), and only neighboring agents can interact. Initially, each agent
is assigned a culture vector uniformly at random (that is, each of the F traits is
given one of the q possible values at random). Thereafter, at each step, a ran-
dom agent is chosen as the focal agent, and one of its neighbors is also chosen
at random. The probability of the two agents interacting and influencing each
other is determined by their cultural similarity, which is defined as the number
of matching features between their culture vectors—that is, the number of cor-
responding features in which the two agents have the same trait value. When
they interact, a random feature on which the two agents’ values differ (if there
is one) on the focal agent is changed to the value of the corresponding feature
on the other agent.
This process is repeated until no more changes are possible, because any two
neighboring agents either have identical culture vectors, or completely different
culture vectors (no features in common, so they cannot interact). At this point,
when no more change is possible, which is known as the absorbing state, there are
two possible results. There is either a monocultural state, in which all agents have
the same culture vector, or a multicultural state, in which any two agents in a
contiguous region have the same culture, but neighboring agents on the region
boundaries have completely different cultures (they can no longer interact as they
have no features in common).
Figure 13.1 shows an example run of an Axelrod model using a visualization
similar to that originally presented in Axelrod (1997), where the similarity of
adjacent sites on the lattice is represented by the darkness of the line between
the two sites. In the initial conditions (Figure 13.1a), adjacent sites are mostly
completely different, although some have some small degree of similarity. In the
intermediate stages (Figures 13.1b and 13.1c), cultural regions have started to
form, and the boundaries between some cultural regions are colored light gray,
indicating they are quite similar. In the absorbing state (Figure 13.1d), which is
a multicultural absorbing state which was reached after approximately 58,000
iterations, there are five cultural regions, the largest of which covers 75% of the
lattice. Note that in this implementation of the model (Weaver, 2010), imple-
mented in NetLogo,2 the lattice “wraps around” so that all sites on the lattice have
288  Kashima, Kirley, Stivala, and Robins

(a) Initial conditions (b) After 25,000 iterations

(c) After 30,000 iterations (d) At absorbing state

FIGURE 13.1  Example of an Axelrod model run with number of features F = 5,

number of traits q = 15, on a 10 × 10 lattice. The darkness of the lines
between sites indicates the cultural similarity (white is identical, black
is completely different).

exactly four neighbors, with no special cases on the edges of the lattice, unlike
the original model specified by Axelrod (1997).
Whether a monocultural or multicultural absorbing state is reached—and if
a multicultural absorbing state is reached, how many cultural regions exist—are
influenced by a number of factors. Increasing q, the number of possible traits of
each feature, results in an increased number of cultural regions, but in a nonlinear
manner. All else being equal, when q is below a critical value, a monocultural
absorbing state is the result. Increasing q above the critical value results in a multi­
cultural absorbing state, with an increasing number of regions as q increases.
Increasing the size of the neighborhood of agents with which the focal agent can
interact decreases the number of cultural regions. Surprisingly, however, increas-
ing the number of features, F, decreases the number of cultural regions, as does
increasing the size of the lattice (which in the original Axelrod model also increases
the number of agents, each lattice site being occupied) after a certain point.
So even this simple model demonstrates complex behavior, showing how
homophily and social influence alone, in which two agents can only ever become
more similar, does not necessarily result in a monoculture. In the language of sta-
tistical physics, there is a nonequilibrium phase transition at the critical value of q,
separating the ordered (monocultural) phase from the disordered (multicultural)
phase (Castellano, Marsili, & Vespignani, 2000).
 Modeling Cultural Dynamics  289

Extensions of the Axelrod Model

The original Axelrod model describes cultural transmission in dyadic social inter-
actions in a grid-like social space with culture vectors taken from a fixed cultural
space (F × q matrix). However, it has been extended to investigate the impli-
cations of non-dyadic social interactions in more complex social and cultural
structures.

Non-Dyadic Social Interactions

The effect of mass media or other external cultural influences can be modeled
as an external field acting on the culture vectors in the Axelrod model, caus-
ing a feature to become more similar to the external cultural message vector
with a certain probability (Gandica, Charmell, Villegas-Febres, & Bonalde, 2011;
González-Avella, Cosenza, Eguíluz, & San Miguel, 2010; González-Avella,
Cosenza, & Tucci, 2005; González-Avella et al., 2006). Such an external cultural
influence can, counterintuitively, lead to multicultural states when they would
otherwise not have existed (González-Avella et al., 2005; González-Avella et al.,
2006). A different, earlier, model of the mass media effect was a “generalized
other,” a hypothetical extra neighbor agent with the most preferred value of
each feature (Shibanai, Yasuno, & Ishiguro, 2001). In this model, rather than
mass media being an external cultural influence, it is viewed as a social norm
constructed from the culture vectors of all agents in the model.
The concept of multilateral social influence (i.e., more than two agents interact-
ing) was explicitly introduced by Parisi, Cecconi, and Natale (2003), although
it was implicit in the “generalized other” of Shibanai et al. (2001). Parisi et al.
(2003) show that, in the absence of homophily, stable multicultural states (neither
monocultural nor anomic) can be sustained even in the presence of noise when
social influence is multilateral. Kuperman (2006) extends the Axelrod model by
having an agent adopt a neighbor’s trait only when it will then become similar to
more of its neighbors. Extensions of the Axelrod model which incorporate multi-
lateral social influence are thoroughly examined by Flache and Macy (2011), who
show that such stable multicultural states can be sustained for wide ranges of noise
levels even when homophily is assumed, as long as social influence is multilateral
and not just dyadic. Rodríguez and Moreno (2010) incorporate both multilateral
social influence and a mass media effect, again finding that multilateral social
influence can maintain diversity in the presence of noise.

Social and Cultural Spaces

Rather than the standard lattice structure, cultural dynamics on more complex
social networks have been investigated. One of them is a small-world network,
which consists of densely connected clusters with sparse links between the clus-
ters. Klemm, Eguíluz, Toral, and San Miguel (2003) find that the critical value of
290  Kashima, Kirley, Stivala, and Robins

q increases with increasing density of long-distance connections. Guerra, Poncela,

Gómez-Gardeñes, Latora, and Moreno (2010) explore the cultural dynamics on
scale-free graphs (i.e., degree distribution follows a power law), which can be
constructed by a process of preferential attachment: each new node is added to
the graph by choosing an existing node to which to connect it with probability
proportional to the existing node’s degree (Barabási & Albert, 1999). Although
these models have static social networks on which cultural dynamics unfold,
more recent models have the social network co-evolve with culture (Centola,
González-Avella, Eguíluz, & San Miguel, 2007; Vazquez, González-Avella,
Eguíluz, & San Miguel, 2007), or allow for empty sites on the lattice and migra-
tion of agents (Gracia-Lázaro, Lafuerza, Floría, & Moreno, 2009), or both (Pfau,
Kirley, & Kashima, 2013).
The cultural space in the original Axelrod model is simple and unstructured—a
F × q matrix. Most of the subsequent models used a uniform random distribution
of culture vectors as the initial conditions. As noted by Pace and Prado (2014),
however, our understanding of the Axelrod model depends on this assumption,
and different initial conditions can lead to very different behavior. An important
exception to this general assumption is a model in which empirical opinion data
is used as the initial culture vectors, which are found to have an approximately
ultrametric distribution, leading to increased diversity at the absorbing state com-
pared to random initial conditions (Valori, Picciolo, Allansdottir, & Garlaschelli,
2012). The effects of ultrametricity and other statistical properties of the initial
culture vectors were further explored in Stivala, Robins, Kashima, and Kirley
(2014). They found that when the agents have culture vectors generated by add-
ing random noises to prototype culture vectors, they showed cultural dynamics
similar to those generated by real opinion data, suggesting that the structure of
cultural space needs to be further explored.
Nonetheless, these models still assume that different cultural features can
change independently from each other.To the best of our knowledge, no research
has investigated the cultural dynamics when cultural representations are truly
configural, so that a change in one cultural element may precipitate a change in
other elements.

Sources of Cultural Dynamics in the Axelrod Model

Some extensions of the Axelrod model addressed the cultural dynamics stemming
from one of the sources of cultural dynamics: cultural drift (e.g., De Sanctis & Galla,
2009; Grauwin & Jensen, 2012; Klemm et al., 2003; Klemm, Eguíluz, Toral, &
San Miguel, 2005; Parisi et al., 2003). In the Axelrod model, random perturbations
can be added in the culture vectors as a form of drift. When a drift rate is small,
the monocultural absorbing state becomes more likely; however, for large noise rates,
a state of “anomie” is more likely to ensue, in which stable cultural regions fail to
form (Centola et  al., 2007; Flache & Macy, 2011; Mäs, Flache, & Helbing, 2010).
 Modeling Cultural Dynamics  291

A high drift rate in this context implies that cultural transmission is noisy or
inaccurate, and the findings here imply that inaccurate cultural transmissions are
likely to result in a monocultural state. Hence, the model suggests that cultural
information needs to be transmitted fairly accurately in order to maintain cultural
diversity.
Arguably, the investigation of external cultural influences reviewed above taps
a source of cultural dynamics akin to importation, although further extensions
including social identity and the like may be possible (e.g., Yamamoto, 2015).
However, this research tradition has not investigated other sources of cultural
dynamics such as cultural invention. That is to say, as noted by Pace and Prado
(2014), Axelrod models (without noise) lack a method for the creation of nov-
elty, since traits can only ever become extinct, and cannot be created. A possible
major extension of the Axelrod model would be a mechanism to introduce new
features or new traits, and compare the behavior of trait change compared to the
creation of a whole new trait or feature. Furthermore, an obvious gap in this lit-
erature is an investigation of selection processes in which cultural information is
selected in or out due to its adaptiveness or a lack thereof. A different tradition of
research—evolution of cooperation—has addressed selection as its central focus,
to which we now turn.

Adaptiveness: Evolution of Cooperation

In the models of opinion dynamics and cultural dissemination discussed in the
preceding section, agents interact with each other and transmit cultural informa-
tion as a matter of course regardless of the social and psychological consequences
of doing so. Nonetheless, as we noted earlier, one of the sources of cultural
dynamics is selection (Table 13.1). That is, the more adaptive cultural information
is (i.e., its retention, transmission, and use are less costly and bring about more
benefits at the psychological, social, and practical levels), the more likely it is to
be selected and retained in the group’s culture (e.g., Kashima, 2014; Kashima,
2016a, 2016b). However, the models of opinion dynamics and cultural dissemi-
nation do not take the consequences of cultural transmission into consideration
in describing cultural dynamics.
In contrast, the literature on the evolution of cooperation is the most devel-
oped research tradition that addresses selection as a source of cultural dynamics. To
be sure, there is a long tradition in social psychology which investigates collective
action in social dilemmas (e.g., Dawes, 1980; Kollock, 1998; Parks, Joireman, & Van
Lange, 2013; Van Lange, Joireman, Parks, & Van Dijk, 2013). A social dilemma
is defined as a situation in which individually beneficial actions produce collec-
tively costly consequences. However, research on the evolution of cooperation
addresses a question somewhat different from the typical experimental social psy-
chological question, “How do humans behave in social dilemmas?” Rather, it
can be construed as asking, “Under what circumstances can cooperation as a
292  Kashima, Kirley, Stivala, and Robins

cultural practice or a genetic characteristic remain in the population of agents?”

In so doing, this research tradition attempts to answer under what circumstances
can the cultural practice of cooperation be adaptive, and therefore selected in to
remain in human culture.
This section describes some of the key concepts and main findings to pro-
vide an introduction to this voluminous literature. In particular, we review the
standard game theoretic treatment of the Prisoner’s Dilemma (PD) as a prototypi-
cal game in evolution of cooperation, introduce an evolutionary game theoretic
approach, and describe some of the key findings in this literature.

Cooperation in Social Dilemma

The standard mathematical approach used to investigate the evolution of coop-
eration is based on game theory (Von Neumann & Morgenstern, 1953), whose
fundamental assumptions are those of rationality and strategic interactions. That
is, a rational player takes into account all the available information in the situation
(i.e., the game) and tries to maximize the expected payoff by identifying an optimal
sequence of choices of strategies (e.g., cooperation or defection). One of the best-
known examples of this approach is the classic Prisoner’s Dilemma game (Axelrod,
1984; Rapoport & Chammah, 1965), a type of symmetric two-player simultaneous
nonzero-sum game. Each of the two agents (typically called players), players A and B,
makes a choice once from two potential strategies: cooperation (C) or defection (D).
There are four possible outcomes for the game for each player, depending on the
combination of what each player decides to do (Table 13.2).
Typically, the outcomes are displayed in the form of a payoff matrix, as shown
in Table 13.3.

TABLE 13.2  Players A and B’s choices and outcomes for A.

A’s choice B’s choice A’s outcome

C C Reward of mutual cooperation (R)

C D Sucker’s payoff (S)
D C Temptation to defect (T)
D D Punishment for mutual defection (P)

Note: T > R > P > S.

TABLE 13.3  Payoff matrix for a two-person game.

B
C D
A C R S
D T P
 Modeling Cultural Dynamics  293

The Prisoner’s Dilemma payoff matrix is defined by the constraint T > R >
P > S. Looking at this situation from player A’s perspective, it is easy to see that
D always brings a better outcome than C. If B chooses C, the outcome of A
choosing D (T) is better than choosing C (R); if B chooses D, the outcome of
A choosing D (P) is better than choosing C (S). In other words, D dominates C.
Each player then choosing D is what is known as a Nash equilibrium—the state in
which unilaterally altering one’s behavior does not improve one’s payoff (Gintis,
2009; M. A. Nowak, 2006a). Assuming that the players behave so as to increase
benefits and reduce costs, this is an equilibrium, in that it is a stable state in which
neither player is willing to change his or her behavior unilaterally—the system
of interlocking behaviors would remain stable. Nevertheless, given the payoff
matrix, both players choosing D (outcome P) brings an outcome worse than their
both choosing C (outcome R). Therein lies a dilemma—a strategic choice that
brings a better outcome for each individual locks them into a mutually worse
outcome.
This situation can be generalized to the case in which a game involves more
than two players. One such game is the Public Goods game (e.g., Dawes, 1980;
Hardin, 1968; also see Camerer, 2003; Gintis, 2009). Here, each player can either
cooperate or defect, but the payoff matrix is generalized to the n person situation
where the payoffs for cooperators and defectors are a function of the numbers
of cooperators and defectors in the group. A player is always better off defecting
than cooperating, but payoff is higher if everyone cooperates than if everyone
defects. This is the essence of a social dilemma—an individually beneficial action
resulting in a collectively unpalatable outcome.

Iterated Games
So far, we have considered the social situation in which agents make a choice
only once (one-shot game). However, when agents play a game repeatedly (iterated
game), the strategy to defect does not necessarily generate the highest payoff. For
instance, imagine a Prisoner’s Dilemma game where players A and B repeat-
edly play a PD as shown in Table 13.2. If they decide to take turns, so that A
cooperates and B defects in the first round, A defects and B cooperates in the
second round, A cooperates and B defects in the third round, etc., every two
rounds both players cumulate the payoff S+T. If S+T > 2R, then the strategy
of taking turns to cooperate and defect can maximize their outcomes (so, for a
repeated PD, there is usually another constraint, S+T < 2R). Thus, in an Iterated
Prisoner’s Dilemma game, the strategy of always defecting is not necessarily the
best option.
More generally, if the number of rounds of the game is unknown (or at the
very least, the probability of playing additional rounds of the game is high), there
are a variety of conditional cooperation strategies whose outcomes are better than
the always-defect strategy in the long run. As we will see later, Trivers (1971)
294  Kashima, Kirley, Stivala, and Robins

was perhaps the first to describe these types of strategies as driving the evolution
of cooperation within a Prisoner’s Dilemma; Axelrod (1984) extended this line
of thinking. Put more broadly, the game theoretic formulation enables us to
examine the adaptiveness of these game playing strategies, whether the long-term
consequences of adopting cooperative strategies are beneficial or costly for the
players. The foregoing discussion amounts to saying that a game playing strategy
to cooperate is better for both players than a defective one under some circum-
stances, even if the social situation that dictates the outcome of their enactment
constitutes a difficult social dilemma to resolve.

Evolutionary Game Theory

Evolutionary game theory (e.g., Axelrod, 1984; Axelrod & Hamilton, 1981;
Maynard Smith, 1982; M. A. Nowak, 2006a) extends the foregoing idea of itera-
tive game plays by considering a population of players and their selections of
strategies. Some terminological clarification is in order here. Although it is the
standard practice in evolutionary game theory to call different ways of playing
a game “strategies,” this seems to us to be a reflection of the conceptual roots
of this research tradition, the fact that game theory was developed as a princi-
pled consideration of strategic moves when players are trying to outsmart each
other. There is an unfortunate connotation that these game playing strategies are
deliberate attempts to outperform each other. However, there is no reason why
they must involve such deliberative foresights. In order to avoid the deliberative
connotation, we will call strategies genetic traits, or genetic tendencies if they are
genetically transmitted, whereas, we will call them cultural traits, cultural practices,
or simply practices if such strategies are socially transmitted.
This terminological distinction should clarify what is being modeled. When
a strategy like cooperation and defection is construed as a genetically transmitted
trait, the resultant dynamics can be thought of as modeling biological evolution.
However, when a strategy is interpreted as a cultural practice—a type of cultural
information that describes a pattern of behavior—the resultant dynamics can be
regarded as modeling cultural evolution. In either case, the payoff an individual
receives corresponds to the notion of Darwinian fitness—average reproductive
success in biological evolution and an average level of rewards in some sense in
cultural evolution. The distribution of players’ strategies defines a population
state, which is formally equivalent to a notion of mixed strategy in game theory.
Thus, evolutionary game theory investigates the emergent population dynamics
and strategy distribution.
Now imagine that a large group of players play the Iterated Prisoner’s Dilemma
over multiple rounds of the game.3 Furthermore, suppose that players can alter
their strategy in response to the action played by the other players in previous
rounds, such that a strategy that brings a better outcome is more likely retained.
In any population of a mixture of players who always choose to cooperate or
 Modeling Cultural Dynamics  295

defect (i.e., non-conditional co-operators or defectors), defection produces a

higher average payoff than cooperation. Over time, cooperation tends to disap-
pear from the population, and the relative proportion of defection increases as a
result of the simulated selection process, with all the group members eventually
defecting—this happens even if there is only one defector, and the rest are all
cooperators. In this sense, defection is an evolutionarily stable strategy (ESS).4
Nevertheless, as in the one-shot Prisoner’s Dilemma Game, if all players
choose to cooperate, everyone will be better off than if all players defect. By every-
one cooperating, the entire group reaps benefit; by defecting, everyone will be
worse off in the long run. But there is an incentive for each player to defect—the
so-called free rider problem. If the Darwinian evolutionary process underlies the
selection of genetically or socially transmittable behavioral patterns (e.g., coop-
eration and defection), how can cooperation evolve in a population? Thus, the
evolution of cooperation is a theoretical puzzle in the Darwinian theory of bio-
logical and cultural evolution. And again, the answer is some form of conditional
cooperation—a search for genetic or cultural mechanisms that enable conditional
cooperative strategies to produce beneficial outcomes in the long run, even when
the social situation that dictates the outcome of combinations of strategy choices
involves social dilemmas like the Prisoner’s or Public Goods Dilemma.

Mechanisms Promoting Cooperation

The past theoretical research has identified a number of mechanisms that pro-
mote the evolution of cooperation. It is now well established that promotion
and maintenance of cooperation within social dilemmas depends on the positive
assortment of cooperators (West, Griffin, & Gardner, 2007). That is, there must
be mechanisms or interaction structures to ensure that cooperators help other
cooperators more than defectors. Below, we provide a brief overview of the
five key mechanisms described by Nowak (M. A. Nowak, 2006b; Rand & Nowak,
2013) as well as others, which generally promote positive assortment. M. A. Nowak
(2006a) and McElreath and Boyd (2007) provide more formal treatment of the
topics.

Kin Selection
Kin selection (or inclusive fitness) is a mechanism for the evolution of cooperation
that arises if agents use conditional strategies based on kin relationship. This mech-
anism is based on the idea that cooperative behavior can emerge where the donor
(the agent that benefits another agent) and the recipient (the agent that receives
the benefit) are genetically related in the sense that there is a high probability
of sharing a gene. That is, an individual is more likely to cooperate with closer
relatives as compared to distant relatives or strangers. This tends to increase the
average fitness of those that carry the genetic information that tends to produce
296  Kashima, Kirley, Stivala, and Robins

cooperation among kinds. Hamilton’s rule is typically used as a formalization for

this mechanism (Hamilton, 1964a): the coefficient of relatedness, r, must exceed the
cost–benefit ratio of the act for a cooperative action to be played. Although this
may be based on genetic evolution, it can also be a result of cultural evolution.

Direct Reciprocity
Reciprocity is a mechanism by which agents use information about a history of
agents’ past behaviors to predict the probability of their cooperation. Direct reci-
procity involves iterated encounters where agents play a game repeatedly across
a finite number of rounds. Such repeated encounters between the same agents
allow for reciprocation of cooperation—that is, when one cooperates, the other
can return the favor by cooperating. As we noted earlier, Trivers (1971), Axelrod
and Hamilton (1981), and Axelrod (1984) advanced direct reciprocity as a key
mechanism for evolution of cooperation.
A well-known example of this type of strategy is “tit-for-tat,” developed by
Anatol Rapoport, where a player cooperates in the first round and from then
on always repeats whatever the other player did in the previous round. When
it competed with other strategies in a tournament of strategies, it was the
best-­performing strategy in that it cumulated the highest payoffs of all strate-
gies (Axelrod, 1984). Nonetheless, this strategy is not very tolerant of a mistake
by another player—if players defect by mistake (by rare weakness of will, or
a momentary lapse of judgment), strict tit-for-tat strategists descend to all-
defection. More flexible and tolerant strategies such as “generous tit-for-tat”
(one defection is generously tolerated) and “win–stay, lose–shift” have been
shown to be effective once cooperators emerge in the population (M. A. Nowak,
2006a). Arguably, these are all variants of conditional cooperation, in which an agent
cooperates if other agents cooperate. Indeed, approximately 50% of human
players in experimental public goods games appear to be conditional cooperators
(e.g., Fischbacher, Gächter, & Fehr, 2001).

Indirect Reciprocity
Direct reciprocity relies on the firsthand information about the probability of
others’ cooperation obtained by the direct observation of their past behaviors.
In contrast, indirect reciprocity rests on the information about the probability of
others’ cooperation obtained not from the direct observation of their past behav-
ior, but typically secondhand information based on third parties who directly
observed these others’ past behaviors (e.g., M. A. Nowak & Sigmund, 1998;
Panchanathan & Boyd, 2004). This is basically predicting other agents’ behavior
based on their reputation, which may derive from gossip or other mechanisms of
reputation management (e.g., Michelin ratings, university rankings). Agents can
 Modeling Cultural Dynamics  297

adopt conditional strategies and base their decision on reputation profiles of others
in the population. For indirect reciprocity to be an effective mechanism in the
evolution of cooperation, it requires that reputation information is as accurate
as direct information obtaining from personal experience (e.g., M. A. Nowak &
Sigmund, 1998; Panchanathan & Boyd, 2004).
It is important to note that, in order for indirect reciprocity to work, agents
must have both cognitive and communicative capacities to remember their own
interactions and monitor the ever-changing social network within their group, and
to use language or other symbolic means to communicate reputational information—
that is, to gossip. Some have argued that this was one of the evolutionary bases
of human language (Dunbar, 1998). In addition, stereotypes can serve as a basis
of reputational information. If a group of agents is seen to be warm, communal, or
moral (e.g., Eagly & Kite, 1987; Fiske, Cuddy, Xu, & Glick, 2002; Leach,
Ellemers, & Barreto, 2007; Wojciszke, 2005a, 2005b), its members are likely
inferred to be at least conditional cooperators. In particular, an agent’s ingroup
is often stereotyped as trustworthy (e.g., Brewer, 1979), and this autostereotype
may be the basis of an ingroup favoritism in cooperation (e.g., Balliet, Wu, &
De Dreu, 2014; Yamagishi & Kiyonari, 2000). Of course, this is not to say that
stereotypes are accurate; however, it does present a new perspective on the
evolution of stereotypes as a reputational mechanism.

Network Reciprocity
Human interactions are not random, but are typically structured in social
networks—some individuals interact with each other more often than others.
A social network is represented as a graph structure in which its vertices are
occupied by agents, and its edges determine who interacts with whom. A payoff
structure can be defined for each interaction, and repeated interactions between
agents determine the overall outcome of each agent. Despite the complexity of
the dynamics (for an overview, see Perc, Gómez-Gardeñes, Szolnoki, Floría, &
Moreno, 2013; Perc & Szolnoki, 2010; Szabó & Fáth, 2007), the underlying
principle in network reciprocity is the notion that “neighbors help each other.”
Here, network reciprocity can promote the evolution of cooperation, because
cooperators form clusters within which they cooperate with each other, which
can prevail against exploitation by defectors. Recent work combining evolution-
ary dynamics of group interactions on structured populations provides important
insights into our understanding of the evolution of cooperation. A range of
extensions can also be considered. For example, the population structure can be
dynamic, so that agents can use “active linking,” where individuals can choose to
break unproductive links and establish new ones (Pacheco, Traulsen, & Nowak,
2006; Rand, Arbesman, & Christakis, 2011), or form social network ties with
others based on their reputations (Fu, Hauert, Nowak, & Wang, 2008).
298  Kashima, Kirley, Stivala, and Robins

Group Selection
The idea of group selection (also called multilevel selection) has a controversial
background in evolutionary biology and is frequently misunderstood (West, El
Mouden, & Gardner, 2011). Wilson and Wilson (2008) describe how natural
selection acts not only on individuals, but also on groups. Consider the situ-
ation where a population is subdivided into groups. Individuals interact with
other members of their ingroup in an evolutionary game that determines their
fitness. In a simple scenario, defectors dominate cooperators within groups.
However, if groups compete with each other, groups of cooperators out-
compete groups of defectors. A little more theoretically, if selection processes
operate at the level of groups (i.e., those groups with higher levels of average
fitness are more likely to produce offspring than those with lower fitness levels),
groups that contain more cooperators are more likely to do better than those
with fewer cooperators. Under such conditions, multilevel selection acts as a
powerful mechanism for the evolution of cooperation, especially if there are
many small groups and if the migration rate between groups is not too large
(Traulsen & Nowak, 2006). On the surface, there appear to be many simi-
larities between spatial selection and multilevel selection. However, they are
quite distinct mechanisms. In the former case, selection (competition) occurs
between individuals. In the second case, there is competition between indi-
viduals and competition between groups.

Signaling
As we noted earlier, the above-mentioned mechanisms typically promote posi-
tive assortment—that is, the likelihood that cooperation is met by cooperation,
so that cooperative actions co-occur in interacting agents. An additional mecha-
nism for generating the co-occurrence of cooperation is based on signaling (e.g.,
Gintis, Smith, & Bowles, 2001; Skyrms, 2004). If individuals who are to cooper-
ate in the future can signal their future action to cooperate to each other, they
can coordinate their interaction (i.e., choose each other) and avoid the negative
consequences of interacting with defectors.
There are a variety of signs that can be used to signal cooperation if agents
with a genetic or cultural trait to cooperate have (a) a certain recognizable sign,
(b) a capacity to recognize the sign, and (c) the tendency to cooperate when they
recognize that sign. One such sign may be physical resemblance in appearance
as a sign of kinship (e.g., Hamilton, 1964a, 1964b). Nonetheless, any arbitrary
sign—what Dawkins (1976) called a “green beard”—may work as long as it
satisfies the above-mentioned three properties (e.g., Riolo, Cohen, & Axelrod,
2001). The displaying of this sign may be costly (Gintis et al., 2001) or cheap
(Robson, 1990; Skyrms, 2004); either way, positive assortment can occur under
some circumstances. However, if agents develop a capacity to “fake” the sign
 Modeling Cultural Dynamics  299

(i.e., any individual can display the “green beard,” but not cooperate with other
“green beards”), these agents will have higher fitness than others that cannot, and
the sign will eventually be decoupled from the trait for cooperation.

Additional Mechanisms
In addition to these mechanisms, a number of further mechanisms have also
captured the attention of both social and evolutionary biology researchers. One
is voluntary participation, in which, in addition to the default actions of cooperate
or defect, a player is also provided with a third action of not playing at all. In
multi-player games such as the Public Goods game, significant levels of coop-
eration emerge, often in dynamic oscillations (Hauert, De Monte, Hofbauer, &
Sigmund, 2002). Under specific circumstance, punishment can promote coop-
erative behavior in social dilemmas (Boyd & Richerson, 1992; Fehr & Gachter,
2002; Sigmund, Hauert, & Nowak, 2001; however, see Ohtsuki, Iwasa, &
Nowak, 2009). Typically, punishment is embedded in models based on under-
lying mechanisms such as indirect reciprocity, spatial selection, and multilevel
selection. Punishment may be implemented as a second-stage action with addi-
tional costs to an individual, activated when other individuals in the game opt
for defection. However, this approach can be undermined by the proliferation of
second-order “free riders,” who cooperate but do not punish defectors.

Sources of Cultural Dynamics in Evolution of Cooperation

In this section, we have attempted to describe how evolutionary game theory
can be used to understand the evolution of cooperation largely from the perspec-
tive of adaptation, more specifically in terms of selection. Assuming that there
are some agents who adopt conditional cooperation, when mechanisms are in
place to ensure assortment, so that cooperators are more likely to interact with
each other than with defectors, the enactment of the cultural practice (or genetic
trait) of cooperation is likely to bring about better consequences than the cultural
practice of defection on the average in the long run. Those practices that bring
about greater benefit with less cost are assumed to be more likely to be transmit-
ted and retained by agents. Therefore, provided that conditional cooperation
brings about better outcomes than defection, conditional cooperation becomes
more prevalent within the group. A culture of cooperation may thus emerge
because, under certain conditions where assortment is possible, cooperation is
more adaptive.
Note that the current treatment is rather sketchy, and there is a more nuanced
treatment about social transmission of cultural information in this framework. For
example, cultural transmission may be biased (e.g., conformity—agents may be
more likely to learn a practice that is more prevalent in a group; prestige—agents
300  Kashima, Kirley, Stivala, and Robins

may be more likely to learn a practice that is used by a more prestigious agent) or
unbiased, defective or cooperative tendencies may be both culturally and geneti-
cally inherited, etc. (e.g., Boyd & Richerson, 1985; Chudek & Henrich, 2011;
Henrich & McElreath, 2003). Nonetheless, it is fair to say that the strength of the
evolutionary game theoretic approaches lies in its theoretical treatment of adaptation,
especially, the selection process in cultural dynamics.

Conclusions
Social psychology has a tradition of modeling cultural dynamics. Starting with the
public opinion dynamics that involve a binary choice between pro and con in a
population of agents without any spatial structure (Abelson & Bernstein, 1963) to
dynamic social impact theory within a spatial structure (A. Nowak et al., 1990)
and beyond, there has been a steady increase in the complexity with which social
psychology has theorized cultural dynamics.
Nevertheless, further advances in modeling cultural dynamics have occurred
outside the traditional boundary of social psychology, although there are intriguing
developments in social psychology as well (e.g., Denrell & Le Mens, 2007). One
of the research traditions extended the univariate representation of culture (i.e.,
single opinion) to a multivariate representation (Axelrod, 1997), and the other has
brought the evolutionary game theoretic framework to incorporate the adaptive-
ness of culture as a significant driver of cultural dynamics (e.g., McElreath & Boyd,
2007; M. A. Nowak, 2006a). These prominent research traditions have highlighted
complementary aspects of cultural dynamics: the process and consequences of
cultural dissemination and the process of adaptation by selection. The complemen-
tarity of these developments poses an intriguing question for further exploration.
How can these two research traditions be integrated so that multidimensional and
configural cultural information can be represented and the adaptiveness of the cul-
tural information can be investigated within the same framework?
Theoretical questions aside, there are a number of pressing questions for mod-
els of cultural dynamics. First of all, there is an issue of empirical validation of
models. Although the early challenges were more about how to model empiri-
cally well-established phenomena of opinion polarization (Abelson & Bernstein,
1963) and persistence of diverse opinions (A. Nowak et  al., 1990), the more
recent models are mathematically sophisticated, but are yet to be tested or vali-
dated (however, some empirical tests are underway; for a recent review, see
Rand & Nowak, 2013). Second, from apparently intractable intergroup conflicts
to climate change, there are a number of challenges to the contemporary world
that require a transformation of contemporary cultures (Kashima, 2016b; Wilson,
Hayes, Biglan, & Embry, 2014). How can the models of cultural dynamics be
used to benefit the public discourse and deliberation on planning and policy
development for our common future? These are some of the difficult questions
for the modeling of cultural dynamics which future research will need to tackle.
 Modeling Cultural Dynamics  301

Notes
1 The preparation of this chapter was facilitated by a grant from the Australian Research
Council (DP130100845) to Yoshihisa Kashima, Michael Kirley, and Garry Robins.
2 U. Wilensky, NetLogo, available from http://ccl.northwestern.edu/netlogo/.
3 Typically, in the Iterated Prisoner’s Dilemma game, an additional constraint on the payoff
matrix values is employed, 2R > (T + S), so that players are not collectively better off if
they simply alternate between playing cooperate and defect actions.
4 Here, the ESS constitutes a Nash equilibrium, but they are not strictly equivalent.
For detailed discussions about the relationship between ESS and Nash equilibria, see
M. A. Nowak (2006a).

References
Abelson, R. P. (1964). Mathematical models of the distribution of attitudes under con-
troversy. In N. Freriksen & H. Gulliksen (Eds.), Contributions to mathematical psychology
(pp. 142–160). New York: Holt, Rinehart, & Winston.
Abelson, R. P., & Bernstein, A. (1963). A computer simulation model of community ref-
erendum controversies. Public Opinion Quarterly, 27(1), 93–122. doi:10.1086/267152
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Axelrod, R. (1997). The dissemination of culture: A model with local convergence
and global polarization. Journal of Conflict Resolution, 41(2), 203–226. doi:10.1177/
0022002797041002001
Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211(4489),
1390–1396. doi:10.1126/science.7466396
Balliet, D., Wu, J., & De Dreu, C. K. W. (2014). Ingroup favoritism in cooperation: A
meta-analysis. Psychological Bulletin, 140(6), 1556–1581. doi:10.1037/a0037737
Barabási, A.-L., & Albert, R. (1999). Emergence of scaling in random networks. Science,
286(5439), 509–512. doi:10.1126/science.286.5439.509
Bartlett, F. C. (1923). Psychology and primitive culture. Cambridge: Cambridge University Press.
Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Cambridge:
Cambridge University Press.
Bentley, R. A., Hahn, M. W., & Shennan, S. J. (2004). Random drift and culture change.
Proceedings of the Royal Society of London, Series B, 271(1547), 1443–1450. doi:10.1098/
rspb.2004.2746
Boyd, R., & Richerson, P. J. (1985). Culture and the evolutionary process. Chicago, IL:
University of Chicago Press.
Boyd, R., & Richerson, P. J. (1992). Punishment allows the evolution of coopera-
tion (or anything else) in sizable groups. Ethology and Sociobiology, 13(3), 171–195.
doi:10.1016/0162-3095(92)90032-Y
Brewer, M. B. (1979). In-group bias in the minimal intergroup situation: A cognitive-
­motivational analysis. Psychological Bulletin, 86(2), 307–324. doi:10.1037/0033-2909.86.2.307
Brulle, R. J., Carmichael, J., & Jenkins, J. C. (2012). Shifting public opinion on climate change:
An empirical assessment of factors influencing concern over climate change in the U.S.,
2002–2010. Climatic Change, 114(2), 169–188. doi:10.1007/s10584-012-0403-y
Camerer, C. F. (2003). Behavioral game theory: Experiments on strategic interaction. Princeton,
NJ: Princeton University Press.
Campbell, D. T. (1960). Blind variation and selective retentions in creative thought as
in other knowledge processes. Psychological Review, 67(6), 380–400. doi:10.1037/
h0040373
302  Kashima, Kirley, Stivala, and Robins

Campbell, D. T. (1975). On the conflicts between biological and social evolution and
between psychology and moral tradition. American Psychologist, 30(12), 1103–1126.
doi:10.1037/0003-066X.30.12.1103
Carley, K. M. (1991). A theory of group stability. American Sociological Review, 56(3),
331–354.
Castellano, C., Marsili, M., & Vespignani, A. (2000). Nonequilibrium phase transition in
a model for social influence. Physical Review Letters, 85(16), 3536–3539.
Cavalli-Sforza, L. L., & Feldman, M. W. (1981). Cultural transmission and evolution.
Princeton, NJ: Princeton University Press.
Centola, D., González-Avella, J. C., Eguíluz, V. M., & San Miguel, M. (2007).
Homophily, cultural drift, and the co-evolution of cultural groups. Journal of Conflict
Resolution, 51(6), 905–929.
Chudek, M., & Henrich, J. (2011). Culture-gene coevolution, norm-psychology and
the emergence of human prosociality. Trends in Cognitive Sciences, 15(5), 218–226.
doi:10.1016/j.tics.2011.03.003
Dawes, R. M. (1980). Social dilemmas. Annual Review of Psychology, 31(1), 169–193.
doi:10.1146/annurev.ps.31.020180.001125
Dawkins, R. (1976). The selfish gene. Oxford: Oxford University Press.
De Sanctis, L., & Galla, T. (2009). Effects of noise and confidence thresholds in nominal
and metric Axelrod dynamics of social influence. Physical Review E, 79, 046108.
DellaPosta, D., Shi, Y., & Macy, M. (2015). Why do liberals drink lattes? American Journal
of Sociology, 120(5), 1473–1511.
Denrell, J., & Le Mens, G. (2007). Interdependent sampling and social influence.
Psychological Review, 114(2), 398–422. doi:10.1037/0033-295X.114.2.398
Dunbar, R. I. M. (1998). The social brain hypothesis. Evolutionary Anthropology: Issues,
News, and Reviews, 6(5), 178–190. doi:10.1002/(SICI)1520-6505(1998)6:5<178::AID-
EVAN5>3.0.CO;2-8
Eagly, A. H., & Kite, M. E. (1987). Are stereotypes of nationalities applied to both women
and men? Journal of Personality and Social Psychology, 53(3), 451–462. doi:10.1037/0022-
3514.53.3.451
Fehr, E., & Gachter, S. (2002). Altruistic punishment in humans. Nature, 415(6868),
137–140. doi:10.1038/415137a
Fischbacher, U., Gächter, S., & Fehr, E. (2001). Are people conditionally coopera-
tive? Evidence from a public goods experiment. Economics Letters, 71(3), 397–404.
doi:10.1016/S0165-1765(01)00394-9
Fiske, S. T., Cuddy, A. J. C., Xu, J., & Glick, P. (2002). A model of (often mixed)
stereotype content: Competence and warmth respectively follow from perceived
status and competition. Journal of Personality and Social Psychology, 82(6), 878–902.
doi:10.1037//0022-3514.82.6.878
Flache, A., & Macy, M. W. (2011). Local convergence and global diversity: From
interpersonal to social influence. Journal of Conflict Resolution, 55(6), 970–995.
doi:10.1177/0022002711414371
Fu, F., Hauert, C., Nowak, M. A., & Wang, L. (2008). Reputation-based partner
choice promotes cooperation in social networks. Physical Review E, 78(2), 026117.
doi:10.1103/PhysRevE.78.026117
Gandica, Y., Charmell, A., Villegas-Febres, J., & Bonalde, I. (2011). Cluster-size entropy
in the Axelrod model of social influence: Small-world networks and mass media.
Physical Review E, 84(4), 046109. doi:10.1103/PhysRevE.84.046109
 Modeling Cultural Dynamics  303

Gintis, H. (2009). The bounds of reason: Game theory and the unification of the behavioral
sciences. Princeton, NJ: Princeton University Press.
Gintis, H., Smith, E. A., & Bowles, S. (2001). Costly signaling and cooperation. Journal of
Theoretical Biology, 213(1), 103–119. doi:10.1006/jtbi.2001.2406
González-Avella, J. C., Cosenza, M. G., Eguíluz, V. M., & San Miguel, M. (2010).
Spontaneous ordering against an external field in non-equilibrium systems. New Journal
of Physics, 12, 013010. doi:10.1088/1367-2630/12/1/013010
González-Avella, J. C., Cosenza, M. G., & Tucci, K. (2005). Nonequilibrium transi-
tion induced by mass media in a model for social influence. Physical Review E, 72(6),
065102(R). doi:10.1103/PhysRevE.72.065102
González-Avella, J. C., Eguíluz, V. M., Cosenza, M. G., Klemm, K., Herrera, J. L., &
San Miguel, M. (2006). Local versus global interactions in nonequilibrium transi-
tions: A model of social dynamics. Physical Review E, 73(4), 046119. doi:10.1103/
PhysRevE.73.046119
Gracia-Lázaro, C., Lafuerza, L. F., Floría, L. M., & Moreno, Y. (2009). Residential
segregation and cultural dissemination: An Axelrod-Schelling model. Physical Review
E, 80(4), 046123. doi:10.1103/PhysRevE.80.046123
Grauwin, S., & Jensen, P. (2012). Opinion group formation and dynamics: Structures that
last from nonlasting entities. Physical Review E, 85(6), 066113.
Guerra, B., Poncela, J., Gómez-Gardeñes, J., Latora, V., & Moreno, Y. (2010). Dynamical
organization towards consensus in the Axelrod model on complex networks. Physical
Review E, 81(5), 056105.
Hahn, M. W., & Bentley, R. A. (2003). Drift as a mechanism for cultural change: An
example from baby names. Proceedings of the Royal Society of London, Series B, 270
(Suppl. 1), S120–S123. doi:10.1098/rsbl.2003.0045
Hamilton, W. D. (1964a). The genetical evolution of social behavior. I. Journal of Theoretical
Biology, 7(1), 1–16. doi:10.1016/0022-5193(64)90038-4
Hamilton, W. D. (1964b). The genetical evolution of social behavior. II. Journal of
Theoretical Biology, 7(1), 17–52. doi:10.1016/0022-5193(64)90039-6
Hardin, G. (1968). The tragedy of the commons. Science, 162(3859), 1243–1248.
doi:10.1126/science.162.3859.1243
Harton, H. C., & Bullock, M. (2007). Dynamic social impact: A theory of the origins
and evolution of culture. Social and Personality Psychology Compass, 1(1), 521–540.
doi:10.1111/j.1751-9004.2007.00022.x
Hauert, C., De Monte, S., Hofbauer, J., & Sigmund, K. (2002). Volunteering as red queen
mechanism for cooperation in public goods games. Science, 296(5570), 1129–1132.
Henrich, J., & McElreath, R. (2003). The evolution of cultural evolution. Evolutionary
Anthropology: Issues, News, and Reviews, 12(3), 123–135. doi:10.1002/evan.10110
Herskovits, M. J. (1948). Man and his works: The science of cultural anthropology. New York:
Knopf Doubleday.
Hutchins, E. (1995). Cognition in the wild. Cambridge, MA: MIT Press.
Kashima, Y. (2000). Conceptions of culture and person for psychology. Journal of Cross-
Cultural Psychology, 31(1), 14–32. doi:10.1177/0022022100031001003
Kashima, Y. (2014). How can you capture cultural dynamics? Frontiers in Psychology,
5(995), 1–16. doi:10.3389/fpsyg.2014.00995
Kashima, Y. (2016a). Cultural dynamics. Current Opinion in Psychology, 8, 93–97.
Kashima, Y. (2016b). Culture and psychology in the 21st century: Conceptions of culture
and person revised. Journal of Cross-Cultural Psychology, 47(1), 4–20.
304  Kashima, Kirley, Stivala, and Robins

Kashima, Y., & Gelfand, M. J. (2012). History of culture and psychology. In A. Kruglanski &
W. Stroebe (Eds.), Handbook of the history of social psychology (pp. 499–520). New York:
Psychology Press.
Kashima, Y., Peters, K., & Whelan, J. (2008). Culture, narrative, and human agency. In
R. M. Sorrentino & S. Yamaguchi (Eds.), Handbook of motivation and cognition across
cultures (pp. 393–421). San Diego, CA: Academic Press.
Klemm, K., Eguíluz, V. M., Toral, R., & San Miguel, M. (2003). Global culture: A noise-
induced transition in finite systems. Physical Review E, 67(4), 045101(R).
Klemm, K., Eguíluz, V. M., Toral, R., & San Miguel, M. (2005). Globalization, polari-
zation and cultural drift. Journal of Economic Dynamics and Control, 29(1), 321–334.
doi:10.1016/j.jedc.2003.08.005
Kollock, P. (1998). Social dilemmas: The anatomy of cooperation. Annual Review of
Sociology, 24(1), 183–214. doi:10.1146/annurev.soc.24.1.183
Kroeber, A. L., & Kluckhohn, C. (1952). Culture: A critical review of concepts and definitions.
New York: Vintage Books.
Kuperman, M. N. (2006). Cultural propagation on social networks. Physical Review E,
73(4), 046139. doi:10.1103/PhysRevE.73.046139
Latané, B. (1981). The psychology of social impact. American Psychologist, 36(4), 343–356.
Latané, B. (1996). Dynamic social impact: The creation of culture by communication.
Journal of Communication, 46(4), 13–25.
Latané, B., & L’Herrou, T. (1996). Spatial clustering in the conformity game: Dynamic
social impact in electronic groups. Journal of Personality and Social Psychology, 70(6),
1218–1230.
Latané, B., Liu, J. H., Nowak, A., Bonevento, M., & Zheng, L. (1995). Distance matters:
Physical space and social impact. Personality and Social Psychology Bulletin, 21(8), 795–805.
doi:10.1177/0146167295218002
Leach, C. W., Ellemers, N., & Barreto, M. (2007). Group virtue: The importance of
morality (vs. Competence and sociability) in the positive evaluation of in-groups. Journal
of Personality and Social Psychology, 93(2), 234–249. doi:10.1037/0022-3514.93.2.234
Lewenstein, M., Nowak, A., & Latané, B. (1992). Statistical mechanics of social impact.
Physical Review A, 45(2), 763–776. doi:10.1103/PhysRevA.45.763
Markus, H. R., & Kitayama, S. (1991). Culture and the self: Implications for cognition,
emotion, and motivation. Psychological Review, 98(2), 224–253. doi:10.1037/0033-
295x.98.2.224
Mäs, M., Flache, A., & Helbing, D. (2010). Individualization as driving force of clustering
phenomena in humans. PLoS Computational Biology, 6(10), e1000959.
Maynard Smith, J. (1982). Evolution and the theory of games. Cambridge: Cambridge
University Press.
McElreath, R., & Boyd, R. (2007). Mathematical models of social evolution: A guide for the
perplexed. Chicago, IL: University of Chicago Press.
Nowak, A., de Raad, W., & Borkowski, W. (2012). Culture change: The perspec-
tive of dynamical minimalism. In M. J. Gelfand, C.-Y. Chiu & Y.-Y. Hong
(Eds.), Advances in culture and psychology (Vol. 2, pp. 249–314). New York: Oxford
University Press.
Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitudes to public opinions: A
dynamic theory of social impact. Psychological Review, 97(3), 362–376.
Nowak, A., & Vallacher, R. R. (1998). Dynamical social psychology. New York: Guilford.
Nowak, M. A. (2006a). Evolutionary dynamics: Exploring the equations of life. Cambridge,
MA: Harvard University Press.
 Modeling Cultural Dynamics  305

Nowak, M. A. (2006b). Five rules for the evolution of cooperation. Science, 314(5805),
1560–1563. doi:10.1126/science.1133755
Nowak, M. A., & Sigmund, K. (1998). Evolution of indirect reciprocity by image scoring.
Nature, 393(6685), 573–577. doi:10.1038/31225
Ohtsuki, H., Iwasa, Y., & Nowak, M. A. (2009). Indirect reciprocity provides only
a narrow margin of efficiency for costly punishment. Nature, 457(7225), 79–82.
doi:10.1038/nature07601
Pace, B., & Prado, C. P. (2014). Axelrod’s model with surface tension. Physical Review E,
89(6), 062804. doi:10.1103/PhysRevE.89.062804
Pacheco, J. M., Traulsen, A., & Nowak, M. A. (2006). Active linking in evolutionary
games. Journal of Theoretical Biology, 243(3), 437–443. doi:10.1016/j.jtbi.2006.06.027
Panchanathan, K., & Boyd, R. (2004). Indirect reciprocity can stabilize cooperation with-
out the second-order free rider problem. Nature, 432(7016), 499–502. doi:10.1038/
nature02978
Parisi, D., Cecconi, F., & Natale, F. (2003). Cultural change in spatial environments:
The role of cultural assimilation and internal changes in cultures. Journal of Conflict
Resolution, 47(2), 163–179.
Parks, C. D., Joireman, J., & Van Lange, P. A. M. (2013). Cooperation, trust, and antago-
nism: How public goods are promoted. Psychological Science in the Public Interest, 14(3),
119–165. doi:10.1177/1529100612474436
Perc, M., Gómez-Gardeñes, J., Szolnoki, A., Floría, L. M., & Moreno, Y. (2013).
Evolutionary dynamics of group interactions on structured populations: A review.
Journal of the Royal Society Interface, 10(80), 20120997. doi:10.1098/rsif.2012.0997
Perc, M., & Szolnoki, A. (2010). Coevolutionary games: A mini review. Biosystems, 99(2),
109–125. doi:10.1016/j.biosystems.2009.10.003
Pfau, J., Kirley, M., & Kashima, Y. (2013). The co-evolution of cultures, social network
communities, and agent locations in an extension of Axelrod’s model of cultural dis-
semination. Physica A, 392(2), 381–391. doi:10.1016/j.physa.2012.09.004
Railsback, S. F., & Grimm, V. (2012). Agent-based and individual-based modelling: A practical
introduction. Princeton, NJ: Princeton University Press.
Rand, D. G., Arbesman, S., & Christakis, N. A. (2011). Dynamic social networks promote
cooperation in experiments with humans. Proceedings of the National Academy of Sciences
of the United States of America, 108(48), 19193–19198. doi:10.1073/pnas.1108243108
Rand, D. G., & Nowak, M. A. (2013). Human cooperation. Trends in Cognitive Sciences,
17(8), 413–425. doi:10.1016/j.tics.2013.06.003
Rapoport, A., & Chammah, A. M. (1965). Prisoner’s Dilemma: A study in conflict and coop-
eration. Ann Arbor, MI: University of Michigan Press.
Richerson, P. J., & Boyd, R. (2005). Not by genes alone: How culture transformed human
evolution. Chicago, IL: University of Chicago Press.
Riolo, R. L., Cohen, M., & Axelrod, R. (2001). Evolution of cooperation without reci-
procity. Nature, 414(6862), 441–443.
Robson, A. J. (1990). Efficiency in evolutionary games: Darwin, Nash and the secret
handshake. Journal of Theoretical Biology, 144(3), 379–396. doi:10.1016/S0022-
5193(05)80082-7
Rodríguez, A. H., & Moreno, Y. (2010). Effects of mass media action on the Axelrod model
with social influence. Physical Review E, 82(1), 016111. doi:10.1103/PhysRevE.82.016111
Shibanai, Y., Yasuno, S., & Ishiguro, I. (2001). Effects of global information feedback on
diversity: Extensions to Axelrod’s adaptive culture model. Journal of Conflict Resolution,
45(1), 80–96.
306  Kashima, Kirley, Stivala, and Robins

Sigmund, K., Hauert, C., & Nowak, M. A. (2001). Reward and punishment. Proceedings
of the National Academy of Sciences of the United States of America, 98(19), 10757–10762.
doi:10.1073/pnas.161155698
Simonton, D. K. (2010). Creative thought as blind-variation and selective-retention:
Combinatorial models of exceptional creativity. Physics of Life Reviews, 7(2), 156–179.
doi:10.1016/j.plrev.2010.02.002
Skyrms, B. (2004). The stag hunt and the evolution of social structure. Cambridge: Cambridge
University Press.
Sperber, D. (1996). Explaining culture: A naturalistic approach. Oxford: Blackwell.
Stivala, A., Robins, G., Kashima, Y., & Kirley, M. (2014). Ultrametric distribution of
culture vectors in an extended Axelrod model of cultural dissemination. Scientific
Reports, 4, 4870. doi:10.1038/srep04870
Szabó, G., & Fáth, G. (2007). Evolutionary games on graphs. Physics Reports, 446(4–6),
97–216. doi:10.1016/j.physrep.2007.04.004
Traulsen, A., & Nowak, M. A. (2006). Evolution of cooperation by multilevel selection.
Proceedings of the National Academy of Sciences, 103(29), 10952–10955. doi:10.1073/
pnas.0602530103
Triandis, H. C. (1964). Cultural influences upon cognitive processes. In L. Berkowitz
(Ed.), Advances in experimental social psychology (Vol. 1). New York: Academic Press.
Triandis, H. C. (1989). The self and social behavior in differing cultural contexts.
Psychological Review, 96(3), 506–520. doi:10.1037/0033-295X.96.3.506
Triandis, H. C. (1994). Culture and social behavior. New York: McGraw-Hill.
Triandis, H. C. (1996). The psychological measurement of cultural syndromes. American
Psychologist, 51(4), 407–415. doi:10.1037/0003-066X.51.4.407
Trivers, R. L. (1971). The evolution of reciprocal altruism. Quarterly Review of Biology,
46(1), 35–57. doi:10.1086/406755
Valori, L., Picciolo, F., Allansdottir, A., & Garlaschelli, D. (2012). Reconciling long-term
cultural diversity and short-term collective social behavior. Proceedings of the National
Academy of Sciences of the United States of America, 109(4), 1068–1073. doi:10.1073/
pnas.1109514109
Van Lange, P. A. M., Joireman, J., Parks, C. D., & Van Dijk, E. (2013). The psychology
of social dilemmas: A review. Organizational Behavior and Human Decision Processes,
120(2), 125–141. doi:10.1016/j.obhdp.2012.11.003
Vazquez, F., González-Avella, J. C., Eguíluz, V. M., & San Miguel, M. (2007). Time-scale
competition leading to fragmentation and recombination transitions in the coevolution
of network and states. Physical Review E, 76(4), 046120.
Von Neumann, J., & Morgenstern, O. (1953). Theory of games and economic behavior (3rd ed.).
Princeton, NJ: Princeton University Press.
Weaver, I. (2010). Dissemination of culture. NetLogo user community models. Retrieved from
http://ccl.northwestern.edu/netlogo/models/community/Dissemination%20of%20
Culture
West, S. A., El Mouden, C., & Gardner, A. (2011). Sixteen common misconceptions
about the evolution of cooperation in humans. Evolution and Human Behavior, 32(4),
231–262. doi:10.1016/j.evolhumbehav.2010.08.001
West, S. A., Griffin, A. S., & Gardner, A. (2007). Evolutionary explanations for cooperation.
Current Biology, 17(16), R661–R672. doi:10.1016/j.cub.2007.06.004
Wilson, D. S., Hayes, S. C., Biglan, A., & Embry, D. D. (2014). Evolving the future:
Toward a science of intentional change. Behavioral and Brain Sciences, 37(4), 395–416.
doi:10.1017/S0140525X13001593
 Modeling Cultural Dynamics  307

Wilson, D. S., & Wilson, E. O. (2008). Evolution “for the good of the group.” American
Scientist, 96(5), 380–389. doi:10.1511/2008.74.1
Wojciszke, B. (2005a). Affective concomitants of information on morality and competence.
European Psychologist, 10(1), 60–70. doi:10.1027/1016-9040.10.1.60
Wojciszke, B. (2005b). Morality and competence in person- and self-perception. European
Review of Social Psychology, 16(1), 155–188. doi:10.1080/10463280500229619
Yamagishi, T., & Kiyonari, T. (2000). The group as the container of generalized reciprocity.
Social Psychology Quarterly, 63(2), 116–132. doi:10.2307/2695887
Yamamoto, K. (2015). Mobilization, flexibility of identity, and ethnic cleavage. Journal of
Artificial Societies and Social Simulation, 18(2), 8. doi:10.18564/jasss.2669
PART IV

Transforming Social
Psychology
14
MODELS ARE STUPID, AND
WE NEED MORE OF THEM
Paul E. Smaldino

All social science research must do some violence to reality in order to reveal
simple truths.
(Lazer & Friedman, 2007)

Despite numerous efforts extolling the virtues of formal modeling (Epstein,

2008; Schank, 2001; Smith & Conrey, 2007; Marewski & Olsson, 2009; Farrell &
Lewandowsky, 2010; Weinhardt & Vancouver, 2012; Smaldino, Calanchini, &
Pickett, 2015), there remains widespread resistance among social and behavioral
scientists to adopt formal modeling in their general research approach. In addi-
tion to the technical challenge posed by the mathematical and programming skills
required to understand and develop models, a common point of resistance appears
to stem from the perception of models as crude, overly simplistic, and unrealistic.
The conclusion is that models are largely useless as anything but a formal exercise,
and unnecessary for most scientists to engage with.
Rather than argue against this perception, I enthusiastically embrace the per-
spective of the resistance, at least in part. Models are, by and large, stupid. My point
of contention is with the conclusion that stupid models are not useful. Quite the
contrary. Stupid models are extremely useful. They are useful because humans are
boundedly rational and because language is imprecise. It is often only by formal-
izing a complex system that we can make progress in understanding it. Formal
models should be a necessary component of the behavioral scientist’s toolkit.
Models are stupid, and we need more of them.

We Are Stupid
Down to the very name of our species, Homo sapiens, we humans love to empha-
size our intelligence relative to other species. We can certainly solve many
312  Paul E. Smaldino

complicated problems. And yet we are often very stupid animals who make
foolish choices. This isn’t a raw failing on our part. We are limited beings,
with finite resources with which to compute a coarse model of our world and
with which to invent options and evaluate their consequences. Moreover, our
world, and the ecological and social environments in which we find ourselves,
are changing rapidly, far too rapidly for our brains to possibly adapt via genetic
evolution. We do the best we can.
Humans appear to have particular difficulty understanding complex sys-
tems. Mitch Resnick, in his book Turtles, Termites, and Traffic Jams, details his
experiences teaching gifted high school students about the dynamics of com-
plex systems using artificial life models (Resnick, 1994). He showed them how
organized behavior could emerge when individuals responded only to local
stimuli using simple rules, without the need for a central coordinating authority.
Resnick reports that even after weeks spent demonstrating the principles of
emergence, using computer simulations that the students programmed them-
selves, many students still refused to believe that what they were seeing could
really work without central leadership.
We who study complex systems for a living may feel a certain smugness
here. The average person may have difficulty understanding the forces that drive
behavior, we think, but through our powerful intellects, our education, and our
hefty experience pondering the deep mysteries, we can trust our intuition when
it comes to understanding the psychological and social forces that make people
do what they do. Unfortunately, my own experience working with complex
systems and working among complexity scientists suggests that we are hardly
immune to such stupidity. Indeed, even seemingly simple puzzles can pose a
challenge.
Consider the case of Marilyn Vos Savant and the Monty Hall problem. Vos
Savant, famous for her record high score on standard IQ tests, has written a
weekly puzzle column in Parade Magazine since 1986. In 1990, she wrote about
a puzzle commonly known as the Monty Hall problem. The problem goes as
follows. You are on a game show and given the choice to open one of three
doors. Behind one of the doors is a fabulous cash prize, and behind the others,
goats (the assumption is that no one would prefer the goats to the cash). You
choose a door, say Door #1. The host, who knows where the cash really is,
opens one of the other two doors, say #3, and shows you a goat behind it. The
host now offers you the option to switch to Door #2. The question is whether
it is to your advantage to do so.
The answer is that, although you are never guaranteed to be correct, you
should probably switch. The cash is twice as likely to be behind Door #2 instead
of Door #1. This is not an easy result for most people to wrap their heads around,
though it follows quite definitively from the assumptions of probability theory (if
you are in doubt of the problem’s trickiness, I suggest that you pose it the next
time you are at a dinner party). Strikingly, Vos Savant’s answer was challenged
 Models Are Stupid, and We Need More of Them  313

not only by lay readers, but also by many with advanced mathematical training.
Indeed, she received many letters from professional mathematicians insisting that
she was mistaken, even after she published a follow-up column with a detailed
proof. The letters were often written in a smug, knowing tone; Vos Savant details
many of these in an article posted to her website (http://marilynvossavant.com/
game-show-problem/). One, written after the publication of the follow-up column
and signed by a Georgetown University professor, reads:

You are utterly incorrect about the game show question, and I hope this
controversy will call some public attention to the serious national crisis in
mathematical education. If you can admit your error, you will have con-
tributed constructively towards the solution of a deplorable situation. How
many irate mathematicians are needed to get you to change your mind?

It is my belief that the widespread inability to grasp the solution to the Monty
Hall problem stems from a failure to properly model the scenario. You should
switch doors because regardless of which door you picked initially, the host can
always show you one with a goat. Being shown a goat therefore has no bearing on
the probability that your initial choice was correct. Since that probability is 1/3,
there is a 2/3 chance that you were wrong and the cash is behind the remaining
door. Thus, two out of three times, switching is the right move. The common
intuition that the choice is instead a 50/50 split between two options is erroneous.
Readers of this chapter are likely to be interested in social behaviors and their
underlying psychological mechanisms. These systems tend to be quite a bit more
complicated than a simple game show problem. This should concern us. Being an
expert does not inoculate us from the failure of our limited imaginations, which
evolved to solve problems quite different from those of interest to behavioral
scientists. We could use some help.

Models to the Rescue?

I am, of course, going to argue that we should turn to models, and particularly
formal models, for help. Specification of a formal model delineates the parts of a
system and the relationships between those parts, allows us to examine the logical
conclusions of our assumptions, and as a byproduct, examine the appropriateness
of those assumptions in the first place. But first, I need to take a brief detour,
because when it comes to explaining any behavior, the first question we need to
ask is: What are we talking about?

Articulating a System and Its Parts

As behavioral and social scientists, we want to understand some system related
to individual or social behavior. Maybe we are interested in how social identity
314  Paul E. Smaldino

manifests when individuals feel threatened, or how individuals coordinate in joint

activities, or how racially charged language is interpreted by individuals of differ-
ent racial and socioeconomic backgrounds. These examples obviously represent
a miniscule selection from among the questions we might ask. The important
thing to note is how each question is subject to myriad interpretations. What
aspect of the behavior are we interested in, specifically? Are we interested in the
neurophysiology of joint attention, down to the way neural spike trains inform
action programs? Or are we more interested in a “higher” level of organization,
perhaps one in which we can ignore physiology and instead simply consider the
temporal relationships between individually designated behavioral units? These
are not trivial questions. For any given behavior, there are many questions we can
ask related to its development, mechanism, and adaptive function, none of which
are obviously favored from a scientific perspective (Tinbergen, 1963).
Once we specify the level of organization and the kind of explanation we
are looking for, we still need to do additional work to specify the exact question
under investigation. Human beings are complex beings. It’s not just that we exist
at many levels of organization. Of course, we are made of organs, which are made
of tissues, which are made of cells, which communicate using molecules and
ions; above the level of the individual, we are enmeshed in local social networks,
communities both corporate and categorical, economies, and nations. A further
problem arises when we consider that these levels interact—the causal arrows flow
both ways (Campbell, 1974; Wimsatt, 1974). The problem is not insurmountable,
but needs to be acknowledged. Any explanation of individual and social behavior
must necessarily ignore important causal relationships both within and between
levels of organization.We must become comfortable with ignoring those relation-
ships, and this comfort is achieved partly through acknowledging their existence.
Part of specifying our research question involves the articulation of the parts
of the system of inquiry. Kauffman’s 1971 essay still provides the best discussion
of this important but overlooked issue. Notice that I do not say that we should
specify our question and then articulate the parts of the system. The two are
parts of a single process. What is our question? To understand joint attention
in coordinated behavior, perhaps. But what is our question right now? We must
decompose the system into explicit parts. We must postulate properties of those
parts and the relationships between them. In some sense, this is the essence of
all scientific inquiry into behavior. All well-formed scientific research questions
concern the properties of parts, the relationships between them, and the con-
sequences of those relationships. The articulation of parts and relationships will
necessarily be overly simplified and ignore details of physical reality. But much
like a map is only useful because it ignores irrelevant detail, so is a well-formed
scientific question useful when it captures only those features of reality most rel-
evant to a useful answer.
To make myself perfectly clear: To ask a scientific question about individual
or social behavior, we must specify the parts of a system and the relationships
 Models Are Stupid, and We Need More of Them  315

between them. The question at hand may be about the nature of these parts or
their relationships, and so we may designate a distribution of parts or relationships
from which to sample, but it amounts to the same thing. The precise specifica-
tion of parts and relationships is what defines a scientific question and separates
it from wishy-washy pseudotheory that is unfalsifiable and distracting (Popper,
1963; Gigerenzer, 1998; Smaldino, 2016).

Building Models, Formal and Otherwise

Let us assume that we have articulated, in words, the parts of our system and the
relationships between them. Perhaps we say, as do the adherents of optimal dis-
tinctiveness theory (Brewer, 1991; Leonardelli, Pickett, & Brewer, 2010), that
individuals have social identities that correspond to different contexts and different
levels of inclusivity, and that they express these identities in order to balance internal
drives for assimilation and differentiation. The parts are obviously the individuals,
each of whom has the property of possessing an array of identities and the ability to
express one of these at any given time. The relationships between the parts manifest
as perceptions of others’ identities, which dictate how individuals update their own
expression. The theory suggests how this updating might occur: individuals should
express more exclusive identities when their currently expressed identity is very
inclusive, and vice versa.
I have just described what is often called a “verbal model.” As Epstein (2008,
para. 1.2) phrases it, “Anyone who ventures a projection, or imagines how a social
dynamic . . . would unfold is running some model.” Most behavioral and social
scientists are quite comfortable with this sort of model. However, look closer.
You’ll see that the parts of the system are not particularly well articulated, and
neither are their relationships. What does it mean to possess an identity, let alone
an array of them? How do individuals choose between their identities when it
comes time to express them? Is the expression of a new identity costly, perhaps
in terms of time or social capital? How do individuals take stock of the identities
of their fellows? Are their perceptions accurate? Are all identities equally easy to
perceive? There are additional related questions as well, concerning the nature of
system. Where do identities come from, and how might an individual gain a new
identity or lose an existing one? What is the adaptive function of expressing an
identity in the first place, since, to be preserved, identities must serve some purpose
other than internal contentment?
This is not to pick on optimal distinctiveness theorists. Social psychology,
and the social and behavioral sciences more generally, are replete with similar
cases. This is the limitation of verbal models. They are often a good way to
begin an inquiry when the available evidence suggests only some broad type of
relationship that might be further refined. The danger with most verbal models
is that there are many ways to specify the parts and relationships of a system that
are consistent with such a model. Scientific inquiry stalls when data is used to
316  Paul E. Smaldino

simply support rather than refine a verbal model. Because many different data
sets are consistent with a vague verbal model, researchers using such techniques
risk lapsing into positing theories that are, by and large, unfalsifiable (Popper,
1963; Gigerenzer, 1998).
The articulation of the parts of a system and the relationships between them
always involves incurring some violence upon reality. Science is an iterative
process, and pragmatically, we must ignore some details about complexity and
organization to make any headway. That said, it’s not a terrible goal to try and
be a bit more precise. This is where formal modeling comes in. A formal model
instantiates the verbal model as a collection of mathematical relationships and/
or algorithmic processes. Rather than saying an individual has something like an
array of social identities, we can model an individual as a computation object
that has precisely an array of social identities, which in turn might be modeled
as simple numerical values for the sake of comparisons between individuals. My
colleagues and I have made models of this type (Smaldino, Pickett, Sherman, &
Schank, 2012; Smaldino & Epstein, 2015). More than anything, we have learned
that we have a long way to go in understanding the nature and social significance
of social identity.
To paraphrase Gunawardena (2014), a model is a logical engine for turn-
ing assumptions into conclusions. By making our assumptions explicit, we can
clearly assess their implied conclusions. These conclusions will inevitably be
flawed, because the assumptions are ultimately incorrect, or at least incomplete.
By examining how they differ from reality, we can refine our models, and thereby
refine our theories, and so gradually we might become less wrong (Wimsatt,
1987; Schank, May, & Joshi, 2014; Smaldino et al., 2015). Making formal models
of the systems we study is the only way to make this possible.

A Brief Note on Statistical Models

When I talk about formal models, I am primarily talking about models whose
purpose is to elucidate the mechanisms underlying psychological and behavioral
phenomena. Another category of formal model, more familiar to many readers,
I’m sure, is the type of model often used in statistical analysis, such as a path
model or a linear model. Statistical models are both important and limited, and
therefore worth commenting upon, but as they are not my focus here, I will keep
my discussion of them brief.
Statistical analyses are necessary and often well-motivated, but we should never
forget that they too have models at their core. The generalized linear model,
the work horse of the social sciences, models data as being randomly drawn
from a distribution whose mean varies according to some parameter. The linear
model is so obviously wrong yet so useful that the mathematical anthropologist
Richard McElreath has dubbed it “the geocentric model of applied statistics,”
in reference to the Ptolemaic model of the solar system that erroneously placed
 Models Are Stupid, and We Need More of Them  317

the Earth rather than the Sun at the center, but nevertheless produced accurate
predictions of planetary motion as they appeared in the night sky (McElreath,
2015). Such models usually assume that one’s data are generated by randomly
sampling from some distribution—perhaps a Gaussian distribution whose mean
tracks some conditional variable. These models are terrifically important in estab-
lishing relationships between variables in empirical data sets, and thus for guiding
the development of increasingly strong theories. However, many of these models
say little about the processes that actually generated the data, or about the mecha-
nistic nature of relationships between variables. This is the domain of the kinds of
formal models I am principally discussing in this chapter. Such models, if suffi-
ciently precise, may utilize data for validation and calibration (e.g., Schank, 2008;
Moussaïd, Helbing, & Theraulaz, 2011; Hills, Jones, & Todd, 2012), but this is
not strictly necessary for such models to be useful (Wimsatt, 1987; Bedau, 1999;
Epstein, 2008; Gunawardena, 2014).

Models Are Stupid

A common objection to formal modeling in the behavioral and social sciences is
that they are grossly unrealistic. This is, in general, quite correct. Formal models
are often fantastically unrealistic. They ignore huge swaths of reality, including
details of individual behavior and environmental complexity. However, framing
this fact as a downside is a serious error, particularly if the alternative is to rely
instead on verbal models. Verbal models can appear superior to formal models
only by employing strategic ambiguity (sensu Eisenberg, 1984), giving the illu-
sion of understanding at the cost of actual understanding. That is, by being vague,
verbal models simultaneously afford many interpretations from among which any
reader can implicitly, perhaps even unconsciously, choose his or her favorite.
I will illustrate this point with a simple parable.

The Parable of the Cubist Chicken

One evening long ago, when I was an undergraduate student, a friend and I found
ourselves waiting in the basement of a theater for a third friend, an actor about
to finish his play rehearsal. There was a large collection of LEGOs in the room,
and being of a jaunty disposition and not entirely sober, we amused ourselves by
playing with the blocks. One of us—precisely who has been lost to memory—
constructed an assembly of red, white, black, and yellow blocks and declared,
“Look! It’s a Cubist chicken!” The other one of us laughed and heartily agreed
that it most definitely looked like a Cubist chicken. We were extremely satisfied
with ourselves, not only because it was very silly, but because if in fact we both
understood the design to be a Cubist chicken, then it surely was one. We had
identified something true about our little masterpiece, and had therefore, inad-
vertently perhaps, created art. This is how liberal arts students amuse themselves.
318  Paul E. Smaldino

Our conversation moved on to other topics, but we continued to occasionally

comment on the Cubist chicken. After some time had passed, our actor friend
entered the room. “Check it out,” we said, “a Cubist chicken!” Our friend smiled
bemusedly and asked us to explain exactly how the seemingly random constel-
lation of LEGOs represented a chicken. “Well,” I said, pointing to various parts
of the assemblage, “Here is the head. And here is the body and the legs, and here is
the tail.” “No!” cried my co-conspirator. “That’s all wrong.The whole thing is just
the head. Here are the eyes, and the beak, and here is the crest,” for my friend had
envisioned our chicken as a rooster. And thus the illusion of our shared reality was
shattered. We thought we had been talking about the same thing. But when more
precision was demanded, we discovered we had not.

Stupidity Is a Feature, Not a Bug

As many a late-night dorm room conversation can attest, humans are capable of
very elaborate theories about the nature of reality. The problem is that, as sci-
entists, we need to clearly communicate those theories so that we can use them
to make testable predictions. In the social and behavioral sciences, the search
for clarity can present a problem for verbal models, and can lead to a depressing
recursive avalanche of definitions. What is a preference? A preference is a ten-
dency for certain behaviors. What are those behaviors? It depends on the context.
What is a context? This can go on for a while.
Formal models provide a means of escape from the recursive abyss. By restrict-
ing our discussion to the model system, we can clearly articulate all the parts
of that system and the relationships between those parts, leaving nothing out.
This generally leaves us with something that, on the surface, often appears to be
pretty stupid. What I mean is that not only are all models wrong, as George Box
famously noted; they are obviously wrong. However, the stupidity of a model is
often its strength. By focusing on some key aspects of a real-world system (i.e.,
those aspects instantiated in the model), we can investigate how such a system
would work if, in principle, we really could ignore everything we are ignoring.
This only sounds absurd until one recognizes that, in our theorizing about the
nature of reality—both as scientists and as quotidian humans hopelessly entangled
in myriad webs of connection and conflict—we ignore things all the time. We can’t
function without ignoring most of the facts of the world. Our selective atten-
tion ignores most of the sensory input that nevertheless innervates our neurons
(as indicated by the well-known “cocktail party effect”). This ignorance is fun-
damentally adaptive; the bounds to our rationality are severe, and dedication of
cognitive resources entails balancing benefits and costs. Causal explanations work
in much the same way. By ignoring all but the most relevant information, we are
able to impose some modicum of order upon the world. Problems arise when
we try to communicate our systems for ordering the world, as each of us has
decomposed the world into a somewhat different set of parts and relationships.
 Models Are Stupid, and We Need More of Them  319

Formal models solve this problem by systematizing our stupidity, and ensuring
that we are all talking about the same thing.
In the following section, I will provide several concrete examples of how
seemingly stupid models help scientists do their science. Before doing that,
however, it is worth taking a moment to discuss some general ways in which
models that are obviously wrong can nevertheless inform our thought. For
example, studying computational models of complex systems can help us to
build mental models of some emergent phenomena whose dynamics are oth-
erwise difficult to visualize (Nowak, Rychwalska, & Borkowski, 2013), and
the process of model construction can illuminate core uncertainties in one’s
knowledge of a system (Epstein, 2008). The clearest delineation I have found is
William Wimsatt’s (1987) list of 12 “functions served by false models,” with the
understanding that all models are false. I therefore reproduce this list, with only
light editing, in Table 14.1.

TABLE 14.1  Twelve functions served by false models. Adapted with permission from
Wimsatt (1987).

  (1) An oversimplified model may act as a starting point in a series of models of

increasing complexity and realism.
  (2) A known incorrect but otherwise suggestive model may undercut the too ready
acceptance of a preferred hypothesis by suggesting new alternative lines for the
explanation of the phenomena.
  (3) An incorrect model may suggest new predictive tests or new refinements of an
established model, or highlight specific features of it as particularly important.
  (4) An incomplete model may be used as a template, which captures larger or otherwise
more obvious effects that can then be “factored out” to detect phenomena that
would otherwise be masked or be too small to be seen.
  (5) A model that is incomplete may be used as a template for estimating the magnitude
of parameters that are not included in the model.
  (6) An oversimplified model may provide a simpler arena for answering questions about
properties of more complex models, which also appear in this simpler case, and
answers derived here can sometimes be extended to cover the more complex models.
  (7) An incorrect simpler model can be used as a reference standard to evaluate causal
claims about the effects of variables left out of it but included in more complete
models, or in different competing models to determine how these models fare if
these variables are left out.
  (8) Two false models may be used to define the extremes of a continuum of cases in
which the real case is presumed to lie, but for which the more realistic intermediate
models are too complex to analyze or the information available is too incomplete to
guide their construction or to determine a choice between them. In defining these
extremes, the “limiting” models specify a property of which the real case is supposed
to have an intermediate value.
(continued)
320  Paul E. Smaldino

TABLE 14.1  (continued)

  (9) A false model may suggest the form of a phenomenological relationship between the
variables (a specific mathematical functional relationship that gives a “best fit” to the data,
but is not derived from an underlying mechanical model).This “phenomenological law”
gives a way of describing the data, and (through interpolation or extrapolation) making
new predictions, but also, because its form is conditioned by an underlying model, may
suggest a related mechanical model capable of explaining it.
(10) A family of models of the same phenomenon, each of which makes various false
assumptions, has several distinctive uses: (a) One may look for results which are true
in all of the models, and therefore presumably independent of different specific
assumptions which vary across models. These invariant results are thus more likely
trustworthy or “true.” (b) One may similarly determine assumptions that are irrelevant
to a given conclusion. (c) Where a result is true in some models and false in others,
one may determine which assumptions or conditions a given result depends upon.
(11) A model that is incorrect by being incomplete may serve as a limiting case to test
the adequacy of new, more complex models.
(12) Where optimization or adaptive design arguments are involved, an evaluation of
systems or behaviors which are not found in nature, but which are conceivable
alternatives to existing systems, can provide explanations for the features of those
systems that are found.

Some (Not So) Stupid Models

Compiling a list of all the interesting and useful models in the sciences is a fool’s
errand. Let it suffice to say that such a list would be vast. Instead, I want to merely
illustrate via a few pointed examples how simple, stupid models can be not only
useful, but fundamental to good science. I will start with four well-known exam-
ples of models that changed our understanding of basic concepts in the physical,
biological, and social sciences. I will then give two examples of how I have used
formal models in my own work, focusing on topics that should be of interest to
social psychologists: (1) social identity and distinctiveness and (2) hypothesis testing
and replication.

Newton’s Model of Universal Gravitation

In 17th-century Europe, the field of astronomy faced a great challenge. Following
the pioneering work of Copernicus and building on the meticulously collected
data of Tycho Brahe, Johannes Kepler had definitively showed that not only do
the Earth and the other planets revolve around the Sun, their orbital paths describe
ellipses rather than perfect circles. It was a great mystery why this should be. Enter
Isaac Newton. Newton was not the first person to propose that the heavenly
bodies might be attracted to one another with a force that varied with the inverse
square of the distance between them, but he was the first to build a model based
on that proposition (Gleick, 2004). His model was startlingly simple, consisting of
 Models Are Stupid, and We Need More of Them  321

only two objects—the Sun and the Earth (Figure 14.1). The model ignored the
Moon as well as the five other known solar planets, not to mention all the celestial
bodies that were unknown in Newton’s time. The size and topology of the Sun
and Earth were also ignored; they were modeled as points identified only by their
mass, position, and velocity. Nevertheless, the model’s strength lies in its simplic-
ity. By restricting the analysis to only two bodies, the resulting planetary orbit was
mathematically tractable. Using a simple rule stating that the force of gravitation
was proportional to the product of the objects’ masses and inversely propor-
tional to the square of the distance between them, Newton was able to show that
the resulting orbits would always take the form of conic sections, including the
elliptical orbits observed by Kepler. And because he could show that the same
law explained the motion of falling objects on Earth, Newton provided the first
scientific unification of the Terrestrial with the Celestial. Newton’s theory of
Universal Gravitation rested on a model that, to naïve eyes, can easily appear
quite stupid. Ultimately, the theory has been shown to be incorrect, and has been
epistemically replaced by the theory of General Relativity. Nevertheless, the the-
ory is able to make exceptionally good approximations of gravitational forces—so
good that NASA’s Moon missions have relied upon them.

FIGURE 14.1   graphical representation of Newton’s model of planetary gravitation.

A
The Earth has a forward velocity v, which is continuously altered by the
gravitational attraction of the Sun, Fg, resulting in an elliptical orbit. In
reality, the model is even simpler than implied here, because the Sun
and Earth were represented as point masses rather than spheres.
322  Paul E. Smaldino

The Lotka-Volterra Model of Predator–Prey Relations

For many years, fur trapping organizations like the Hudson’s Bay Company in
Canada kept meticulous records on the pelt-producing animals in the regions
where they trapped.These records illustrated that linked predator and prey species,
like the Canada lynx and the snowshoe hare, tended to have cyclical population
levels whose dynamics were tightly correlated. How to explain this? In the early
20th century, Alfred Lotka and Vito Volterra, working independently, applied ideas
from the chemistry of autocatalytic reactions to generate a simple model of two
interrelated populations, which can be instantiated as a pair of coupled differential
equations. This model specifies two animal species: a prey species with a positive
rate of growth in the absence of predators, and a predator species with a negative
growth rate in the absence of prey. The number of predators negatively influences
the number of prey, and the number of prey animals positively influences the
number of predators. The model can produce correlated oscillations in the two
populations that bear a striking resemblance to data on many predator–prey systems.
The model also identifies conditions under which the two growth rates can give
rise instead to more stable equilibria as well as yielding complete population
collapse—predictions that have since been borne out empirically. However, the
model is extremely simplistic. It assumes perfect mixing, so that the probability of
a prey animal encountering a predator is simply the relative frequency of preda-
tors in the population. It ignores seasonality, circadian cycles, migration, density
dependence in the growth rate of the prey species, development, and interactions
with other species. Thus, when these features matter, the model may fail to align
with empirical fact (Luckinbill, 1973; Berryman, 1992). Nevertheless, the core
assumptions of the model often hold. This provides opportunities for extensions
and refinements of the model when additional features cannot be ignored. By pro-
viding a foundational structure, the Lotka-Volterra model remains one of the core
tools for understanding the relationship between predator and prey populations.

Hopfield’s Model of Content-Addressable Memory

Memory—the ability to store information for later recall—is a wondrous prop-
erty of neural networks that makes possible all but the most rudimentary forms
of cognition. By the early 1980s, long-term potentiation—the process by which
Donald Hebb’s theory that “neurons that fire together wire together” occurs—was
relatively well described. It was believed that the formation of such associations
was intrinsic to more complex forms of memory, such as that by which a person’s
face is encoded and then later recognized, but the mechanism was unclear. How
could a brain possibly use partial information, like an occluded face, to reconstruct
information encoded in memory? To begin to answer this question, the biophysi-
cist John Hopfield (1982) constructed a simple model of two-state neurons in a
fully connected network. Edge weights were determined by a process of Hebbian
learning assumed to have already occurred, so that a number of configurations
 Models Are Stupid, and We Need More of Them  323

(or states) of “on” and “off ” neurons were encoded in the network, with edges
assumed to be bidirectionally symmetrical (i.e., undirected). Using mathematics
derived from statistical physics, Hopfield showed firstly that in such a system,
encoded states would be stable, and secondly that if initialized in a non-encoded
state, the network would self-organize into the encoded state that most closely
matched the initialized state. In other words, he had a model for how memory
retrieval could emerge spontaneously in a simple neural network. This model is
almost absurdly simplistic, even stupid in its assumptions. Neurons are either on
or off, ignoring subtleties of firing rates or even graded activation. Directionality
is also ignored; links between neurons are equally strong in each direction. Exactly
how the network is presumed to first arrive in its initial state is left a mystery.
Yet analysis of the model showed that something like biological neural networks
could produce content-addressable memory. Hopfield himself later showed that
the model’s functioning was robust to the relaxation of some of his strict assump-
tions (Hopfield, 1984), and the work has laid the foundation for much subsequent
work in understanding the neurobiology of memory.

Bass’s Model of the Diffusion of Innovations

How do new products diffuse in a population? In the early 1960s, Everett
Rogers (1962) provided a near-exhaustive study of this question. He showed
that cumulative adoption very often corresponds to an S-shaped curve in which
adoption starts slowly, accelerates, and then plateaus. Although Rogers showed
that this pattern of product diffusion is common to a startlingly wide variety of
domains, he could not explain it. Instead, he merely identified five tautological
categories of adopters, defined in terms of their timing of adoption. This expla-
nation is rather unsatisfying and raises many additional questions, including why
individuals would fall into a particular category of adopter and how robust the
adoption curves are to different proportions of each of those categories. Shortly
after Rogers’ book was published, Bass (1969) introduced a simple model that
provided a strikingly parsimonious explanation of Rogers’ data. Suppose, said
Bass, that instead of five discrete types, there is only a single type of individual,
who with some small probability spontaneously adopts the new innovation (i.e.,
becomes an innovator) and otherwise adopts with a probability proportional to
the number of other adopters he or she encounters. In other words, suppose that
innovations spread like diseases. Bass constructed a mathematical model based
on these assumptions, and showed not only that they resulted in S-shaped adop-
tion curves, but that by fitting the model to empirical data on the diffusion of
different products, characteristics of a given population concerning the rate of
observation and the propensity to adopt could be inferred. The Bass model is still
the core model for studying the diffusion of products used in communication,
technology, and marketing research today (Bass 2004). The Bass model is, of
course, extremely simplistic. It ignores real differences between individuals, such
324  Paul E. Smaldino

as network position (Valente, 1996) or the propensity to adopt based on social

group membership (Berger & Heath, 2008), which may influence the dynamics
of diffusion. Nevertheless, the Bass model provides critical structure for developing
theory and guiding data collection related to the diffusion of innovations.

The Dynamics of Distinctiveness

Some of my own work has concerned the population dynamics resulting from
individual preferences for distinctiveness. Though much of human social behav-
ior stems from conformity—that perfectly reasonable heuristic to copy others
“when in Rome”—it is also quite common to actively differentiate ourselves
from others (at least in the large, complex societies in which most of us find
ourselves; see Smaldino in press). I first became involved in this research in grad-
uate school, when I was approached by two social psychologists working within
the domain of optimal distinctiveness theory (ODT; Brewer, 1991; Leonardelli
et al., 2010). This theory has long had at its core the sort of vague verbal model
I discussed in the subsection “Building Models, Formal and Otherwise.” The
presumption is that individuals have traits called social identities, and that, all else
being equal, they will “identify” with whichever identity optimally balances the
opposing needs for assimilation (to be similar to others) and differentiation (to be
different from others). It is never stated exactly what does or does not constitute a
social identity, what it means to identify as one thing, how the needs for assimila-
tion and differentiation are calibrated, or how one optimizes a balance between
them. Empirical tests have shown that US college students do prefer to express,
at least on paper, a more exclusive part of their social identity when the initially
proposed identity (e.g., being a student of their college) is described as being
non-noteworthy (Brewer & Pickett, 2002). However, many questions remain,
and the theory remains largely lacking in precision.
One assumption of the ODT is that deviations from optimality will be cor-
rected as individuals change their expressed identities to ones that more optimally
balance their opposing needs, and that this will result in a stable equilibrium in
which individuals are satisfied in their relative distinctiveness (Leonardelli et al.,
2010). To test this, my colleagues and I decided to model a simple scenario based
on one possible interpretation of ODT (Smaldino et al., 2012). We assumed a
population of individuals who could each express one of some number of discrete
identities at any given time. We also assumed that each individual had a preference
for some optimal level of distinctiveness, where an individual’s distinctiveness
was defined as the proportion of neighbors also expressing the same identity. One
at a time, agents would consider the distinctiveness of their currently expressed
identity, and if a better option was available, switch to that identity (agents were
updated one at a time because synchronous updating is unrealistic, eliminates
the possibility of behavioral cascades, and can generate peculiar model artifacts;
see Huberman & Glance, 1993). The result was that individuals always ended
 Models Are Stupid, and We Need More of Them  325

up expressing identities that were far too popular to satisfy their preferences for
differentiation. I later learned that this result echoed earlier work by ecologists
considering animals joining groups of varying size, who had reached similar
conclusions (Sibly, 1983).
Our model makes extremely simplistic assumptions about individuals’ abilities
to observe, express, and change identities. Nevertheless, the model accomplishes
something that no previous work on ODT had: it defined all of the parts of
the system and their relationships explicitly. Based on a set of assumptions that is
entirely consistent with the verbal model, we produced a model that provided
several initial conclusions and prompted two broad questions. First, is it true
that individuals are perpetually more similar to others than they would prefer?
This could, in fact, be the case. Several other models have recently shown that
even explicit preferences for anti-conformity or distinctiveness can nevertheless
result in local conformity (Muldoon, Smith, & Weisberg, 2012; Touboul, 2014;
Smaldino & Epstein, 2015). Second, if it is instead the case that individuals are
generally satisfied with the distinctiveness of their expressed identity, then what
key factors related to the dynamics of identity expression were missing from our
model? Several possibilities present themselves, including factors such as network
structure, interdependence between identities, behavioral inertia, and transaction
costs to switching identities. We examined the first of these, network structure,
by situating individuals on a square lattice and having them only respond to
nearby neighbors. We found that for a wide range of conditions, this kind of
network structure solved the problem: individuals could maintain identities that
maximized their preferences for distinctiveness. Our implementation of network
structure was itself quite unrealistic—real social networks rarely approximate
square lattices. Nevertheless, the model represents a step, if only a small one,
toward a more precise theory linking individual preferences for distinctiveness
with the social organization that results from those preferences.

Turning the Modeling Lens on the Scientific Process

As a final example, I want to explore how formal models can help us better
understand the larger endeavor in which we are engaged: science. Recently, con-
troversy has raged over the roles of replication and publication policy in improving
the reliability of research (Open Science Collaboration, 2015). Some propose that
all results should be published, to ensure that a “file drawer effect” doesn’t lead to
over-representation of positive results (Franco, Malhotra, & Simonovits, 2014),
while others are skeptical of the value of failed replications because replication
studies may have diminished power (Kahneman, 2011; Bissell, 2013; Schnall,
2014). All acknowledge the importance of replication, but opinions vary widely
on how much is needed and what its evidential value might be. Until now, each
view has been based on intuition and lacked concrete rationale. And empirical
analysis is inherently limited, both by the incompleteness of the published record
326  Paul E. Smaldino

and by the lack of internally consistent models of the scientific process that would
allow us to usefully interpret extant data.
To remedy this dearth, Richard McElreath and I developed an analytical
model of the population dynamics of science (McElreath & Smaldino, 2015).
The model represents a population of scientists who, with regularity, select a
hypothesis for investigation, investigate it using the standard methods of their
field, and then attempt to communicate their results to the scientific community.
We built on previous work by Ioannidis (2005), who introduced a simple model
of scientific investigation that highlights the importance of the base rate—that is,
the a priori probability that a novel hypothesis is true. When the base rate is low,
even the most stringent experimental methods may produce more false positives
than true positives. Our model extends this discussion to consider the fact that
scientists may replicate their own and each other’s work, but also that results must
also run the gauntlet of peer review, with negative results being less likely to be
published than positive ones. We conclude that regardless of how much replica-
tion is done, the biggest impediments to the effectiveness of science are low base
rate and high false positive rate. I know of no better way to improve the base
rate than to make sure that hypotheses stem from well-validated, precise theories.
Such theories, in turn, are often developed at least partly through the extensive
use of formal modeling. The model also speaks directly to the debate over the
meaning of failed replications. We show that replications are informative even
when they have substantially lower power than the initial investigations. Perhaps
counterintuitively, we also find that suppression of negative findings may be ben-
eficial, at least when such findings are tests of novel hypotheses and the base rate
is low. Under those conditions, most novel results will be correct rejections of
incorrect hypotheses. As these will not be surprising, we may want to avoid fill-
ing our journals with such results, or at least delegate them to a distinct location.
Our model of science is extremely simple. It frames hypothesis testing in a
standard but unsatisfying true/false classification, rather than considering practical
significance and effect size estimation. It ignores researcher bias, multiple test-
ing, and data snooping. It ignores the incentives that drive scientists in choosing
and publishing results, as well as differences in exclusivity and impact between
journals. Nevertheless, our model provides, for the first time, specific quantitative
evaluations of many verbal arguments. As I have noted throughout this chapter,
all models, whether formal or verbal, ignore some factors. The difference is that,
with a formal model, it is precisely clear which factors are being considered and
which are being excluded.

Modelers Are Stupid (Sometimes)

Models can help us to specify theories of how a complex system works, and
to assess the conclusions of our assumptions when they are made precisely.
However, I want to be careful not to elevate modelers above those scientists
 Models Are Stupid, and We Need More of Them  327

who employ other methods. This is important for at least two reasons, the first
and foremost of which is that science absolutely requires empirical data. Those
data are often painstaking to collect, requiring clever, meticulous, and occasion-
ally tedious labor. There is a certain kind of laziness inherent in the professional
modeler, who builds entire worlds from his or her desk using only pen, paper,
and computer. Relatedly, many scientists are truly fantastic communicators, and
present extremely clear theories that advance scientific understanding without a
formal model in sight. Charles Darwin, to give an extreme example, laid almost
all the foundations of modern evolutionary biology without writing down a
single equation. That said, evolutionary biology would surely have stagnated
without the help of formal modeling. Consider that Darwinism was presumed to
be in opposition with Mendelian genetics until modelers such as R. A. Fisher and
Sewall Wright showed that the two theories were actually compatible.
The second reason is that having a model is not the same thing as having a
good model, or a model that is well presented, well analyzed, or well situated in
its field. I want to focus on presentation and analysis. A model’s strength stems
from its precision. I have come across too many modeling papers in which the
model—that is, the parts, all their components, the relationships between them,
and mechanisms for change—is not clearly expressed. This is most common with
computational models (such as agent-based models), which can be quite compli-
cated, but also exists in cases of purely mathematical models. I am not a big fan
of standardized protocols for model descriptions, as the population of all models
is too varied and idiosyncratic to fit into a one-size-fits-all box. I will simply ask
modelers to make an effort in their reporting. Make sure your model description
is clear. The broad strokes, which may stem from verbal theory, should come
first, followed by a filling in of details. When possible, make code available as
soon as your paper is published, if not before. Clarity reveals how well the model
really represents the system it purports to represent. Obfuscation is the refuge of
the poor or insecure modeler.
This is not the place to go into great detail about the best practices for model
analysis. I will only say that a major benefit of a model is the ability to ask all
manner of “what if” questions. The assumptions of a model, including but not
limited to its parameter values, should be explored extensively. After all, obtain-
ing the conclusions that follow from those assumptions is the entire purpose of
modeling. If you forgive the indulgence, I’ll pick one small nit here concerning
methods for analyzing computational models. Where differences between condi-
tions are indicated, avoid the mistake of running statistical analyses as if you were
sampling from a larger population. You already have a generating model for your
data—it’s your model. Statistical analyses on model data often involve modeling
your model with a stupider model. Don’t do this. Instead, run enough simulations
to obtain limiting distributions.
Finally, it is important to always evaluate whether the conclusions of our
model rely on reasonable assumptions. For example, it has been claimed that
328  Paul E. Smaldino

some economists have fallen prey to a sort of theory-induced blindness, giving

too much credence to their models—which are generally based on the theory
of the rational actor—and ignoring the fact that the core assumptions of the
model are based on severe distortions of human psychology (Thaler, 2015).
Microeconomic models based on rational choice theory are useful for developing
intuition, and may even approximate reality in a few special cases, but the history
of behavioral economics shows that standard economic theory has also provided
a smorgasbord of null hypotheses to be struck down by empirical observation.

Conclusion
Humans, scientists included, are limited beings who are bad at forming intuitions
about the organization and behavior of complex systems. Verbal models, while
critical first steps in scientific reasoning, are necessarily imprecise. Overreliance
on verbal models can impede precision and, by extension, impede progress in our
understanding of complex systems. Formal models are explicit in the assumptions
they make about how the parts of a system work and interact, and moreover are
explicit in the aspects of reality they omit. This has the potential disadvantage
of making formal models appear stupid. And of course, they are stupid, because
we are limited beings and stupid models are the best we can do. As Braitenberg
writes, fiction will always be part of science “as long as our brains are only minis-
cule fragments of the universe, much too small to hold all the facts of the world
but not too idle to speculate about them” (Braitenberg, 1984, p. 1).
An old adage holds that it is better to stay silent and be thought a fool than
to speak and remove all doubt. As scientists, our goal is not to save face, but in
fact to remove as much doubt as possible. Formal models make their assumptions
explicit, and in doing so, we risk exposing our foolishness to the world. This
appears to be the price of seeking knowledge. Models are stupid, but perhaps
they can help us to become smarter. We need more of them.

References
Bass, F. M. (1969). A new product growth for model consumer durables. Management
Science, 15, 215–227.
Bass, F. M. (2004). Comments on “A new product growth for model consumer durables”:
The Bass model. Management Science, 50, 1833–1840.
Bedau, M. A. (1999). Can unrealistic computer models illuminate theoretical biology? In
Wu, A. S. (Ed.), Proceedings of the 1999 Genetic and Evolutionary Computation Conference
Workshop (pp. 20–23). Orlando, FL: GECC.
Berger, J., & Heath, C. (2008). Who drives divergence? Identity signaling, outgroup
dissimilarity, and the abandonment of cultural tastes. Journal of Personality and Social
Psychology, 95, 593–607.
Berryman, A. A. (1992). The origins and evolution of predator–prey theory. Ecology, 73,
1530–1535.
Bissell, M. (2013). Reproducibility: The risks of the replication drive. Nature, 503, 333–334.
 Models Are Stupid, and We Need More of Them  329

Braitenberg, V. (1984). Vehicles: Experiments in synthetic psychology. Cambridge, MA: MIT Press.
Brewer, M. B. (1991). The social self: On being the same and different at the same time.
Personality and Social Psychology Bulletin, 17(5), 475–482.
Brewer, M. B., & Pickett, C. L. (2002). The social self and group identification: Inclusion
and distinctiveness motives in interpersonal and collective identities. In J. Forgas &
K. Williams (Eds.), The social self: Cognitive, interpersonal, and intergroup perspectives.
Philadelphia, PA: Psychology Press.
Campbell, D. T. (1974). “Downward causation” in hierarchically organised biological systems.
In F. Ayala & T. Dobzhansky (Eds.), Studies in the philosophy of biology (pp. 179–86).
Oakland, CA: University of California Press.
Eisenberg, E. M. (1984). Ambiguity as strategy in organizational communication.
Communication Monographs, 51(3), 227–242.

Epstein, J. M. (2008). Why model? Journal of Artificial Societies and Social Simulation,
11(4), 12.
Farrell, S., & Lewandowsky, S. (2010). Computational models as aids to better reasoning
in psychology. Current Directions in Psychological Science, 19: 329–335.
Franco, A., Malhotra, N., & Simonovits, G. (2014). Publication bias in the social sciences:
Unlocking the file drawer. Science, 345, 1502–1505.
Gigerenzer, G. (1998). Surrogates for theories. Theory & Psychology, 8(2), 195–204.
Gleick, J. (2004). Isaac Newton. New York: Vintage Books.
Gunawardena, J. (2014). Models in biology: “Accurate descriptions of our pathetic thinking.”
BMC Biology, 12, 29.
Hills, T. T., Jones, M. N., & Todd, P.M. (2012). Optimal foraging in semantic memory.
Psychological Review, 119, 431–440.
Huberman, B. A., & Glance, N. S. (1993). Evolutionary games and computer simulations.
Proceedings of the National Academy of Sciences USA, 90, 7716–7718
Hopfield, J. J. (1982). Neural networks and physical systems with emergent collective compu-
tational abilities. Proceedings of the National Academy of Sciences USA, 79, 2554–2558.
Hopfield, J. J. (1984). Neurons with graded response have collective computational prop-
erties like those of two-state neurons. Proceedings of the National Academy of Sciences
USA, 81, 3088–3092.
Ioannidis, J. P. A. (2005). Why most published research findings are false. PLoS Medicine,
2(8), e124.
Kahneman, D. (2011). A new etiquette for replication. Social Psychology, 45, 310–311.
Kauffman, S. A. (1971). Articulation of parts explanation in biology and the rational search
for them. In R. C. Buck & R. S. Cohen (Eds.), PSA 1970 (pp. 257–72). Irvine, CA:
Philosophy of Science Association.
Lazer, D., & Friedman, A. (2007). The network structure of exploration and exploitation.
Administrative Science Quarterly, 52, 667–694.
Leonardelli, G. L., Pickett, C. L., & Brewer, M. B. (2010). Optimal distinctiveness
theory: A framework for social identity, social cognition, and intergroup relations.
In M. P. Zanna & J. M. Olson (Eds.), Advances in experimental social psychology (Vol. 43,
pp. 66–115). New York: Academic Press.
Luckinbill, L. S. (1973). Coexistence in laboratory populations of Paramecium aurelia and
its predator Didinium nasutum. Ecology, 54, 1320–1327.
Marewski, J. N., & Olsson, H. (2009). Beyond the null ritual: Formal modeling of
psychological processes. Zeitschrift für Psychologie, 217, 49–60.
McElreath, R. (2015). Statistical rethinking: A Bayesian course with R examples. New York:
Chapman & Hall.
330  Paul E. Smaldino

McElreath, R., & Smaldino, P.E. (2015). Replication, communication, and the population
dynamics of scientific discovery. PLoS One, 10(8), e0136088.
Moussaïd, M., Helbing, D., & Theraulaz, G. (2011). How simple rules determine
pedestrian behavior and crowd disasters. Proceedings of the National Academy of Sciences,
108, 6884–6888.
Muldoon, R., Smith, T., & Weisberg, M. (2012). Segregation that no one seeks. Philosophy
of Science, 79, 38–62.
Nowak, A., Rychwalska, A., & Borkowski, W. (2013). Why simulate? To develop a
mental model. Journal of Artificial Societies and Social Simulation, 16(3), 12.
Open Science Collaboration (2015). Estimating the reproducibility of psychological
science. Science, 349, aac4716.
Popper, K. (1963). Conjectures and refutations. New York: Routledge.
Resnick, M. (1994). Turtles, termites, and traffic jams: Explorations in massively parallel
micro-worlds. Cambridge, MA: MIT Press.
Rogers, E. M. (1962). Diffusion of innovations. New York: The Free Press.
Schank, J. C. (2001). Beyond reductionism: Refocusing on the individual with individual-
based modeling. Complexity, 6(3), 33–40.
Schank, J. C. (2008). The development of locomotor kinematics in neonatal rats: An
agent-based modeling analysis in group and individual contexts. Journal of Theoretical
Biology, 254, 826–842.
Schank, J. C., May, C. J., & Joshi, S. S. (2014). Models as scaffold for understanding. In
L. R. Caporael, J. R. Griesemer, & W. C. Wimsatt (Eds.), Developing scaffolds in evolu-
tion, culture, and cognition (pp. 147–167). Cambridge, MA: MIT Press.
Schnall, S. (2014). Clean data: Statistical artefacts wash out replication efforts. Social
Psychology, 45(4), 315–320.
Sibly, R. M. (1983). Optimal group size is unstable. Animal Behaviour, 31, 947–948.
Smaldino, P. E. (2016). Not even wrong: Imprecision perpetuates the illusion of under-
standing at the cost of actual understanding. Behavioral and Brain Sciences, 39, e163.
Smaldino, P. E. (in press). The evolution of the social self: Multidimensionality of social
identity solves the coordination problems of a society. In W. C. Wimsatt & A. C. Love
(Eds.), Beyond the meme: Articulating dynamic structures in cultural evolution. Minneapolis,
MN: University of Minnesota Press.
Smaldino, P. E., Calanchini, J., & Pickett, C. L. (2015). Theory development with agent-
based models. Organizational Psychology Review, 5(4), 300–317.
Smaldino, P. E., & Epstein, J. M. (2015). Social conformity despite individual preferences
for distinctiveness. Royal Society Open Science, 2, 140437.
Smaldino, P. E., Pickett, C., Sherman, J., & Schank, J. (2012). An agent-based model of
social identity dynamics. Journal of Artificial Societies and Social Simulation, 15(4), 7.
Smith, E. R., & Conrey, F. R. (2007). Agent-based modeling: A new approach for theory
building in social psychology. Personality and Social Psychology Review, 11, 87–104.
Thaler, R. H. (2015). Misbehaving: The making of behavioral economics. New York:
W. W. Norton.
Tinbergen, N. (1963). On aims and methods of ethology. Zeitschrift für Tierpsychologie, 20,
410–433.
Touboul, J. (2014). The hipster effect: When anticonformists all look the same. arXiv,
1410.8001.
Valente, T. W. (1996). Social network thresholds in the diffusion of innovations. Social
Networks, 18, 69–89.
 Models Are Stupid, and We Need More of Them  331

Weinhardt, J. M., & Vancouver, J. B. (2012). Computational models and organizational

psychology: Opportunities abound. Organizational Psychology Review, 2(4), 267–292.
Wimsatt, W. C. (1974). Complexity and organization. In K. Schaffner & R. S. Cohen
(Eds.), PSA 1972 (pp. 67–86). Irvine, CA: Philosophy of Science Association.
Wimsatt, W. C. (1987). False models as means to truer theories. In M. H. Nitecki &
A. Hoffman (Eds.), Neutral models in biology (pp. 23–55). New York: Oxford University
Press.
15
BIG DATA IN PSYCHOLOGICAL
RESEARCH
David Serfass, Andrzej Nowak,1
and Ryne Sherman

Just as the microscope allowed us to look at smaller and smaller parts of physical
objects, big data allows us to zoom in on our minds.
(Stephens-Davidowitz, 2015)

We live in the age of information. With the increased use of computers, more
data is being recorded than ever. Exabytes of data are created daily (McAfee &
Brynjolfsson, 2012). Cell phone providers record call histories, banks record
transaction histories, Google records search histories, and Facebook records status
updates. In fact, data is the fastest-growing commodity in the modern world, and
businesses have been quick to capitalize on these digital records. Netflix pro-
vides recommendations for its users based on what they have previously viewed,
improving the user experience. Amazon tracks not only what books people
buy, but also what they view, allowing it to make better recommendations for
customers and sell more products. Retailers (e.g., Target, CVS) track customer
purchases through rewards cards to better target their advertising. Google, one
of the most valued companies in the world, is on a self-proclaimed mission to
gather, organize, and make accessible all of the information (data) in the world.2
Beyond the business world, the Big Data revolution is also transforming social
science. This chapter has three goals. The first is to describe the ways in which
Big Data is transforming the social sciences. As will be seen, Big Data is chang-
ing not only the ways social scientists gather and analyze data, but even the way
they design their studies. The second goal of this chapter is to discuss some of
the opportunities Big Data affords social scientists. We do so by detailing a recent
example of Big Data use in the social sciences. Finally, the third goal of this
chapter is to describe the new set of challenges that Big Data has created for social
scientists and how we may go about overcoming them.
 Big Data in Psychological Research  333

The New Social Sciences in the Big Data Revolution

The empirical revolution of the early 20th century changed the social sciences
by emphasizing quantitative methods, surveys, and statistics, separating the social
sciences from the rest of the humanities. Similarly, the modern Big Data revolu-
tion is transforming the social sciences. To see how, consider the way in which
a social scientist might traditionally answer a research question. First, he or she
might begin by evaluating the various theories related to his or her question. Such
theories might yield one or more hypotheses. Next, the scientist would design
a study to test these hypotheses, including the operationalization, manipulation,
and measurement of key variables of interest. An experimental study might con-
tain two to four experimental conditions with around n ≈ 30 in each, while a
large survey study might gather reports from several hundred respondents. The
researcher would execute the study by gathering data in his or her laboratory and
analyze the results using basic inferential tests such as ANOVA, correlation, or
regression.
In contrast, Big Data scientists take a different approach to thinking about
how research questions can be answered. Rather than considering various ways
to gather data to examine the research question, Big Data scientists begin by
thinking about the kinds of data that are already available and how they could
be used to answer the research question. Once a possible data source is located,
Big Data scientists consider how they might acquire that data, what analyses
they would conduct on the data, and what format the data would need to be in
to conduct such an analysis. This different approach to thinking about how to
answer research questions is the foundation for at least four fundamental differ-
ences between Big Data and traditional approaches to social science.

Data Sources
The Big Data approach is nothing without data sources. As we noted in the
introduction to this chapter, though, data are being produced and stored at
greater rates than ever before. As such, the data scientist has many data sources
available. Some of the most popular data sources are social media platforms
(e.g., Facebook, Twitter), Web search records (e.g., Google Trends), and
mobile devices. Because data sources are so essential for the Big Data approach
to social science, we will provide some examples of popular data sources and
highlight some studies using them.

Facebook
With over one billion users (Sedgi, 2014), Facebook has been used in several
social science studies. One study examined the voting behavior of Facebook
users in response to an experimental manipulation. The researchers manipulated
whether or not Facebook users saw pictures of their friends who reported voting
334  Serfass, Nowak, and Sherman

or a banner that said “Today is Election Day.” Users who saw pictures of friends
who voted actually voted more than those who only saw the banner (Bond et al.,
2010). Another experimental Facebook study manipulated the extent to which
users were exposed to positive or negative status updates from their friends. The
results indicated that people who were exposed to more negative statuses also
posted more negative statuses themselves, thereby demonstrating emotional con-
tagion via large-scale social networks (Kramer, Guillory, & Hancock, 2014).
Finally, three observational studies examined the relationship between personal-
ity and Facebook usage. Two found that the words that people use in their status
updates are related to their personalities (Schwartz et al., 2013; Park et al., 2015),
and the third found that the things that people “like” on Facebook are related to
their personalities (Youyou, Kosinski, & Stillwell, 2015).

Twitter
Another social media site that has facilitated a considerable amount of research
is Twitter. Twitter’s API allows for easy access of publicly streamed Tweets and
numerous researchers have used this tool for research purposes. In one study,
researchers mapped the patterns of people’s affective experiences throughout the
course of a day by analyzing millions of individual Tweets and comparing the
number of positive and negative words used (Golder & Macy, 2011). They found
that positive affect peaks both in the early morning and around 11 p.m., whereas
negative affect is highest in the late night hours between midnight and 4 a.m. In
another Twitter study, researchers used machine learning to predict psychopathic
tendencies based solely on the words that people used in their Tweets (Wald,
Khoshgoftaar, Napolitano, & Sumner, 2012).

Google Trends
Nearly everyone knows that Google is a powerful search engine enabling mil-
lions of people to instantly access a huge network of information. However, far
fewer people may realize that Google records these searches and that aggregated
historical searches can be accessed through the tool Google Trends.3 The Google
Trends repositories have been utilized in several recent studies.
One study used Google Trends to examine Web search frequencies for the
words “jobs,” ”welfare,” and “unemployment” to better estimate real-time job-
less claims. Traditional economic measures like jobless claims are usually several
months delayed due to data collection processes. However, search frequencies
from Google Trends improved real-time predictions of joblessness (Choi &
Varian, 2012). Another study used Google Trends to explore seasonal differences
in sexual behavior by examining search frequencies for mate-seeking terms as well
as pornographic material. The results demonstrated a six-month pattern of mate-
seeking behavior with peaks in winter and early summer (Markey & Markey,
2013). Another analysis using Google Trends showed that people are more likely
 Big Data in Psychological Research  335

to search for terms like “news” and “prayer” in the morning, but words like
“suicide” and “symptoms” in the late night hours (Stephens-Davidowitz, 2015).

Transcripts
Historically, finding past newspaper articles required going to a library and
searching through microfilm records. Today, however, many news repositories,
and even newscasts, are kept online, making them a particularly juicy data source
for Big Data scientists. For instance, different American news sources have repu-
tations for their political leanings. Fox News is considered conservative, MSNBC
is considered more liberal, and CNN is considered more moderate. One study
used historical news transcripts from a 12-month period to assess the political bias
of these different new outlets via semantic preferences. Consistent with expec-
tations, the study found that Fox was the most conservative news source and
MSNBC was the most liberal (Holtzman, Schott, Jones, Balota & Yarkoni, 2011).

Communication Networks
Researchers have also explored the nature of relationships using digital com-
munication records. Utilizing network theory (e.g., Barabási, 2012a, 2012b),
researchers have examined underlying organizational structures using corporate
e-mail records (Adamic & Adar, 2005). Others have examined communication
patterns in response to emergency events using cellular phone records (Candia
et al., 2008). Such digital communication records provide a concrete method for
understanding how people actually interact with one another in their daily lives.

Mobile Devices
Cell phones, smart watches, and other mobile devices provide continuous streams
of data related to the activity of the user (Miller, 2012). Most mobile devices con-
tain a GPS module that can locate the user and track movements in space, as well
as acceleration sensors that can assess the user’s movements. Many mobile phones
and wearable electronics also monitor physiological parameters such as heart rate.
These data are used by companies for various purposes (e.g., to inform coffee
drinkers about their proximity to the nearest Starbucks), but social scientists have
begun tapping these data sources as well. Lakens (2013) used cellular phones to
measure changes in heart rate during relived happiness and anger. Others have
used apps controlling the microphones in mobile devices to make audio recordings
of participants’ daily lives (Mehl & Robbins, 2012).

Other Data Sources

Of course, this list is not exhaustive, but it serves to illustrate the types of data
that are available to researchers. New sources of data are constantly becoming
336  Serfass, Nowak, and Sherman

available. Large amounts of data are collected by other techniques such as aerial
surveys, cameras, microphones, and sensors. Further, with the new development
of the Internet of Things, where objects equipped with sensors and microprocessors
communicate directly with each other, data concerning objects and their use are
becoming available. As we have seen, these new data sources are rapidly advanc-
ing research capabilities of psychologists and other social scientists.

Getting Outside the Lab

While the examples just described showcase various sources available to Big Data
social scientists, they also emphasize a second fundamental difference between tra-
ditional and Big Data approaches to research. Big Data approaches are much more
likely to gather data outside of laboratory settings. While the laboratory experiment
will probably always have its uses for social scientists, advancing technology and
a desire for increased ecological validity have moved more and more social sci-
ence research outside the laboratory (Reis, 2012). Indeed, one strength of the Big
Data approach to social science is that the data gathered are more likely to reflect
actual human behavior (e.g., a Google search, a text message) than self-reports of
behavior on a survey or reaction time in a laboratory (Baumeister, Vohs, & Funder,
2007). Moreover, because Big Data often come from real-world data sources (as
opposed to artificial sources in the laboratory), we can gain a better understanding
of the natural relationships among variables of interest. Laboratory experiments
are regularly tightly controlled, often even controlling variables that are naturally
correlated with each other. This set of circumstances eliminates the possibility of
actually estimating the true (population-level) relationships between parameters
(Baumeister, 2016). The Big Data approach embraces natural covariation.

Description and Measurement Precision

Traditional approaches to social science involve gathering a sample and using
statistical induction (i.e., inferential statistics) to generalize to some larger popula-
tion. Of course, even the logic of these methods is tenuous given that the heavy
reliance on convenience samples (e.g., college sophomores) of the WEIRD
(Western, educated, and from industrialized, rich, and democratic countries)
samples variety severely limits generalizability (Henrich, Heine, & Norezayan,
2010). In contrast, Big Data approaches regularly concern larger, more repre-
sentative, and more diverse samples, if not entire populations. As a result, Big
Data approaches tend to emphasize descriptive rather than inductive approaches.
Indeed, if one has the entire population (or close to it), there is nothing left to
which one can generalize. Because of their emphasis on effect sizes (Cohen,
Cohen, West, & Aiken, 2003), Big Data researchers tend to prefer statistical
regression (and related techniques) over traditional techniques such as ANOVA,
which emphasize p-values and statistical significance. Indeed, the sample sizes
 Big Data in Psychological Research  337

used in Big Data research often render all relationships “statistically significant.”
As a result, Big Data researchers focus more on precision of measurement as
embraced by the so-called “New Statistics” (Cumming, 2012).

Tolerance for Complexity

A fourth fundamental difference between traditional approaches to social science
and Big Data approaches concerns complexity. A laboratory experiment or a
survey of 100 or so undergraduate students makes the detection of complex rela-
tionships (e.g., higher order interactions, non-linear relationships) much more
difficult. Indeed, published interaction effects in the social sciences are notori-
ously difficult to replicate. Because of the sheer amount of the data available, Big
Data researchers can readily explore higher-level interactions, the shape of non-
linear relationships, and temporal dynamics without fear of making Type-I errors.
In practice, Big Data researchers often employ machine learning or deep learning
approaches to identify complex relationships and simultaneously employ cross-
validation to ensure the robustness of their results. Such tools allow researchers to
both explore and confirm relationships between variables that may not have been
related to the initial goals or theory behind the study.

Big Data Opportunities in the Social Sciences

Big Data offers a huge number of possibilities for social science researchers.
Ultimately, we believe that the Big Data approach will allow us to understand
and predict human behavior better than ever before. In this section, we detail
an example of a study we recently conducted using the Big Data approach. Our
intent is to use this example to illustrate some of the opportunities Big Data pro-
vides for social science researchers and some of the issues involved in using a Big
Data approach.

Measuring Situation Experiences via Twitter

In a recent study, we used Twitter to examine what people all over the United
States experience in their everyday lives (Serfass & Sherman, 2015). This study
employed machine learning to approximate human ratings of the situational
characteristics present in over 20 million Tweets. We began with two research
questions: What kinds of situations do people experience in everyday lives, and
how do such situation experiences vary across time?
A traditional approach to these questions might involve distributing survey
questionnaires to a group of people representative of our population of inter-
est. Such questionnaires could ask people to recall the things they did yesterday,
following a technique known as the “day reconstruction method” (Kahneman,
Krueger, Schkade, Schwarz, & Stone, 2004). Alternatively, we could have taken
338  Serfass, Nowak, and Sherman

advantage of modern advances in mobile technology to conduct an experience

sampling study. While either of these approaches may have been useful, they
would ultimately be labor-intensive, requiring the recruitment of participants,
developing questionnaires, and software for distributing these questionnaires.
Further, even with our best efforts, the sample size for such a study may have
never really approached Big Data numbers.
Taking a Big Data approach, we wondered what data might already be available
that would allow us to rapidly assess daily life situations from millions of people. We
ultimately decided to use Twitter, for several reasons. First, Twitter has approxi-
mately 271 million users who are responsible for over 500 million Tweets (i.e., social
media posts of 140 characters or less) per day (Seward, 2014). Second, the content of
such Tweets reflects what people are currently thinking, feeling, and experiencing
(i.e., situations). Third, Tweets are often posted in real time, meaning that they tend
to reflect experiences that are occurring right now or just occurred recently. Fourth,
the Twitter API allows one to rapidly gather millions of Tweets from all across the
United States. Thus, Twitter provided an excellent platform for rapidly gathering
lots of data aimed at addressing our two research questions.
We used the streamR and twitteR packages in the statistical computing lan-
guage R to gather over 42 million Tweets posted during a two-week period.4
Because posts from public social media accounts are often spam or corporate
advertisements (see our discussion of data quality below), we limited their impact
by removing Tweets from users who (a) posted more than 165 times during the
two-week period, (b) had more than 2926 followers, or (c) had more than 40,358
total account statuses (Tweets and Retweets). These figures represented the top
2.5% of each metric. This resulted in a final sample of 20,239,179 Tweets from
1,347,499 users for analysis.
Next, we needed to score each of these 20 million Tweets for their situa-
tion characteristics. We used the DIAMONDS model of situation characteristics,
which consists of eight core situation dimensions: Duty, Intellect, Adversity,
Mating, pOsitivity, Negativity, Deception, and Sociality (for more details, see
Rauthmann et  al., 2014). Clearly, it would be far too labor-intensive to have
research assistants read and rate all 20 million Tweets on the DIAMONDS
dimensions. As such, we employed machine learning to build predictive models
of the DIAMONDS in each Tweet (based on Tweet content) that approxi-
mate human judgments. To do so, we first needed what computer scientists
call “ground truth.” We had research assistants rate a (different) sample of 5000
Tweets on the Situational Eight DIAMONDS (Rauthmann et al., 2014). Each
Tweet was rated by four coders, and composite judgments were used as the
criteria for the machine learning models.
After research assistants rated this sample of Tweets, the frequencies with
which different words were used in the sample were tallied using the Linguistic
Inquiry Word Count computerized text analysis program (LIWC).5 The default
LIWC dictionary contains 64 different word categories. These categories include
 Big Data in Psychological Research  339

standard linguistic information (e.g., personal pronouns, prepositions), psycho-

logical constructs (e.g., anxiety, anger, cognitive mechanism), personal concern
categories (e.g., work, leisure), paralinguistic dimensions (e.g., “um”), and
general descriptive categories (e.g., word count, six-letter words). The LIWC
dictionary provides a validated means for analyzing written and spoken word use,
and has been used to analyze textual data in numerous studies (e.g., Chung &
Pennebaker, 2008; Yarkoni, 2010).
The word frequencies for each category in the LIWC dictionary were used as
the predictors for the machine learning models. We employed various training
methods, including linear models, support vector machine, and random forest.
Based on both internal and external cross-validation results, we selected models
using the random forest method because those models were most accurate at
replicating human judgments with Model R values ranging between .26 and .70,
depending on the DIAMONDS dimension. We also analyzed the number of
coded Tweets necessary to build accurate prediction models. Figure 15.1 shows
the cross-validated Model R values for the model predicting Duty based on the
number of Tweets that coders rated. As can be seen, Model R values asymptote
around 4000 Tweets. This indicates that by having coders rate 5000 Tweets, we
captured as much information as possible for our machine learning models.

FIGURE 15.1   odel R values at different sample sizes. Five models were built
M
at each sample size from 100 to 5000 Tweets. Model Rs for each
iteration are shown, along with a locally weighted regression line
depicting the predicted Model R at each number of Tweets.
340  Serfass, Nowak, and Sherman

With these models now built, we applied them to our original sample of
20 million Tweets, effectively simulating what we would find if research assistants
had read and rated all 20 million Tweets. In other words, we scored millions of
Tweets for their psychologically relevant situation characteristics in just a few
hours. In contrast, we estimate that attempting to rate all 20 million Tweets by
hand would have taken about 400,000 hours at a rate of 50 Tweets per hour.
Using these 20 million Tweets that were scored on the Situational Eight
DIAMONDS, we explored the daily patterns of situation experience. We found
that people experience more pOsitivity over the weekend and more Negativity
during the work week. We also found that people experience, on average,
more Duty during the workday and more Sociality in the evenings. Figure 15.2
shows weekly and daily trends in experienced Duty, Sociality, pOsitivity, and
Negativity on Twitter.
Further, we also found that women reported more emotional situations,
scoring higher on both pOsitivity and Negativity, than men. Women also
Tweeted about situation experiences that were, on average, higher on Mating
(i.e., romance) than men. Finally, using the geotagged location of the Tweets,

FIGURE 15.2   verage situation experiences by time and day. The average
A
DIAMONDS dimension score for each minute of each weekday is
shown, along with the generalized additive model smoothed lines
for each weekday (N = 17,005,376 statuses, 1,347,499 Twitter users,
males = 324,244, females = 310,372, unknown = 167,051).
 Big Data in Psychological Research  341

we compared three categories of population density: Urban Areas (population >

50,000), Urban Clusters (population between 2500 and 50,000), and Rural Areas
(population < 2500). To our surprise, we found virtually no differences between
the situation experiences that people in these areas Tweet about. This suggests
that, perhaps, people experience psychologically similar situations regardless of
their local population density.
Overall, we hope the discussion of this study serves as an example of the type
of research that is enabled by the Big Data approach to social science. Gathering
20 million self-reports of participants’ situation experiences, as one might attempt
to do employing traditional approaches, would take years, if not decades. Further,
as noted above, rating 20 million Tweets manually would be completely unfea-
sible. However, we want to emphasize that the method outlined here is not
limited to situation experiences, or even Twitter. Researchers interested in any
psychological construct (e.g., social support, aggression, fear) can follow a similar
approach: (1) identify or obtain some sort of ‘ground truth’ criteria, (2) quantify
the stimuli (e.g., word usage, objects in images, etc.), (3) use machine learning
techniques to build models to predict the criteria of interest, and (4) apply predic-
tive models to a larger corpus of interest.

Challenges for Big Data in the Social Sciences

While the Big Data revolution offers many opportunities for social scientists, it is
not without its challenges and limitations. This section describes five challenges
facing social scientists employing a Big Data approach, and offers potential solu-
tions to these challenges.

Data Analysis and Storage

First, Big Data challenges prior models of data analysis and storage. In tradi-
tional social science research, the researcher collects some data, stores the data
in a spreadsheet, and analyzes that data using statistical software on a laptop or
personal computer. However, the exponential growth of the amount of data
generated and collected every day has rendered these traditional techniques of
data gathering, storage, and analysis inadequate, if not completely obsolete. In the
aforementioned Twitter study, the raw data file (JSON format) was over 100 GB
in size. Files of this size make storage and memory capabilities a necessary consid-
eration. Most laptops and personal computers are not powerful enough to work
with files like this. In fact, as the size of the data to be analyzed becomes larger,
it may actually be more efficient to move the analysis to the server storing the
data, rather than the other way around. This is the idea behind the Hadoop-based
distributed computing paradigm that many companies use for their data storage
and analytics (e.g., Facebook, Google).
342  Serfass, Nowak, and Sherman

Causal Inferences
A second challenge for social scientists employing a Big Data approach is that
experimental methods are often unavailable. The obvious drawback is that mak-
ing causal claims is more difficult, if not impossible, with Big Data approaches.
However, unlike typical survey researchers, social scientists employing Big
Data approaches have two advantages. First, it is often the case that the social
scientist is not attempting to make causal claims, but rather only interested in
building models that best predict an outcome of interest. For instance, in the
Twitter example described previously, we were not really interested in under-
standing the causal processes behind how raters of Tweets made judgments
of situational characteristics. Instead, we were only interested in building and
applying models that best reflected the ratings, no matter how they were made.
Second, Big Data approaches often measure not only a large number of obser-
vations, but also a large number of variables. Doing so allows one to rule out
many alternative explanations for relationships by including them in the model
(Cohen et al., 2003).

Imperfect Data
A third challenge for social scientists employing Big Data approaches is that the
data available to Big Data social scientists are often unintended. In traditional
social science research, the scientist carefully determines what to measure and
how to measure it. Data that are available to the Big Data social scientist, on
the other hand, are often unplanned and consist of things that happen to be
measured for no particular (scientific) reason. For example, data from Facebook
profiles include the user’s relationship status. As a result, researchers with access
to Facebook profiles can use this information to study romantic relationships.
However, Facebook did not include a question about relationship status to study
romantic relationships per se, but rather to enhance user experience. Thus, the
data available to the Big Data social scientist are not always exactly the data one
might like to have. However, Big Data social scientists are often willing to trade
a small amount of perfect data for a vastly larger amount of imperfect data. Big
Data social scientists are counting on the fact that larger and more representative
samples of data are more likely to replicate and generalize to the population, even
if the data are imperfect.
Relatedly, because the data available to the Big Data social scientist are rarely
by his or her design, they are often messy, include many outliers, lots of missing-
ness, and/or lack a common format. In response, Big Data social scientists must
equip themselves with tools for data wrangling and generally dealing with the
problems associated with messy data. It is perhaps no surprise then that many
social scientists have begun teaming up with computer scientists, who often
have expertise in acquiring and managing Big Data. In the Twitter example we
described earlier, we had to eliminate nearly 50% of the Tweets prior to analysis,
 Big Data in Psychological Research  343

to reduce the impact of Tweets produced by spammers and corporate accounts.

However, even using this conservative data cleaning procedure, we had more
than enough Tweets left for analysis. Whereas experimental social scientists use
their creative talents to devise sophisticated experiments, Big Data social scien-
tists must use their creativity to identify potentially useful data and methods for
extracting and analyzing such data.

Theoretical Considerations
The fourth challenge for social scientists using Big Data approaches is how
to combine Big Data with psychological theory. For the most part, Big Data
approaches are purely descriptive. As noted previously, they are usually less inter-
ested in why and how variables are connected to each other than in which and to
what degree variables are connected. Such an emphasis on purely practical mat-
ters can be offputting to social scientists who are interested in explanation and
building theory. Although the descriptive approach has been proven to be very
effective in short-timescale prediction, it cannot replace the theory. The value
of the descriptive knowledge holds only as long as the conditions stay the same.
Without the theory, one cannot say what will happen in a novel situation. Also,
the value of the theory is to compress a large pool of observations to a small set of
causal rules. Although the complex relationships discovered by machine learning
algorithms may be useful for prediction by computers, their usefulness in allow-
ing humans to better understand the world they live in is limited. Thus, whether
the potential of Big Data will be realized by social scientists depends on the abil-
ity of social scientists to use Big Data in developing theories and in its ability to
show the added value of understanding how human mind works beyond purely
descriptive patterns of human behavior.

Privacy Concerns
The fifth challenge for social scientists using Big Data approaches is privacy risks.
From Facebook status updates to credit card transactions to text message records,
Big Data is inherently personal. Both corporations and social scientists have
already received considerable pushback from perceived Big Data overreach. One
Facebook study, already mentioned in the subsection on “Data Sources” earlier
in this chapter, manipulated users’ news feeds with either more positive or nega-
tive news. The researchers found small, but statistically significant, evidence for
emotional contagion such that emotions seen on Facebook news feeds seem to
impact users’ own expressed emotions (Kramer et al., 2014). Another Facebook
study, also previously mentioned, manipulated users news feeds about voting,
which seemed to have a real impact on voting behavior (Bond et  al., 2012).
Both studies received considerable public and academic backlash (e.g., Dumas,
Serfass, Brown, & Sherman, 2014; Gleibs, 2014), and have largely resulted in
344  Serfass, Nowak, and Sherman

Facebook keeping its research findings internal (i.e., not publishing them). More
recently, researchers released a public data set containing user names, age, gender,
location, personality traits, and responses to over 2000 data profiling questions
compiled by the dating website OkCupid (Kirkegaard & Bjerrekær, 2016; see
also Zimmer, 2016). These data were gathered, or scraped, from public profiles
on the OkCupid website. This study received tremendous public backlash con-
cerning privacy and ethical issues in Big Data use, and has since been removed
by the authors.
Both researchers and the public need to be aware of such privacy and ethi-
cal issues in using Big Data. Researchers must take great care to protect the
privacy of those who provide them with data, or else they risk losing access to
such data (e.g., users leaving websites, having access to data sources revoked).
Simultaneously, the public need to be aware of their own privacy rights and how
their reactions to Big Data science can alter scientific advancement. While com-
panies, such as Facebook, may have the right to experiment with their news feed
algorithm, they will be reluctant to share the results of such experiments with the
public if it will only lead to negative publicity. Data scientists should work with
the public to help them understand how Big Data can be used to positively affect
the lives of the data providers.

Concluding Thoughts on Big Data

The majority of research presented in this volume utilizes computational mod-
eling to understand social processes. The use of simulations facilitated a new
research paradigm for social scientists. The advent of computers capable of pro-
cessing intensive simulations allowed researchers to simulate societies (Nowak,
Szamrej, & Latané, 1990), evolution (Axelrod & Hamilton, 1981), mental
processes (Nowak, Vallacher, Tesser, & Borkowski, 2000) and interpersonal pro-
cesses. Computational modeling led researchers to numerous new insights that
were previously unavailable. Big Data research is similarly poised to facilitate a
revolution in psychological research.
Big Data has the potential to transform social psychology from a qualitative to
quantitative science. By using the data from extremely large samples, or even from
the populations, the Big Data approach concentrates on describing the strength
of the relationships, rather than their significance. This moves social psychology
from qualitative science, concentrating on whether a relationship exists, to a quan-
titative science, where the goal is the precise description of the relationship. This
paves the way to describe theories using formulas, or computational models rather
than qualitative verbal descriptions. The mathematical or computational models
can be used for practical applications. The qualitative nature of the theories and
models makes it possible to precisely predict effects of different types of interven-
tions and to find out the most effective ways to achieve desired changes.
 Big Data in Psychological Research  345

Finding complex, also nonlinear, relationships involving many variables and

their interactions is another potential of the Big Data approach. The size of the
sample makes it possible to use machine learning algorithms to find very complex
relationships in the data and to express the relationships by mathematical for-
mulas. For traditional social psychological research, even if data analysis revealed
significant higher than third-order interactions, interpreting them and describ-
ing the nature of high-order interactions was a daunting task, because verbal
description of the nature of relationship between more than four variables where
each variable modifies the effects of the other variables become too complex.
Quantitative models based on Big Data analysis have no problem with describing
types of relationships that are too complex to be expressed by verbal description,
by expressing them in mathematical formulas.
The Big Data approach also naturally leads to more comprehensive models.
One of the strongest advantages of the Big Data approach is the potential to
merge multiple data sets in a single analysis, with these data sets representing
many levels of psychological reality. For example, questionnaire data concerning
individuals can be merged with the data describing the organizations where they
are employed and the demographic and statistical data concerning the counties
where they live (e.g., crime rates, income, etc).
In the traditional experimental approach in social psychology, the number of
variables that could be manipulated in a single study is severely limited by the
growth of the number of cells with the addition of each new factor. In the cor-
relational approach, the limit on the number of variables was limited by the time
the participants could spend answering the questions. In most cases of Big Data
studies, the data are collected automatically, without significant effort from the
participant, which allows one to collect much more data than in the research
requiring subjects to answer questions. Also, it is possible to place psychologi-
cal data concerning an individual in the context of aggregate multiple-level data
(e.g., the organization to which the individual belongs, the county, and the state).
Hierarchical multilevel regression models can explore relationships both within
and between levels, which allows one to build and test comprehensive, complex,
multilevel models, where the psychological processes can be understood in the
context of the environmental effects.
Adding dynamics to existing theories in social psychology and developing new
theories in the tradition of Dynamical Social Psychology (Nowak & Vallacher,
1998; Vallacher & Nowak, 1997; Vallacher, Read, & Nowak, 2002) represents
another potential of Big Data approach. Because most of the data in this approach
are gathered continuously and contain time stamps, it allows the researchers,
as we have also shown in this chapter, to analyze patterns of change in time.
Dynamical theories thus became empirically meaningful and testable. Observing
temporal patterns in psychological and social phenomena also inspires new
models and theories in social psychology.
346  Serfass, Nowak, and Sherman

In sum, Big Data provides an empirical platform for the development and
testing of a new kind of theories in social psychology. Together with com-
puter simulations, the Big Data approach paves the way for the development of
theories that are quantitative, complex, comprehensive, multilevel, and dynamic.
In doing so, it facilitates integration with computational theories in other social
sciences, and also with natural sciences.

Notes
1 Andrzej Nowak acknowledges the support from the grant of Polish Committee for
Scientific Research (DEC-2011/02/A/HS6/00231).
2 More pessimistically, we might say that Google’s goal is to “monetize all of the informa-
tion in the world” (Morozov, 2013).
3 See https://www.google.com/trends/.
4 For streamR, see https://cran.r-project.org/web/packages/streamR/; for twitteR, see
https://cran.r-project.org/package=twitteR; for R, see www.r-project.org/.
5 See http://liwc.wpengine.com/.

References
Adamic, L., & Adar, E. (2005). How to search a social network. Social Networks, 27(3),
187–203.
Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211,
1390–1396
Barabási, L. (2012a). Introduction. In Network science: Interactive textbook. Retrieved from:
http://barabasilab.neu.edu/networksciencebook/
Barabási, L. (2012b). Graph theory. In Network science: Interactive textbook. Retrieved from:
http://barabasilab.neu.edu/networksciencebook/
Baumeister, R. F. (2016). Charting the future of social psychology on stormy seas:
Winners, losers, and recommendations. Journal of Experimental Social Psychology, 66,
153–158.
Baumeister, R. F., Vohs, K. D., & Funder, D. C. (2007). Psychology as the science of
self-reports and finger movements: Whatever happened to actual behavior? Perspectives
on Psychological Science, 2(4), 396–403.
Bond, R. M., Fariss, C. J., Jones, J. J., Kramer, A. D., Marlow, C., Settle, J. E., & Fowler, J. H.
(2012). A 61-million-person experiment in social influence and political mobilization.
Nature, 489(7415), 295–298.
Candia, J., González, M. C., Wang, P., Schoenharl, T., Madey, G., & Barabási, A. L.
(2008). Uncovering individual and collective human dynamics from mobile phone
records. Journal of Physics A: Mathematical and Theoretical, 41(22), 224015.
Choi, H., & Varian, H. (2012). Predicting the Present with Google Trends. The Economic
Record, 88(S1), 2–9. doi:10.1111/j.1475-4932.2012.00809.x
Chung, C. K., & Pennebaker, J. W. (2008). Revealing dimensions of thinking in
open-ended self-descriptions: An automated meaning extraction method for natural
language. Journal of Research in Personality, 42(1), 96–132.
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied Multiple Regression/
Correlation Analysis for the Behavioral Sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum.
 Big Data in Psychological Research  347

Cumming, G. (2012). Understanding the new statistics: Effect sizes, confidence intervals, and
meta-analysis. New York: Routledge.
Dumas, G., Serfass, D. G., Brown, N. A., & Sherman, R. A. (2014). The evolving nature
of social network research: A commentary to Gleibs (2014). Analyses of Social Issues and
Public Policy, 14(1), 374–378.
Gleibs, I. (2014). Turning virtual public spaces into laboratories: Thoughts on conducting
online field studies using social network sites. Analyses of Social Issues and Public Policy,
14(1), 352–370.
Golder, S. A., & Macy, M. W. (2011). Diurnal and seasonal mood vary with work, sleep,
and daylength across diverse cultures. Science, 333(6051), 1878–1881.
Henrich, J., Heine, S. J., & Norezayan, A. (2010). The weirdest people in the world?
Behavioral and Brain Sciences, 33(2–3), 61–83.
Holtzman, N. S., Schott, J. P., Jones, M. N., Balota, D. A., & Yarkoni, T. (2011).
Exploring media bias with semantic analysis tools: Validation of the Contrast Analysis
of Semantic Similarity (CASS). Behavior Research Methods, 43, 193–200.
Kahneman, D., Krueger, A. B., Schkade, D. A., Schwarz, N., & Stone, A. A. (2004).
A survey method for characterizing daily life experience: The day reconstruction
method. Science, 306(5702), 1776–1780.
Kirkegaard, E. O. W., & Bjerrekær, J. D. (2016). The OkCupid dataset: A very large
public dataset of dating site users. Unpublished manuscript.
Kramer, A. D., Guillory, J. E., & Hancock, J. T. (2014). Experimental evidence of massive-
scale emotional contagion through social networks. Proceedings of the National Academy of
Sciences, 111(24), 8788–8790.
Lakens, D. (2013). Using a smartphone to measure heart rate changes during relived
happiness and anger. IEEE Transactions on Affective Computing, 4(2), 238–241.
Markey, P. M., & Markey, C. N. (2013). Seasonal variation in internet keyword searches:
A proxy assessment of sex mating behaviors. Archives of Sexual Behavior, 42(4), 515–521.
McAfee, A., & Brynjolfsson, E. (2012, October). Big Data: The management revolution.
Harvard Business Review. Retrieved from https://hbr.org/2012/10/big-data-the-
management-revolution/ar
Mehl, M. R., & Robbins, M. L. (2012). Naturalistic observation sampling: The
Electronically Activated Recorder (EAR). In M. R. Mehl & T. S. Conner (Eds.),
Handbook of research methods for studying daily life (pp. 176–192). New York: Guilford
Press.
Miller, G. (2012). The Smartphone Psychology Manifesto. Perspectives on Psychological
Science, 7, 221.
Morozov, E. (2013, November 11). Why we are allowed to hate Silicon Valley. Frankfurter
Allgemeine Feuilleton. Retrieved from www.faz.net/aktuell/feuilleton/debatten/the-
internet-ideology-why-we-are-allowed-to-hate-silicon-valley-12658406.html
Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitude to public opinion: A
dynamic theory of social impact. Psychological Review, 97(3), 362.
Nowak, A., & Vallacher, R. R. (1998). Dynamical social psychology. New York: Guilford
Press.
Nowak, A., Vallacher, R. R., Tesser, A., & Borkowski, W. (2000). Society of self: The
emergence of collective properties in self-structure. Psychological Review, 107, 39–61.
Park, G., Schwartz, H. A., Eichstaedt, J. C., Kern, M. L., Kosinski, M., Stillwell,
D. J., Ungar, L. H., & Seligman, M. E. (2015). Automatic personality assessment
through social media language. Journal of Personality and Social Psychology, 108(6), 934.
348  Serfass, Nowak, and Sherman

Rauthmann, J. F., Gallardo-Pujol, D., Guillaume, E. M., Todd, E., Nave, C. S., Sherman,
R. A., Ziegler, M., Jones, A. B., & Funder, D. C. (2014). The Situational Eight
DIAMONDS: A taxonomy of major dimensions of situation characteristics. Journal of
Personality and Social Psychology, 107(4), 677–718.
Reis, H. T. (2012). Why researchers should think “real-world”: A conceptual rationale.
In M. R. Mehl & T. S. Connor (Eds.), Handbook of research methods for studying daily life
(pp. 3–21). New York: Guilford.
Schwartz, H. A., Eichstaedt, J. C., Kern, M. L., Dziurzynski, L., Ramones, S. M., Agrawal,
M., Shah, A., Kosinski, M., Stilwell, D., Seligman, M. E. P., & Ungar, L. H. (2013).
Personality, gender, and age in the language of social media: The open-vocabulary
approach. PLoS One, 8(9), e73791
Sedgi, Amy (2014, February 4). Facebook: 10 years of social networking, in numbers.
The Guardian. Retrieved from www.theguardian.com/news/datablog/2014/feb/04/
facebook-in-numbers-statistics
Serfass, D. G., & Sherman, R.A. (2015). Situations in 140 characters: Assessing real-
world situations on Twitter. PLoS One, 10(11), e0143051. doi:10.1371/journal.
pone.0143051
Seward, Z. M. (2014, July 31). How many of Twitter’s active users are actually human?
Quartz. Retrieved from http://qz.com/242483/how-many-of-twitters-active-users-
are-actually-human/
Stephens-Davidowitz, S. (2015, July 4). Days of our digital lives. New York Times. Retrieved
from www.nytimes.com/2015/07/05/opinion/sunday/seth-stephens-davidowitz-days-
of-our-digital-lives.html?ref=topics&_r=0
Vallacher, R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 8, 73–99.
Vallacher, R. R., Read, S. J., & Nowak, A. (2002). The dynamical perspective in person-
ality and social psychology. Personality and Social Psychology Review, 6, 264–273.
Wald, R., Khoshgoftaar, T, M., Napolitano, A., & Sumner, C. (2012). Using Twitter con-
tent to predict psychopathy. In Machine Learning and Applications (ICMLA), 2012: 11th
IEEE International Conference on Machine Learning and Applications (Vol. 2, pp. 394–401).
Piscataway, NJ: IEEE. Retrieved from http://ieeexplore.ieee.org/abstract/document/
6406768/
Yarkoni, T. (2010). Personality in 100,000 words: A large-scale analysis of personality and
word use among bloggers. Journal of Research in Personality, 44, 363–373.
Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments
are more accurate than those made by humans. Proceedings of the National Academy of
Sciences, 112(4), 1036–1040.
Zimmer, M. (2016, May 14). OkCupid study reveals the perils of Big-Data science.
Wired. Retrieved from https://www.wired.com/2016/05/okcupid-study-reveals-
perils-big-data-science/
16
THE FUTURE OF COMPUTATIONAL
SOCIAL PSYCHOLOGY
Andrzej Nowak1 and Robin R. Vallacher

The contributions to this volume have each made a strong case for exploiting
the power of modern computer technology and advanced statistical techniques
to capture the dynamics and complexity of human experience. Traditional
experimental social psychology has created an important foundation for under-
standing how people think, feel, and interact with one another, of course, but
this approach could only take us so far. In recent years, the limitations and
problematic features of mainstream social psychological research have begun to
receive more attention than have the findings generated by such research—a
state of affairs that has generated both concern and defensiveness in the field.
Whether by coincidence or in response to this crisis of confidence, the compu-
tational approach has emerged as a viable path forward in the quest to create a
truly scientific social psychology.
In so doing, we not only get around the limitations of traditional approaches,
we also are equipped to address the defining features of mind and action that
were of central concern in the field’s early decades. Phenomena such as nonlin-
earity, intrinsic dynamics, self-organization, emergence, and reciprocal causality,
which were recognized implicitly, if not explicitly, by early theorists such as
Lewin (1936) and James (1890), have resurfaced as crucial considerations in the
computational approach. And in focusing on the expression of these phenomena
across the topical landscape of social psychology, there is renewed hope that an
integrated social psychology is within reach—an integration that is consistent
with the principles of dynamics and complexity that have been established in
other areas of science.
This rosy future is not a foregone conclusion. New ideas and research strate-
gies have punctuated the history of social psychology, many of them met with
350  Andrzej Nowak and Robin R. Vallacher

enthusiasm as the new way forward, only to falter in their attempt to provide
an integrated and nuanced template for the diversity of human experience.
Psychodynamics, social learning, general systems theory, and catastrophe theory,
for example, all have had their moments in the spotlight, but none has succeeded
in providing the unifying set of principles that the field strives to attain. This is
not to denigrate the importance of these frameworks—each has generated impor-
tant insights into the human condition—but rather to remind ourselves that big
ideas can fall short of their promise in providing coherence for a fragmented and
often contentious field.
Will it be different this time? Can the computational approach live up to its
promise as both the solution to the limitations of traditional social psychology
and the basis for integrating the diverse surface structure of the field? The con-
tributions highlighted in this book are cause for optimism, but they should be
looked upon as initial forays with unclear points of similarity and divergence.
This is to be expected in the early stages of a new perspective, of course, but it
also serves as a reminder that much remains to be done before computational
social psychology can be rightly embraced as the field’s source of integration and
a viable path forward.
Our aim in this concluding chapter is to speculate on the future of compu-
tational social psychology. We begin by noting recent advances in technology
that are likely to enhance the salience of computational models and make this
approach more accessible to lay people as well as psychologists. We then discuss
the advances to be expected in theory construction, especially the potential of
computational modeling to provide theoretical integration for the complexity
and nuance of human social experience. In a concluding section, we bring tradi-
tional social psychology back into the spotlight, noting its added value as the field
matures as a truly scientific endeavor.

Advances in Technology
The pace of innovation in technology has accelerated in recent decades, and this
growth shows no signs of diminishing. Several of these advances are poised to
refine the computational approach and extend its reach to existing and yet to be
defined topics in social psychology. Incorporating and building on these develop-
ments is likely to have tremendous payoff for both theory and research.

Computers and Software

In parallel with the development of computational social science (Conte &
Giardini, 2016; Conte et al., 2012), advances in computational social psychol-
ogy will be based on the increasing capacity of information and communication
technologies to acquire and handle large amounts of data. Computers are
 Future of Computational Social Psychology  351

becoming faster and more powerful, making them better equipped not only to
process large data sets, but also to merge such data sets for use in a single analysis.
These advances in technology should pave the way for the discovery of complex
patterns in diverse sources of data, and the construction of models that replicate
these patterns.
For the potential of advances in computing technology to be realized, soft-
ware must become increasingly powerful and sophisticated. The information
generated by text uploaded on the internet, communication over social media
(e.g., Twitter, Facebook), and queries on search engines provide rich sources of
data to be mined for insights into personal concerns, shared interests, and the
spread of social influence. Software designed for this purpose has been devel-
oped in recent years—for example, the Linguistic Inquiry Word Count algorithms
(Chung & Pennebaker, 2008)—and growth in this area is likely to accelerate in
the years to come. New computer methods, such as Deep Learning (LeCun,
Bengio, & Hinton, 2015), provide a way to find meaningful representations
from unstructured Big Data. New analytic algorithms can automatically extract
features from signals in different modalities (e.g., text, pictures, and sounds).
Speech recognition software can, with high accuracy, convert speech to writ-
ten text. Software can quite accurately recognize emotions from pictures and
movies. All of this makes data acquisition automatic. Meanwhile, the increasing
popularity of mobile devices—smart phones and tablets—creates the potential
for tracking and identifying patterns of movement. So, rather than being limited
to counting the frequency of specific semantic content expressed in electronic
media, for example, software on the horizon will be able to identify how com-
munication patterns are related to the extent and rate of people’s movement in
physical space.
In like manner, programs to simulate social processes will become increasingly
sophisticated, to take advantage of the increasing power of computers. Computer
modeling packages are being developed for the social sciences that simulate com-
plex phenomena on many timescales, and these can be adapted to topics in social
psychology. The standardization and availability of such packages means that
implementing increasingly sophistical theoretical models in a computer simula-
tion will ironically require less programming expertise and low-level technical
skills, allowing social psychologists without such training to focus on how abstract
concepts and principles are expressed.

Online Information and Communication

Communication and information technology have dramatically expanded the
range and availability of data of interest to social psychology. The internet, social
networks, and mobile devices are especially likely to play important roles in the
development of computational models of social processes.
352  Andrzej Nowak and Robin R. Vallacher

The Internet
The internet has become an important environment in which social life takes
place. Interaction over social media, in particular, has become increasingly
important in recent years, rivaling the importance of face-to-face interactions
in the real world. Social media communication clearly provides direct insight
into individuals’ personal lives. Thus, researchers can mine social media data to
investigate people’s moods, preferences, and modes of self-presentation (e.g.,
modesty versus boastfulness). And the recent advent of Big Data (Chapter 15 in
this volume) has enabled researchers to emphasize descriptive rather than induc-
tive approaches to identify processes and patterns in human interaction. Whereas
laboratory research in social psychology must rely on relatively small convenience
samples (most often, college students), the Big Data approach focuses on larger
and more representative samples—sometimes approaching entire populations.
With such large samples, any observed relationship can essentially be considered
“statistically significant,” obviating the need to employ inferential statistics.
The internet has also become a popular forum for playing n-person games.
The behavior of individuals in such games can provide insight into a variety of
social psychological phenomena, including strategic decision-making, collabora-
tion, competition, trust, defection, and aggression. This source of data concerning
social processes is destined to increase in the years to come, thus making available
the online gaming records of thousands of people from different cultures and
with different demographic backgrounds.
It should be noted that the internet can also be employed to conduct
traditional social psychological research. Rather than coming to a laboratory, par-
ticipants can take part in experiments and complete surveys online. Survey Monkey
and Qualtrics for example, are widely utilized platforms for measuring people’s
attitudes, values, reactions to news events, personal preferences, and personality
traits. Several widely used computer-administered assessment devices, such as the
Implicit Association Test, have been implemented in online versions. Especially
popular among social psychologists is the service of Mechanical Turk, where volun-
teers, for a small compensation, complete questionnaires or participate in online
experiments.

Social Networks
All interactions on social media leave a trace, which greatly expands the capac-
ity to investigate the structure and dynamics of social networks. Instead of being
limited to tracing friendships structures in a classroom (e.g., Moreno, 1934),
researchers can look for the connections among people in an entire country, based
on records of calls and text messages (e.g., Onnela et al., 2007). Analyses exploit-
ing the terabytes—or even petabytes—of such data have already yielded new
insights into social networks. It has been shown, for example, that communication
 Future of Computational Social Psychology  353

patterns are not uniformly distributed in time, but rather occur in bursts, such
that different subsets of a network are activated at a given time (Barabási, 2010;
Karsai et al., 2011). Large-scale social networks, meanwhile, can be constructed
as association networks, where individuals are connected if they belong, or have
belonged, to the same entity. Two individuals are connected, for example, if they
are both members of the same supervisory board or if they went together on a trip.
Perhaps most noteworthy, analyses of massive data sets have revealed that
social networks have several common properties, including scale-free distribution,
small-world structure, homophile, and giant component. Scale-free distribution means that
that the distribution of the number of links in a network is highly skewed and
follows a Pareto distribution, such that most individuals have a small number
of connections and very few individuals have a disproportionally high num-
ber of links. In a small world, whereas most social links of an individual connect
to his or her neighbors, a few links connect the individual to distant others.
Because of these distant links, the average distance between any two individuals
is short, consisting of very few links (Watts, 1999). Homophile refers to the fact
that an individual tends to be connected to others who are similar with respect
to demographic factors, such as age, education, and income, as well as personal-
ity characteristics and values. Giant component means that all the individuals in a
community are connected to each other, forming one big network, rather than
several disconnected networks.

Mobile Devices
Mobile technology is becoming a dominant means of interaction. This emerg-
ing reality is relevant to social psychology because mobile phones are routinely
equipped with a variety of sensors, such as a geo-location sensor for GPS, a
compass, an acceleration sensor that can precisely detect movement, a light
sensor, two cameras allowing one to see and be seen by others, and as many as
three microphones that can collectively detect the direction of a sound. These
devices can thus collect rich data not only about the communication of the user,
but also about many aspects of his or her behavior and the situation in which the
behavior occurs. For example, based on the data from a set of microphones and
sensors, one can recognize with high accuracy the type of activity a person is
engaged in (Ward, Lukowicz, Tröster, & Starner, 2006). As another example, data
from mobile phones can be used to detect an individual’s chronic or momentary
mood, such as depression (Grünerbl et al., 2015).
Because modern mobile phones are essentially small computers, various sensors
(e.g., electroencephalogram) can be connected to them. And wearable devices
that collect sensory and physiological data, such as Google Glasses or sport brace-
lets, can transmit such data to mobile phones. This clearly amplifies the ability of
mobile technology to collect and integrate rich geo-located personal data con-
cerning the user. Current health applications, for example, not only automatically
354  Andrzej Nowak and Robin R. Vallacher

measure a person’s physical activity such as the number of steps he or she has
taken, but also ask the user to input data such as his or her current mood. Other
applications ask the person to make ratings of objects and events on a number of
dimensions and to judge other aspects of his or her experience. User-provided
data, combined with automatically collected data, can generate a precise and
psychologically meaningful stream of information concerning the user and his or
her current environment (e.g., Lukowicz, Kirstein, & Tröster, 2004).
Mobile devices are increasingly employed in psychological research. Because
they have the capabilities of computers but can be carried around by people in
their daily lives, such devices (e.g., smart phones) can be used to record people’s
actions and subjective experience in natural settings rather than in artificial labo-
ratory environments. This approach is proving useful, for example, for gaining
insight into the perception of situations in daily life and the influence of situ-
ational factors in shaping how people think and behave (Sherman, Rauthmann,
Brown, Serfass, & Jones, 2015). And because mobile devices enable people to be
contacted at different points in time, they are well suited to generate time series
data that can be analyzed for dynamic properties in action and subjective experi-
ence (e.g., fixed-point attractors, periodic structure, ruptures in routine due to
perturbing events). This avenue of research remains unexplored at present, but
is likely to become a major focus in light of the emerging interest in the intrinsic
dynamics of psychological processes (Vallacher, Van Geert, & Nowak, 2015).

Cameras and Sensors

Cameras and sensors are becoming ubiquitous in daily life, whether placed strate-
gically at busy intersections in a metropolitan area or worn by individuals as they
move about in the course of their everyday activities. They collect a steady stream
of images and other types of sensory information (e.g., sound, temperature), pro-
viding a continuous record of individual and group experiences. A light sensor,
for example, can detect the presence and movement of people, since a passing
person casts a shadow that is detected by the device. The utility of cameras and
sensors is limited only by our ability to make sense of the voluminous data they
generate.
The amount and quality of data will be amplified in the near future with the
rapidly developing implementation of the Internet of Things (IoT). This refers
to the network of physical devices, vehicles, and other items that are embed-
ded with sensors, software, and network connectivity, enabling these objects
to collect and exchange data. The IoT can be looked upon as the infrastruc-
ture of the information society (Howard, 2015), in that it allows objects to be
sensed and remotely controlled across existing networks. Devices in the house
or in offices, for example, can sense events in their environment, react to these
events, and even communicate with one another. They can also be designed
to communicate with users and to detect the intentions of nearby individuals.
 Future of Computational Social Psychology  355

It is estimated that the IoT will consist of close to 50 billion objects by 2020
(Evans, 2011), making this a potential gold mine for data collection and inter-
pretation regarding behavior as it is embedded in everyday contexts.
To date, the IoT has been primarily employed to enhance the efficiency of
physical systems, as in the creation of smart electrical grids, washers and dryers
that use wifi for remote monitoring, and energy management systems. One can
envision, however, the exploitation of IoT technology to facilitate human inter-
action by means of a common link to an IoT device. Whether the nature of
human interaction will be qualitatively changed by such developments or simply
extended in reach remains to be seen.

Technological Advances in Perspective

The availability of modern technology clearly has added value for research, but
some pause is warranted before embracing it wholeheartedly. The internet, social
media, and the use of cameras and sensors all provide an abundance of ecologi-
cally valid (“real-world”) information about individuals and their social lives, but
the collection and use of such information raises two concerns—one practical, the
other ethical.
The practical issue stems, ironically, from the rich abundance of information
that can be generated with modern technology. For one thing, the exponential
growth of the amount of data generated and collected on a daily basis has ren-
dered traditional techniques of data gathering, storage, and analysis inadequate,
if not completely obsolete. Beyond the data management problem, the massive
amounts of data available for analysis can make it difficult to determine whether
observed patterns in the data are meaningful. With enough information at one’s
disposal, one can always find patterns—the trick is to distinguish the “signal”
from the “noise” (Silver, 2012). This problem is analogous to the persistent pop-
ularity of conspiracy theories. If one believes the terrorist attacks on September
11, 2001, were really an “inside job” coordinated by the Bush administration,
for example, the information relevant to that event is so enormous that one can
find a way to “connect the dots” and maintain that belief. The advanced statisti-
cal methods employed in Big Data research get around this problem for the most
part, but there is always a risk of identifying a pattern that is illusory rather than
real. The future is certain to see the development of increasingly sophisticated
means of discerning true patterns in large, interconnected data sets.
The ethical issue is not as easy to resolve, as it centers on subjective stand-
ards of privacy and personal autonomy. We live in an increasingly transparent
world where every personal preference, attitude, and social interaction is open to
inspection by those who have not been given explicit permission to collect and
collate such information. Mobile phones track our every move and every contact
we make. The content on most social networking sites, meanwhile, is penetrated
by crawlers or sold by providers. And the true value of most mobile apps in not
356  Andrzej Nowak and Robin R. Vallacher

in the service they provide, but in the data they collect. Often the apps are free
and the developers make profits by selling the data. A flashlight app for mobile
phones, for example, asks for access to the person’s list of contacts, contact dates,
stored photos, and texts. None of this information is needed to turn on the light
in the phone; the company providing the app for “free” makes money by selling
the information to marketing companies.
Privacy concerns are not confined to the collection and use of personal data by
hackers and those with profit motives. The myriad sources of information about
people’s personal and interpersonal lives raise ethical concerns regarding the
collection and use of such information in the name of science. Ethical considera-
tions are salient enough in laboratory studies for which participants volunteer;
consider the amplification of such considerations in the age of social media,
Twitter feeds, and internet searches by people who are unaware that their behav-
ior is likely to be scrutinized by total strangers. Although the mining of Big Data
is designed to identify patterns that reflect generic social processes, it can also
be used to identify patterns in an individual’s behavior, preferences, and atti-
tudes, and do so without the individual’s knowledge, let alone his or her consent.
Facebook posts, for example, reveal a great deal of information about people’s
personal lives and their social relationships, and such information can be traced
to all the individuals in a person’s social network. The use of cameras in public
places, meanwhile, can generate important insight into social interaction and the
synchronization of movement among large numbers of people, but cameras can
also be used to identify individuals—and do so without their permission.
It is important to emphasize that the collection and use of personal data by
those with commercial or nefarious interests presents a greater threat to privacy
than does the mining of such data by scientists. In order to collect and analyze
data obtained from social media, cameras and sensors, and internet searchers,
researchers must obtain approval from the Internal Review Boards (IRBs) at
their institutions. IRBs are charged with protecting the welfare of the participants
in research, and have developed clear standards that must be met to justify the
collection of personal data. It is standard practice in science to use anonymized
data, for example, so that the identities of “participants” are not known to the
researchers. Nonetheless, the trade-off between scientific merit and poten-
tial harm can prove tricky to determine, and is often a source of contention.
Resolutions to the ethical dilemmas posed by Big Data must continue to evolve
as this branch of computational social psychology continues its ascendance as an
accepted approach to exploring the social dynamics of everyday life.

Advances in Theory
The enormous complexity of human thought and behavior presents a daunting
challenge to the construction of a parsimonious and generalizable theoreti-
cal account of human experience. After all, if complexity is assumed to reflect
 Future of Computational Social Psychology  357

complex interactions among a large number of variables, with the salience and
magnitude of these variables changing on different timescales in the absence of
external forces, how can one hope to develop an account that does not incorpo-
rate such complexity and associated dynamism into its principles? This is a serious
conundrum for traditional social psychology, which of necessity isolates a small
number of variables in an artificial setting stripped of the ecological context in
which they naturally occur, and measures the effect of these variables on a single
variable of interest at one point in time.
As noted in Chapter 1 of this volume, although early theorists were quite
sensitive to the complexity, embeddedness, and dynamism of psychological pro-
cesses, the methods and tools were not available at that time to construct theory
and conduct psychological research in these terms. In effect, experimental social
psychology had little choice but to focus its efforts on cause–effect relations and
one-step processes, and to do so in laboratory settings—in effect, investigating
thought and behavior “where the light is better.” With the advent of compu-
tational approaches, we are in a position to recapture and explore the insights
that launched social psychology in the first place. Because computational social
psychology is in its nascent stage, however, it is not clear what will eventually
emerge as a result of this paradigmatic shift in research—but this does not stop us
from speculating.

Dynamics
Even in the most stable setting, a person’s thoughts, feelings, and actions are never
static, but undergo constant change on various timescales. This defining feature
of human experience was recognized by the founding fathers of social psychology
and laid the theoretical foundation for the field. James (1890) coined the term
“stream of consciousness” to capture the ever-changing nature of mental process,
with thoughts, images, and feelings succeeding each other on a moment-to-
moment basis. Cooley (1902) stressed the constant press for action, even in the
absence of external forces. Lewin’s psychological field theory (1936) emphasized
the constant interplay of internal and external forces that gave rise to change
and stability at the intrapersonal and interpersonal level. Asch (1946) theorized
about the dynamic interplay of thoughts and feelings that give rise to unique
Gestalts that cannot be reduced to the component mental elements. Krech and
Crutchfield (1948) portrayed social experience as the constant reconfiguration
of thought and action in response to a conflicting field of forces in everyday life.
Despite this focus on the dynamic underpinnings of human experience, the
lack of adequate theoretical and methodological tools at the time rendered
the dynamical approach unproductive. Subsequent theory and research tended
instead to concentrate on the static aspects of thought and behavior, with
dynamism essentially reduced to a one-step process. In the absence of external
forces that presumably “caused” people to change the way they think and act,
358  Andrzej Nowak and Robin R. Vallacher

variability over time in a phenomenon of interest was considered noise, and

relegated to error variance in statistical analyses. The “power of the situation”
reigned supreme as the guiding assumption in theory and research.
The 1990s saw renewed appreciation of the dynamic nature of human expe-
rience, largely due to rapid advances in computational tools appropriate for
studying human experience. Dynamical social psychology (Nowak & Vallacher,
1998; Vallacher & Nowak, 1994a, 1997; Vallacher, Read, & Nowak, 2002)
provided an explicitly dynamical framework for rethinking the topical landscape
of the field, and showcased methods for investigating the interaction of external
factors and intrinsic dynamics in shaping intrapersonal, interpersonal, and col-
lective processes. Computer simulations (e.g., Nowak, Szamrej, and Latané,
1990; Nowak, Vallacher, Tesser, & Borkowski, 2000) and new empirical tools
such as the mouse paradigm (Vallacher, Nowak, & Kaufman, 1994; Vallacher,
Nowak, Froehlich, & Rockloff, 2002) made dynamic properties tractable and
subject to empirical investigation. During this period, too, connectionist mod-
els were developed to investigate the dynamics of social thinking and behavior
(Read & Miller, 1998; Smith, 1996).
This redirection of the field—from a focus on external forces and one-step
processes to the investigation of the interplay of context and intrinsic dynamics—
is likely to become increasingly prominent in the future. As computational tools
become a mainstream tool familiar to succeeding generations of social psycholo-
gists, existing theories will be reframed to accommodate dynamic properties, and
new theories will be developed with such properties at their core—much as the
field’s early theorists had in mind.

Differentiation
As evident in the preceding chapters, computational social psychology is not
a single approach, but rather represents a broad umbrella for several distinct
research strategies, each with its own set of methods and tools. This is appropri-
ate, even necessary, at this point in the development of this perspective. In fact,
we can expect the computational paradigm to become more differentiated in the
near future, with different strategies being developed to exploit the potential of
computer and information technology.
In part, this differentiation in research strategies will mirror the differentia-
tion of the topical landscape of social psychology. Big Data, mobile technology,
computer simulations, and time series analyses are each better suited for some
concerns than for others. The study of social networks, for example, will advance
along with developments in the mining of interactions over social media, while
research on societal dynamics and culture will continue to rely on computer
simulations of formal models, and research agendas concerning interpersonal
coordination and the dynamics of self-evaluation are likely to progress with an
emphasis on the dynamic properties inherent in time series data. Within each
 Future of Computational Social Psychology  359

approach, the methods and tools are likely to become increasingly fine-tuned
to the subject matter at hand, enabling enhanced precision in data gathering and
greater generalizability of the results obtained.
The differentiation of computational strategies, moreover, is likely to promote
collaborations between these strategies and other scientific disciplines. Researchers
employing computational approaches to cultural evolution, for example, may
forge ties with researchers in evolutionary biology, anthropology, and sociology.
Those who develop computational models of interpersonal synchronization,
meanwhile, may find increasingly common ground with those who study cog-
nitive neuroscience or collective dynamics in biological and physical systems.
Social psychology has been described as a “hub” science (Cacioppo, 2007), with
links to other disciplines, providing a foundation for such disparate fields as soci-
ology, education, sports, business, criminal justice, and international relations.
The ascendance of the computational approach is likely to amplify the central
role of social processes in understanding otherwise disconnected facets of human
experience.

Theoretical Integration
Despite the prospect of increased differentiation of specific research strategies
in the future, the computational perspective is likely to bring about increasing
theoretical integration to a field sorely in need of such efforts. Traditional social
psychology has done much to illuminate the mechanisms associated with a wide
array of topics, from self-concept to intergroup relations. In so doing, how-
ever, the field has become increasingly fragmented with few unifying principles
(Vallacher & Nowak, 1994b). This fragmentation is evident in four forms: the
topical landscape of the field, with each topic developing in relative isolation
from other topics; the sealing off of different levels of psychological reality, with
little or no consideration of how macro-level phenomena are related to micro-
level processes; the unclear relation between social psychology and other human
sciences; and the independence of social psychology from other fields of science.
In each case, there is reason to think that the computational approach can create
a foundation that allows for the missing integration.

Topical Integration
The diverse topical landscape of social psychology is mirrored in the diversity of
theoretical concepts that have been generated over the years. Established princi-
ples of attitude change, for example, have little intersection with the principles said
to underlie self-concept change or the dissolution of close relationships. Because
the computational approach emphasizes formal principles that characterize the
expression of social phenomena generally, it holds potential for reframing seem-
ingly distinct topics in the same terms. These foundational principles derive in
360  Andrzej Nowak and Robin R. Vallacher

large part from complexity science and dynamical systems theory (e.g., Schuster,
1984; Strogatz, 1994). In recent years, these principles have been adapted to the
subject matter of social psychology (Guastello, Koopmans, & Pincus, 2009; Read &
Miller, 1998; Vallacher & Nowak, 1997; Vallacher, Read, & Nowak, 2002),
creating the potential for framing different topics in the same language. The
formation of attitudes, self-concepts, and close relationships, for example, can
each be understood in terms of self-organization, the emergence of higher-order
structures, and the resistance of higher-order structures to disruption. Change
in each case, meanwhile, can be understood as the disassembly of higher-order
structures into their respective lower-level elements.
Progress toward topical integration is likely to accelerate with the increasing
use of computer simulations to reveal the expression of psychological processes
over relevant timescales (e.g., Liebrand, Nowak, & Hegselman, 1998; Read &
Miller, 1998). Natural language is useful for expressing the nature of a process,
of course, but words and metaphors lack the precision of formal rules that can be
investigated for their long-term consequences or for their generality across dif-
ferent domains. The translation of verbal theories into computational algorithms
provides a common platform for comparing and hopefully integrating different
topics. Two processes may have very different surface features, but the underly-
ing rules leading to the emergence of their properties may be the same. Public
opinion—a collective process—and self-concept—a personal process—are clearly
different phenomena, for example, but research has established that remarkably
similar rules underlie the emergence of both (see Nowak et al., 1990, 2000).
One might think that the enormous complexity associated with each fact of
human experience would necessitate an equally complex computer simulation
model in order to capture its nuance. Even with the aid of computer technology,
then, one must develop a complex theory with many rules, and wind up with an
explanation that is as complex and nuanced as the phenomenon one is trying to
explain. This is not the case, however. The study of nonlinear dynamical systems
has shown, in fact, that extremely complex phenomena can be understood with
recourse to very few simple rules. This approach is referred to as dynamical mini-
malism (Nowak, 2004). It is minimalist because it attempts to identify the most
basic rules governing a process; it is dynamical because it assumes that the process
evolves in time due to the interaction among the basic rules. In effect, dynami-
cal minimalism achieves parsimony in theory construction without stripping the
process of its complexity and nuance.

Developing Links between Micro- and

Macro-Level Phenomena
We know intuitively that there are links between different levels of social reality.
Group dynamics is clearly different from the thought or behavior of any of the
individuals in the group, but the former would not be possible without the
 Future of Computational Social Psychology  361

latter. Establishing the nature of such linkage, however, has proven to be a chal-
lenge for social psychology, so much so that different levels of social reality are
usually treated as though they had nothing in common, and thus are discussed
in separate journals, in separate sections of a journal, and in separate chapters in
a social psychology text. If attempts are made to relate micro and macro phe-
nomena, reductionism is commonly assumed, such that the properties at the
macro level reflect analogous properties of the elements comprising the micro
level. To explain frustration and anti-social behavior in a group, for example,
a theorist might reduce this phenomenon to frustration and readiness for anti-
social behavior among the individuals in the group.
The computational perspective is explicitly concerned with the link between
micro and macro phenomena, but does so without assuming reductionism. The
focus instead is emergence, a central concept in the study of dynamical systems
(e.g., Holland, 1995; Johnson, 2001; Vallacher & Nowak, 1997). The idea that
“the whole is more than the sum of its parts”—that something new somehow
emerges at a global level from the basic components of a system—seems mys-
terious, if not implausible, but it has been repeatedly confirmed in computer
simulations based on the dynamical minimalism approach. The researcher speci-
fies the properties of the elements at the micro level, and the rules of interaction
among these elements. As the elements interact over time, properties appear at
the macro level that were not programmed—or sometimes not even imagined—
at the micro level. In effect, a theory constructed at a basic level of psychological
reality can be investigated for its consequences at a higher level of psychologi-
cal reality. Simple rules concerning the interaction of basic and often mutually
contradictory thoughts about oneself that arise in the stream of thought give
rise to global properties of the self-concept such as self-esteem and self-concept
certainty (Nowak et al., 2000), for example, and simple rules of social interac-
tion among individuals promote the emergence of public opinion and customs
(Nowak et al., 1990).
Computer simulations within the dynamical minimalism approach are also
useful in distinguishing between properties that are critical to the emergence
of higher-order processes and those that are trivial and can be ignored in con-
structing a theory concerning micro–macro linkages. So although psychological
processes are complex and involve a wide variety of factors, it is not necessary
to incorporate all—or even many—of them in a theory concerning a given pro-
cess or the emergent consequences of the process. Such distinctions may not
be obvious when attempting to explain a phenomenon, so one runs the risk of
generating an incomplete account if certain factors are omitted for the sake of
designing a manageable study within traditional social psychology.
In the computer simulation approach, however, a researcher can systematically
vary the properties of basic factors and their interactions, and then observe which
properties and factors generate meaningful changes at the macro level. Those
properties that have trivial consequences can then be eliminated from the model.
362  Andrzej Nowak and Robin R. Vallacher

In this way, computer simulations enable one to distill the minimal set of
features—the elements and rules of interaction among them—that are neces-
sary to capture the essence of a process. A simple explanation can therefore be
developed and verified without forfeiting a nuanced and comprehensive under-
standing of the topic under consideration.

Establishing Connections with Other Human Sciences

As noted earlier, a strong case can be made that social psychology provides a
meaningful hub for other human sciences (Cacioppo, 2007). Against this back-
drop, it is ironic that many scholars in these disciplines have begun to argue that
there is no need for social psychology, and that it has outlived its usefulness and
relevance as a distinct discipline. It is noteworthy that this disturbing argument
is coming from those in the social sciences who have adopted the computational
paradigm (Conte et al., 2012). Their point is that with its emphasis on cause–
effect relations investigated in laboratory as opposed to real-world settings and
its reliance on statistical inference from small convenience samples as opposed to
population description, social psychology is essentially a trivial pursuit that does
not contribute meaningful and ecologically valid understanding.
The adoption of the computational perspective blunts this dismissive judg-
ment of what social psychology has to offer. When researchers are no longer
bound by the constraints of experimental design, the focus on unidirectional
cause–effect relations, and statistical inference from small convenience samples,
they can develop and test models that reflect the complexity, nonlinearity, and
dynamic patterns of human behavior in ecologically valid contexts. And because
social psychology focuses on the individual, computational models can identify
the mechanisms of thought and action that mediate or moderate the social pro-
cesses identified by computational researchers in economics, political science,
and other social sciences. The computational approach, then, is not merely an
adjunct discipline for computational science, but rather provides unique insight
into the mutual feedback between individual and collective processes.

Integration with Other Areas of Science

Computational social psychology has a set of methods that are unique to human
experience, but its conceptual foundation reflects assumptions and principles
that are common to other areas of science. Like models of physical phenomena,
the various models presented in the preceding chapters emphasize the complex-
ity and dynamism underlying social phenomena of all kind, from those that are
intrapersonal in nature to those that are interpersonal and collective. Indeed, the
specific features of complexity and dynamism at work in computational mod-
els correspond with the features of phenomena in such disparate fields as physics,
chemistry, and biology. Thus, human dynamics are increasingly framed in
 Future of Computational Social Psychology  363

terms of such notions as self-organization, emergence, attractor dynamics, and

synchronization—notions that are central to everything in nature from sand piles
to galaxy formation.
Of course, the subject matter of social psychology is decidedly different from
that of the physical sciences. People are not atoms or gravitons. But the principles
by which individuals develop attitudes or form social groups are similar in formal
terms to the principles observed throughout the known universe. By viewing
the topical landscape of social psychology as complex systems governed by basic
processes by which the world operates, the potential for developing an integra-
tive theory of human experience becomes a realistic (though yet unattained) goal.
It is ironic that a true appreciation of human uniqueness ultimately derives from
what humans have in common with everything else in nature.

The Future of Social Psychology

We hasten to add that the computational approach does not supplant the traditional
experimental approach that has been the mainstay of social psychology through-
out the field’s history. To the contrary, a comprehensive understanding of human
experience will entail a synergistic relationship between computational modeling
and experimental methods. Big Data, for example, cannot directly test the causal
factors responsible for the patterns that are discovered in large data sets. Knowing
that a high volume of Twitter feeds devoted to a topic promotes a change in pub-
lic opinion concerning the topic does not tell us why this change occurs. Perhaps
awareness of the high volume creates a bandwagon effect, with everyone wanting
to be part of the conversation or to feel solidarity with others, or perhaps the con-
tent of the Tweets has information value that clarifies the meaning of the topic for
those who previously did not know what to think. Research conducted in social
psychology labs can help to decide among these alternative mechanisms.
Beyond that, traditional social psychology has an important role to play in
theory construction. This role is especially apparent in the use of computer
simulations to model psychological processes. We noted earlier that describing
psychological processes in terms of formal rules affords greater precision than does
a verbal description that reflects intuition or metaphors. This is true, but it is also
true that the rules must come from somewhere. Before a researcher can specify in
formal terms the ways in which a process unfolds, he or she must have a guiding
vision and set of assumptions about what the process looks like. Social psycholo-
gists have generated a rich repertoire of basic visions of what shapes and drives
human behavior, and research has functioned to verify a handful of these guiding
visions, while eliminating others. The verbal descriptions that have survived this
selection process motivate the attempt to identify the rules that enable the process
in question to occur.
Consider, for example, one of the most inspiring metaphors of all: James’ (1890)
stream of consciousness.This enduring image of mental function was not articulated
364  Andrzej Nowak and Robin R. Vallacher

in formal terms by James, but its ring of truth has provided the inspiration for such
depictions in recent years (e.g., Nowak et al., 2000). It is one thing to recognize that
thoughts tend to unfold over time without the need for external forces or triggers,
but how and why does this happen? In Nowak and colleagues’ (2000) compu-
tational model, mental process is driven in part by a press for integration, so that
independent thoughts come to acquire common higher-order meaning (e.g., evalu-
ation) as they co-occur in thinking.Thoughts that differ in initial meaning influence
one another to achieve a shared meaning, and thoughts that have similar meaning
tend to call each other to mind. The process of developing a common meaning for
one’s thoughts may involve repeated iterations occurring over long periods of time,
with momentary oscillation between conflicting coherent states if the thoughts have
contradictory meaning (Vallacher, Nowak et al., 2002). This scenario ensures that
mental process is dynamic rather than static.
It is also the case that computer simulations alone may not be sufficient
to verify a theoretical model. The use of computer simulations is critical in
identifying the properties that are central to the model and investigating the
consequences of these properties for the higher-order functioning of the sys-
tem under investigation. Once these consequences have been identified, they
provide the basis for hypotheses that need to be tested in empirical research.
The comparison of patterns inherent in experimental data and produced by
computer simulation of a model provides a new means of verifying a theory.
Sometimes the relation between computer simulations and empirical verifica-
tion is reversed. Thus, data collected in laboratory studies can suggest the basic
properties and rules of a model, which are then implemented in a computer
model to test whether the anticipated emergent properties are observed. The
results of such simulations, in turn, can provide the bases for new hypotheses
to be tested empirically, and so on, in an iterated approach that holds potential
for generating progressive understanding of the phenomenon in question. The
reciprocal feedback over time between computer simulations and empirical data
is central to theory construction in computational social psychology.
The rapidly growing field of computational social science (Conte et al., 2012;
Gilbert, 2007) provides an important context for the development of computa-
tional social psychology. This interdisciplinary field, which is populated more by
computer scientists and physicists than by social scientists, embraces the subject
matter of social psychology, but does so without the benefit of social psycho-
logical theory. Indeed, this field is largely atheoretical, focusing more on the
description of social processes based on very large data sets and sophisticated
agent-based models than on the factors mediating the patterns that are revealed.
So, although computational social science is currently more advanced in its
methods, it can benefit from the explicit theoretical focus that has defined social
psychology throughout its history.
It is noteworthy that computational social science began with a focus on agent-
based models (Conte et  al., 2012; Epstein, 2006; Gilbert, 2007; Nowak, 2009),
 Future of Computational Social Psychology  365

concentrating on the emergence of group properties from individual behavior

and the self-organization of groups and societies. The focus shifted over time
to social networks, however, largely due to the availability of large-scale data
on communication on the internet and with mobile phones (Lazer et  al.,
2009; Wyatt, Choudhury, Bilmes, & Kitts, 2011). The current focus is on data
analysis, mostly provided by Big Data (Alvarez, 2016; Wallach, 2016). This
progression provides a road map for the likely evolution of computational social
psychology. Although agent-based models provide the predominant focus at
present—as indicated by the majority of the chapters in this volume—there
is increasing attention devoted to social network analyses, communication
patterns over the internet and electronic media, and Big Data. The future
is likely to see an acceleration in this trajectory, with greater overlap in the
respective research methods and concerns of computational social science and
computational social psychology. The background in theory provided by social
psychology, meanwhile, should facilitate new and deep insight into the nature
of human experience as these two approaches utilize common methods to focus
on issues of shared interest.
In sum, the computational approach does not replace canonical approaches in
social psychology, but rather provides a critical complement to traditional means
of generating and testing insight into human experience. Indeed, all the methods
available to social psychologists—experiment, field study, computer simulation,
Big Data, model fitting to time series data—are optimally viewed as having posi-
tive feedback loops, with the results of one approach providing the inspiration
and hypotheses for another approach. Because the computational perspective is
relatively new, it is competing for attention in the research community. Once
it has secured its rightful place in the conduct of social psychological science, its
reciprocal connection to traditional approaches will be appreciated and acted
upon. Accordingly—and with an optimistic mindset—we can expect to reap the
benefits of this budding relationship in the years to come.

Note
1 Andrzej Nowak acknowledges the support from the grant of Polish Committee for
Scientific Research (DEC-2011/02/A/HS6/00231).

References
Alvarez, R. M. (Ed.) (2016). Computational social science. Cambridge: Cambridge University
Press.
Asch, S. E. (1946). Forming impressions of personalities. Journal of Abnormal and Social
Psychology, 41, 258–290.
Barabási, A. L. (2010). Bursts: The hidden patterns behind everything we do, from your e-mail to
bloody crusades. New York: Penguin.
Cacioppo, J. T. (2007). Psychology is a hub science. APS Observer, 20, 1–3.
366  Andrzej Nowak and Robin R. Vallacher

Chung, C. K., & Pennebaker, J. W. (2008). Revealing dimensions of thinking in

open-ended self-descriptions: An automated meaning extraction method for natural
language. Journal of Research in Personality, 42, 96–132.
Conte, R., & Giardini, F. (2016). Towards computational and behavioral social science.
European Psychologist, 21(3), 131–140.
Conte, R., Gilbert, N., Bonelli, G., Cioffi-Revilla, C., Deffuant, G., Kertesz, J., Loreto, V.,
Moat S., Nadal, J. P., Sanchez, A., Nowak, A., Flache, A., San Miguel, M., & Helbing, D.
(2012). Manifesto of Computational Social Science. European Physical Journal Special
Topics, 214(1), 325–346.
Cooley, C. H. (1902). Human nature and the social order. New York: Scribner.
Epstein, J. M. (2006). Generative social science: Studies in agent-based computational modeling.
Princeton, NJ: Princeton University Press.
Evans, D. (2011). The Internet of Things: How the next evolution of the internet is changing
everything. White paper, Cisco Internet Business Solutions Group.
Gilbert, N. (2007). Computational social science: Agent-based social simulation. In
D. Phan & F. Amblard (Eds.), Agent-based modelling and simulation (pp. 115–134).
Oxford: Bardwell.
Grünerbl, A., Muaremi, A., Osmani, V., Bahle, G., Oehler, S., Tröster, G., Mayora, O.,
Haring, C., & Lukowicz, P. (2015). Smartphone-based recognition of states and state
changes in bipolar disorder patients. IEEE Journal of Biomedical and Health Informatics,
19(1), 140–148.
Guastello, S., Koopmans, M., & Pincus, D. (Eds.) (2009). Chaos and complexity in psychology:
The theory of nonlinear dynamical systems. New York: Cambridge University Press.
Holland, J. H. (1995). Emergence: From chaos to order. Reading, MA: Addison-Wesley.
Howard, P. N. (2015). Pax Technia: How the Internet of Things may set us free, or lock us up.
New Haven, CT: Yale University Press.
James, W. (1890). Principles of psychology. New York: Holt.
Johnson, S. (2001). Emergence: The connected lives of ants, brains, cities, and software. New
York: Scribner.
Karsai, M., Kivelä, M., Pan, R. K., Kaski, K., Kertész, J., Barabási, A. L., & Saramäki, J.
(2011). Small but slow world: How network topology and burstiness slow down
spreading. Physical Review E, 83(2), 25–102.
Krech, D. & Crutchfield, R. S. (1948). Theory and problems of social psychology. New York:
McGraw-Hill.
Lazer, D., Pentland, A., Adamic, L., Aral, S., Barabási, A. L., Brewer, D., Christakis, N.,
Contractor, N., Fowler, J., Gutman, M., Jebara, T., King, G., Macy, M., Roy, D., &
Van Alstyne, N. (2009). Computational social science. Science, 323(5915), 721–723.
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444.
Lewin, K. (1936). Principles of topological psychology. New York: McGraw-Hill.
Liebrand, W., Nowak, A., & Hegselman, R. (Eds.) (1998). Computer modeling of social
processes. New York: SAGE.
Lukowicz, P., Kirstein, T., & Tröster, G. (2004). Wearable systems for health care
applications. Methods of Information in Medicine/Methodik der Information in der Medizin,
43(3), 232–238.
Moreno, J. L. (1934). Who shall survive? Beacon, NY: Beacon House.
Nowak, A. (2004). Dynamical minimalism: Why less is more in psychology. Personality
and Social Psychology Review, 8, 183–192.
Nowak, A. (Ed.) (2009). Applications physics and mathematics in social science. In R A. Meyers
(Series Ed.), Springer encyclopedia of complexity. Heidelberg: Springer-Verlag.
 Future of Computational Social Psychology  367

Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitude to public opinion: A
dynamic theory of social impact. Psychological Review, 97, 362–376.
Nowak, A., & Vallacher, R. R. (1998). Dynamical social psychology. New York: Guilford
Press.
Nowak, A., Vallacher, R. R., Tesser, A., & Borkowski, W. (2000). Society of self: The
emergence of collective properties in self-structure. Psychological Review, 107, 39–61.
Onnela, J. P., Saramäki, J., Hyvönen, J., Szabó, G., Lazer, D., Kaski, K., Kertesz, J., &
Barabási, A. L. (2007). Structure and tie strengths in mobile communication networks.
Proceedings of the National Academy of Sciences, 104(18), 7332–7336.
Read, S. J., & Miller, L. C. (Eds.) (1998). Connectionist models of social reasoning and social
behavior. Mahwah, NJ: Lawrence Erlbaum.
Schuster, H. G. (1984). Deterministic chaos. Vienna: Physik Verlag.
Sherman, R. A., Rauthmann, J. F., Brown, N. A., Serfass, D. G., & Jones, A. B. (2015).
The independent effects of personality and situations on real-time expressions of
behavior and emotion. Journal of Personality and Social Psychology, 109, 872–888.
Silver, N. (2012). The signal and the noise: Why so many predictions fail—but some don’t. New
York: Penguin.
Smith, E. R. (1996). What do connectionism and social psychology offer each other?
Journal of Personality and Social Psychology, 70, 893–912.
Strogatz, S. (1994). Nonlinear dynamics and chaos. Cambridge, MA: Perseus Books.
Vallacher, R. R., & Nowak, A. (Eds.) (1994a). Dynamical systems in social psychology. San
Diego, CA: Academic Press.
Vallacher, R. R., & Nowak, A. (1994b). The chaos in social psychology. In R. R. Vallacher &
A. Nowak (Eds.), Dynamical systems in social psychology (pp. 1–16). San Diego, CA: Academic
Press.
Vallacher, R. R., & Nowak, A. (1997). The emergence of dynamical social psychology.
Psychological Inquiry, 4, 73–99.
Vallacher, R. R., Nowak, A., Froehlich, M., & Rockloff, M. (2002). The dynamics of
self-evaluation. Personality and Social Psychology Review, 6, 370–379.
Vallacher, R. R., Nowak, A., & Kaufman, J. (1994). Intrinsic dynamics of social judgment.
Journal of Personality and Social Psychology, 67, 20–34.
Vallacher, R. R., Read, S. J., & Nowak, A. (Eds.) (2002). The dynamical perspective in
social psychology. Personality and Social Psychology Review, 6, 264–273.
Vallacher, R. R., Van Geert, P., & Nowak, A. (2015). The intrinsic dynamics of psycho-
logical process. Current Directions in Psychological Science, 24, 58–64.
Wallach, H. (2016). Computational social science: Discovery and prediction. Cambridge:
Cambridge University Press.
Ward, J. A., Lukowicz, P., Tröster, G., & Starner, T. E. (2006). Activity recognition of
assembly tasks using body-worn microphones and accelerometers. IEEE Transactions on
Pattern Analysis and Machine Intelligence, 28(10), 1553–1567.
Watts, D. (1999). Networks, dynamics, and the small-world phenomenon. American
Journal of Sociology, 105, 493–527.
Wyatt, D., Choudhury, T., Bilmes, J., & Kitts, J. A. (2011). Inferring colocation and
conversation networks from privacy-sensitive audio with implications for computa-
tional social science. ACM Transactions on Intelligent Systems and Technology, 2(1), 7.
INDEX

Abelson, R. P. 285 Allport, Floyd 258

ABM see agent-based modeling anchoring 187
absorbing states 287–288, 290 animals 190–191, 212, 322, 325
abstraction, levels of 55 Approach system: attachment neural
ACT-R 100 network 70–71; personality 16, 18, 20,
action identification 221 23–25; primary appraisal process 68
adaptation: attachment neural network arousal 168, 185; control parameters 214;
73–74; cultural dynamics 291–292, fractal scaling 119; synchronization 211
300; game theory 294; temporal artificial intelligence 250
interpersonal emotional systems 128 Asch, S. E. 65, 193, 262, 265–267, 272,
affective matching 218 276, 281, 357
affordances: behavior settings 198–199; associative learning 42
cooperation 180, 193–197; perception assumptions 316, 320, 325, 327–328
190–192; personality 16, 23, 27, 28–30, asymmetric connections 39, 43–46, 53
33–34; use of the term 200n3 Atkinson, J. W. 15, 27
agency 191, 197 atomisms 185, 186, 196
agent-based modeling (ABM) 7, 82, 231, attachment: attachment styles 70, 72,
270–272; computational social science 74–75; attachment theory 68, 73;
364–365; cultural dynamics 282, 287; co-regulation 145; neural network
public health 99; social networks and 70–74, 75; social interactions 77
goal striving 239, 241 attention 40, 49–52, 55, 197
agreeableness 16, 24, 26 attitudes 5, 7; dynamical systems approach
Ainsworth, M. D. S. 75 84, 85–86; formation of 363; group
algorithms 4, 351, 360; Bayesian dynamics 9–10; health behavior 81;
generative models 133; dynamical homophily 115; intention formation
systems approach 83, 85; free 86–87; social experiments 99–100;
parameters 275; impression formation social networks 96; as states 88; topical
56; Leabra architecture 19; machine integration 359–360
learning 343, 345; network learning attractor 7, 8–9, 69, 92, 119, 148,
44; neural networks 69; social influence 152–153, 156–160, 165–167, 169,
259; social networks 248, 250, 252n17, 172, 183, 195, 212–214, 221–222,
252n18 244, 246–248, 250–252, 257, 354,
 Index  369

362–363; control parameters 214; behavioral systems theory 65, 67, 73

dynamical systems models 148, 150, Behrends, Arwen A. 144–177
152–153, 156–160, 165–167, 172; beliefs: collective 111; evaluation and
network goal graphing 244, 246, 47; health behavior 81, 86–89, 94;
247–248; orders of equations 169; impression formation 41; information
social interaction 221, 222 flows 86; personality tests 16; social
attributes 47–48, 51, 54, 55 networks 96; valences 87–89
authorization levels 252n9 Berezow, A. B. 62
autism 187 Bernstein, A. 285
Avoidance system: attachment neural Berridge, K. C. 28
network 70–71; personality 16, 18, bias: group deliberation 268; impression
20, 23–25, 29, 32; primary appraisal formation 47; news sources 335; social
process 68 judgment 197
Axelrod, R. 281, 284, 286–291, bidirectional burden of proof (bBOP)
294, 296 265–267, 270, 272, 273, 276–277
Big Data 5, 332–346, 352, 355, 363, 365;
Barker, Roger 198, 199 causal inferences 342; data analysis and
Barnard, Kobus 127–143 storage 341; data sources 333–336;
Baron, R. M. 191 Deep Learning 351; description and
BAS see Behavioral Approach System measurement precision 336–337;
basins of attraction 213 differentiation in research strategies 358;
Bass, F. M. 278n2, 323–324 imperfect data 342–343; privacy issues
Baucom, Brian R. 144–177 343–344, 356; situation experiences
Bauer, D. J. 162 on Twitter 337–341; social influence
Bayesian generative models 130, 133, 260; theoretical considerations 343;
135–137 tolerance for complexity 337
bBOP see bidirectional burden of proof Big Five personality traits 15, 16–17, 20,
behavior: attachment styles 74; causal 23–27, 34, 149
relationships 314; changes over time “Big Theories” 65
and situation 27–32; children 222; binding problem 189
complex behavioral system 67–76; biological temperament 18
fractal scaling 116–117, 118; group biotensegrity 182
dynamics 9–10; impression formation Birch, D. 27
54; information flows 86; interaction- BIS see Behavioral Inhibition System
dominant dynamical systems 105, 106, Boccaletti, S. 114
107, 119–120; motivational systems bodily states 28, 29–30, 31–32, 33
17; multi-agent systems 66–67; body mass index (BMI) 131, 134,
personality tests 16; quantum behavior 136–140
change theory 90, 93; responses Bombeck, E. 144
to social and non-social stimuli boot-strapping methods 6
258; social aspects of 61–62; social BOP see burden of proof
networks 81–82, 96; synchronization Borkowski, W. 286
209–211, 215–217, 218–219, 223, bounded rationality 311, 318
224; time series methodology 6; Bowlby, John 65, 67, 68, 71, 73
Virtual Personality model 21, 23, Box, George 318
24–27; within-person versus between- Boyd, R. 295
person variability 32–33; see also brain 69, 107, 113, 120, 181, 322–323
health behavior Braitenberg, V. 328
behavior settings 198–199 Brau, Ernesto 127–143
Behavioral Approach System (BAS) 18, Buckminster Fuller, Richard 182
70–71 burden of proof (BOP) 264–267, 270,
Behavioral Inhibition System (BIS) 18, 272, 273, 275, 276–277
70–71 Burnstein, E. 10n2
370 Index

Butler, Emily A. 127–143 complex systems 7, 61, 63, 65–78,

Butner, Jonathan E. 144–177 105; difficulty understanding 312;
health behavior 90, 91–93; holism
cameras 353, 354, 355, 356 64; interaction-dominant dynamical
Campbell, D. T. 262 systems 120; intuitions about 328;
caregiving 68 mental models 319; social networks
Castelfranchi, C. 232 and goal striving 236; social structures
“catastrophe flags” 93 105; topical landscape of social
causality 4, 183, 314; Big Data 342; psychology 363; see also dynamical
circular 119, 120n2; reciprocal 7, 349; systems psychology; nonlinear
self-organization 181–182 dynamical systems
CDDS see component dominant complexity 356–357, 360, 362; Big Data
dynamical systems 337; complexity matching 118; social
Cecconi, F. 289 62, 66–67, 76, 77
cellular automata models 231, 270, complexity science 7, 61–63, 76–77, 105,
271–272 312; holism in 63–64; interaction-
central nervous system (CNS) 107 dominant dynamical systems 108;
centrality 233, 251n1 multi-agent systems 66–67; topical
Cesta, A. 232 integration 359–360
chameleon effect 210 component dominant dynamical systems
change 66–67, 76, 78; dynamical (CDDS) 106
systems models 145, 147, 148, 160, computational models: complex systems
170; health behavior 81; linear and 62–63, 77; cultural dynamics 282;
nonlinear equations 168–169; nonlinear embedding into social networks 98–99;
dynamical systems 183, 213; quantum free parameters 275–276; health
behavior change theory 90, 93; topical behavior 81, 82, 83, 90–94, 99–100;
integration 360 impression formation 40–56; new
charts 237–238 research paradigm 344; personality
Chen, Daniel 81–102 19–32; precision 327; social influence
children: attachment theory 68, 75; 258; social networks and goal striving
autistic 187; behavior settings 199; 233, 239–243, 244, 249–250; social
behavioral systems theory 67; crawling relationships 144–145; software
experience 200n7; moving objects 194; packages 351; theoretical integration
personality development 222 350; see also agent-based modeling;
CHL see contrastive Hebbian learning dynamical systems models; models;
CLOs see coupled linear oscillators simulations
clustering 272, 273–274, 285, 286 computational social psychology 3–5;
CNS see central nervous system differentiation 358–359; dynamics
co-action 109, 118 357–358; future for 349–365;
co-regulation 9, 145–148, 150, 158, 166, integrative potential 9–10; methods of
167–168, 171, 172 5–6; technological advances 350–356;
cognition: dynamical systems approach theoretical integration 359–363; theories
83; fractal performance variability in 6–9; see also models; simulations
118; health behavior 99; integrated computational social science 364–365
models 77; synergistic social motor computer simulations see simulations
coordination 114 computers 4, 5, 7, 350–351
cognitive capacity 49–52 conditional probabilities 43–44
cognitive psychology 61–62, 179 Condon, D. M. 27
cognitive science 61–62, 83, 85, 181, 359 Condorcet, M. 261–262, 263, 268
collective behavior 76, 258; see also conformity 262, 264, 265–267, 272, 299,
groups; social networks 324
communication networks 335, 351–354 connectionist models 38–39, 40–41, 42,
competition 7, 209, 352 53, 358
 Index  371

conscientiousness 16, 18, 24, 26, 47 damping 129, 130, 132, 134
consensus 260, 285 Darwin, Charles 191, 327
constraint satisfaction 86, 88, 89, 93 Darwinian evolution 73, 294, 295
Conte, R. 232 data analysis and storage 341, 355
content-addressable memory 322–323 data sources 333–336
context dependencies 182 Davis, J. H. 262–263
contrastive Hebbian learning (CHL) 69 Dawkins, R. 298
control parameters 8; dynamical systems de Raad, W. 286
models 169, 171–172; parent-child decentralization 72
relationships 222; slow-to-fast timescale decision making 61, 78; collective
enslavement 108; synchronization behavior 76; dynamic network theory
213–214, 215–217, 218–219, 223; see 234; evaluation system 46–47; multi-
also free parameters agent systems 66; online gaming 352;
control systems theory 67, 82, 86 production systems 100; social decision
Conway, John 270 scheme framework 262–264, 276
Cooley, C. H. 357 Deep Learning 351
cooperation 178–179, 180, 189, 209; deliberation 264, 265, 267
affordances 193–197; computer descriptive approach 336, 343, 352
simulations 7; environment-based developmental psychology 73
features 199; evolution of 284, DeYoung, C. G. 24
291–300; slow-to-fast timescale differential equations 147, 156, 159,
enslavement 109 169–170, 322
coordination 9; environment-based features differentiation 358–359
199; synchronization 178–179, 186–188, diffusion of innovations 270, 278n2,
209–211, 215–217, 219–220; synergistic 323–324
social motor coordination 111–114; diffusion paradigm 96–98
tensegrity 182; see also joint action digital communication networks 335,
correspondent inference theory 56n1 351–354
Couple Satisfaction Index 148 dimensional compression 112
coupled linear oscillators (CLOs) 131–140 direct reciprocity 296
coupled logistic equations 209, 215, discrimination 110–111
223, 224 disease spread 116, 268–269, 323
coupled oscillator models 183, 184, 322 dispositions 15, 33, 218
coupling: dynamical systems models distinctiveness 315, 324–325
158, 161, 166, 171, 172; parent- DNI see dynamic network intelligence
child relationships 222; second-order DOA see Dynamics of Action model
equations 170; synchronization 8, 187, Droutman, Vita 15–37
215, 216–217, 218–220, 223; temporal DSMs see dynamical systems models
interpersonal emotional systems 129, dynamic network intelligence (DNI) 249,
132, 134 252n15
covariation 149–151, 152 dynamic network theory 232–233,
cross-validation 138, 140 234–239, 240, 243, 245, 248, 250–251,
Crutchfield, R. S. 357 252n12
Cubist chicken 317–318 dynamical minimalism 223, 360, 361–362
culture 281–300; Axelrod model of dynamical social psychology 178, 345, 358
cultural dissemination 286–291; cultural dynamical systems models (DSMs) 145,
drift 290–291; cultural evolution 283, 146–172; covariation 149–151; dynamical
359; cultural psychology 73; definition systems as maps 151–156; estimating the
of 282; evolution of cooperation map 158–162; inferred map 156–158;
291–300; hierarchical nestedness linear and nonlinear equations 168–169;
64; inaccuracies in social judgment moderation and multistability 171–172;
196–197; social psychology of cultural orders of equations 169–170; results
dynamics 284–286 162–167; time 170–171
372 Index

dynamical systems psychology 7, 8, 63, 65, ethology 67

212–215, 358; adaptation 74; Big Data Eubank, S. 116
345; cooperation 193, 195–196; health evaluation feedback 40, 46–49
behavior 82, 83–86; interpersonal evolution 73, 99; cooperation 179,
emotional systems 129; multi-agent 291–300; cultural 283, 359
systems 67; synchronization 208, 209, evolutionary biology 327, 359
222–224; topical integration 359–360; evolutionary game theory 284, 294–300
see also complex systems; interaction- evolutionary psychology 68, 73
dominant dynamical systems; nonlinear evolutionary tasks 18
dynamical systems excitation 21–22
dynamics 72–73, 357–358, 362–363; executive function 18, 188
interaction-dominant 105, 106–108, 119; expectancies 74
intrinsic 9, 106, 149, 216, 349, 354, 358 experimental design 4, 6, 333, 345, 362
Dynamics of Action (DOA) model 27–28 extraversion 16, 18, 24, 26

ecological psychology 200n2 Facebook 332, 333–334, 341, 342,

economic models 327–328 343–344, 351, 356
Eguíluz, V. M. 289–290 facial feedback 218
Eiler, Brian A. 105–126 factor analysis 23, 25, 26
embeddedness 64 false models, functions served by 319–320
emergence 69, 212–213, 312, 349, 361, Farber, Ilya 38–60
362–363; attachment neural network fear 29, 72–73
72; computational social science 365; feed-forward 41, 69, 71, 86
control parameters 214; dynamical feedback 7, 71; dynamic network theory
systems theory 147; interaction- 243; evaluation 40, 46–49; facial 218;
dominant dynamical systems 106, 119; network rippling 243; reciprocal 364;
slow-to-fast timescale enslavement 105; research methods 365; social 221–222;
social networks and goal striving 233; synchronization 223; temporal
topical integration 360 interpersonal emotional systems 129
emotions: attachment neural network 71, Festinger, L. 65
72, 74, 75; behavioral systems theory 67; Feynman, Richard 258, 277
emotional arousal 119, 168; emotional field theory 357
contagion 243, 268, 334, 343; individual Fisher, R. A. 327
differences 66; integrated models 77; fit statistics 138
network rippling 243; patterns of change fitness 294, 295–296, 299
9; personality tests 16; social networks Flache, A. 289
and goal striving 251n7; synchronization Fleeson, W. 17, 27
208–209; temporal interpersonal flows 185
emotional systems 127–141; time series forecasting 248–250, 252n15, 252n17
methodology 6 formal models 4, 316, 317, 318–319,
empathy 210, 219 325–326, 328, 358
empirical data 6 fractal scaling 116–119
energy 185, 186 fractals 7, 8–9
environment 179, 189, 192, 197–199 free parameters 85, 275–276
Epstein, J. M. 315 free rider problem 295, 299
equilibrium 88; basins of attraction 213; French, J. P. R. 259
Nash 293, 301n4; near-equilibrium Freud, Sigmund 65, 67
systems 185 Friedman, A. 311
ergodicity 149 funding 63, 76, 77
error-correcting learning 22–23, 56 Fusaroli, R. 113
errors 138–139; error variance 357–358;
measurement 153–154 Galton, Francis 267
ethical issues 344, 355–356 game theory 179, 193, 284, 292–300
 Index  373

games, online 352 Hebb, Donald 322

Gestalt psychology 64–65, 357 Hebbian learning 22, 56, 69, 322–323
giant component 353 Heft, H. 199
Gibson, J. J. 190, 191, 198, 200n2 Heider, F. 65
Gilbert, Margaret 193 heterogeneity 90–93, 96
goals: attachment neural network 71, HEXACO model 16, 17
72–73; complexity matching 118; hierarchical nestedness 64, 71
dynamic network theory 234–239; Hinsz, V. B. 267
goal affordances 16, 23; impression holism 63–67, 77
formation 41; motivational conflict 214; homeostasis: dynamical systems models
network goal graphing 233, 244–250, 145–146, 148, 150–151, 157, 166, 171,
251, 256–257; personality traits 17–18; 172; temporal interpersonal emotional
social interactions 209; social networks systems 128, 129
231–257; synchronization 211; Whole homophily 115, 249, 286, 288, 353
Trait Theory 27 Hopfield, John 322–323
Gómez-Gardeñes, J. 290 human body 182
Google 332, 334, 341 hypothesis testing 326, 333, 364
Google Trends 333, 334–335 hysteresis 184, 195, 210
gossip 297
graphical models 135–137 Iberall, A. S. 185
graphs: cultural dynamics 290; network IDDS see interaction-dominant dynamical
goal graphing 244–250, 251, 256–257 systems
gray-box system identification 85 identity 315, 324–325
group selection 298 ignorance 45–46
groups: computational social science 365; image perception 189
continuous judgments 267–268; culture imperfect data 342–343
distinction 282–283; formation of 363; impression formation 38–57, 65;
fractal scaling 118; group dynamics 9–10, asymmetric connections 39, 43–46;
360–361; polarization 268; social decision base rates 41–43; current limits and
scheme framework 262–264, 276; social future paths 53–54; description of
influence 260, 261–262, 276–277; social model 40–41; empirical test of model
judgment 196, 197; stereotypes 297; 52–53; evaluation feedback 40,
synchrony 188; synergistic social motor 46–49; limited attention 40, 49–52;
coordination 111–114 neural network models 38–39; prior
growth modeling 268–270 probabilities 39, 41–42
Guan, Jinyan 127–143 indirect reciprocity 296–297, 299
Guerra, B. 290 individual differences 16, 17, 66
guiding visions 5, 363 individuals 61, 64, 65–66, 76; health
Gunawardena, J. 316 behavior 81, 82; responses to social
Gupta, Swati 38–60 stimulus 258; as unit of analysis 199
inference: Bayesian generative models
habit 243 130, 133; Big Data 342; correspondent
Hamilton, W. D. 296 inference theory 56n1; dynamical
Harrison, J. R. 231, 242–243 systems models 156–158; impression
health behavior 81–100; drivers of formation 40, 41, 42–45, 53, 54, 55;
heterogeneity 90–93; dynamical systems mind reading 179; neural networks
approach 83–86; future directions 39; social judgment 197; statistical
99–100; neural networks 86–90; past 362; temporal interpersonal emotional
experience 90, 93–94, 95; social networks systems 137–138
96–99, 115; temporal interpersonal inferential statistics 336, 352
emotional systems 137, 139–140 influence 5, 258–277; burden of proof
heart rate 148, 149–152, 156–158, framework 264–267; cultural dynamics
159–161, 165–168 281, 288; free parameters 275–276; key
374 Index

ideas 259–262; modeling continuous social judgment 196–197; modeling

judgments 267–268; modeling growth continuous 267–268
over time 268–270; multilateral 289; Jung, C. G. 65
nonlinear dynamical systems 212; juries 262, 264, 267, 268, 276–277
parameter space 276–277; simulating jury theorem 261
communities of interacting agents
270–272; social decision scheme k-Winners-Take-All (kWTA) 69
framework 262–264; synchronization Kahneman, D. 268
215, 216–217, 219–220, 223, 224 Kallen, Rachel W. 105–126
information: cultural 282, 283, 284, Kashima, Yoshihisa 281–307
286–287, 291, 299; informational Kauffman, S. A. 314
synchrony 186 Kelman, H. C. 259
information processing 40, 49–52, 105 Kepler, Johannes 320, 321
inhibition 18, 21–22, 70–71, 73 Kerr, N. L. 264, 266
innovations 270, 278n2, 323–324 kin selection 295–296
instability 109 kinematic specification of dynamics
integrated models 77, 78 (KSD) 192
intention 86–87, 90, 91–92, 93–94, 95, Kirley, Michael 281–307
96–98 Kitayama, S. 281
interaction-dominant dynamical systems Klemm, K. 289–290
(IDDS) 105–120; fractal scaling knowledge, schematic 40, 48, 55, 57n1
116–119; slow-to-fast timescale Krech, D. 357
enslavement 108–111; social networks KSD see kinematic specification of
114–116; synergistic social motor dynamics
coordination 111–114 Kuhn, T. S. 9
interaction-dominant dynamics 105, Kunda, Z. 53
106–108, 119 Kuperman, M. N. 289
interdisciplinarity 63, 130, 359, 364 kWTA see k-Winners-Take-All
Internal Review Boards (IRBs) 356
internal states, synchronization of 8, Laine, Tei 38–60
211–212, 216, 217–220, 221, 223, 224 Lakens, D. 335
internet 5, 351, 352, 355, 365; see also Latané, B. 259, 285, 286
social media Latent Change Score (LCS) Modeling
Internet of Things (IOT) 336, 354–355 161, 163
interpersonal emotional systems 127–141 Latent Variable Factor models 25
interpersonal synchrony see Latora, V. 290
synchronization/synchrony lattice networks 271–272, 285, 287–288, 325
intolerance 45–46 “law of mass action” 268–269
intrinsic dynamics 9, 106, 149, 216, 349, layers 19, 20–21, 25, 28; attachment
354, 358 neural network 70–71; impression
Ioannidis, J. P. A. 326 formation 41, 54
IOT see Internet of Things Lazarus, R. S. 68
IRBs see Internal Review Boards Lazer, D. 311
iterated games 293–294, 301n3 LCS see Latent Change Score Modeling
LDS see linear dynamical systems
James, William 4, 8, 349, 357, 363–364 Leabra architecture 19, 21, 23, 69, 72–73
Järvilehto, T. 181 leak conductance 21–22
John Templeton Foundation 77, 78n1 learning: associative 42, 69, 181;
joint action 180, 184–185, 189, 193, attachment neural network 70;
194–196; see also cooperation decentralized 72; error-correcting
Jordan, J. S. 107 22–23, 56; health behavior 83, 90,
judgments: impression formation 39–40, 93–94, 95; Hebbian 22, 56, 69,
42–48, 50, 52, 54–56; inaccuracies in 322–323; intention formation 87;
 Index  375

machine learning 6, 337–341, 343, 345; mindful modeling 130–131

neural networks 69; over time 71; self- minority opinions 7
organized 68, 75 Mischel, W. 15
Lewin, Kurt 4, 65, 272, 349, 357 Mitchell, Melanie 63
Lichtenstein, M. 50 mixture modeling 159, 161–162, 169
limit cycles 158 mobile devices 333, 335, 351, 353–354,
linear dynamical systems (LDS) 355–356, 358, 365
131–132, 134 models 311–328; articulating a system
linear equations 168–169 313–315, 316; Bass’s model of the
linear models 6, 316–317 diffusion of innovations 278n2, 323–324;
Linguistic Inquiry Word Count (LIWC) Bayesian generative 130, 133, 135–137;
338–339, 351 blueprints for 277; building 315–316;
links 19 connectionist 38–39, 40–41, 42, 53, 358;
LIWC see Linguistic Inquiry Word Count dynamics of distinctiveness 324–325;
logistic equations 209, 214, 215, 223, 224 formal 4, 316, 317, 318–319, 325–326,
Lorge, I. 262 328, 358; functions served by false models
Lotka, Alfred 322 319–320; growth modeling 268–270;
Lowe, R. 86 Hopfield’s model of content-addressable
memory 322–323; Lotka-Volterra model
MacCoun, Robert J. 258–280 of predator-prey relations 322; machine
machine learning 6, 337–341, 343, 345 learning 339–340, 341; mindful modeling
Macy, M. W. 289 130–131; “model grounding” 250;
majority amplification 262, 268 Newton’s model of Universal Gravitation
MANN see Multi Agent Neural Network 320–321; research publication and
platform replication 325–326; statistical 316–317;
maps 151–162 as “stupid” 311, 317–319, 326–328;
Markov-Chain Monte Carlo (MCMC) synchronization 211–212, 215–220,
sampling 137–138 223–224; verbal 315–316, 318, 325,
Markus, H. R. 281 326, 328; see also agent-based modeling;
Marsh, Kerry L. 178–207 computational models; simulations
Maslow, A. H. 65 moderators 171–172
mass media 289 Molenaar, P. C. 149
mating 7 Monroe, Brian M. 38–60
McAdams, D. P. 15 Monte Carlo simulation 270
McArthur, L. Z. 191 Monty Hall problem 312–313
McClelland, J. L. 99 moods: computer simulations 7; group
McElreath, Richard 295, 316–317, 326 dynamics 9–10; mobile phone data 353;
MCMC sampling see Markov-Chain synchronization 211
Monte Carlo sampling mooring hypothesis 187
meaning 192, 364 Moreno, Y. 289, 290
measurement error 153–154 Morris, M. W. 47, 48, 49
memory: belief valences 87–88; Hopfield’s motivation: attachment neural network
model of content-addressable 322–323 70–71, 72–75; behavioral systems theory
mental models 319 67; dynamics of 18, 27–32; goals and
mentalization 179 18; integrated models 77; motivational
meta-stable states 109, 117 conflict 214–215; motivational systems
Miceli, M. 232 15, 16–17, 23–27, 34; network 235,
micro and macro phenomena 360–362 238, 241–243, 244; person-situation
Milgram, S. 265, 266, 267 interactions 33; personality tests 16; social
Miller, Lynn C. 15–37, 198 interactions 209; Virtual Personality
mimicry 184–185, 210 model 20, 23–27; within-person versus
mind 61–62, 64–65, 69, 75, 213 between-person variability 32–33
“mind reading” 178–179 mouse paradigm 8, 358
376 Index

movement: kinematic specification 183–184; health behavior 90; hysteresis

of dynamics 192; perception of 210; multi-agent systems 67; multistability
momentum 110; synchronization 178, 221; social networks and goal striving
184, 186–188, 209–210, 211; synergistic 242; synchronization 209, 223; see also
social motor coordination 111–114 dynamical systems psychology
Mullen, B. 260 nonlinear equations 168–169
Multi Agent Neural Network (MANN) nonlinearity 7, 117, 349, 362
platform 96–97, 99 Norman, D. 200n3
multi-agent systems 62–63, 66–67, 76, norms: computer simulations 7;
77–78 descriptive versus injunctive 259;
multilateral social influence 289 group dynamics 9–10; intention
multilevel modeling 6, 148–149; formation 86–87; slow-to-fast timescale
multilevel regression 132–133, 345 enslavement 109; social interactions
multilevel selection 298, 299 209; synchronization 224
multimodality 93 Nowak, Andrzej 3–12, 63, 349–367; Big
multistability 8, 108, 169, 172, 195, 221 Data 332–348; cultural dynamics 285,
multivariate emotional observations 286, 287; synchronization 208–228
132–133, 135 Nowak, M. A. 295
Murray, H. A. 65 NSF see National Science Foundation

Nash equilibrium 293, 301n4 ODT see optimal distinctiveness theory

Natale, F. 289 Open Science Collaboration (OSC) 3
National Institutes of Health 76–77 openness to experience 16, 24, 26
National Science Foundation (NSF) 63, 99 opinions 260, 284–286, 300; see also public
natural selection 298 opinion
near-equilibrium systems 185 optimal distinctiveness theory (ODT) 315,
negativity 232, 236, 238, 240–241 324–325
neo-diffusionism 283 order effects 49–50, 54
nestedness 64, 71, 77 O’Reilly, R. C. 19
network density 233 Orr, Mark G. 81–102
network goal graphing 233, 244–250, 251, OSC see Open Science Collaboration
256–257 oscillations: emotional 128, 129–130,
network reciprocity 297 131, 140–141; limit cycles 158; logistic
networks see social networks equations 214; second-order equations
neural networks: attachment 70–74, 170; see also coupled oscillator models
75; complex behavioral system Other-Total Ratio 264, 265
67–68; complexity matching 118;
connectionist models 38–39; health Pace, B. 290, 291
behavior 82–83, 86–90, 94, 96, Page, S. E. 90
100; memory 322–323; modeling paradigms 9
psychological complexity 68–69; parallel processing 39
personality 15, 16, 17, 19–27 parameter space 276–277
neuroticism 16, 18, 24, 26 parents 222
news sources 335 Parisi, D. 289
Newton, Isaac 320–321 path dependence 90, 92
nodes 19, 20, 21–23, 24, 25, 28; patterns of change 9, 147, 148
attachment neural network 71, 72; goal PD see Prisoner’s Dilemma
232, 236, 237–238, 239, 245–246, 248; peer influence 222
impression formation 41–45, 47, 48, perception 180, 189–193, 200n2;
53, 55, 56; k-Winners-Take-All 69; behavioral systems theory 67; dynamical
social networks 114, 233 systems approach 83; inaccuracies in
nonlinear dynamical systems 7, 8, 65, social judgment 196–197; perception-
105, 212–215, 231, 360; adaptation 74; action research 193, 194, 196;
cooperation 193, 195–196; features of synergistic social motor coordination 114
 Index  377

performance 116–119 prior probabilities 39, 41–42; social

person-context system 81, 82, 96 decision scheme framework 263; social
person perception research 38, 39, 46, 49 influence 261
person-situation interactions 19, 33–34 production systems 100
personality 15–34; behavior patterns 75; psyche 61, 65, 66, 73, 76, 77–78
changes over time and situation 27–32; psychology: in complexity science 62,
dynamic approach 15, 16, 34; dynamics 63, 76; holism in 64–67; integrated
of motivation 18; Facebook usage psychological science 77–78; see also
334; impression formation 41; neural computational social psychology; social
network model 19–27; person-situation psychology
interactions 19, 33–34; personality psychometric approach 15, 16
tests 16; structural approach 15, 16, public health 99
34; synchronization 211, 220–222; public opinion 7, 284–286, 300, 360, 361,
within-person versus between-person 363
variability 17, 19, 32–33 punishment 299
perturbations 108, 247–248; cultural
dynamics 290; dynamical systems quantum behavior change theory 90, 93
models 148, 153–154, 157–158; fractal
scaling 117; resilience to 183; temporal Rapoport, Anatol 296
interpersonal emotional systems 128 Raven, B. 259
Peters, K. 283 Read, Stephen J. 3–12, 15–37, 78n1, 198
phase portraits 151–152, 157, real-world systems 244–250
162, 165 recency effects 50
“phenomenological law” 320 reciprocal causality 7, 349
physical appearance 54 reciprocal compensation 112
physical settings 179, 180, 197–199 reciprocity 296–297, 299
physics 62, 78, 362; atoms 63; dynamical reductionism 9, 76, 144, 361
systems theory 212; ergodicity Reed, E. S. 191
149; Newton’s model of Universal regression models 130, 131, 132–133;
Gravitation 320–321; parameter Big Data 336, 345; dynamical systems
space 276 models 160, 161
pi number prediction 193, 194–195, 196 reinforcement 74, 75, 222
pink noise 117–118 relationships: between parts of a system
polarization 268, 285, 300 314–315, 325; between variables 317,
polynomials 168, 169 320, 337, 343, 345; see also social
Poncela, J. 290 relationships
positive focus ratio 235, 239, 247, repellers 7, 8–9, 119; dynamical systems
256–257 models 148, 156–158, 159; orders of
positivity 232, 233–234, 238–239 equations 169
power: behavioral systems theory 68; replication crisis 131
network 240–242, 251; social 259 reputation 296–297
Prado, C. P. 290, 291 research ethics 344, 355–356
predator-prey relations 322 research, publication and replication
Predoehl, Andrew 127–143 325–326
prestige 299–300 research questions 333
primacy effects 50, 57n5 resistance: constructive 251n6; to
primary appraisal process 68, 71 modeling 311; network 235, 236,
prior probabilities 39, 41–42 240–241, 242, 243; supportive 234,
Prisoner’s Dilemma (PD) 292–293, 236, 238, 240, 244, 256–257
294–295, 301n3 Resnick, Mitch 312
privacy issues 343–344, 355–356 Resnicow, K. 90
probability: Bayesian generative models Revelle, W. 27
133, 135; conditional probabilities rhythm 184, 186, 187, 188
43–44; Monty Hall problem 312–313; Richardson, Michael J. 105–126
378 Index

Rigoli, L. M. 118 SEM see structural equation modeling

Rivera, Daniel 82 sensors 353, 354, 355
Rivera Lab 82, 83–86, 100 Serfass, David 332–348
Robins, Garry 281–307 serial processes 39
Rodríguez, A. H. 289 set points 148, 160
Roese, N. J. 47, 48, 49 settings 197–199
Rogers, Everett 323 sex behavior 93–94, 95, 334
role-playing 218 Sherif, M. 281
romantic relationships: co-regulation Sherman, Ryne 332–348
145–146, 167–168; Facebook Shibanai, Y. 289
status 342; modeling 148–168; Shon, DaHee 231–257
synchronization 211; temporal signaling 298–299
interpersonal emotional systems Simek, Kyle 127–143
131–140; see also social relationships similarity 287
Romero, V. 112 Simmel, M. 65
Rosenwein, R. E. 262 simplicity 275
simulations 6, 7–8, 231; analysis
sampling: Big Data 336–337, 338, 339, of computational models 327;
345, 352; Markov-Chain Monte communities of interacting agents
Carlo 137–138; social decision scheme 270–272; complex systems 312;
framework 263 differentiation in research strategies
San Miguel, M. 289–290 358; dynamical minimalism 361–362;
Santa Fe Institute 62, 63, 77 dynamical systems approach 84, 358;
scale-free distribution 353 health behavior 90–94, 96; impression
Schelling, Thomas 265, 270 formation 40, 42–43, 45, 50–52; new
schemas 50, 55 research paradigm 344; personality
schematic knowledge 40, 48, 55, 57n1 23–27, 28–32, 34; public opinion
Schkade, D. A. 268 285; social networks and goal striving
Schmidt, R. C. 188 233, 239–243, 244; synchronization
Schoggen, P. 198, 199 211–212, 216–217, 218–219, 223,
science 63, 314–315, 316, 320, 325–327, 224; topical integration 360; valence
362–363 48–49; verification of theory 364;
scientific method 3, 6 vertex architecture 97–99; see also
SCT see Social Cognitive Theory computational models
SDS see social decision scheme framework situations 197–198
self-concept 7, 8–9, 359–360, 361 slaving principle 120n2; see also timescale
self-efficacy 81, 84, 236 enslavement
self-esteem 9, 119, 361 slow-to-fast timescale enslavement 105,
self-evaluation 8–9 106–107, 108–111, 112–113, 115–116,
self-organization 7, 9, 62, 180–184, 118–119, 120n2
212–213, 349, 362–363; computational Smaldino, Paul E. 311–331
social science 365; control parameters small-world networks 97, 115–116, 119,
214; fractal scaling 117–118, 119; health 289–290, 353
behavior 92; interaction-dominant smart phones 5; see also mobile devices
dynamical systems 106; learning 68, Smith, C. A. 68
75; memory 323; neural networks smoking 115
68–69; self-organized criticality 109, Snijders, T. A. B. 239
117; slow-to-fast timescale enslavement social capital 242
105; small-world networks 115; social social cognition 40, 105, 108; culture
connection 199; synergistic social 281; health behavior 83, 94; impression
motor coordination 113, 114; system formation 54
embeddedness 64; topical integration 360 Social Cognitive Theory (SCT) 83–85
self-regulation 116, 128 social complexity 62, 66–67, 76, 77
 Index  379

social connection 178–199; interpersonal dynamical systems 114–116; network

synchrony 184–188; perceptual goal graphing 244–250, 251, 256–257;
processes 189–193; self-organization network reciprocity 297; network
180–184; see also cooperation; social structures 271–272, 325; traditional social
interaction; synchronization/synchrony network frameworks 233–234, 238–239
social context: belief valences 88, 89; social perception 191–192; attachment
cooperation 179; health behavior 90–91, styles 74, 75; inaccuracies 196–197;
93–94, 95; intention formation 87; interaction-dominant dynamical
interaction-dominant dynamical systems systems 114
107, 108, 120; meta-stable roles 109 social power 259
social decision scheme (SDS) framework social psychology: adaptation 73; Big
262–264, 276 Data 344–346; complex systems
social dilemmas 7, 291, 292–294, 295, 299 perspective 105; cultural dynamics
social-ecological approach 179, 197, 198 281–282, 284–286, 300; data in
social growth modeling 268–270 258; dynamical systems approach 85;
social identity 315, 324–325 future of 363–365; growth modeling
social impact theory 260, 264, 265, 275, 270; health behavior 83; as a “hub”
285, 286, 300 science 359, 362; interaction-dominant
social influence 5, 258–277; burden of dynamical systems 119, 120; models
proof framework 264–267; cultural 277; new ideas and research strategies
dynamics 281, 288; free parameters 349–350; nonlinear dynamical systems
275–276; key ideas 259–262; modeling 213; online research 352; replication
continuous judgments 267–268; crisis 131; scientific method 3; social
modeling growth over time 268–270; dilemmas 291; theoretical integration
multilateral 289; parameter space 359–363; traditional 349, 357, 359,
276–277; simulating communities 361, 363, 365; see also computational
of interacting agents 270–272; social social psychology
decision scheme framework 262–264 social relationships: behavior settings
social interaction: cameras in public 199; co-regulation 145–148, 158, 166,
places 356; cultural dynamics 282, 289; 167–168, 171, 172; interdependence
interaction-dominant dynamical systems 144; modeling 144–145, 148–168;
107, 114; mimicry 210; nonlinear nonlinear dynamical systems 213;
dynamical systems 213, 215; personality personality influence on 220–221;
influence on 220–222; public opinion synchronization 8, 208–209, 211,
361; social networks and goal striving 222–223, 224; temporal interpersonal
252n16; synchronization 208, 209, 211 emotional systems 127, 129, 131–140
social media 5, 99, 260, 333–334, social roles 109, 196
337–341, 351–352, 355–356, 358 social sciences 130–131, 311, 318; Big
social motor coordination 111–114 Data 332–346; computational social
social network roles 234–235, 236–237, science 364–365; dynamical systems
239, 250–251; changes over time 248; models 172; ergodicity 149; funding
forecasting 249, 250; network goal 76; individuals 61; lack of simulation
graphing 245–246, 247, 256–257 research 231; multi-agent systems
social networks 314, 352–353; adding 62–63; social influence 259
complexity 242–243; computational social systems 7, 61, 62, 77; goals 232;
environment 240–242; computational health behavior 81, 82, 96–99;
social science 365; cultural dynamics interaction-dominant dynamical systems
289–290; differentiation in research 106, 107–108; network power 241;
strategies 358; dynamic network theory slow-to-fast timescale enslavement 111
232–233, 234–239, 240, 243, 245, 248, sociality 120, 180, 199; culture 281–282;
250–251, 252n12; fractal scaling 119; synchrony 188; Twitter 340
goal striving 231–257; health behavior societal change 7
81–82, 94, 96–99; interaction-dominant software 4, 7, 350–351
380 Index

Solomon, H. 262 temperament 18, 211, 220

sources 259, 260, 261, 265–267; see also temporal interpersonal emotional systems
data sources (TIES) 127–141
space 285, 286, 290 temporality 4, 6, 7, 8; co-regulation 146;
spatial selection 298, 299 dynamical systems models 84, 147, 152,
Srull, T. K. 50 170–171; fractal scaling 117; health
stability 213 behavior 90; impression formation 54;
state-space representation 134–135 neural networks 69; synchronization
stationarity 130, 133 209; timescale enslavement 105,
statistics 6; Big Data 336–337, 352; fit 106–107, 108–111, 112–113, 115–116,
138; inferential 336, 352; statistical 118–119, 120n2; see also time series
models 316–317 methodology
Stephens-Davidowitz, S. 332 tensegrity 182
stereotypes 7, 49, 109, 297 Thagard, P. 44, 53
Stivala, Alexander 281–307 theory 6–9, 272, 356–363; Big Data
Strange Situation 75 combined with 343; “Big Theories”
stream of consciousness 357, 363–364 65; construction and verification 4;
stress: control parameters 214; theoretical integration 350, 359–363;
coordination 210; interpersonal traditional social psychology 363;
synchrony 185; weight loss 246, 257 verification of 364
structural equation modeling (SEM) 6, Theory of Planned Behavior (TPB)
131, 149; dynamical systems models 83–84, 85, 86
159, 161, 162; free parameters 275; Theory of Reasoned Action (TRA)
social networks 252n18 86–88, 90, 93, 95, 97–98, 100
Sunstein, C. R. 268 thermodynamics 185–186
synchronization/synchrony 8, 178, thought experiments 40, 42–43, 45
179–180, 184–188, 208–224, 359, thoughts 5, 357; control parameters
362–363; anti-phase and in-phase 214; mental process 364; patterns
183, 210, 213–214, 216; coordination of change 9; suppression of 215;
of behavior 209–211, 215–217, 223; synchronization 208–209, 211; time
coupled logistic equations 215; emotions series methodology 6, 8
130; internal states 8, 211–212, 216, threats 29, 32, 73
217–220, 221, 223, 224; personality TIES see temporal interpersonal emotional
220–222; small-world networks 116; systems
social-ecological perspective 197 time lags 171
synergy 180, 182–183, 184–185, 187–188, time series methodology 6, 8;
196; cooperation 189, 193; social- differentiation in research strategies
ecological perspective 197; synergistic 358; dynamical systems models 145,
social motor coordination 111–114 152, 154, 159; fractal scaling 117;
system embeddedness 64 health behavior 91–92; mobile devices
systems dynamics models 231; see also 354; temporal interpersonal emotional
dynamical systems models systems 128
systems theory 147, 149, 151, 153; see also timescale enslavement 105, 106–107,
dynamical systems models; dynamical 108–111, 112–113, 115–116, 118–119,
systems psychology; nonlinear 120n2
dynamical systems Tobias, R. 243
topical integration 359–360
Takens, F. 172n1 topographical features 156–158, 159,
Talevich, Jennifer Rose 61–80 161–162, 165–167, 169
targets 259 topology 151, 171
technology 7, 332, 358; advances in 4, Toral, R. 289–290
350–356; analysis of synchrony 186; TPB see Theory of Planned Behavior
see also Big Data; computers; mobile TRA see Theory of Reasoned Action
devices transcripts 335
 Index  381

Triandis, H. C. 281 visual perception 189–190

Trivers, R. L. 293–294, 296 visualization 5, 244, 245, 247
Twitter 333, 334, 337–341, 342–343, Volterra, Vito 322
351, 356, 363 voluntary participation 299
Tylén, K. 113 Vos Savant, Marilyn 312–313

unidirectional burden of proof (uBOP) “wanting” 27, 28–30

265–267, 276 Warren, W. H. 194
unity states 178, 180 wearable technologies 353–354
Universal Gravitation, Newton’s model weight loss 232, 238, 246–247, 249–250,
of 320–321 256–257
Westaby, James D. 116, 231–257
valence 41, 46–49, 53, 87–89; fractal Whelan, J. 283
scaling 119; health behavior 90–91, 92, Whole Trait Theory 27
93, 94, 95; social networks 97 Wilder, D. A. 260
validity 100 Wilson, D. S. 298
Vallacher, Robin R. 3–12, 63, 208–228, Wilson, E. B. 268
247–248, 349–367 Wilson, E. O. 298
values 5, 9–10, 41, 211, 220 Wimsatt, William 319
Vancouver, J. B. 243 wisdom of crowds (WOC) effect 267, 268
vector plots 154–156, 157–158, 162–165 women 116
verbal models 315–316, 318, 325, 326, 328 Worcester, J. 268
Verhulst, P.-F. 268–269, 270 Wright, Sewall 327
vertex architecture 96–99
Virtual Personality model 17, 19–27 Yang, Y. 198