University of Southern California Dissertations and Theses

Information sharing, deliberation, and collective decision-making: A computational model of collaborative governance

(USC Thesis Other)

Information sharing, deliberation, and collective decision-making: A computational model of collaborative governance

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

INFORMATION SHARING, DELIBERATION, AND COLLECTIVE DECISION-
MAKING: A COMPUTATIONAL MODEL OF COLLABORATIVE GOVERNANCE

by

Taehyon Choi

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POLICY, PLANNING, AND DEVELOPMENT)

May 2011

Copyright 2011 Taehyon Choi

ii
DEDICATION

To my family, the anchor of my life.

iii
ACKOWLEDGMENTS

At this moment in finishing this dissertation, I realize that I have been surrounded
by unbelievably supportive and friendly people, including faculty, friends, and family
members. Completion of this dissertation may not have been possible without them. This
space is too small to deliver my deep gratitude to them, but I hope to at least convey
some measure of my appreciation here.
I extend my deepest gratitude to my advisor, Dr. Peter Robertson. I met him first
in his class, where he was delivering an impassioned speech on humans and human
values in organizations. His lecture fascinated me, and he was the reason I chose USC for
my doctoral work. The last four years I have been working with him was one of the
happiest periods in my life. He helped me to grow as an independent scholar, to find
interesting research questions, to learn how to organize ideas and papers, and to pursue
happiness. If I can say this dissertation is my child, he is its grandparent, especially
regarding Chapter 2 for which we worked closely together to develop a paper. He should
receive credit for the contents in Chapter 2, but errors are mine. He encouraged me to
consider different levels of analysis for this dissertation. His comments were a significant
source for many of the improvements in this dissertation. He was always encouraging,
helpful, instructive, and friendly. I wanted to be like him, and that was sufficient
motivation for me to complete this dissertation. Further, Dr. Robertson has been my true
friend, if I may say so. I will never forget the moments we shared at Montreal in 2010,
where we received an award and walked here and there together. It was also a very
special experience we traveled together to Milan, which was somewhat bumped, but

iv
that‟s why the memory is even more special. No doctoral student would receive his or her
degree without critical situations. I experienced few of them because he was always fully
supportive to my development and to my life. I was able to consult him regarding
everything, which I believe made all the things different, from dark to bright, from weak
to strong, from obscure to obvious, and from sad to joyful.
I was extremely fortunate when I met Dr. Eric Heikkila in the first doctoral class.
I didn‟t realize how much help, instruction, and support I would derive from him
throughout my doctoral program. He inspired me to be a scholar with his stunning
imagination and to challenge to my own work. I have received compliments I do not
deserve from him, and I know they were to let me know what I was doing well and to
instruct me in what I had to do better. If this dissertation has something to contribute to
academia, especially regarding agent-based modeling and theoretical backgrounds of the
model, it is because he inspired and pushed me to go upfront and he believed in me. He,
in particular, should receive credit with respect to future research based on this
dissertation for his encouragement to consider the future.
I must have done something good because I was extremely fortunate to meet Dr.
Janet Fulk. She is one of the most considerate and noble persons I have ever known. It
was a blessing to me to meet her – few doctoral students are ever lucky enough to have
such a helpful outside committee member like her. She opened a whole new academic
world to me when I took her class on organizational communication and knowledge
management, which formed the theoretical foundation of this dissertation. She has so
keen an insight that with a few comments from her, I was able to significantly improve

v
this dissertation. Without her instruction and inspiration, Chapter 3 and the following
framework would not have been possible.
Many faculty members in the School of Policy, Planning, and Development at
USC helped me a lot. I would like to give thanks to Dr. Catherine Burke. I cannot forget
the generous help she provided for me at the outset of the doctoral program here, and her
guidance for my qualifying exam. Dr. Robert Myrtle is one of the best teachers and
networkers I have ever met. His class on teaching has been an invaluable resource for me
as I struggle to become a good teacher. I would like to particularly acknowledge Dr. Yan
Tang, Dr. Terry Cooper, and Dr. Genevieve Giuliano. It was always joyful and helpful to
be actively engaged in discussions with Dr. Yan Tang and to see him comment on others‟
research. His keen insight was always challenging to me. I still regret that I could not
invite him to my dissertation committee, but his teaching on institutionalism is immersed
in my dissertation. The very small seed of my dissertation was planted in Dr. Terry
Cooper‟s class. I was able to develop an early version of my dissertation proposal
through Dr. Genevieve Giuliano‟s class. Her invaluable comments on my proposal made
the seed grow strong. I also appreciate Dr. Dan Mazmanian, Dr. Anthony Bertelli, and Dr.
Keith Provan (University of Arizona). Their comments on my paper in the Doctoral
Conference of the Consortium on Collaborative Governance stimulated me to think about
the big picture regarding this dissertation. I was also lucky to receive valuable lessons
from Dr. Gerald Caiden, who raised a lot of intriguing questions, Dr. Gary Painter, who
helped me understand basic ideas of public sector economics, Dr. Susan Webb-Yackee
(University of Wisconsin, Madison), who provided me with an opportunity to perform a

vi
case study on collaborative governance, and Dr. Greg Hise (University of Nevada, Las
Vegas), whose class on qualitative methods is a masterpiece, and once again, Dr. Dan
Mazmanian, who always shares his wisdom with students in his class and with colleagues
in seminars. Dr. Nicole Esparza and Dr. David Suarez are always supportive to doctoral
students including me. I am grateful to Dr. Dowell Myers, who gave valuable comments
on a preliminary plan of my doctoral research, and stimulated my research interest in
immigration. Although I haven‟t had many individual contacts, Dr. Martin Krieger was a
source of lessons and insights via his e-mails. I believe Dean Jack Knott is the pride of
doctoral students at SPPD; I see SPPD has been even stronger during his tenure, which
benefits students like me. Last but not least, I am grateful to Dr. Melissa Lopez and Dr.
Tom D‟Agnes, who were fully supportive to my trip for job interviews. Special thanks to
Melissa for the lovely cake celebrating my defense at the classroom.
I cannot forget the instruction and help I received from Dr. Peter Monge and Dr.
Andrea Hollingshead in the Annenberg School for Communication, whom I met through
Dr. Janet Fulk. Dr. Peter Monge continuously challenged me to reflect on what I could do
with agent-based modeling. When I was writing this dissertation, I was constantly
reminded of his comments and challenges. I also liked his humor very much. I owe to Dr.
Andrea Hollingshead part of the theoretical basis of my dissertation. Her influence on this
dissertation, through the class and through her research, is significant. I don‟t think I
could have written this dissertation without knowing her and her works.
I also have to thank friends. My special thanks to Dr. James Polk, who became my
sincere friend. Discussions of this dissertation and other social issues with him were

vii
always pleasant and inspiring, and he became a wall to lean on when I was struggling in
the job market. June Muranaka and Christine Wilson are among the best administrators I
have ever met. I cannot emphasize enough the importance of the help I received from
them. June helped pave the way for all the processes entailed in finishing my degree,
which otherwise might have been very rugged. Many thanks to my colleagues at
Annenberg and SPPD, who all gave me valuable insight and ideas through discussion. I
appreciate Jason Weiner and Tim Walsh, with whom I conducted a case study mentioned
in Chapter 9. It was my pleasure to interview people with them. I would like to give
thanks to Sangmin Kim, who introduced me to the concept of social learning. I would
also like to give special thanks to Jong-soo Baek in the Department of Psychology, who
taught me the basics of Matlab. I also have a long list of friends who have been sincere
supporters during the time I was writing this dissertation. I am afraid I cannot list all of
them.
I would like to give my special respect and gratitude to my previous advisor, Dr.
Yong-duck Jung at Seoul National University. He is like my second father. My journey to
the study of public administration began by and through him. I cannot imagine another
life without him. If my study abroad in the U.S. can be considered successful, he should
get the full credit for it. Not only did he form my scholarly development through
intensive training in my early ages, but he also walked with me when I was weary. When
I was on the roaring, dark ocean, he was a guiding beam from a lighthouse.
Finally, from the bottom of my heart I thank my family. None of this would have
been possible without the love and patience they have given me. My parents always stood

viii
calm when my heart fluctuated. They were the anchor of my life. My wife has been with
me in the most difficult time to finish this dissertation. I cannot thank her enough for her
many sacrifices in those days. She deserves the same congratulations I received. My
daughter is my inspiration. When she laughs, I feel I can accomplish anything. I dedicate
this dissertation to my family.

ix
TABLE OF CONTENTS

Dedication

Acknowledgments

List of Tables

List of Figures

Abstract

Chapter 1: Introduction

Chapter 2: Collaborative Governance: The Context
Characteristics of Collaborative Governance
Measuring Performance of Collaborative Governance
Conclusion

Chapter 3: Group Information-Processing and Decision-Making: A
Behavioral Foundation of Collaborative Governance
The Hidden Profile Paradigm
Motivated Information-Processing in Group Decision-Making
Information Asymmetries Model
Conclusion

Chapter 4: Conceptual Framework
Collective and Egalitarian Characteristics of Collaborative
Governance and Managerial Interventions
Deliberative Characteristics of Collaborative Governance,
Motivation, and Information Sharing
Consensus-Orientation of Collaborative Governance and Decision-
Making
Mutually Beneficial Interaction, Motivation, Interdependence of
Interest, and Performance

Chapter 5: Methods
Agent-Based Computational Modeling
Overview of the Simulation
Constituents and Participants
Rules of Deliberation
Experimental Design
Dependent Variables
ii

iii

xi

xiii

xv

1

8
9
15
18

19

20
27
30
33

35
36

38

43

44

47
47
52
58
70
73
76

x

Chapter 6: Manipulation Check of the Computational Model
Manipulation Check Points
Comparison of Consequences between Symmetric and Asymmetric
Conditions
Results

Chapter 7: Effect of Social and Epistemic Motivation on the Performance of
Group Deliberation
Success
Global Responsiveness of the Collective Decision
Local Responsiveness of the Collective Decision
Learning
Satisfaction

Chapter 8: Effect of Group Level Variables on the Performance of Group
Deliberation
Success
Global Responsiveness
Local Responsiveness
Learning

Chapter 9: Discussion
Exemplary Cases for Analysis
Propositions Regarding the Effect of Motivated Information-
Processing
Propositions Regarding Group-Level Variables

Chapter 10: Conclusion
Information-Processing Perspective on Collaborative Governance
Limitations and Future Research

Bibliography

79
80
87

89

94

97
101
104
108
112

116

117
124
128
132

137
138
141

167

187
187
194

201

xi
LIST OF TABLES

Table 4-1: Type of Information-Processing Motivation and Related Bias

Table 5-1: Calculation of Constituents‟ and their Representative‟s Policy
Opinion in One Sub-Dimension: An Example

Table 5-2: The Design of Social Motivation

Table 5-3: The Design of Epistemic Motivation

Table 5-4: Calculation of the Level of Satisfaction: An Example

Table 6-1: Parameter Settings for Manipulation Check

Table 6-2: Results for Manipulation Check

Table 7-1: Parameter Setting for the Virtual Experiment about the Effect of
Motivation

Table 7-2: Success Rate and Time: Main Effect

Table 7-3: Success Rate and Time: Interaction Effect

Table 7-4: Global Responsiveness: Main Effect

Table 7-5: Global Responsiveness: Interaction Effect

Table 7-6: Local Responsiveness: Main Effect

Table 7-7: Local Responsiveness: Interaction Effect

Table 7-8: Learning: Main Effect

Table 7-9: Learning: Interaction Effect

Table 7-10: Level of Satisfaction by the Type of Motivation

Table 7-11: Satisfaction: Main Effect

Table 7-12: Satisfaction: Interaction Effect

41

63

68

70

78

89

90

95

99

99

102

103

106

107

110

111

114

115

115

xii
Table 8-1: Effect of Group Variables on Success Rate and Time to Reach
Consensus

Table 8-2: Effect of Group Variables on Global Responsiveness

Table 8-3: Effect of Group Variables on Local Responsiveness

Table 8-4: Effect of Group Variables on the Degree of Learning

Table 9-1: Contingency of the Performance of Motivated Information-
Processing on Measurement
118

124

129

132

164

xiii
LIST OF FIGURES

Figure 4-1: The Conceptual Framework of Deliberation in the Context of
Collaborative Governance

Figure 5-1: The Contents and Flow of the Computational Model

Figure 7-1: Success Rate

Figure 7-2: Time to Reach Consensus

Figure 7-3: Global Responsiveness

Figure 7-4: Global Responsiveness: Main Effect

Figure 7-5: Local Responsiveness

Figure 7-6: Newly Obtained Data Points through Information Sharing

Figure 7-7: Ratio of the Focal Dimensions Changed

Figure 8-1: Interaction between Interdependence and the Rule of Speaker
Turn: Success Rate

Figure 8-2: Interaction between the Rule of Speaker Turn and Forum Size:
Success Rate

Figure 8-3: Interaction between Interdependence and the Rule of Speaker
Turn: Time to Reach Consensus

Figure 8-4: Interaction between Interdependence and Forum Size: Time to
Reach Consensus

Figure 8-5: Interaction between the Rule of Speaker Turn and Forum Size:
Time to Reach Consensus

Figure 8-6: Interaction between Interdependence and the Rule of Starting
Alternative: Global Responsiveness

Figure 8-7: Interaction between Interdependence and the Rule of Speaker
Turn: Global Responsiveness

36

56

98

100

101

102

105

109

111

121

121

122

123

123

127

128

xiv
Figure 8-8: Interaction between Interdependence and Forum Size: Newly
Obtained Data Points

Figure 8-9: Interaction between Interdependence and the Rule of Starting
Alternative: Change of Policy Opinion
135

136

xv
ABSTRACT

In recent decades, there has been an increase in theoretical attention to
collaborative governance as a deliberative decision-making process among stakeholders.
Meanwhile, relatively less attention has been paid by scholars to behavioral and
procedural aspects of the decision-making process in collaborative governance. The
following study was based on group information-processing and decision-making
literature in social psychology to develop a decision-making model of collaborative
governance. This model conceptualizes collaborative governance as a collective,
egalitarian, deliberate, and consensus-oriented decision-making process. Further, this
model considers different types of human motivation and biases in information-
processing and group decision-making. This researcher employed an agent-based
modeling method to design a computational model of the deliberation process in the
context of collaborative governance to identify relationships among actors‟ social and
epistemic motivations; managerial interventions including agenda setting, speaking turn,
and forum size; and performance of collaborative governance such as success in reaching
consensus, responsiveness of the decision, and mutual learning. The results of the
simulation are discussed via nineteen propositions regarding the effects of motivated
information-processing behaviors, responsiveness to the public and specific stakeholder
groups, authentic dialogue, and managerial intervention. This study concludes by
discussing the prospect of an information-processing perspective on collaborative
governance.

1
CHAPTER 1: INTRODUCTION

Based on numerous case studies of collaboration among actors from the public,
private, and nonprofit sectors worldwide, researchers have increasingly recognized the
need for a theory of collaborative governance. Accordingly, theoretical frameworks of
collaborative governance have recently been proposed (e.g., Ansell & Gash, 2008;
McGuire, 2006; Tang & Mazmanian, 2009; Thomson & Perry, 2006; Thomson, Perry, &
Miller, 2009). Some efforts have focused on the overarching framework of collaborative
governance that includes contextual, structural, procedural, and outcome variables
(Ansell & Gash, 2008; Bryson, Crosby, & Stone, 2006), as well as common factors found
in successful collaborative governnance systems such as institutional conditions, human
and financial resources, social capital, and leadership (Leach & Pelkey, 2001; Thomson
& Perry, 2006). Further, research in this line has aimed to identify major factors and
their relationships that influence performance of collaborative governance (Bryson et al.,
2006). At the same time, other researchers have focused on the deliberation process
embedded in collaborative governance (Bouwen & Taillieu, 2004; Dryzek, 2000; Fung &
Wright, 2001; Roberts, 2002). As such, these researchers have argued that a deliberation
process that excludes political biases negative to authentic dialogue among stakeholders
is a critical factor for the success of participatory democracy (Beierle & Konisky, 2001;
Fung & Wright, 2001; Innes & Booher, 2010). In an ideal deliberation condition,
participants in a collaborative governance system may have equal opportunity to input
their opinions and knowledge into the group decision-making process. In other words, all
participants‟ policy opinions may be equally considered by other participants and all

2
information from participants may be contemplated by others in the process. Researchers
assume that when the collective decision-making process is not affected by a political
power structure, participants in the deliberation process will reach a decision that best
reflects the collective interests of their constituents. Similarly, researchers have begun to
examine collaborative governance as a collective learning process through which
participants come to understand the bigger picture of the social problem they collectively
confront, increase their knowledge and technology necessary to solve the problem, and
form social memories of success and failure (Blatner, Carroll, Daniels, & Walker, 2001;
Bouwen & Taillieu, 2004; Hahn, Olsson, Folke, & Johansson, 2006). As such, collective
learning occurs during the deliberation process in a manner that all participants better
understand the public problem and they come to believe that they can collectively solve
complex social problems if they recognize the interdependence of their interests and
share their knowledge with mutual trust (Booher, 2004; Folke, Hahn, Olsson, & Norberg,
2005; Thomson & Perry, 2006).
A theoretical problem in the literature of collaborative governance is that many
researchers have taken an institutional approach to describing the structure and process of
collaborative governance systems while paying little attention to deliberation behaviors at
the individual and group levels (Tang & Mazmanian, 2009). Although empirical research
has found that participants tended to be more collaborative and deliberative when
conditions such as the existence of facilitative leaders or interdependence of interests are
met (Himmelman, 1996; Thomson & Perry, 2006), public administration research has

3
seldom asked how participants actually behave during deliberation. Further, research has
not questioned the results of such behaviors concerning the quality of collective decisions.
Representation of diverse citizen interests, especially minority interests, by
participants in collaborative governance is a critical issue in determining the
responsiveness of the collective decision (Sowa & Selden, 2003). However, it is doubtful
whether participants in a collaborative governance system are able to perform an
unbiased collective deliberation at the group level and reach a decision by considering all
the information pertinent to the public issue. When participants are engaged in self-
interested information sharing, meaning that participants only share information
consistent with their own preferences to maximize their benefits from the collective
decision, high responsiveness of the collective decision is not guaranteed.
The development of a theory of collaborative governance as a deliberative
learning system requires a behavioral foundation of group deliberation because the
fundamental unit of learning is either individual or group (Huber, 1991; Kim, 1993;
Senge, 1990). Researchers on group information-processing and decision-making in
social psychology have investigated theoretical issues pertinent to the challenges of
collaborative governance mentioned above. According to previous research, group
members tend to mention and discuss commonly shared information more frequently
than unique information. That is, members fail to consider critical unique information in
a group decision-making setting (Stasser & Titus, 1985). Therefore, when information is
disparately distributed among group members, they frequently conclude with inferior
alternatives to an ideal due to biased information sharing. More recent research on group

4
information sharing behavior has focused on the behaviors of group members with
different social and epistemic motivations for information sharing and decision-making
(Brodbeck, Kerschreiter, Mojzisch, & Schulz-Hardt, 2007; De Dreu, Nijstad, &
Knippenberg, 2008; Wittenbaum, Hollingshead, & Botero, 2004). One conclusion from
this previous research is that groups often fail to be effective information processors
when information is disparately distributed and processed among members (Brodbeck et
al., 2007; Stasser & Titus, 2003), which interacts with members‟ motivations (De Dreu et
al., 2008).
The implication of the group decision-making models in research of collaborative
governance is clear: Participants in a collaborative governance system may not reach a
conclusion that fully reflects diverse citizen preferences and relevant information even
when they are not influenced by a political power imbalance during the deliberation
process. Because of behavioral tendencies that are inherent in human mind and group
dynamics, deliberation may still be subject to biases.
Based on the literature of collaborative governance and that of group decision-
making, the current research aimed to explore answers to the following questions: What
are the effects of human motivation in deliberation on the quality of the collective
decision-making in the context of collaborative governance? What institutional
conditions or managerial interventions, in a group deliberation process, might improve
the quality of the decision in collaborative governance? Further, this study aims to
answer the research questions by developing an integrated theoretical framework of
collaborative governance and group information sharing and decision-making model by

5
executing virtual experiments via a computational model of deliberation in collaborative
governance.
The literature of collaborative governance and group information-processing can
benefit each other. First, group information-processing models can provide a micro
foundation for a model of collaborative governance. When collaborative governance is
conceived of as a collective decision-making system in which people gather and share
knowledge about a social problem and solution, then group information-processing
models may explain the process of knowledge sharing among participants as well as the
quality of the decision made (Moreland, 1999; Lewis, 2004; Wittenbaum & Park, 2001).
Second, collaborative governance models can provide group information-processing
models with institutional contexts (Arrow, McGrath, & Berdahl, 2000) in which group
members are embedded to make a collective decision. Differing from groups in an
organization whose members are assumed to have a given, common goal, groups of
stakeholders in collaborative governance are composed of representatives who are
presumed to advocate their own interests, which are often driven by self-oriented
motivations (De Dreu et al., 2008). In addition, although social psychological research
has accumulated repeated findings about human information sharing behaviors, previous
findings are almost exclusively based on laboratory experiments with nominal groups
detached from any real world context. Therefore, consideration of the context of
collaborative governance in group information processing would contribute to the
enrichment of group decision-making theories.

6
The current study employed an agent-based modeling method to design a
computational model of collaborative governance to investigate the relationships among
information distribution, actors‟ information-processing behaviors, managerial
interventions in deliberation, and the performance of collaborative governance.
Complexity and the dynamic nature of deliberation processes were obstacles to theorizing
the deliberative dynamics of collaborative governance systems. It is extremely difficult
to collect multi-level data about the relationship between deliberative behaviors of
stakeholders and its consequences, in the natural environment, while controlling for other
variables that might affect the relationship between them. In addition, although
traditional case studies add knowledge about collaborative governance, a theory-based
approach to collaborative governance systems is necessary for a systematic understanding
of collaborative governance (Tang & Mazmanian, 2009). In this sense, agent-based
modeling was an apt alternative to investigate the research questions raised in this
dissertation. Agent-based modeling can provide researchers with extensive data about
the relationships among variables, which renders the opportunity to investigate
unexplored areas of theory beyond the limitations of empirical research. In addition, via
agent-based modeling, researchers can develop a comprehensive model of collaborative
governance that integrates group-level behaviors and system-level contextual variables.
In the following chapter, literature on collaborative governance is reviewed.
Through this review, collaborative governance is characterized as a group deliberation
process with four distinct characteristics through which participants build consensus on a
policy and search for mutually beneficial solutions. The issue of measuring the

7
performance of collaborative governance at the individual, group, and system-level is
also discussed. Next, the literature of group information-processing and decision-making
is reviewed to provide a theoretical background to understand the collective deliberation
process in collaborative governance. The chapter focuses on hidden profile models and
motivated information-processing models. Following the literature review, this
researcher proposes an integrated conceptual framework of the deliberation process in
collaborative governance, as the basis for the development of a computational model of
collaborative governance. In the methods chapter, details of the agent-based
computational model are described including the purpose of agent-based modeling to
study collaborative governance as well as deliberation, design of agents, rules,
environments, and other relevant measures. Details of a series of experiments and their
designs aimed to identify the main and interaction effects of agents‟ information-sharing
behaviors and managerial interventions on the performance of a collaborative governance
system are provided. Following, results of the simulation are presented and organized in
three ways: results of the manipulation check, results of experiments on information-
processing behaviors, and results of experiments on managerial interventions. In the
discussion chapter, theoretical implications of the results are discussed and theoretical
propositions are developed. Finally, in the conclusion chapter, a summary discussion
from an information-processing perspective on collaborative governance is provided and
the limitations of this research and future research directions are discussed.

8
CHAPTER 2: COLLABORATIVE GOVERNANCE: THE CONTEXT

In the public sector, decision-makers use various rules fur such collective
decisions including representative democracy, quasi-market, and participatory policy
formation and planning. Collaborative governance is a form of participatory democracy
that has gained research interest in recent years (Ansell & Gash, 2008; Bryson et al., 2006;
Cooper, Bryer, & Meek, 2006; McGuire, 2006; O‟Leary, Gerard, & Bingham, 2006;
Thomson & Perry, 2006; Thomson et al., 2009). The notion of collaborative governance
generally refers to a group of interdependent stakeholders, usually from multiple sectors
(public, private, and nonprofit sectors), who work together to develop and implement
policies to address a complex, multi-faceted problem or situation. The essential reason
that collaborative governance systems are increasingly used is that constituents and
policy makers expect that they can make a better decision by virtue of information and
knowledge shared by stakeholders. Specifically, collaborative governance can enhance
the responsiveness of a decision by ensuring equal participation of stakeholders. Further,
in collaborative governance systems, each stakeholder has more opportunity to influence
the collective decision compared to traditional bureaucratic decision-making processes.
The theoretical and practical expectations are that stakeholders‟ equal power in
collaborative governance (Fung & Wright, 2001; 2003) is conducive to more responsive
policy decisions concerning needs and interests than with policy solutions from top-down
approaches (Beierle & Konisky, 2000; Booher, 2004; Healey, 1996).
The purpose of this chapter is to develop a conceptual understanding of
collaborative governance from the viewpoint of information processing. Specifically,

9
literature on collaborative governance is reviewed from the perspective that conceives
collaborative governance as a deliberative group decision-making process. Conceptual
elements of collaborative governance in the literature are examined and key
characteristics describing collaborative governance as a deliberative group decision-
making process are discussed. In essence, collaborative governance entails a collective,
egalitarian decision-making process in which stakeholders seek a mutually satisfactory
outcome through deliberations to reach consensus on the matter at hand. The
characteristics of collaborative governance discussed in this chapter also call attention to
responsiveness as a primary performance measure of collaborative governance. However,
the concept of responsiveness invokes difficult theoretical issues concerning
collaboration among stakeholders who voluntarily participate in a collaborative
governance system on behalf of their own constituents. This chapter discusses the issue
of measuring the performance of collaborative governance while considering these
theoretical challenges.
Characteristics of Collaborative Governance
Collective and egalitarian decision-making process. Based on a
comprehensive review of literature on collaborative governance, Ansell and Gash (2008)
defined collaborative governance as a “governing arrangement where one or more public
agencies directly engage non-state stakeholders in a collective decision-making process
that is formal, consensus-oriented, and deliberative and that aims to make or implement
public policy or manage public programs or assets” (p. 544). Similarly, Thomson and
Perry (2006) suggested the following definition of collaborative governance

10
Collaboration is a process in which autonomous actors interact through formal
and informal negotiation, jointly creating rules and structures governing their
relationships and ways to actor decide on the issues that brought them together; it
is a process involving shared norms and mutually beneficial interactions. (p. 23)

These definitions highlight the notion that collaborative governance is a collective and
egalitarian process in that participants are endowed with substantive authority to make
collective decisions and each stakeholder has an equal opportunity to reflect his or her
preferences in the collective decision. Ansell and Gash (2008) point out that participation
without decision-making authority is not collaborative governance, rather simply a
consultation or hearing. Thus, collaborative governance is particularly valuable when
minority preferences need to be accounted for in the decision-making process, whether
this is due to a desire for information diversity (Booher & Innes, 2002; Feldman &
Khademian, 2007), a need for accuracy in decision-making (Postmes, Spears, & Cihangir,
2001), or political concerns (Fossett & Thompson, 2006). From an egalitarian point of
view, decisions in a collaborative governance system aim to increase the average level of
satisfaction among stakeholders and decrease the variance of the level of satisfaction,
with a particular emphasis on improving acceptability of the decision, stakeholders whose
interests and opinions have not been addressed effectively under other rules (cf. Rawls,
1958).
Deliberation. Collaborative governance is a deliberative process (Booher &
Innes, 2002; Bouwen & Taillieu, 2004; Dryzek, 2000; Roberts, 2002; Yankelovich, 1999).
Ideally, the behavioral manifestations of deliberation may include participants sharing
knowledge and information openly, considering all opinions as equally important, and

11
considering all information available before reaching a collective conclusion (Beierle &
Konisky, 2001). The deliberation process is often characterized as dialogue (Booher &
Innes, 2002; Roberts, 2002; Yankelovich, 1999). Roberts (2002) summarized the essence
of dialogue as “a process of mutual understanding that emerges when participants treat
each other with equality, not coercion, and when they listen empathically to one another‟s
concerns in order to probe their fundamental assumptions and worldviews” (p. 660).
Booher and Innes (2002) argued that, without authentic dialogue, stakeholders in
collaborative governance might not benefit from their diversity and interdependence.
Therefore, through deliberation, the broad range of interests and perspectives relevant to
the decision being made can be clarified (Healey, 1996), which would lead to a better
understanding of what different stakeholders really care about and why. This can also
increase the possibility of developing an alternative response to particular concerns and
priorities. In this way, collaborative governance can potentially enable collective
decisions that satisfy most stakeholders on the issues of most importance.
Consensus-orientation. Supporting the premise that it can help reach a
collective decision that satisfies a greater range of stakeholders, collaborative governance
is also a consensus-oriented decision-making process. In contrast to hierarchical systems,
where information flows up to the highest level and decision-makers integrate
information and choose an alternative (Carley, 1991; 1992) or a judge-advisor system,
where a judge seeks information and opinions from diverse advisors and the final
decision is up to the judge (Van Swol & Ludutsky, 2007), the goal of collaborative

12
governance is to reach consensus among participating stakeholders (Ansell & Gash, 2008;
Booher, 2004).
Consensus orientation is particularly important for two reasons. First, given that
stakeholders possess relatively equal power in the decision-making process, consensus
may be the only viable way to make a decision. Second, consensus may be a critical
factor during the implementation stage of a decision (Thomson & Perry, 2006). A
deliberative process may take more time when reaching conclusion compared to what a
hierarchical decree or majority voting system would require. However, the legitimacy
obtained from consensus among stakeholders may encourage them to take responsibility
for the implementation of the decision as well as voluntarily monitor the process (Booher,
2004).
The orientation toward reaching consensus raises the question of how to
determine that a consensus has been reached among stakeholders. On one hand, in a
collaborative governance system with a large number of stakeholders and considerable
complexity of interests, it may be nearly impossible to reach unanimity regarding the best
course of action, such that a standard of 100% percent agreement is too high to achieve.
On the other hand, the criterion of a simple majority is too low since collaborative
governance would not be clearly distinguishable from other types of collective decision-
making processes that operate according to the principle of majority rule. Many political
and social institutions use a criterion of two-thirds or three-quarters when making their
most important or fundamental decisions and a large majority like this comes closer to
the premise behind consensus that there is considerable, if not overwhelming, support for

13
the proposed alternative. Ideally, consensus is reached when all participants indicate that
they are willing to support the proposal, even if it is not their preferred alternative or that
they are not in full agreement with the proposed decision (Schein, 1999).
Mutually beneficial interaction. Finally, collaborative governance can enable
mutually beneficial interactions among stakeholders who bring different preferences and
interests to the policy discussion (Thomson & Perry, 2006). Stakeholders participate in
collective, collaborative decision-making because their interests are interdependent.
Collaboration is critical in this type of interdependent situation in which the achievement
of one stakeholder‟s goal depends on other stakeholders‟ actions (Beierle & Konisky,
2001; Fung & Wright, 2003; Thomson & Perry, 2006). In this situation, stakeholders can
develop relationships with each other to exchange useful resources (Pfeffer & Salancik,
1978) such as knowledge, information, funds, legitimacy, and human capital (Argote,
1999; Donahue & Nye, 2002; Leach & Pelkey, 2001). Further, such exchanges can
benefit relevant stakeholders in the short-term and help build social capital that has
potential value in the future. Another type of interdependence exists when two or more
stakeholders‟ interests are in conflict but there is room for coordination among them. For
example, a common pool resource management system aims to solve conflict among
resource users in a manner that gives them the right to use the resource while also
regulating their activities so as not to infringe on other‟s rights to use the same resource
as well as to prevent resource depletion (Ostrom, 2005; Wade, 1987). Wood and Gray
(1991) made it clear that, in collaborative governance, every stakeholder can develop
mutually beneficial relationships, gain what is needed from others, and provide what is

14
possessed; specifically, “Collaboration can occur if stakeholders can satisfy one
another‟s differing interests without loss to themselves…[T]he achievement of any
benefit is contingent on mutual action” (p. 161).
In collaborative decision-making, it is helpful for stakeholders to clearly
acknowledge their primary interests and reveal which aspects of the issue have the
highest priority (Fisher & Ury, 1981). Hollingshead (1996) found that when group
members had to determine the rank order of alternatives, group members were more
successful in performing deeper deliberation. In other words, they thoroughly used the
information they possessed and were able to choose the best alternative more frequently.
Clarifying their preferences and the order of their preferences may help stakeholders
recognize the area and degree of their interests. Therefore, each stakeholder may be able
to strategically act to pursue what is most important to them and yield to the other
stakeholders, what is less important. Literature on collaboration points to the importance
of a negotiated order (Gray, 1985; O‟Toole & O‟Toole, 1981) that emphasizes both the
clarification of preferences for each party and recognizes the interdependence of their
interests. In other words, clarification of preferences and priorities can help stakeholders
better understand their shared interests and the areas of agreement and disagreement
among them, thus, facilitating the search for mutually beneficial alternatives. This
process may also help stakeholders view the situation as win-win, rather than maintaining
a zero-sum orientation that can undermine the effectiveness of collaborative processes.
Ultimately, stakeholders in collaborative governance may reach a point that their “needs

15
and interests are not defined in terms of a single organization but in terms of the
interdependencies among the stakeholders” (Gray, 1985, p. 915).
Measuring Performance of Collaborative Governance
Responsiveness, effectiveness, efficiency, equity, accountability, and
responsibility are values to be pursued in public administration in a democratic society
(Stivers, 1994; Vigoda, 2000). According to Vigoda (2002), responsiveness implies that
public servants are sensitive to their duties and committed to serving citizens. Accuracy
or the degree of fit between citizens‟ needs and service providers‟ responses and
timeliness have been suggested as two indicators of responsiveness (Thomas & Palfrey,
1996; Vigoda, 2002).
In collaborative governance, responsiveness may be one of the most important
criteria by which to judge a system‟s performance. Because decision-making authority in
collaborative governance is distributed among participants, it is not an exclusive function
of the government to ensure system responsiveness. Rather, as suggested by complexity
theory (Connick & Innes, 2003), the responsiveness of the system as a whole is an
emergent phenomenon derived from the behaviors of interconnected actors who try to
achieve their own goals. In other words, the actions of individual actors may affect the
system‟s overall responsiveness, but the level of responsiveness may not be attributable
to specific actors (Arrow et al., 2000).
Of the various difficulties associated with the concept of responsiveness
(Saltzstein, 1992), one problem is the challenge of clearly defining to whom a
governance system should be responsive. On the one hand, stakeholders who participate

16
in a collaborative governance system on behalf of their constituents may want to focus on
their constituents‟ preferences and try to affect the collective decision toward those
preferences. This is common, at least at the early stages of collaborative governance.
Thus, maintaining the constituents‟ interests may lead to a failure to reach consensus and
thus dissolution of the collaborative process (Connick & Innes, 2003). However, an
obvious merit of collaborative governance is this process allows diverse voices to be
heard by all relevant actors in an effort to find a mutually beneficial solution. Indeed,
responsiveness to disadvantaged actors is an important political and administrative
concern of collaborative governance (Fossett & Thompson, 2005) and may be an
indicator how egalitarian the collaborative process is. Furthermore, diverse voices are the
source of unshared information that enhances the quality of the deliberation process
among people when information is disparately distributed (Stasser & Titus, 2003). From
this viewpoint, responsiveness to actors‟ own constituents may facilitate information
sharing and collective learning (Brodbeck et al., 2007).
On the other hand, collaborative governance systems can benefit from the
participation of actors such as entrepreneurial governmental representatives or non-
partisan nonprofit actors who are focused on the pursuit of public interest rather than
specific interests of particular constituents. One problem here is the concept of public
interest that guides collaborative governance deliberations, which is often vague or
misguided (Saltzstein, 1992). While all citizens have their own unique preferences
regarding desired public services, a collective entity such as the public may not exist in
reality, rather simply be a convenient mental construct. This construct is sometimes used

17
to refer to the majority voting coalition and professionals and powerful actors often
influence the articulation of its meaning (Rourke, 1992). In spite of these limitations, the
construct of general citizen is useful in collaborative governance to call attention to a
broad range of citizen interests that might otherwise be ignored. This is the case
especially when government agencies or key interest groups are co-opted by those who
can better mobilize resources, influence, and voice opinions in the political system, which
systematically restricts the capability of the system to recognize and respond to the full
array of citizens‟ needs.
This contrast between global responsiveness (i.e., the public) and local
responsiveness (i.e., each participant‟s own constituents), raises a question regarding the
nature of the relationship between these two goals in the context of a collaborative
governance system. While some tension between the two can be expected, another
possibility is that they may be reconciled via a deliberative collaborative governance
process that involves diverse actors, interests, and information. From a knowledge
perspective, reconciliation of the tension between the pursuit of narrow local interests and
broader general interests can be the result of “knowledge-as-participation” (Bouwen &
Taillieu, 2004, p. 146), that is, knowledge formed in the interaction among different
actors or knowledge is situated in the coordinated actions among the actors. In other
words, what is important is not the knowledge retained by individual participants but the
knowledge shared among them. It also may be that a deliberation process in
collaborative governance is a process of devising mutually beneficial alternatives that
may be unknown to individual stakeholders while assuming selfish motivations in

18
information-sharing and decision-making. That is, the successful outcomes of a
deliberative process within a collaborative governance system should meet three
requirements, (1) realization of mutual benefit among stakeholders, (2) local
responsiveness of the collective decision to each stakeholder‟s own constituents, and (3)
global responsiveness of the collective decision to the general constituents.
Conclusion
In this chapter, the conceptualization of collaborative governance, as a collective
decision-making process, was discussed. With a focus on collective learning,
collaborative governance is conceptualized by four characteristics, (1) collective and
egalitarian decision-making process, (2) deliberative decision-making process, (3)
consensus-oriented decision-making process, and (4) mutually beneficial interactions.
Performance measures of collaborative governance systems were also discussed. Further,
it was suggested that the performance of collaborative governance systems should be
measured from three perspectives, (1) whether the collective decision is satisfactory to as
many stakeholders as possible, (2) whether the collective decision is responsive to each
stakeholder‟s constituents, and (3) whether the collective decision is responsive to the
general constituents. The conceptualization of collaborative governance suggested in this
chapter provides the conceptual background upon which to design a computational model
of collaborative governance, which is discussed in more detail in the methods chapter.

19
CHAPTER 3: GROUP INFORMATION-PROCESSING AND DECISION-
MAKING: A BEHAVIORAL FOUNDATION OF COLLABORATIVE
GOVERNANCE

In recent years, public and private sector organizations have increasingly applied
team-management practices (Cohen, Ledford, & Spreitzer, 1996; Katzenbach & Smith,
2003; Lawler, Mohrman, & Ledford, 1992; Semler, 1989; Srivastava, Bartol, & Locke,
2006). Particularly, as interest about learning organizations has increased, researchers
have pointed out the virtue of groups as an appropriate learning unit in diverse social and
organizational contexts (Argyris, 1999; Edmondson, Dillon, & Roloff, 2007; Huber, 1991;
Kim, 1993; Senge, 1990).
In hierarchies, people have less incentive to share information because
information is seen as a source of power; each department becomes a silo of information
(Adler, 2001). In markets, information is exchanged as individual property. However,
pricing mechanisms have failed to properly estimate the price of information (Adler,
2001). In contrast, groups facilitate information flow among members based on trust and
reciprocity. Argote (1999) argued that groups are micro underpinnings of organizational
learning. Similarly, but with a different focus, Senge (1990) emphasized dialogue as “the
capacity of members of a team to suspend assumptions and enter into genuine „thinking
together‟” (p. 10). Through dialogue, groups may discover insights not individually
attainable. Dialogue at a deeper level may only be possible at the group level. Thus,
Senge (1990) considered groups as the fundamental units of learning in organizations.
Perspectives that view groups as information processors prevail in the study of
groups (Levine & Moreland, 1994). For example, research on transactive memory

20
(Wegner, 1987), hidden profile (Stasser & Titus, 1985), and groups as information
processors (Hinsz, Tindale, &Vollrath, 1997), have triggered studies on various methods
of group information-processing and decision-making. Specifically, group information-
processing and decision-making models in distributed knowledge conditions are
discussed. Further, this chapter includes a discussion of the hidden profile model, a well-
developed paradigm of research in studying group information-processing and decision-
making that explores cognitive and motivational biases that group members demonstrate
during group decision-making processes. This paradigm sheds light on the theoretical
development of stakeholder behavior in collaborative governance systems at the micro
level. The purpose of this chapter is to identify and describe conceptual and theoretical
frameworks provided by group information-processing literature relevant to modeling the
deliberation process of collaborative governance. In the following sections, the research
in the hidden profile paradigm is reviewed, followed by a review of more recent research
that explicitly considers human motivation and bias in information-processing. The
relevance of the models discussed in this chapter to the study of collaborative governance
is briefly examined in the conclusion.
The Hidden Profile Paradigm
Overview of the hidden profile paradigm. Among diverse theoretical
approaches to group information-processing, the hidden profile model provides a way to
understand how information distribution among group members affects the quality of
group decision-making. In their seminal research about information sharing behavior
among group members, later called the hidden profile model, Stasser and Titus (1985)

21
raised an intriguing question as to whether groups are really effective in using the
information available and reaching a better decision compared to the decision an
individual could make alone. The reason this model is called the hidden profile model is
that some unique, critical information necessary to recognize the best alternative is
hidden to some or all of group members. This means two things. First, unique
information is not distributed to all group members. Second, unique information is less
likely to be mentioned during deliberation. In Stasser and Titus‟ (1985) work, when
these conditions were met, somewhat unexpected effects of group information sharing
behavior were found. Specifically, group discussions were biased in favor of an
alternative indicated by commonly shared information among group members because
this shared information had a higher probability, compared to unshared information, of
being mentioned and, once mentioned, was repeated during the discussion. Consequently,
groups frequently failed to effectively use the information the group members possessed.
After Stasser and Titus‟ (1985) study, following studies replicated or expanded their
findings (e.g., Hollinghead, 1996; Larson, Christensen, Abbott, & Franz, 1996; Larson,
Christensen, Franz, & Abbott, 1998a; Stasser & Titus, 1987; Wittenbaum, 1998).
A typical situation manipulated in the hidden profile experiments was that group
members must find the most desirable candidate for student body presidency, which was
defined as the candidate group members would choose if they possessed all necessary
information about the candidates (Stasser & Titus, 1985; 1987). Variations were
implemented to find the guilty suspect in a homicide investigation (Stasser & Stewart,
1992), the best job candidate (Wittenbaum, 1998), and the best diagnosis of a patient‟s

22
condition (Larson et al., 1996; 1998a). Typically, from two to six persons participated in
the experiments. Participants were first given information about alternatives to form a
pre-discussion preference and were directed to engage in a group discussion. All or some
of the group members received information that supported alternatives other than the
most desirable option. Therefore, at the beginning of the discussion, all members had
both shared and unique information and different opinions on each alternative. There
was also no guarantee that the group members would select the same best alternative that
they would have if they had had all the information. For example, in Stasser and Titus‟
original experiment (1985), participants were assigned information such that members
preferred the same less desirable alternative (consensus) or they preferred different
alternatives (conflict). In summary, a hidden profile situation occurs when the following
conditions are met (1) information is distributed in a manner that group members possess
only part of the information about each alternative so that they may form a pre-discussion
preference for a suboptimal alternative, (2) information that supports the best alternative
is unshared or partly shared among group members, and (3) the best alternative can be
chosen if group members share all the unique information they possess. In a hidden
profile situation, unique or unshared information that one or some of the group members
possess may be critical in determining the best alternative. Further, shared information
may also be important. In research, however, this condition has been manipulated such
that using only shared information, the group members could not reach the best decision.

23
Basic findings from studies on the hidden profile have been well replicated.
1
First,
shared information was more likely to be mentioned during discussions than was
unshared information (e.g., Steward & Stasser, 1995; Stasser & Titus, 1987; Wittenbaum,
1998; 2000). Second, shared information was mentioned earlier in the discussion than
unshared information (e.g., Larson et al., 1996; 1998a). Unshared information was
usually mentioned for the first time when shared information was sufficiently discussed
and when the discussion time was long enough. Third, shared information was more
likely to be repeated during discussion after it was mentioned than was unshared
information (e.g., Larson et al., 1996; 1998a). Finally, because of these biases,
discussing shared information more frequently than unshared information in group
decision-making, groups failed to use the unique information that was critical to their
decision, which resulted in the failure to select the alternative they would have chosen if
they had had all the information.
Researchers found various factors that could facilitate sharing unique information
among group members and improve their decision-making quality under a hidden profile
situation. Wittenbaum et al., (2004) summarized the factors examined in the various
studies of the hidden profile into seven categories (1) information type and distribution,
(2) task features, (3) group structure and composition, (4) temporal features, (5) member
characteristics, (6) discussion procedures, and (7) communication technology.

1
For detailed references about each finding, see Wittenbaum et al. (2004) who published a comprehensive
literature review on hidden profile research. Although there has been following research in this venue since
the review was published, it still offers a clear and useful summary of related studies. Additionally,
although more recent studies are reviewed with earlier studies in this chapter, the framing of the review of
hidden profile research in this chapter owes much from their review.

24
Theoretical explanations for information-sharing behaviors. Findings from
the studies reviewed by Wittenbaum et al. (2004) raise a significant question about the
usefulness of group decision-making. One merit of group decision-making, compared to
individual decision-making or majority voting, is that groups can use a broad range of
information available, better validate the information provided by a member, correct
errors in information-processing, and be more creative than are individuals alone
(Hollenbeck et al., 1995; Vroom & Jago, 1988; West & Anderson, 1996). However,
findings have indicated that groups are less likely to use the best of their knowledge stock
in decision-making than are typically expected to. When groups fail to use all the
information available, their collective decision may not necessarily be better than the
most knowledgeable individual‟s decision. Various attempts have been made to provide
theoretical explanations for the bias in group information-processing. Stasser and Titus
(2003), in a comprehensive review of studies regarding this issue, summarized the three
ways a theoretical explanation developed.
First, the information-sampling model argues that shared information is more
frequently mentioned because it is widespread among group members such that its
probability to be mentioned is higher than that of unshared information. Stasser and
Titus (2003) summarize the idea; “[I]nformation that is widely available before
discussion has a sampling advantage over information that is available to one or a few” (p.
306). This model was later expanded to consider the dynamic that unshared information
has higher probability to be mentioned at a later stage in the discussion (Larson, Foster-
Fishman, & Keys, 1994), which included the effect of discussion flow that enhances the

25
probability information related to previously discussed information will be mentioned
(Stasser & Taylor, 1991). Finally, Stasser and colleagues (i.e., Hastie & Stasser, 2000;
Stasser, 1988; 2000) developed a computational model of information sampling behavior.
Second, the social cost model has tried to explain information-processing bias in
the hidden profile situation by focusing on the fact that mentioning unique information
may incur social costs, both to the discussant and to the whole group. Social costs
include the need of and difficulty in evaluating the validity of unique information (Larson
et al., 1996) as well as the risk of breaking the consensus norm (Postmes et al., 2001).
Larson et al. (1996) examined discussions among medical students, who diagnosed
patients‟ conditions, and found that lower status students refrained from repeating unique
information during the discussion, while higher status students repeated unique
information more frequently. Stasser and Titus (2003) pointed out that another line of
studies also found that recognition of formal or informal group member status, in the
domains pertinent to the unique information they mentioned, helped facilitate the sharing
of unique information. Therefore, the status of the information bearer was an important
cue of credibility for the information the individual provided and reduced other group
members‟ efforts for validation of the unique information (Larson et al., 1996; Stasser &
Titus, 2003). Status includes expertise (Stasser, Stewart, & Wittenbaum, 1995; Stasser,
Vaughan, & Stewart, 2000; Stewart & Stasser, 1995), leadership (Larson, Foster-Fishman,
& Franz, 1998b), and experience (Wittenbaum, 1998; 2000) in the domain pertinent to
the unique information.

26
Finally, it has been argued that group members tend to repeat shared information
more frequently than unshared information because discussion of shared information
“enhances impressions of both self and peer-competence and credibility” (Stasser & Titus,
2003, p. 311), which is known as mutual enhancement (Wittenbaum, Hubbell, &
Zuckerman, 1999). Specifically, Wittenbaum et al. (2004) pointed out that, in situations
of high uncertainty, group members depend on each other to evaluate the relative
importance of their information. They also noted that mutual enhancement better
explains why group members repeat shared information compared to how they explain
why they first mention this type of information. In fact, research suggests that shared
information is regarded as more important, relevant, and accurate than unshared
information; therefore, is worth mentioning (Postmes et al., 2001). Other studies have
argued that group members strategically mention shared information before they mention
unique information, which is a way to build their credibility (Wittenbaum & Park, 2001;
Toma & Butera, 2009). That is, in the absence of high credibility at the early stage of
discussion, group members strategically choose to mention shared information to
accumulate credibility.
Wittenbaum et al. (2004) added another explanation for information-processing
bias in the hidden profile situation. Specifically, group members tend to evaluate
information consistent with their preferences as more favorable (Greitemeyer & Schulz-
Hardt, 2003). Based on the findings, Wittenbaum et al. (2004) summarized the logic of
explanation as follows:
Because shared information largely supports members‟ initial preferences, it is
evaluated as more important and therefore more worthy of discussion than

27
unshared information. Even exposing members to all information after having
read a hidden profile does not improve decision quality because members engage
in biased information-processing to maintain their initial preferences. (p. 297)

The idea that the hidden profile situation cannot be solved because group members tend
to communicate information that is consistent with their preferences has been
sophisticated in the line of research that has resulted in the motivated information-
processing model (De Dreu et al., 2008; Toma & Butera, 2009; Wittenbaum et al., 2004)
and the information asymmetries model (Brodbeck, Kerschreiter, Mojzisch, & Schulz-
Hardt, 2002; Brodbeck et al., 2007). In the following section, the models are reviewed in
more detail.
Motivated Information-Processing in Group Decision-Making
Motivated information-processing model originated from the idea that real world
group decision-making processes are more complex than in experimental situations of the
hidden profile paradigm. From the viewpoint of motivated information-processing,
hidden profile research has been based on unrealistic assumptions. De Dreu et al. (2008)
argued that group information-processing and decision-making research that has focused
on the cognitive or, in their term, epistemic side of information-processing, has assumed
1) that group members are cooperative to the goal of finding the best alternative and they
are not motivated to opportunistically gain personal benefits from the process and
outcome of group decision-making but motivated to maximize the benefit of the group
and 2) that group members‟ contributions are indispensable for better decision-making.
Similarly, Wittenbaum et al. (2004) argued that the hidden profile research
“carries with it strong theoretical assumptions that bear little resemblance to many

28
naturalistic group decision-making situations and limit the applicability of the hidden
profile paradigm” (p. 298). Similar to De Dreu et al.‟s (2008) argument, Wittenbaum et
al. specified the assumptions of group information-processing research. First, they
argued that this research assumes information sharing is unbiased in the sense that group
members have no motive to screen out the recalled information and members are
cooperative. Specifically, the information sampling model‟s explanation of group
information-processing biases is parsimonious. However, it deviates from the reality of
human motivation to selectively communicate information that is consistent with their
preferences (Toma & Butera, 2009; Van Swol, Savadori, & Sniezek, 2003). Further,
when there is discrepancy between the group goal and their preferences, group members
may not act cooperatively (Tjosvald & Field, 1983; Toma & Butera, 2009).
Second, Wittenbaum et al. (2004) argued that group information-processing
research is based on the assumption that groups can outperform individuals or voting
schemes when unshared information is more important than shared information. This
assumption is based on the notion that “unshared information in work groups is held by
experts with valuable, unique knowledge to share.” (Wittenbaum et al., 2004, p. 301).
However, these assumptions may not apply to all cases (Fraidin, 2004). Wittenbaum et al.
(2004) appropriately pointed out that “favoring shared information during discussion
may hamper group members‟ ability to discover the correct solution to a hidden profile (a
task goal), but it may help members to develop trust and interpersonal closeness (a social
goal)” (p. 301). In other words, unshared information may be more important from the
perspective of task goals (achieving the task). In contrast, shared information brings a

29
wider range of benefits to the group (social goals). This argument has an important
implications for decision-making groups, such as public policy deliberation forums in
which political relationships are of concern. In decision-making groups (e.g., forums of
collaborative governance systems), achieving social goals before achieving task goals
may be more important because trust building is usually an urgent issue in groups that
have often experienced a long history of conflict (Booher, 2004).
De Dreu et al. (2008) summarized the motivation observed in the group decision-
making literature in two categories: epistemic motivation and social motivation. First, De
Dreu et al. defined epistemic motivation as “the willingness to expend effort to achieve a
thorough, rich, and accurate understanding of the world, including the group task or
decision problem at hand” (p. 23). They specified that the level of epistemic motivation
varies, depending on the perceived sufficiency of the information available to group
members. If group members perceive that the available information is insufficient to
make a final decision, they will engage in more information seeking, sharing, and
validating. Second, De Dreu et al. (2008) defined social motivation as “the individual
preference for outcome distributions between oneself and other group members” (p. 23).
This definition applies to two specific categories of social motivation: pro-self (motivated
by concerns with the decision maker‟s own interests) and pro-social (motivated by
concerns with group outcomes and fairness). The eminent characteristic of pro-self
social motivation involves spinning preference-consistent information. In other words,
group members who are motivated by pro-self tendencies may develop strong ownership
of their preferences and ideas by investing energy in developing their ideas (Abelson,

30
1986). Ultimately this will lead to “vehemently argue for their own perspective and
against new counterevidence provided by other group members” (De Dreu et al., 2008, p.
34). In contrast, pro-socially motivated group members may develop group ownership
for information and demonstrate less adversary to preference-inconsistent information
(De Dreu et al., 2008). In summary, De Dreu et al. (2008) developed a 2 (high/low
epistemic motivation) by 2 (pro-self/pro-social motivation) model regarding epistemic
and social motivation and argued that epistemic and social motivation may interact to
influence group information-processing as well as the quality of group decision.
Information Asymmetries Model
As a result of a series of experiments focusing on the effect of group members‟
pre-discussion preferences (Brodbeck et al., 2002; Greitemeyer, Schulz-Hardt, Brodbeck,
& Frey, 2006; Schulz-Hardt, Brodbeck, Mojzisch, Kerschreiter, & Frey, 2006), Brodbeck
et al. (2007) proposed the information asymmetries model, which described different
types of asymmetries in information-processing during group decision-making. They
viewed groups as “a vehicle for combining and integrating different knowledge, ideas,
and perspectives into high-quality decisions and innovations” (p. 459). Brodbeck et al.
(2007) further argued that asymmetric information-processing during the discussion
phase impedes groups from working as effective information processors. Additionally,
they identified three categories of asymmetries in information-processing: negotiation
focus, discussion bias, and evaluation bias.
First, Brodbeck et al. (2007) defined negotiation focus as an asymmetric
information-processing condition where “group members focus on exchanging and

31
negotiating opinions and preferences so that the dominant or majority position can be
identified and settled within the group” (p. 463). Similarly, Postmes et al. (2001),
examined consensus-oriented and critical group norms and found that, under the
consensus-oriented group norms, which is related to low epistemic motivation (De Dreu
et al., 2008), group members were less likely to solve the hidden profile problem. Under
the group norms condition, which typically hampers critical thinking, group members‟
epistemic motivation may be low, thus prompting consensus formation without fully
examining the information group members possess.
Second, discussion bias is composed of sampling bias and repetition bias that
favors shared and preference-consistent information (Brodbeck et al., 2007). Sampling
and repetition bias favoring shared information is well replicated in the hidden profile
paradigm. What is unique is that sampling and repetition bias favor preference-consistent
information.
2
Further, group members tend to share, that is, mention and repeat,
information that is consistent with their own preferences more often than information that
is inconsistent with their own preferences. This bias may be highly related to social
motivation (De Dreu et al., 2008).
Finally, evaluation bias is composed of ownership bias, which is a bias toward
shared information that can be socially validated, and evaluation bias toward preference-
consistent information (Brodbeck et al., 2007). First, ownership bias refers to the
tendency that group members perceive their own information as more valid than

2
Shared information and preference-consistent information are conceptually different constructs. However,
there will be significant overlap between them when information that group members form their
preferences on is mostly shared among them, which was demonstrated in as experimental condition called
“hidden profile/consensus” (Stasser & Titus, 1985).

32
information held by other group members (Van Swol et al., 2003). Because of this bias,
shared information is preferred. Specifically, shared information is owned by multiple
group members and shared information is more prone to social validation than is
unshared information (Brodbeck et al., 2007). Second, findings suggest that group
members judge preference-consistent information as more credible and important than
preference-inconsistent information (Greitemeyer & Schulz-Hardt, 2003), which is
referred to as a preference consistency effect.
The concluding argument for the information asymmetries model is that the
combination of asymmetric distribution and symmetric processing of information is the
key to groups‟ outperformance over individuals and voting schemes (Brodbeck et al.,
2007). In other words, when information is distributed among group members in the
form of a hidden profile (asymmetric distribution of information), one condition for the
group to outperform individuals or voting schemes is fulfilled. For this possibility to be
realized, however, there should be no information-processing asymmetries (i.e.,
negotiation focus, discussion bias, and evaluation bias). Brodbeck et al. (2007) left
specific interaction effects among the different types of asymmetries for further study.
For example, one context in which the interaction effects can be analyzed is a
collaborative governance system. It is quite likely that, in a certain context, such as
collaborative governance, where stakeholders bring diverse and intense pre-discussion
policy opinions to a group deliberation process and negotiate their policy opinions, strong
interactions among the diverse asymmetries will occur.

33
Conclusion
Raised early in this study, a question regarding collaborative governance is
whether the decision-making process in real collaborative governance systems can be as
deliberative as normative theory expects. Collaborative governance confronts practical
challenges and information about pertinent social issues is often disparately distributed
among stakeholders, which results in different opinions and perspectives about the issues.
For example, diverse organizations, such as government, housing developers, and
universities, may bring with them different data about affordable housing issues that
could contrast each other. Disparate distribution of information about a policy issue
among social groups may then prevent those groups from recognizing the bigger picture
of the issue and, thus, better solutions. Of course, according to the information
asymmetries model, information disparity may not be detrimental if stakeholders are
ready to learn and modify their preferences. However, the diverse motivations of
stakeholders involved in a collaborative governance process may aggravate the problem
of disparate information distribution. Although collaborative governance cannot be
successful without stakeholders who are motivated to pursue mutual benefits with other
stakeholders (Thomson & Perry, 2006), stakeholders are frequently motivated to act
exclusively for their own interests. In addition, although many policy issues brought to a
collaborative governance system require high level of intellectual investigation of the
problem and solution (Berkes, 2008), stakeholders are not always motivated to share
knowledge because of a low level of mutual trust. Consequently, it is quite uncertain
whether stakeholders are open enough to learn from each other and adapt their

34
perspectives accordingly. Nevertheless, the deliberation process has been treated like a
black box, the internal process of which has not been well investigated either
theoretically or empirically (Thomson & Perry, 2006). Further, few researchers have
approached the success and failure of collaborative governance systems from the
perspective of information distribution among stakeholders or the quality of deliberation
among them. As a result, the field of public management and policy does not yet have a
good understanding of group level interactions among stakeholders that might shape the
quality of the deliberative process or the outcomes of decisions achieved via collaborative
governance. The implication of the argument in the group decision-making literature is
that the group deliberation process, in the context of collaborative governance, may fall
short of the normative expectation of authentic dialogue (Innes & Booher, 2003; Roberts,
2002; Yankelovich, 1999). The group information-processing and decision-making
models reviewed in this chapter may serve as a theoretical foundation for this
deliberation process in the context of collaborative governance. In the next chapter, a
conceptual framework that integrates the context of collaborative governance and group
decision-making models is developed.

35
CHAPTER 4: CONCEPTUAL FRAMEWORK

In the previous chapters, collaborative governance was characterized as (1) a
collective and egalitarian decision-making process, (2) a deliberative decision-making
process, (3) a consensus-oriented decision-making process, and (4) mutually beneficial
interaction among stakeholders. In collaborative governance literature, researchers have
treated the process like a black box (Thomson & Perry, 2006) due to the difficulty of
empirical investigation of the internal dynamics of deliberation processes. This study
reviewed the conceptual framework of group information-processing and decision-
making literature. Finally, the issue of measuring outcomes of collaborative governance
was discussed. The purpose of this chapter is to provide an integration of the conceptual
elements of collaborative governance and group information-processing into a
comprehensive conceptual model, which is the basis to establish a framework of a
computational model of collaborative governance. In the conceptual framework and
computational model developed for this study, deliberation among stakeholders is
contextualized by the four aforementioned characteristics of collaborative governance. In
this chapter, a series of statements were established to connect the context of
collaborative governance and the behavioral model of the deliberation process. Outcome
measures, pertinent to the context and corresponding process of deliberation, are also
discussed. Figure 4-1 summarizes the overall conceptual framework of deliberation in
the context of collaborative governance.

36

Figure 4-1. The Conceptual Framework of Deliberation in the Context of Collaborative
Governance

Collective and Egalitarian Characteristic of Collaborative Governance and
Managerial Interventions

The literature on group decision-making processes has found that the minority
opinions are often ignored, or insufficiently discussed; that is, mentioned only near the
end of the discussion (Larson et al., 1996; 1998a). Finding highlights the importance of a
facilitative leader in collaborative governance systems (Thomson & Perry, 2006), whose
role is to maintain the process of a collaborative governance system. From this
leadership, participants remain in the process and their mutual benefit is realized.
The emphasis on facilitative leadership implies that a collaborative governance
system, in practice, may not necessarily be egalitarian in the absence of managerial

37
intervention. Rather, some managerial interventions, during the deliberation process,
may be necessary for the process to be egalitarian. Specifically, some intervening
measures can be employed to ensure that every stakeholder‟s interest is considered
equally. For example, the egalitarian characteristic of a collaborative governance system
may be realized by giving less satisfied stakeholders more opportunities, such as
speaking initiatives, to influence the deliberation process. In addition, according to the
implications of the information sampling model, another means to encourage appropriate
consideration of minority stakeholders‟ opinions is to allow stakeholders to begin the
deliberation with a proposal that supports their opinion. Finally, forum size also
influences how effectively the needs of the minority are dealt with in a forum. In large
forums, the voices of each stakeholder will not be reflected well in the collective decision
(Connick, 2006). In short, the collective and egalitarian characteristic of collaborative
governance is reflected in the computational model in the form of a managerial
intervention that gives an advantage to less-satisfied stakeholders. This managerial
leadership may help better reflect the interests of minority stakeholders in the decision.
In this context, the responsiveness of the collective decision to stakeholders, especially to
the relative minority, is a relevant measure of the outcome of a collaborative governance
system.
Statement 1-1. The collective and egalitarian characteristics of a collaborative
governance system are reflected in the form of a managerial intervention during the
deliberation process that offers the minority an advantage.

38
Statement 1-2. A relevant measure of the performance of collaborative
governance is the responsiveness of the collective decision to stakeholders.
Deliberative Characteristic of Collaborative Governance, Motivation, and
Information Sharing

In this study, it was assumed that the deliberative characteristic of a collaborative
governance system can be specified by the theoretical models of group information
sharing and decision-making processes. Deliberation in collaborative governance is a
process by which pertinent information is shared and a collective decision is made based
on mutual learning. Although the normative claim about the deliberation process
encourages an authentic dialogue situation (Yankelovich, 1999), which should be as
immune to cognitive, motivational, social, or political biases, empirical findings from
group decision-making literature indicate that the deliberation process is more likely far
from ideal (Stasser & Titus, 2003). Further, the information sampling model states that
human beings have an inherent tendency to discuss shared information more frequently
than to discuss unshared information (Stasser & Titus, 1985). The motivated
information-processing model states that human beings are affected by their self-oriented
motivations (De Dreu et al., 2008; Toma & Butera, 2009). Finally, research on the
effects of group norms has indicated that when people share information they are
constrained by social influence such as status difference among group members or
consensus-oriented norms (Larson et al., 1996; Postmes et al., 2001).
These tendencies in information sharing and decision-making shed light on the
understanding of the deliberation process in the context of collaborative governance.
According to the implications of group decision-making literature (Brodbeck et al., 2007),

39
first, it may be that stakeholders mention and repeat information that is already common
among them more frequently than they do unique information (information sampling bias;
e.g., starting discussion with well-known information about the pertinent social issue and
repeating it without assigning an appropriate time to discuss unique information that may
reflect minority opinions or may be important in finding a solution). Second, it may be
that individuals evaluate their preference-consistent information as more favorable and
reliable than preference-inconsistent information (evaluation bias; e.g., developmentalists
may evaluate survey data that shows local residents‟ preferences for gentrification of
their environment as more valuable and reliable than do community activists). Third,
individuals may also prefer negotiating their interests (e.g., log-rolling) to a critical
examination of alternatives by sharing information (negotiation focus). When avoiding
critical examination of alternatives, group members refrain from mentioning new
information that criticizes the current or majority opinion, rather they prefer the status
quo in order to quickly reach consensus (Brodbeck et al., 2007). Finally, it may be that
stakeholders strengthen their confidence on a piece of information or their opinion via
mutual enhancement (Wittenbaum et al., 1999). Rather than critically examining the
validity of information, group members tend to increase the credibility of shared
information simply because it is mentioned more frequently.
Different combinations of social and epistemic motivations may simultaneously
incur some biases (Brodbeck et al., 2007; De Dreu et al., 2008). Table 4-1 summarizes
motivation and corresponding biases. First, following De Dreu et al. (2008), this study
categorized the social motivation into pro-self and pro-social motivation. Pro-self

40
motivation, in information-processing, indicates that stakeholders try to more frequently
mention and discuss information consistent with their own opinions and tend to ignore
information that is inconsistent with their own opinions. Pro-social motivation indicates
that stakeholders do not discriminate between information that is consistent or
inconsistent with their own opinions. Second, this study categorized epistemic
motivation as either high or low. High epistemic motivation leads stakeholders to learn
as much information as possible, while low epistemic motivation discourages
stakeholders from learning and leads them to repeat the same information and prefer the
status quo. Information sampling bias is a group-level phenomenon (Brodbeck et al.,
2007) that implies that whatever the motivation of stakeholders in information-processing,
information sampling occurs because shared information has a higher probability of
being sampled due to its universality. Therefore, theoretically, information sampling bias
can be observed regardless of specific individual motivation. Information sampling bias
may not be as salient in the condition of pro-self motivation as in the condition of pro-
social motivation. Pro-self motivation encourages stakeholders to share information that
is consistent with their interests. Therefore, even unique information, as long as it is
preference-consistent, will have a higher probability of being mentioned by an
information bearer with pro-self motivation than by an information bearer with pro-social
motivation. Finally, mutual enhancement may be found under any motivation, and it may
be salient in individuals with pro-self and low epistemic motivations, both of which
prevent stakeholders from discussing a wide range of information. As they mention the
same information repeatedly, the degree of mutual enhancement increases.

41
Table 4-1
Type of Information-Processing Motivation and Related Bias
Social motivation
Pro-self Pro-social
Epistemic
motivation
Low Evaluation bias
Negotiation focus
Information sampling bias
Mutual enhancement
No evaluation bias
Negotiation focus
Information sampling bias
Mutual enhancement
High Evaluation bias
No negotiation focus
Information sampling bias
Mutual enhancement
No evaluation bias
No negotiation focus
Information sampling bias
Mutual enhancement

At one extreme, stakeholders may be motivated in a pro-self and low epistemic
manner in which negotiation of opinions among self-oriented members is the major
feature of the group decision-making process. This context reflects a highly fragmented
group of stakeholders who strive to assert their own interests and seldom learn additional
information. In this context, evaluation bias and negotiation focus may prevail. At the
other extreme, stakeholders may be motivated in a pro-social and high epistemic manner.
This context is close to an authentic dialogue situation where stakeholders remain open-
minded to others‟ opinions and readily learn additional information. In this context,
evaluation bias and negotiation focus may be minimized.
In summary, the computational model assumes that the deliberation process of
collaborative governance can be characterized as a group information-sharing and
decision-making process where stakeholders bring and share diverse information and
opinions, driven by different motivations and related information-processing biases. In
the process, different degrees of mutual learning among stakeholders occur, and because

42
of learning, group consensus on a policy is formed. The computational model typifies the
motivational patterns of stakeholders via the combination of pro-self and pro-social
motivation and low and high epistemic motivation and related information processing
biases, which combines De Dreu et al.‟s (2008) and Brodbeck et al.‟ s framework (2007).
The ultimate goal of deliberation in the context of collaborative governance is to
enhance responsiveness to constituents (Vigoda, 2002) and ensure that as many
stakeholders as possible are satisfied with the collective decision. Both global and local
responsiveness of the collective decision to constituents should be measured based on
results of the computational model of collaborative governance. In addition, since the
deliberation process is a collective learning process (Blatner et al. 2001; Hahn et al.,
2006), how much learning occurs among stakeholders during the deliberation process is
another relevant measure.
Statement 2-1. The deliberative characteristic of collaborative governance can be
represented by a group information-sharing and decision-making process where
stakeholders bring and share diverse information and opinions that are driven by different
motivations and related information-processing biases.
Statement 2-2. Specific combinations of human motivation in information-
processing may cause different degrees of information-processing biases.
Statement 2-3. Relevant measures of the performance of collaborative governance
include global and local responsiveness of the collective decision to the general or
specific constituents and the degree of mutual learning among stakeholders.

43
Consensus-Orientation of Collaborative Governance and Decision-Making
The premise that collaborative governance is a consensus-oriented decision-
making process implies that stakeholders, to declare that a collective decision is made,
should reach a significantly high level of consensus through deliberation. The
deliberation process in a collaborative governance system usually starts with a low level
of consensus among stakeholders (Ansell & Gash, 2008), as is the typical initial
experimental condition of the hidden profile paradigm (Wittenbaum et al., 2004).
Without reaching consensus, the performance of collaborative governance cannot be
evaluated as positive except to say that some degree of trust-building or mutual learning
could occur even in unsuccessful cases (Booher, 2004; Connick & Innes, 2003).
Different combinations of information-processing motivations may facilitate or impede
consensus formation. For example, it may be difficult to reach consensus when
stakeholders are motivated in a pro-self and high-epistemic manner. First, pro-self social
motivation means that stakeholders are in a competitive rather than a cooperative
condition in negotiating opinions (Toma & Butera, 2009). In such a condition, the rate of
reaching consensus decreases (Tjosvald & Field, 1983). Second, high epistemic
motivations mean that stakeholders do not conform to the status quo, rather perform
critical examinations of incoming information. As long as stakeholders remain critical
about the current proposal, more input will be necessary to reach consensus. In contrast,
stakeholders with pro-social and low epistemic motivations may reach consensus more
easily because they are open to others‟ opinion and they conform to the status quo.

44
In summary, the computational model assumes that stakeholders reach a high
level of consensus in order to declare that a collective decision is made in the context of
collaborative governance. From this perspective, whether stakeholders succeed in
reaching consensus is an important measure of the performance of collaborative
governance. In addition, for stakeholders to reach consensus, their level of satisfaction
should be high enough to agree on the decision. Therefore, the stakeholders‟ level of
satisfaction is also an important measure of the performance of a collaborative
governance system.
Statement 3-1. Stakeholders, in the context of collaborative governance, need to
reach a high level of consensus to achieve a collective decision at more than a simple
majority vote.
Statement 3-2. A relevant measure of the performance of a collaborative
governance system is whether stakeholders succeed in reaching consensus. Since the
extent to which stakeholders agree on the collective decision depends on their level of
satisfaction with the decision, stakeholders‟ level of satisfaction is another relevant
measure.
Mutually Beneficial Interaction, Motivation, Interdependence of Interest, and
Performance

The premise that collaborative governance is a mutually beneficial interaction
among stakeholders has significant implications regarding social motivations in
information-processing. In group information-processing literature, tasks assigned to
subjects were not directly associated with subjects‟ interests (e.g., „jurors‟ task to find the
guilty suspect (Hollingshead, 1996)) or subjects‟ interests were manipulated to explicitly

45
conflict with each other (e.g., Toma & Butera, 2009). The real context of collaborative
governance is far more complex than the simplified experimental conditions:
Stakeholders may find a mutually beneficial alternative through a complex negotiation
process in which stakeholders realize that the achievement of their goals depends on
others‟ actions (Thomson & Perry, 2006). Whether stakeholders internalize the goal of
finding a mutually beneficial alternative may depend on their motivation. In other words,
the recognition of the interdependence of their interests is a necessary condition for the
success of a collaborative governance system. Even when they recognize the
interdependence, however, stakeholders may not always be motivated to explicitly pursue
mutual benefits. Likewise, whether stakeholders can find a mutually beneficial solution
may depend on how much overlap or conflict of interest exists among them. For example,
too low a degree of interdependence may negate the need to depend on the collaborative
governance system to make a collective decision. Conversely, too high a degree of
interdependence may not be manageable by collaborative governance since the situation,
by definition, comes close to a zero-sum game. In short, the computational model in this
study assumed that stakeholders were in the condition of interdependence with different
degrees, which indicates that stakeholders‟ interests overlap to some degree, while
maintaining that there is room to find a mutually beneficial solution via deliberation.
The stakeholders‟ level of satisfaction measures whether they reach a mutually
beneficial decision. If stakeholders succeed in finding a mutually beneficial alternative,
the overall level of satisfaction of the stakeholders should increase compared to their
initial level of satisfaction. In addition, the responsiveness of the collective decision to

46
constituents is also an indicator of whether stakeholders reach a mutually beneficial
solution. Finally, the process of recognizing the interdependence of their interests and
devising mutually beneficial alternatives among stakeholders is, itself, a learning process.
Therefore, the degree of collective learning is another relevant measure in this context.
Statement 4-1. The interests of stakeholders participating in a collaborative
governance system are interdependent.
Statement 4-2. The context of collaborative governance yields a condition under
which stakeholders can mutually benefit through deliberation.
Statement 4-3. Relevant measures of the performance of a collaborative
governance system include the level of stakeholder satisfaction, responsiveness of the
collective decision, and degree of learning.
In conclusion, the statements proposed in this chapter serve as a theoretical
foundation of the computational model developed in this study. The context of
collaborative governance is more complex than typical experimental conditions designed
in group decision-making literature. Therefore, the conceptual elements of the group
decision-making literature were revised and extended to reflect the context of
collaborative governance. In addition, the context of collaborative governance
determined the development of various measures that reflect different dimensions of
performance of such systems. Consequently, the computational model integrates the two
levels of perspective into one model, which contains conceptual elements of context,
process, and outcomes.

47
CHAPTER 5: METHODS

Agent-Based Computational Modeling
The methodology of the research was agent-based computational modeling.
Since Cohen, March, and Olsen (1972) published an article on the garbage can model
using a simulation method, computational organization theory has developed as an area
of research that views organizations as computationally intelligent systems (March, 2001).
Likewise, computational modeling and simulation has provided a fourth leg for the social
science research stool; the three of which are theory, empirical data from natural
experiments, and empirical data from synthetic experiments (Levitt, 2004, p. 130).
According to Carley (2002, p. 7257), computational organization science is a neo-
information-processing approach to the study of social, organizational, and policy
systems that combines social science, computer science, and network analysis. Under
this perspective, social systems, including organizations, are multi-agent, distributed
intelligent systems (Weiss, 1999). The multi-agent modeling approach used in a what-if
fashion has improved our understanding of how different technologies, decisions, and
policies influence the performance, effectiveness, flexibility, and survivability of
complex social systems (Carley, 2002, p. 7257). In addition to theoretical developments,
agent-based modeling offers a way of using virtual experiments to predict the possible
results of real-world interventions. Levitt (2004, p. 127) argued that computational
emulation models of organizations are evolving into powerful new kinds of
organizational design tools for predicting and mitigating organizational risks and flexible

48
new kinds of organizational theorem-provers for validating extant organization theory
and developing new theory.
Computational organization theory (Carley & Gasser, 1999; Carley & Prietula,
1994; Lomi & Larsen, 2001; Prietula, Carley, & Gasser, 1998) is a field of study where
researchers depend on computational and mathematical models to study organizations.
Carley (1995) defined this field as follows:
Computational and mathematical organization theory is an interdisciplinary
scientific area whose research members focus on developing and testing
organizational theory using formal models. The community shares a
theoretical view of organizations as collections of processes and intelligent
adaptive agents that are task oriented, socially situated, technologically
bound, and continuously changing. (p. 39)

Computational organization theory has especially depended on simulation modeling
methods including procedural and agent-based computational modeling methods
(Axelrod, 1997; Burton, 2003; Carley, 1996; Epstein, 1999; Harrison, Lin, Carroll, &
Carley, 2007). Harrison et al. (2007) defined a computer simulation as a computational
model of system behavior coupled with an experimental design. They also identified
three types of simulation models, including agent-based, systems dynamics, and cellular
automata models (Harrison et al., 2007). Computational modeling, especially agent-
based modeling, is composed of three components: agent, rule, and environment (Epstein
& Axtell, 1996). In a simulation model, an agent is defined as having characteristics
derived from pertinent theories or empirical findings. The most important characteristic
of an agent in agent-based modeling is the agent‟s information-processing capabilities.
Carley and Newell (1994) developed a typology of artificial agents from omniscient to

49
cognitive-emotional, degrading intelligent power, respectively. It is usually assumed that
agents are bounded-rational, meaning that they do not possess all the information extant
in the virtual world; do not search for all alternatives; and behave in a myopic, locally
optimal way. Agents are also to follow rules set by the researcher and may choose an
action, interact with each other, and perform intellectual tasks according to pertinent rules
in the model. Rules should represent the theoretical premises or core dynamics of social
phenomena that a researcher wishes to study. Further, the environment is composed of
resources and demands that provide agents with tasks, information, and other resources
that may change turbulently or remain stable depending on the experimental designs.
With this method, research has demonstrated how local interactions among bounded-
rational agents result in global emergent structures (Epstein & Axtell, 1996; Epstein,
2006; Halley & Winkler, 2008; Holland, 1998; Holland & Miller, 1991; Miller & Page,
2007).
Both computational modeling (e.g., simulation) and mathematical modeling (e.g.,
formal logic) are used in computational organization theory (Carley, 1995). Agent-based
modeling is a powerful tool for hypothesis generation and theory building when more
than a few heterogeneous agents are involved in complex interactions in the model and
the environment is changing (Harrison et al., 2007; Miller & Page, 2007). This type of
modeling allows researchers to overcome the limitations typically existent with formal
models (e.g., assumption of homogeneous agents, limited number of agents (usually 1 or
2), and restricted repertoire of interaction (Miller & Page, 2007)).

50
The merits of computational modeling fall into two categories. First,
computational modeling is a tool to handle complex phenomena. Epstein (1999) argued
that the virtue of computational modeling is to generate complex social phenomena with
a relatively simple set of rules and bounded rational agents. To deal with complex social
phenomena, characterized by nonlinear relationships among variables, methods that
assume linear and deterministic relationships among variables are not adequate.
Therefore, computational modeling can be an alternative to traditional methods in this
sense. Second, computational modeling provides researchers with three ways to
overcome the constraints of empirical research methods. One major problem of
empirical experiments is to control variables. While it is almost impossible to control all
possible variables in social experiments, simulations allow researchers to control
variables that are not of research interest (Harrison et al., 2007). Another merit of
computational modeling is that, in virtual experiments, researchers are not constrained by
time, sample size, and other contamination of samples caused by an experiment and
general history. Further, computational modeling can explore a larger range of
conditions of independent variables.
Collaborative governance is a complex process of interactions among
heterogeneous and interdependent actors in which nonlinear relationships among
numerous factors influence the evolution and outcomes of the process. Given these
features of collaborative governance, agent-based modeling provides many advantages in
theorizing the process of collaborative governance. First, in using an agent-based
simulation, researchers can investigate the complex interaction processes among agents

51
that are designed to act like stakeholders in collaborative governance. Second,
collaboration is affected by both individual motivations and institutional settings. A
comprehensive framework of collaborative governance should then consider both
individual and institutional-level variables. Agent-based modeling can also help
researchers explore such a comprehensive framework by allowing them to include
heterogeneous actors with different motivations and rules of collaboration within a model.
Third, during model development, agent-based modeling encourages researchers to
develop concretely defined concepts and measures of collaborative governance. Because
an agent-based model is numerical and based on algorithms, all objects and functions in
the model should be formally defined. As such, researchers are pushed to go beyond
verbal definitions of core concepts of collaborative governance. Fourth, agent-based
modeling is often combined with an experimental design (Harrison et al., 2007), which is
a more powerful method to investigate causal relationships between variables compared
to regression analyses. Finally, the study of collaborative governance has primarily
involved case studies. Recently, however, there has been increasing interest in the
development of a theory of collaborative governance (Tang & Mazmanian, 2009).
Agent-based modeling is one method dedicated to theory building inductively and
deductively.
Agent-based modeling can be used to simulate the deliberative decision-making
process of collaborative governance. In a deliberative process, participants share their
information and adjust their initial preferences according to new information, which is
the process of building a shared mental model and mutually adjusting interdependent

52
interests. As will be described in detail in this chapter, through agent-based modeling
using learning agents, learning attitudes of individual actors can be simulated. Second,
participants are also constrained by institutional settings or decision-making rules. In the
real world, various institutional rules are mixed and it is difficult to discern the effect of
each rule on the performance of a collaborative governance system. Through agent-
based modeling, researchers can either apply a theory-based set of rules to examine their
effects or design a comparative study on different institutional settings. Finally, some
critical variables are extremely difficult to measure. Agent-based modeling provides a
way to overcome the problem and explore new theories regarding such variables. For
example, interdependence of interest among stakeholders is an important characteristic of
collaborative governance. However, the level of conflict of interest among stakeholders
is not easy to measure in the real world. In virtual experiments, researchers can control
the level of conflict among stakeholders and observe how the level of conflict interacts
with other variables.
In conclusion, agent-based modeling is an apt method by which to examine the
internal dynamics of the deliberation process of collaborative governance while
considering both individual and institutional level variables to develop a comprehensive
theoretical model of collaborative governance.
Overview of the Simulation
The computational model developed in this study was designed as a simplified
representation of the deliberation process in a collaborative governance system. This
study conducted virtual experiments using the computational model to examine the

53
effects of human motivation and interactions surrounding the deliberation process. The
computational model simulated a collective decision-making situation similar to those
situations manipulated in the experiments of group decision-making research
(Wittenbaum et al., 2004). The typical procedure of many experiments began with a
group of two to six decision makers. The task of the decision makers was to collectively
find, from among two or three alternatives, the best or correct alternative indicated by the
information provided. For example, in Stasser and Stewart (1992) and Hollingshead
(1996), participants were assigned a juror role and required to select the guilty suspect
among a few suspects in a homicide investigation. More than 20 data points about the
circumstances and suspects were distributed to the subjects in such a way that individual
subjects could not recognize the correct guilty suspect without receiving additional
information from others via the discussion. That is, all subjects received different subsets
of the complete data points that they formed different pre-discussion beliefs about each
suspect. Subjects shared information they possessed during deliberations and collectively
determined which suspect was guilty. Similar schemes of this experiment have also been
used in other studies in group decision-making literature (cf., Hollingshead, 1996).
This dissertation also used a similar experimental design to develop the
computational model of this study. In general, the model is composed of decision makers
who possess different information and opinions about decision alternatives, rules of
deliberation among the decision makers, and the environment as a group setting in which
decision makers were located to reach a collective decision.

54
In addition to the group decision-making condition, this study contextualized the
group decision-making model under the condition of collaborative governance. The
context of collaborative governance adds features to the group decision-making model
that locate the group setting in a more general political condition. One contextual feature
of the current model is the rule of political representation. Unlike groups formed within
organizations and composed of relatively homogeneous members in terms of goals, a
group of stakeholder representatives in a collaborative governance system is formed out
of a larger society and composed of heterogeneous participants who represent different
constituents. The ideal of political representation is that (1) representatives act on behalf
of the constituents who delegate political authority to the representatives to participate in
public deliberations and (2) the representatives‟ actions are evaluated in terms of the
responsiveness of their actions to the constituents (Miller & Fox, 2007; Vigoda, 2002).
Generally speaking, the context of collaborative governance defines the rule of the
formation of participants as representatives of constituents and the criteria upon which to
measure the responsiveness of their decision. The computational model developed in this
study reflects this rule of political representation in the definition of collaborative forums
and in the measurement of participant performance.
Another contextual feature of the model is the rule of managerial intervention to
the deliberation process. For a collaborative governance system to be egalitarian, for
example, minority stakeholders may be given more opportunities to speak to their
interests in the deliberation process. In addition, a strict criterion upon which to
determine whether the group of participants reaches consensus also reflects the context of

55
collaborative governance. The other contextual feature of the model is the task the group
performs: find a mutually beneficial policy alternative. In collaborative governance,
stakeholders are usually in a non zero-sum game situation (Fisher & Ury, 1981; Thomson
& Perry, 2006). Therefore, the computational model reflects the possibility of mutually
beneficial adjustment of interests among stakeholders in the design of the deliberation
process.
This study built a computational model that reflects the experimental conditions
of preceding group decision-making research and includes additional contextual features
of collaborative governance. Figure 5-1 illustrates the contents and flow of the
computational model. The simulation begins by defining constituents.
3
Each constituent
is given a set of information; that is, data points that contain a positive or negative signal
about a decision alternative. Led by the information signal, each constituent forms a
policy opinion. Next, participants who represent subsets of the constituents are defined.
The constituents are grouped into a given number of parties that could vary in size
according to the similarity of their policy opinions. All the information available in a
party is collected. Again, led by the signal of the information collected from the
constituents, the party forms its collective policy opinion and the representative of the
party advocates the policy opinion as a participant in the following deliberation process.

3
The concepts marked by italic in the following paragraphs are described in more detail in the next sections.

56

Figure 5-1. The Contents and Flow of the Computational Model

Once participants are defined, the group decision-making process starts. The
participants who represent each party gather, set a starting policy alternative to be
discussed, and share information they possess regarding the alternative to affect other
participants‟ opinions. Participants are assumed to participate in the deliberation process
with different motivations. Specifically, some are more self-oriented (social motivation)
and some are active in sharing a wide range of information (epistemic motivation). In

57
accordance with type of motivation, participants show information-processing biases in
that 1) participants tend to provide information consistent with their policy opinions and
evaluate new information provided by other participants based on the consistency of the
information to their own preferences (evaluation bias), (2) shared information among
participants is more frequently mentioned and repeated (information sampling bias), (3)
participants tend to focus on negotiating their policy opinions rather than critically
examining the solution (negotiation focus), or (4) participants tend to mutually strengthen
the credibility of a data point simply because it is mentioned frequently (mutual
enhancement). When a speaker turn, determined by a certain rule (e.g., priority given to
the minority), is given to a participant, the participant provides the group with a data
point selected from the information he or she possesses. The selection of the data point is
determined by the experimental manipulation of participant motivation. Once the data
point is shared with the other participants, they consider the importance of the data point
and determine whether they will revise their policy opinions. As a result of this
collective learning process, participants‟ diverse policy opinions are adjusted. At each
iteration, consensus among the participants is determined; that is, whether the current
alternative is satisfactory among all participants. If consensus is not reached, the
deliberation process continues. If the group reaches consensus, the deliberation process
ends and the outcomes of the deliberation process, such as the time to reach consensus,
the degree of mutual learning, the degree of participant satisfaction, and the level of
responsiveness of the decision to the constituents, are measured.

58
The details of the computational model are provided in the following. First, the
properties of constituents and participants as well as the rules that define their properties
are described. Second, the rules of deliberation are described. Third, three experimental
designs to investigate the effect of motivation in managerial intervention are described,
followed by the measurement of the dependent variables.
Constituents and Participants
In the current computational model, two types of artificial agent were defined:
constituents and participants. Constituents were designed to represent citizens with their
own policy opinions who are affected by policy decisions made by a collaborative
governance system. Participants were designed to act as representatives of subsets of
constituents whose policy opinions were similar and participate in the deliberation
process in a collaborative governance system. Each constituent was designed to possess
two properties: policy opinion and information. The same generic properties comprised
each participant, although the method by which each participant‟s properties were
defined was different from the means by which each constituent‟s properties were
defined. In addition, since participants were the major agents in the simulation, they
were designed to follow the rules of deliberation. In this section, the definition of generic
properties of constituents and participants is described. The rules that determine
constituents and participants‟ properties are also described.
Policy opinion. Constituents and participants were designed to have a policy
opinion, which was numerically expressed by a string of eight binary numbers (e.g.,

59
1/0/1/1/0/1/0/0).
4
This design reflects two theoretical features. First, the design reflects
the fact that constituents‟ policy opinions are diverse. With the eight binary numbers,
256 different policy opinions could be defined. Similarly, the design reflects the fact that
constituents‟ policy opinions are not completely divergent: Some constituents may have
very similar opinions. The value of each digit was determined by the signal of the set of
information, which is described in detail below. Second, the design means that the
computational model assumes a policy is composed of independent sub-dimensions or
opinions that can be independently determined.
5

A constituent‟s policy opinion was used in two ways. First, it was used to
identify constituents who are similar to each other in terms of policy opinions. Similar
constituents formed a party and generated a representative. Second, a policy opinion was
the basis on which to calculate the responsiveness of the collective decision made by a
collaborative forum. A participant‟s policy opinion was something to be adjusted via the
deliberation process. A participant‟s policy opinion was also used to calculate level of
satisfaction with the policy alternative.
Information. An agent‟s policy opinion (for both constituent and participant)
was determined by the characteristics of the information the agent possessed. This design
corresponds to the common means of manipulating subject preferences in group decision-

4
The design of agent characteristics in the form of a binary string has been frequently used in agent-based
modeling literature (Epstein & Axtell, 1996; Holland, 1995).
5
From the complexity perspective (Miller & Page, 2007; Waldrop, 1992), independence among sub-
dimensions of a policy corresponds to the stable status of a system in the spectrum from stability through
the edge of chaos to chaos. As the degree of interdependence among sub-dimensions increases, the system
runs to chaos (Kauffman, 1993), which makes it difficult to discern the effect of the focal variables in this
study. For the purpose of the stepwise development of a theory, this study designed the stable status.

60
making literature (e.g., Hollingshead, 1996) by distributing selective information that
would influence their preferences in a certain direction. This design also reflects the
theoretical idea about the effect of pre-discussion preference affected by disparately
distributed information (Brodbeck et al., 2007).
The experimental procedures commonly used in group decision-making processes
(Stasser & Titus, 1985; Wittenbaum et al., 2004) usually distribute information that is
either positive, negative, or neutral to an alternative. In this study, to simplify the model,
two types of information were defined: positive and negative with regard to a sub-
dimension of a policy. Positive information encouraged the information bearers to set the
value of the dimension as 1. Negative information encouraged the information bearers to
set the value as 0. The maximum number of data points assigned to a constituent for one
sub-dimension of the policy opinion was set to 32 (16 data points for each type of
information), resulting in an 8 (dimensions) by 32 (data points) matrix. The actual
number of data points assigned to each constituent varied across constituents.
Rule of policy opinion formation: Constituents. A simple rule to determine a
constituent‟s policy opinion on a sub-dimension is to count the number of positive and
negative data points he or she possesses and determine what type of data points prevails.
For example, when the number of data points with positive signals exceed the number of
data points with negative signals, the constituent‟s policy opinion in the dimension could
be set as 1. This rule is simple. However, it does not reflect the important theoretical
feature that different group members weigh the same information differently (Brodbeck

61
et al., 2007; Stasser, 2000).
6
That is, it is reasonable to assume that some data points are
perceived as being more important than others are. Thus, the computational model
considers weights constituents placed on data points.
According to the signal of a data point (positive or negative), its weight was
randomly generated and ranges from -1 to 0 (when negative) or 0 to 1 (when positive).
Consequently, a weight contains two types of information: the direction of the signal and
the importance of the data point perceived by a constituent. A constituent‟s policy
opinion in a sub-dimension is then determined as 1 if the sum of the weights of the data
points, pertinent to the dimension, is greater than zero. Conversely, a constituent‟s policy
opinion in a sub-dimension is determined as 0 if the sum of the weights of the data points
is less than zero. The absolute deviation from zero indicates the strength of the opinion.
As such, policy opinions across all dimensions are determined by the same rule.
7

Rule of policy opinion formation: Participants. The formation of participants‟
policy opinion is designed differently compared to that of constituents. Specifically, the
rule of participants‟ policy opinion formation corresponds to the logic of political
representation. In the current study, constituents were grouped into parties in accordance
with the similarity of their policy opinions. Below is the process of constituent grouping.

6
Though not critical, another problem is that the rule cannot exclude the case in which the numbers of the
positive and the negative data points are the same.
7
This rule does not logically exclude the possibility that the sum of the weights equals zero. However,
with randomly generated weights of up to more than four decimal places, the probability becomes nearly
zero and was not realized in the actual simulations.

62
1) Five out of the eight dimensions in the policy opinion were randomly selected.
For example, a vector of the selected dimensions might be [1,2,4,6,8].
8

2) A binary string with five digits was randomly generated, for example,
[1/0/1/1/0]. Each of the digits corresponded to the selected five dimensions.
3) The two vectors, defined above, determined which constituents would be
selected. In the preceding examples, the two vectors indicate that the policy
opinions in the first, second, fourth, sixth, and eighth dimensions should be 1,
0, 1, 1, and 0 respectively. Consequently, all constituents whose policy
opinion is [1/0/*/1/*/1/*/0] were grouped into one party. The asterisk
indicates both 1 and 0 on the dimension.
4) Next, part of the five dimensions, called focal dimensions, were determined.
This design indicates in how many dimensions of the policy opinion
constituents will be exclusively interested. For example, when the number of
focal dimensions is manipulated to three, three positions are randomly
selected out of the party opinion vector [1, 2, 4, 6, 8], for example, [1, 2, 6].
5) Consequently, the party‟s policy opinion to be advocated in deliberation is
determined [1/0/*/*/*/1/*/*], which means that the sub-opinions at position 1,
2, and 6 are activated and proclaimed by the constituents in the party.
6) The same process is repeated as many times as the number of parties required.

8
The reason that the model starts the definition of a party in this way is twofold. First, the definition
implies that parties are designed as local actors. Like stakeholder groups in the real world, they are not
interested in all policy dimensions, rather are only interested in a specific range of policy dimensions.
Second, the definition helps control the size of a party. With five dimensions to define a party, all parties
generated were sized similarly. Size of stakeholder groups may be an important factor that affects the
deliberation process. However, this was not the focus of this study.

63
Once parties were generated, each generated one agent that would represent its
collective policy opinion in the following deliberation process. The collective policy
opinion of a party was predetermined by the rule described above and was accompanied
by the weights that were calculated with the following rule. First, all constituents‟ data
points and their weights were listed on a scoreboard by dimension. Second, all weights
of a dimension on the scoreboard were summed. If the party‟s collective policy opinion
in the dimension was 1, the sum of the weights should have been positive. If the party‟s
collective policy opinion in the dimension was 0, the sum of the weights should have
been negative. Table 5-1 shows the calculations of the constituents and their participant‟s
policy opinions.

Table 5-1
Calculation of Constituents’ and Their Representative’s Policy Opinion in One Sub-
Dimension: An Example
Dimension i
Data
Point 1
Data
Point 2
Data
Point 3 …
Data
Point 32 Sum
Expressi
on of the
opinion
Constituent 1 .20 .05 .50 … .27 1.02
1
Constituent 2 .39 -.36 .42 … .32 .38
1
…

Constituent N .46 -.35 -.06 … .37 .42
1
Representative (sum of
each column) 1.05 -.66 .86 … .96 1.82
1
Note. Each number indicates the weight of each data point. The sign indicates the
direction of the data point (positive or negative). Expression indicates the binary form of
the policy opinion.

These rules of policy opinion formation have some theoretical implications. First,
because each party has a distinct identity concerning member policy opinions, the
participant that represents the party consequently is delegated a set of dimensions of the

64
policy opinion that is most concerned with in the deliberation process (focal dimensions).
The concept of focal dimensions reflects the theoretical premise that participants are
motivated to share, strategically, information that supports their specific policy opinions
(Brodbeck et al., 2007; De Dreu et al., 2008). That is, participants first recognized which
dimensions of their policy opinions were crucial to their satisfaction and then advocated
for them. As the number of focal dimensions increased, participants confronted higher
interdependence of interest on those focal dimensions with other participants.
Accordingly, the need for mutual adjustment of interest became more eminent (Fisher &
Ury, 1981; Thomson & Perry, 2006). This study manipulated three degrees of
interdependence of interest among participants: low (one focal dimension), moderate
(three focal dimensions), and high (five focal dimensions).
9

Second, while a constituent‟s weight vector was generated randomly and ranged
from -1 to 1, a participant‟s weight vector was defined by the sum of the weight vectors
of all constituents it represented. This reflects a situation of strict political representation
in which participants depend entirely on information provided by local constituents and
on the opinions expressed by the constituents when they formed their policy agenda. The
stronger the constituents argue for or against a policy, the greater is the weight given to
the opinion.

9
Note that by design, the maximum number of focal dimensions is less than the length of the policy
opinion; stakeholders in this study are characterized as local actors. That is, no stakeholder‟s interest
stretches to all the dimensions of the policy. This reflects that real world stakeholders in collaborative
governance usually act locally: they seldom try to “dominate” the overall policy decision, and pursue
relatively narrow interests.

65
Finally, in this simulation design, asymmetry of information distribution among
constituents was assumed in defining their policy opinions. The basic assumption in
group decision-making literature is that group members form their pre-discussion
preferences according to the information distributed to them (Brodbeck et al., 2002; 2007;
Greitemeyer & Schulz-Hardt, 2003; Stasser & Titus, 1985). Therefore, empirical
experiments performed in the group decision-making literature have manipulated
information distribution in such a way that participants start discussion with different pre-
discussion preferences on the alternatives. Following this tradition, the current study
asymmetrically distributed information among constituents such that they formed biased
policy opinions toward the signal the information they possessed indicated.
Rule of motivation in information-processing. According to Dreu et al. (2008),
participants are motivated in two ways in information sharing and decision-making:
social (pro-self/pro-social) and epistemic (high/low). In this dissertation, consistent with
the types of motivation, participants were defined to show specific types of information-
processing asymmetry identified by Brodbeck et al. (2007). The detailed design of
motivation and consequent biases are described in the following paragraphs.
Social motivation. The information sharing behavior designed in the
computational model corresponds to two modes of social motivation: pro-self and pro-
social (De Dreu et al., 2008). Participants with pro-self motivation shared and evaluated
information in favor of their policy opinions. In contrast, participants with pro-social
motivation shared and evaluated information without any favor. Social motivation then
resulted in some types of bias. To begin, participants varied in evaluating information

66
provided by other participants (evaluation bias) (Van Swol et al., 2003). A participant
may discount the validity of information provided by others, especially when the
information is not consistent with his or her policy opinion. It is also known that group
members favor preference-consistent information (Edwards & Smith, 1996) that fits
one‟s preference, over preference-inconsistent information. Additionally, in deliberations,
a speaker delivers the content as well as the weight he or she puts on the content (Stasser,
2000). Likewise, during the deliberation process in the current simulation, a participant
provided a data point and its weight together for the other participants. Considering the
evaluation bias, the other participants‟ motivated information-processing occurred in the
following way.
1) When a data point, “A” with the weight α was provided by a participant, all
the other participants were ready to revise their policy opinions considering
the data point and its weight.
2) Three discount rates were defined for each participant.
10
The discount rate
reflected the evaluation of information others possess. When participants with
pro-self motivation received a preference-inconsistent data point, they applied
a large discount rate (for example, .9) to the weight. In contrast, when the
data point was consistent with a participant‟s policy opinion, they applied a
relatively small discount rate (for example, .1). Participants with pro-social

10
This rate can also be conceptualized as the level of openness to others‟ opinion when it is reversed. The
author used the concept of discount rate for explanation purposes. In Chapter 7, the reversed concept was
also used in the discussion of the sensitivity analysis.

67
motivation applied the same, intermediate discount rate (for example, .3)
regardless of the fit between the information and their policy opinion.
3) Provided that if a receiver‟s weight on the data point A was β, the receiver‟s
new weight on the data point would be β + α*d, where α denotes the
information provider‟s weight and d denotes the receiver‟s discount rate. The
accumulative feature of determining a new weight reflects the theoretical
argument that group members experience mutual enhancement of the validity
of the information they share (Wittenbaum et al., 1999).
Another category of bias occurred when participants selected information to share
with other participants. According to the literature on motivated information-processing
(De Dreu et al., 2008; Toma & Butera, 2009), a participant may provide information he
or she possesses in two ways. First, as is assumed in the typical hidden profile paradigm
(Wittenbaum et al., 2004), especially in the information sampling model, a participant
may randomly select information to share regardless of the information‟s consistency
with the participant‟s policy opinion. This implies to pro-social motivation in
information sharing in that participants do not purposefully screen what to share. Second,
a participant may only provide information that is consistent with the participant‟s policy
opinion to influence other participants‟ opinions in support of one' own opinion. This
reflects the pro-self motivation in information sharing.
11

11
According to the implication of the research of strategic information sharing behavior (cf., Toma &
Butera, 2009), pro-self motivation does not necessarily lead people to share only preference-consistent
information. For example, group members need to build credibility of their general expertise before
asserting their unique preferences. In that case, group members may need first to mention shared
information even when the information is not consistent with their preferences. This argument, however,

68
In summary, participants with pro-self motivation demonstrated two self-oriented
behaviors: discount the weight of preference-inconsistent information in receiving
information from others and share only preference-consistent information with other
participants. In contrast, participants with pro-social motivation did not discriminate
between preference-consistent and preference-inconsistent information when they shared
and received information. Table 5-2 summarizes the design of social motivation in the
computational model.

Table 5-2
The Design of Social Motivation
Motivation Preference-consistent
information
Preference-inconsistent
information
Information sharing
Pro-self Low discount rate High discount rate
Preference-consistent
information only
Pro-social Medium discount rate No screening

Epistemic motivation. In addition to social motivation, participants were
designed to act at two levels of epistemic motivation (De Dreu et al., 2008): low and high.
Participants with low epistemic motivation were designed to evaluate information with a
high consensus focus (Postmes et al., 2001) and share information with low intellectual
demand (De Dreu et al., 2008). Two types of bias were designed regarding epistemic
motivation. One was related to the behavior of information reception. Specifically, a
participant with low epistemic motivation was conceptualized as favoring the status quo

needs theoretical elaboration, so the computational model in this study did not consider this feature for
simplicity of the model.

69
and accepting the information consistent with the current proposal. Therefore, the
participant applied a small discount rate to the weight of the data point consistent with the
current proposal. In contrast, when the participant received a data point that opposed the
current proposal and incurred change, he applied a large discount rate to the weight of the
data point. This design reflects the finding that when the dominant norm in a group is
consensus-oriented, group members are resistant to accepting arguing data and tend to
conform to the majority or status quo (Postmes et al., 2001). Contrary to those with low
epistemic motivation, participants with high epistemic motivation did not discriminate
information by its consistency with the current proposal, which resulted in the application
of the same, intermediate discount rate to all incoming information.
The other type of bias was related to the sharing behavior. According to the
literature (De Dreu et al., 2008; Postmes et al., 2001; Wittenbaum et al., 2004), group
members with high epistemic motivation show intellectual propensity to collect as much
information available as possible, while group members with low epistemic motivation
tended to make a decision with a relatively narrow range of information. In other words,
one important feature of epistemic motivation is the breadth of information shared among
group members. In the computational model, accordingly, participants with low
epistemic motivation were designed to share information that was most important to them;
that is, information with the largest weight at the point of speaking. In contrast,
participants with high epistemic motivation were designed to share different information
in each speaking turn by selecting data points one by one in descending order of the

70
weights. Table 5-3 summarizes the design of epistemic motivation in the computational
model.

Table 5-3
The Design of Epistemic Motivation
Motivation Conforming information Refuting information Information sharing
Low epistemic Low discount rate High discount rate Narrow range
High epistemic Medium discount rate Wide range

Rules of Deliberation
Once constituents and participants were defined, the process of deliberation began.
The computational model simulated a situation where participants started with a specific
policy alternative and shared data points in each speaker turn. Participants modified their
policy opinions because of information sharing and they finally reached consensus about
a policy.
12

Rule of defining starting alternatives. In the current study, three types of
starting alternative were manipulated. First, since the majority rule has usually been
applied in decision-making in public spheres, this study set a rule of deliberation starting
with the majority policy opinion among participants. The majority policy opinion among
participants was calculated for each of the eight dimensions of the policy opinion (i.e.,
whether, for each dimension, there is a majority of 0s or 1s). This eight-digit string
served as the starting proposal in one of the experimental conditions. Another type of

12
The basic scheme of the process of deliberation was informed by the DISCUSS model Stasser (1988;
2000) developed as a simulation model of the hidden profile situation.

71
starting alternative was the opposite of the majority policy opinion. If the majority policy
opinion was 1/1/…0/1, the opposite policy opinion was 0/0/…1/0. This may represent an
extreme, unpopular proposal that should be readily amenable to improvement via a
collaborative process. The third rule of defining the starting alternative was, for
comparison purposes, that one of the participants was randomly selected and his or her
policy opinion became the starting proposal for deliberation. This may reflect a situation
in which one particular initiator offers its proposal as the opening point of discussion.
Rule of assigning speaker turns. In this simulation, participants were assumed
to participate in a face-to-face deliberation process. Participation took the form of
speaker turns to provide other participants with the information they possess (Stasser,
1988; 2000). This feature reflected the assumption that deliberation among group
members in the real world can be characterized as information-processing relevant to the
task to be solved (Hinsz et al., 1997; Postmes et al., 2001; Roberts, 2002; Wegner, 1987).
Participants may provide general information (Stasser & Titus, 1985), information
consistent with their preferences (Brodbeck et al., 2002), or information that confirms or
contradicts the most recent argument (Parker, 1988; Stasser & Taylor, 1991).
This study manipulated three rules of assigning speaker turns. One rule was to
assign the next turn randomly, which served as the baseline for comparison. A second
rule was to give priority to the least satisfied participant. This rule was based on the
assumption that the less satisfied a participant with the current proposal, the higher the

72
probability of that participant to take the next speaker turn be.
13
This rule also reflects
the egalitarian characteristic of collaborative governance, which normatively offers an
advantage to less satisfied actors. The other rule of assigning speaker turns was a round-
robin rule, also used for comparison purposes. A round-robin rule is often used to
guarantee that all participants are assigned the same opportunity to speak (Innes &
Booher, 2010). While the second rule is related to substantive equality among
participants, the third rule is related to procedural equality among participants.
Rule of mutual adjustment of policy opinions and the policy alternative.
After a participant shared information during his or her speaking turn, the other
participants modified their policy opinions based on the new information. The specific
rules for how to modify policy opinions were described in the motivation rules above.
Another rule of mutual adjustment was associated with the modification of the policy
alternative. This rule allowed the current speaker to propose a change of the policy
alternative in the dimension pertinent to the information he or she shared to be consistent
with the signal of the information. For example, if the i
th
dimension of the current policy
alternative was 1 and the information shared was negative, then the participant could
propose to change the value of the i
th
dimension to 0. This rule may reflect an extreme
end of the idea that collaborative governance is an egalitarian decision-making process in
which every participant has the same authority to influence the collective decision.

13
As is defined in the following section, the level of satisfaction increased as far as the current proposal
was similar to a participant‟s policy opinion, that is, when sub-dimensions of a participant‟s policy opinion
had the same value as those of the current proposal.

73
Rule of continuing or terminating the deliberation process. After one speaker
turn was used by a participant and the other participants modified their policy opinions
accordingly, each participant‟s level of satisfaction was calculated. The goal of each
participant was to improve his or her satisfaction with a proposed alternative relative to
its level of satisfaction with the original proposal at the start of the simulation.
Participants agreed to accept the new proposal when the following two conditions were
met: A participant‟s level of satisfaction with the new proposed alternative was positive
(absolute criterion) and satisfaction was higher than the level of satisfaction with the
original proposal (relative criterion).
In the absence of sufficient acceptance of the current alternative, there was
another iteration of information sharing. When all participants agreed to accept the
current alternative, consensus was reached and the collective decision was made.
Whereas the traditional notion of majority rule implies that at least 50% of the collective
should agree with an alternative to make a decision, the intention to reach consensus in
collaborative governance suggests the need for a higher threshold in a simulated
environment. In the current study, the criterion of unanimity was employed.
Experimental Design
A series of experiments was designed to calibrate the computational model and
examine the effects of individual motivations in information-processing on the
deliberation process and performance as well as the effects of the rules of deliberation on
performance. As described in the simulation design section above, six independent
variables were manipulated in the model: social motivation, epistemic motivation, the

74
rule of defining starting alternatives, the rule of assigning speaker turns, the level of
interdependence of interest among participants, and forum size.
Experiment 1. The purpose of Experiment 1 was to calibrate the computational
model (Carley, 1996). The calibration was accomplished by confirming the
correspondence between the logical inference and simulation results and by replicating
the results found by empirical research about the group decision-making through the
computational model. Through this experiment, the researcher confirmed that the
computational model developed on the assumptions of the hidden profile and the
motivated information-processing literature, successfully manipulated the asymmetric
information distribution and processing by generating results consistent with empirical
findings. In addition, through this experiment, the researcher confirmed the randomized
experimental conditions.
In this experiment, information distribution was manipulated by two types:
symmetric (in which the ratio of positive and negative data points distributed to each
constituent approximated the ratio of positive and negative data points in the whole
information pool) and asymmetric (in which information was distributed asymmetrically
among constituents such that they formed different policy opinions biased toward the
information they possessed). More details about the experimental design are provided in
Chapter 6.
Experiment 2. The purpose of Experiment 2 was to investigate the main and
interaction effects of the two types of individual information-processing motivations:
social motivation and epistemic motivation (De Dreu et al., 2008). Social motivation was

75
manipulated to two levels: pro-self and pro-social. Epistemic motivation was also
manipulated on two levels: low and high. As such, a 2 (pro-self, pro-social motivation)
by 2 (low, high epistemic motivation) factorial design was manipulated. The other
independent variables were set as follows. First, the type of information distribution was
set to asymmetric, reflecting the general condition of group decision-making and
collaborative governance. Second, the starting alternative was set to random. Third, the
speaker turn was also set to random. Fourth, the degree of interdependence of interest
among participants was set to moderate (three focal dimensions). Finally, the forum size
was set to medium (30 participants). More details about the experimental design are
provided in Chapter 7.
Experiment 3. The purpose of Experiment 3 was to investigate the moderating
effects of different deliberation rules and contexts. While Experiment 2 manipulated
individual-level variables, Experiment 3 manipulated group-level variables. First, the
type of starting alternative was manipulated in three ways: the majority policy opinion,
the opposite policy opinion of the majority policy opinion, and a randomly selected
participant‟s policy opinion. Second, the rule of assigning speaker turns to participants
was manipulated in three ways: random selection of a participant, opportunity given to
the least satisfied participant with the current proposal, and round-robin. Third, the
forum size was manipulated at three levels: small (10 participants), medium (30
participants), and large (50 participants). These three variables were the modes of the
managerial intervention. In addition, the level of interdependence of interest among
participants was manipulated on three levels: low (one focal dimension), moderate (three

76
focal dimensions), and high (five focal dimensions). In summary, a 3 (starting
alternatives) by 3 (speaker turns) by 3 (forum size) by 3 (interdependence) factorial
design was used.
The main effects of the manipulation variables and the interaction effects among
them were analyzed. Because it is too complicated to interpret the results of more than
three-way ANOVA, Experiment 3 set the social and epistemic motivation of participants
as random within a forum. In other words, participants with different motivations
coexisted in a forum. More details about the experimental design are provided in Chapter
8.
Dependent Variables
Four categories of dependent variables were measured: success of the deliberation
process, mutual learning, participant satisfaction, and responsiveness of the decision.
Dependent variables in the first category measured how successfully sample groups
reached consensus. The success rate of reaching consensus was measured via the
percentage of groups that reached consensus in each experimental condition. A high
success rate indicated that participants were more likely to reach consensus in such a
condition. Time to reach consensus was measured as the average number of speaker
turns it took for the successful groups in each condition to reach consensus.
Dependent variables in the second category measured the degree of mutual
learning among participants. First, the average number of data points participants newly
obtained during the deliberation process across participants was measured. Since the set
of data points that the participants possessed at the beginning varied, what data points

77
were new also varied across participants. The more new data points a participant
obtained during the deliberation, the more the participant learned in terms of quantity.
The unit of the measure was the percentage of newly obtained data points to the
maximum number of data points available in the virtual society. Second, the average
number of sub-dimensions of policy opinion that inversely changed through deliberation
was measured across participants. Following this process, quantitative learning should
have resulted in adjusting the mutual interests among participants. Therefore, this
measure indicated the qualitative aspect of learning through the deliberation process. The
degree of change was measured in the focal dimensions. The unit of the measure was the
percentage of the changed focal dimensions to the total number of focal dimensions.
When necessary, the absolute number of changed focal dimensions was reported in the
results.
Dependent variables in the third category measured participants‟ level of
satisfaction. Each participant‟s level of satisfaction was defined as the average degree of
fit for each dimension between its policy opinion and the decision multiplied by the
weight of each dimension. The level of satisfaction was measured in two points: at the
start and at the point where consensus was reached. Table 5-4 illustrates the calculation
of the level of satisfaction. In the example, the agent‟s level of satisfaction is positive
(.25).

78
Table 5-4
Calculation of the Level of Satisfaction: An Example
Focal dimensions 1 3 4
Opinion (A) 1 1 1
Collective Decision (B) 1 0 0
Fit (C) 1 -1 -1
Weight (D) .75 .45 .05
Satisfaction (C * D) .75 -.45 .-05

Dependent variables in the final category measured the responsiveness of the
group decision to the constituents. First, the global responsiveness of the decision was
measured using the average level of satisfaction of all the constituents. Different
measures were used, including the responsiveness of the starting alternative, the final
decision, and the degree of improvement of global responsiveness through deliberation.
The local responsiveness of the decision to specific parties was also measured. The level
of satisfaction of constituents in each party was calculated and all levels of satisfaction
were averaged across parties. The same measures were used for both global
responsiveness and local responsiveness.

79
CHAPTER 6: MANIPULATION CHECK OF THE COMPUTATIONAL MODEL

In this chapter, the results of the manipulation check of the computational model
are reported. The researcher examined whether the calibration of the computational
model was conducted successfully to confirm that the computational model generates the
basic results that are logically inferred and empirically supported. Therefore, the purpose
of the analysis in this chapter is to confirm that the computational model is reliable
concerning the manipulation of the fundamental features of the model, including the
information distribution condition, randomization in experimental treatments, and agent
behaviors corresponding to empirical findings. More specific manipulation checks
regarding the type of motivation and the mode of managerial intervention are also
reported in the following chapters.
The manipulation checkpoints were categorized into four groups. The first group
included variables that describe the distribution of information among constituents and
participants. Specifically, it was necessary to check whether the condition of asymmetric
information distribution was successfully simulated. The second group included
variables that needed to be randomized at the initial condition, such as the global and
local responsiveness of a starting alternative. The third group included variables that
measured the degree of information sampling bias. These checkpoints determined
whether the design of artificial agents to follow the information sampling behavior at the
individual level generated the information-sampling phenomenon that has been found in
empirical research at the group level. In the final group variables measured the supposed

80
effect of information distribution, such as the degree of difference between an ideal
opinion with complete information and the actual opinion with partial information.
These variables were not of serious or direct theoretical interest of this study; the
pertinent results should be straightforward by logic. Unexpected results, which are not
rare in complex computational models, may be due to either a result of unpredicted
internal dynamics of the model or a programming error. Either way, the results helped
the researcher calibrate the computational model to be more reliable. In short, in this
chapter, it is demonstrated that the design of the computational model successfully
manipulated the experimental conditions, which were the basis of the experiments
reported in the Results chapters. The detailed description of the measures used for
manipulation check is provided, followed by the description of the experimental design
and the results.
Manipulation Check Points
Information distribution. In the first category, dependent variables measured
the distribution of information among constituents and participants. To begin, the
average amount of data points each constituent possessed was calculated.
14
To remove
any significant effect of the amount of data points, the researcher controlled the amount
to be the same in collaborative governance systems regardless of the type of information

14
The measure was divided by the maximum number of data points a constituent could possess, which was
256, the number of data points in a dimension (32) multiplied by the number of dimensions in a policy
opinion (8).

81
distribution. In other words, in the experiment, only the distribution of information, not
the volume of information, was to differ.
15

Manipulation 1. The number of data points available among constituents are
identical both under the condition of symmetric information distribution and under the
condition of asymmetric information distribution.
The ratio of the number of data points a participant possessed in their focal
dimensions to the maximum amount of data points a participant could possess was
calculated and averaged across participants. According to the representation rule
described in the Methods chapter, each participant aggregated the data points the
constituents in its party possessed. In other words, each participant knew all of and
nothing but the data points the party constituents possessed. Therefore, the diversity or
distribution of data points a participant possessed depended on the diversity of data points
available among its constituents. When information is asymmetrically distributed among
constituents, the constituents in the same party would possess highly overlapping sets of
data points. In contrast, when information is symmetrically distributed among the
constituents, the constituents in the same party, in spite of the identity of their policy
opinions, would possess fewer overlapping sets of data points.
16
Consequently, if the

15
Note, the actual amount of data points in each collaborative systems were not exactly the same by design.
To ensure the asymmetric information distribution condition, each constituent in a system was supposed to
possess only about 16% of the total sum of data points. The design was done in a way that there was a
small variance in the number of data points each constituent possessed. However, on average, the number
of data points across systems was same. Considering this probabilistic nature of the design, the author
checked the results of the actual simulation.
16
In this condition, for example, although constituents‟ policy opinions were the same in a party, a
constituent determined its policy opinion with data points, for example, A and B, and another constituent
would do so with data points C and D, while the other would do so with A and E. In contrast, under the

82
total sum of data points across constituents is controlled identically across types of
information distribution, the number of data points each participant could possess would
depends on the diversity of data points its constituents possessed. In other words, a
participant under an asymmetric information distribution condition may possess fewer
data points because its constituents provided highly overlapping sets of data points. By
the same token, a participant under a symmetric information distribution condition may
possess more data points. Therefore, the simulation of the computational model should
demonstrate the following relationship.
Manipulation 2. The number of data points a participant possesses regarding its
focal dimensions will be larger under the condition of symmetric information distribution
than under the condition of asymmetric information distribution.
Finally, in this category, the ratio of the list of data points, available in a
governance system as a whole, to the complete list of data points was calculated. For a
comparison purpose, it is good to manipulate this regardless of the type of information
distribution; there was no difference in the availability of data points in the simulated
collaborative governance systems. Although information is asymmetrically distributed
among constituents, each of the 256 data points should have been known to one or more
constituents in the governance system. This manipulation check was to ensure that what
matters in the experiment was not the amount or availability of information among
constituents but the type of information distribution.

condition of asymmetric information distribution, many constituents in a party may possess data points A
and B, and those in another party may possess data points C and D.

83
Manipulation 3. As a default, all the possible data points should exist in a
governance system regardless of the type of information distribution.
Randomized conditions. In the second category, checkpoint variables are the
values that should not statistically differ across the types of information distribution as a
result of the randomized design of the experiment. Randomization of the experiment was
achieved via the algorithmic design of the computational model itself. However, because
the internal dynamics of the computational model are quite complex, which is the nature
of agent-based modeling, one should consider and check the possibility that there could
be an unexpected systematic bias, in spite of the randomization design.
The design was checked with three measures. First, the average global
responsiveness of starting alternatives was measured. When the experimental condition
involves a starting alternative that is derived from a randomly selected participant in a
governance system, it is reasonable to assume that the level of the global responsiveness
of alternatives at the beginning is also a matter of randomness. Likewise, the average
local responsiveness of starting alternatives was measured. Finally, the average level of
satisfaction of participants with the starting alternative was measured. These measures,
in the absence of any systematic bias related to the type of information distribution and
should demonstrate randomness.
Manipulation 4. The average global responsiveness of the starting alternatives
under the condition of symmetric information distribution will not be different from that
under the condition of asymmetric information distribution.

84
Manipulation 5. The average local responsiveness of the starting alternatives
under the condition of symmetric information distribution will not be different from that
under the condition of asymmetric information distribution.
Manipulation 6. The average level of satisfaction of participants at the start under
the condition of symmetric information distribution will not be different from that under
the condition of asymmetric information distribution.
Information sampling bias. Information sampling bias occurs at the group level
simply because shared information is more likely sampled than unshared information
(Stasser, 2000). Information sampling bias appears stronger when group members
widely share common information before the discussion. Since this is a well-investigated
phenomenon in group information-sharing literature, the agents in the computational
model were designed to follow this kind of behavior.
In the computational model, a participant, because many of its data points
provided by constituents overlapped, would come to possess less information under the
asymmetric information distribution condition than they would under the symmetric
information distribution condition. First, under the asymmetric information distribution
condition, constituents possessed a narrow range of data points mostly consistent with
their policy opinions. Second, due to the logic of information aggregation, the
narrowness of data points at the constituent level lead to the narrowness of data points at
the participant level.
Participants were designed to mention data points with the order of the magnitude
of the weights. As explained in the Methods chapter, the magnitude of the weight of a

85
data point a participant possessed was determined by the sum of the weights of the data
point from the constituents which possessed that data point. Therefore, data points with a
larger weight in participants‟ mindset were likely more common data points among the
constituents. Therefore, the design of agent behavior to mention data points with the
order of the magnitude of the weights corresponded to the information sampling behavior
found in real world experiments.
The information sampling bias was measured in two ways. First, it was measured
by the correlation coefficient between the number of constituents that possessed a data
point and the frequency with which the data point was mentioned during the deliberation
process. Second, it was measured by the correlation coefficient between the number of
participants that possessed a data point and the frequency with which the data point was
mentioned during the deliberation process. A large correlation coefficient meant that
common information among constituents and participants was mentioned more
frequently in the deliberation process, which reflected the information sampling bias.
Manipulation 7. Information sampling bias during the deliberation process
among participants will be higher under the condition of asymmetric information
distribution than under the condition of symmetric information distribution.
Effect of information distribution. In studies of group decision-making, group
members form their pre-discussion preferences based on provided information (Brodbeck
et al., 2002; 2007). Through group discussion, they more or less revised their preferences
according to new information shared among them. One assumption in the literature is
that when group members have an access to all the information in advance, they will

86
come to the meeting with more similar preferences than when they have an access only to
part of the information. The public deliberation process in a collaborative governance
system also starts with a group of participants whose policy opinions are, at least ideally,
diverse enough to represent all constituent preferences. The premise is that participants
form diverse pre-deliberation policy opinions due to information asymmetries and that
the pre-deliberation policy opinions are adjusted after deliberation among the participants
in the mutual learning process, which is the basis for reaching consensus on a collective
action plan.
Reflecting the premise of collaborative governance, two measures were developed
to check that the process of the computational model simulated the basic nature of the
deliberation process. First, the degree of fit between a constituent‟s policy opinion,
formed through information on hand, and an ideal policy opinion, formed by having
complete information, was calculated. In the computational simulation, each constituent
could have been provided with a complete set of data points. In this study, only some of
the data points were actually provided. Using this design, the author could compare the
constituent‟s actual policy opinion formed with partial information and ideal but
unrealized policy opinion with complete information. It is known that when group
members are provided with asymmetric information, they tend to form unique
preferences according to the information and it becomes more difficult for them to find
the correct or best alternative (Stasser, 2000). Therefore, it was expected, in the
computational model, that the degree of fit between the actual policy opinion and the

87
ideal policy opinion would be lower under the condition of asymmetric information
distribution than under the condition of symmetric information distribution.
Manipulation 8. The degree of fit between constituents or participants‟ actual
policy opinions and the ideal policy opinion will be lower under the condition of
asymmetric information distribution than under the condition of symmetric information
distribution.
Another measure was developed to check that the computational model simulated
the nature of the deliberation process as mutual learning. The measure of the effect of
information distribution on mutual learning was the ratio of the number of data points
newly obtained through the deliberation process to the total number of data points. When
participants begin deliberations with asymmetric information, many of the data points
shared during the process would be new to each other. In contrast, when information was
symmetrically distributed, a data point shared by a participant was less likely to be new to
other participants in the forum. Therefore, the degree of learning measured by the
amount of data points newly obtained through deliberation would be higher under the
condition of asymmetric information distribution.
Manipulation 9.The amount of data points newly obtained by participants
through the deliberation process will be larger under the condition of asymmetric
information distribution than under the condition of symmetric information distribution.
Comparison of Consequences between Symmetric and Asymmetric Conditions
To investigate the manipulation check points discussed above, a series of
independent sample t-tests were performed on ten measures by dividing the sample

88
groups according to the type of information distribution. The parameter values, other
than the type of information distribution, were set and are summarized in Table 6-1. First,
groups with three levels of interdependence of participant interest, which were
manipulated by the number of focal dimensions, were generated. Second, groups with
two types of social motivation (pro-self and pro-social) were generated. Third, groups
with two types of epistemic motivation (low and high) were generated. Fourth, three
different sizes of groups (10, 30, and 50) were generated. Finally, the rule of selecting a
starting alternative and the rule of selecting a speaker were randomly selected to verify
the randomness of the starting condition of deliberation. In summary, including the two
types of information distribution (symmetric and asymmetric), sample groups were
generated in each of 2 (information distribution) x 3 (interdependence) x 2 (social
motivation) x 2 (epistemic motivation) x 3 (forum size) factorial design condition. The
purpose of this design is to include samples from all the levels of parameter values
regarding interdependence, motivation, and forum size. In each condition, 200 sample
groups were generated to ensure statistically reliable results. Consequently, there were
7,200 sample groups for each type of information distribution and a total of 14,400
sample groups. Independent sample t-tests were conducted by comparing the 7,200
sample groups that included all variations of the parameter values. In each simulation,
500 constituents were randomly generated. Each group was given up to 1,000 speaking
turns to share information and negotiate the decision.

89
Table 6-1
Parameter Settings for Manipulation Check
Parameter Description
Value setting
Degree of
interdependence
The degree of overlap of interest among participants,
defined by the number of focal dimensions
Low, moderate, high
Starting alternative Selecting the initial proposal to start deliberation with Random selection
Speaker turn Selecting the next speaker Random selection
Social motivation Information-processing driven by preference Pro-self, pro-social
Epistemic
motivation
Information-processing driven by intellectual propensity Low, high
Forum size The number of participants in a forum 10, 30, 50

Results
Table 6-2 summarizes the results of the manipulation check. As indicated in the
last column, the experimental manipulation in the computational model was successful.
Two exceptions were found.

90
Table 6-2
Results for Manipulation Check
Variable Manipulation Symmetric Asymmetric t-statistics Check
Information
Volume of information among
constituents sym = asym
15.91
(.10)
15.91
(.08) .75 Yes
Volume of information among
participants sym > asym
89.98
(1.56)
61.46
(1.70) 1046.24** Yes
Completeness of information sym = asym
100.00
(.00)
99.94
(.20) 27.14** No
Randomized conditions
Starting global responsiveness sym = asym
.31
(1.95)
.32
(1.83) -.34 Yes
Starting local responsiveness sym = asym
5.48
(15.69)
5.25
(16.07) .88 Yes
Starting level of satisfaction sym = asym
.75
(2.18)
1.41
(4.36) -11.53** No
Information Sampling Bias
Sampling bias regarding
constituents sym < asym
.10
(.08)
.41
(.18) -132.24** Yes
Sampling bias regarding participants sym < asym
.08
(.07)
.39
(.18) -136.20** Yes
Effect of distribution
Degree of opinion bias sym < asym
30.27
(3.87)
45.33
(4.36) -219.24** Yes
Newly obtained data points sym < asym
5.73
(3.24)
15.20
(10.62) -72.45** Yes
Note. Numbers in parenthesis are standard deviation. ** p < .01.

The measure of the volume of information among constituents (Manipulation 1)
showed that the computational model controlled the amount of data points distributed to
constituents did not to vary across groups with different types of information distribution
(t = .75, p > .05). The measure of the volume of information among participants
(Manipulation 2) showed that, in the computational model, participants started with more
data points under the symmetric information distribution condition (89.98%) than they
did under the asymmetric information distribution (61.46%) (t = 1046.24, p < .01).
Therefore, the computational model successfully simulated the asymmetric information

91
distribution situation among group members, which is the general experimental condition
in the group decision-making literature (Brodbeck et al., 2007) and in collaborative
forums in the real world collaborative governance systems (Thomson & Perry, 2006).
Finally, the measure of the completeness of information showed that there was a
statistically significant (t = 27.14, p < .01), however, practically non-significant
difference between symmetric and asymmetric groups, which confirms Manipulation 3.
As for the check of the randomized conditions, the two measures of global and
local responsiveness showed that the condition of randomness (Manipulation 4 and 5)
was fulfilled. The researcher expected that the level of satisfaction at the beginning may
not be affected by the type of information distribution. However, the measure of the level
of satisfaction at the beginning revealed an important, unpredicted aspect of the internal
dynamics of the computational model. Further investigation of the simulation
demonstrated that the difference was partly due to the magnitude of weights of policy
opinion, which were affected by the type of information distribution. Under the
condition of symmetric information distribution, because constituents and their
representatives were more likely to possess a balanced set of data points between positive
and negative, their strength of policy opinion (that is, the sum of the weights attached to
the focal dimensions) was relatively small. In contrast, under the condition of
asymmetric information distribution, because constituents provided their representative
with a set of data points tilted to the positive or negative, the representatives participated
in the deliberation process with a strong policy opinion toward either positive or negative.
Consequently, each participant‟s level of satisfaction in that condition may be very high

92
or very low, as shown with the large standard deviation (4.36). Given this variance, the
results indicate that the overall level of satisfaction might be higher in groups with
asymmetric information distribution. This may be partly due to the exceptionally high
level of satisfaction of the participant whose policy opinion was selected as the starting
proposal in the asymmetric information distribution condition. Other things being equal,
because of the weights of policy opinion, the participant‟s level of satisfaction in the
asymmetric information distribution was much higher than that of the participant in the
symmetric information distribution condition. These results also indicate that the starting
alternatives, including the majority opinion, the opposite of the majority opinion, and the
random opinion were, on average, more satisfactory than unsatisfactory, though the
degree was determined by the weights.
The results about the information sampling bias show that the computational
model successfully simulated the information sharing behaviors found in group decision-
making research (Manipulation 7). When information was asymmetrically distributed
among constituents, more widely shared information had more opportunities to be
mentioned in the deliberation process (t = -132.24, p < .01). The measure of information
sampling bias regarding participants also shows that, even when the information was
aggregated to the participant level, there was still effective information sampling bias as
was designed to be (t = -136.20, p < .01).
Finally, the results on the effect of information distribution demonstrated what
was expected. First, as for Manipulation 8, the degree of opinion bias was larger among
constituents under the asymmetric information distribution than among those under the

93
symmetric information distribution (t = -219.24, p < .01). Second, concerning
Manipulation 9, participants obtained more data points under the asymmetric information
distribution condition (15.20% point of the total number of data points) than under the
symmetric information distribution condition (5.73% point) (t = -72.45, p < .01).
In conclusion, the computational model developed in this study fulfilled the basic
requirements with which to judge the model‟s reliability regarding its internal dynamics.
The results show that the model calibrated to generate results corresponding to logical
expectations and empirical findings. A computational model can be evaluated based on
relational equivalence of its results with other empirical or computational models; two
models can produce the same internal relationships (Axtell et al., 1996). The relational
equivalence between the results of the computational model and the logical and empirical
consequences demonstrated that the computational model correctly combined the
theoretical assumptions and replicated fundamental findings from empirical research;
therefore is ready to be used for further theoretical investigation.

94
CHAPTER 7: EFFECT OF SOCIAL AND EPISTEMIC MOTIVATION ON THE
PERFORMANCE OF GROUP DELIBERATION

The purpose of the current experiment was to analyze the main and interaction
effects of social motivation and epistemic motivation on the performance of the collective
decision among participants in collaborative governance systems. To focus on these
effects, the other parameters were set as constants as summarized in Table 7-1.
17
In each
cell of the 2 (pro-self and pro-social motivation) by 2 (low and high epistemic motivation)
factorial design, 1,000 sample groups were generated, resulting in a total of 4,000 sample
groups.
18
In each simulation, 500 constituents were randomly generated. Each group
was given up to 1,000 speaking turns to share information and negotiate the decision.

17
Unlike the case of manipulation check, the degree of interdependence was set to the moderate level and
the forum size was set to 30 or medium size. The purpose of including samples from all the three levels of
interdependence and forum size was to confirm that manipulation was successful, in general. In the
experiment described in this chapter, all parameters except social and epistemic motivation were controlled
as constants. Therefore, the moderate level of interdependence and forum size was selected. The specific
effects of interdependence and forum size are investigated in the next chapter.
18
As described, the number of successful forums in reaching consensus, which were used in this
experiment, varied across the experimental conditions. Therefore, the sample size in this experiment was
determined to obtain sample groups large enough to obtain reliable statistics.

95
Table 7-1
Parameter Setting for the Virtual Experiment about the Effect of Motivation
Parameter Description Value setting
Interdependence The degree of overlap of interest among
participants
Moderate
Starting alternative A specific policy opinion with which to start
deliberation
Random
selection
Speaker turn A rule to determine who will take the next
speaker turn
Random turn
Information
distribution
The way how information is distributed
among constituents
Asymmetric
Forum size The number of participants in a forum Medium size

The dependent variables were grouped to five categories (success, global
responsiveness, local responsiveness, learning, and satisfaction) and analyzed. In each
category, data generated with the computational model was analyzed in two ways. First,
as a sensitivity analysis, the effect of social and epistemic motivation was analyzed at
various levels of discount rate. The value of discount rates was determined by two
theoretical concerns: relative size among the types of motivation and overall absolute size
across all types of motivation.
19
First, the relative sizes of the discount rates among the
types of motivation were determined by deductive reasoning from the conceptualization
of motivation as described in the Methods chapter. Second, the absolute size of the
overall discount rate remained arbitrary; in the real world some groups happen to be more
or less open-minded than others. To solve this problem and improve the credibility of the
results, the sensitivity of the results, in addition to the change of the absolute size of the

19
A discount rate was composed of two parts: α and β, where α denotes the relative size of discount rate
described in the methods chapter and β denotes a constant that shifts the overall discount rate up and down.

96
discount rates was analyzed. Since the discount rate was a negative measure on positive
learning attitude, it was not intuitive. Therefore, the absolute size of the discount rate
was reversed to a measure of the level of openness to incoming information for a better
intuitive display of results.
20

Second, the results in a specific level of openness (i.e., the maximum level
reported in this chapter) were analyzed to evaluate the statistical significance of the effect
of social motivation and epistemic motivation.
21
The purpose of the analysis was to
statistically clarify the main and interaction effects of the types of motivation on the
performance of the deliberation groups. Quantitative statistics from the analysis of
variance helps us better understand the magnitude of the effects and the implications of
the simulation results. In this case, the sample groups were limited to those whose levels
of openness were the highest in the given range for three reasons. First, as only
successful collaborative groups that reached consensus were included in the analysis,
there was a large gap of the number of successful groups between pro-self motivation
groups and pro-social motivation groups in the lower range of openness. Furthermore,
the number of successful pro-self motivation groups was too small in the low to medium
level of openness because of their low success rate. In contrast, groups with the highest
level of openness yielded a sufficient number of successful groups to apply an ANOVA

20
As the following sensitivity analysis showed, the range illustrated in the figures was the most reasonable
one to report. Below the minimum level, because the success rate of groups in reaching consensus was too
low, there were few groups to be included in the analysis. Above the maximum level, because most of the
groups succeeded in reaching consensus, no significant variance was found in success rate. Both results
reduced the credibility of the statistical analysis.
21
Note that the usefulness of statistical significance is relatively low in simulation research because the
statistics does not assume a practical meaning (Mezias & Lant, 1994). Rather, the relative size of the
effects is more relevant to discuss. A report of statistical significance should be understood as a reference.

97
test. Second, from the sensitivity analysis it was found that, in the range of high
openness, the performance of groups with different types of motivation showed a
tendency to converge into kind of attraction points. This result suggests that the samples
with the highest level of openness may yield the most conservative comparison result to
analyze the difference among groups. Therefore, if a difference, found at this level, was
statistically significant and the difference was not sensitive to the level of openness, the
difference could safely be inferred to another level. Finally, it was reasonable to assume
that, in the real world, if a collaborative governance system was successful, participants
in the system may be very open to others‟ opinion (Innes & Booher, 2010). Since this
researcher analyzed successful sample groups, those from the highest level of openness
may reasonably reflect the characteristics of successful collaborative governance forums.
Success
Two dependent variables were used to assess how successfully participants in a
forum reached consensus: success rate and time to reach consensus. Figure 7-1 illustrates
that the main effects of social motivation and epistemic motivation were not sensitive to
the level of openness. Participants with pro-social motivation were more likely to reach
consensus than were participants with pro-self motivation. The positive effect of low
epistemic motivation on reaching consensus was also found. In addition, the success rate
was higher among groups with higher level of openness.

98

Figure 7-1. Success Rate

The statistical analysis of the success rate in reaching consensus confirms the
result shown in Figure 7-1 (Tables 7-2; 7-3). First, pro-social motivation was associated
with higher success rate (100.00%) compared to those with pro-self motivation (83.10%),
which was statistically significant (F = 492.83, p < .01). Second, low epistemic
motivation was associated with higher success rate (99.40%) compared to those with high
epistemic motivation (83.70%), which was also statistically significant (F = 425.33, p
< .01). The interaction effect was also significant (F = 425.33, p < .01) due to the
particularly low success rate of pro-self-high epistemic motivation groups (67.40%).

0
20
40
60
80
100
Lowest Highest
pro-self-low pro-self-high
pro-social-low pro-social-high

99
Table 7-2
Success Rate and Time: Main Effect
Social Epistemic
Pro-self Pro-social Low High
Success
M(SD) 83.10(37.49) 100.00(.00) 99.40(7.73) 83.70(36.95)
F 492.83** 425.33**
N 2000 2000 2000 2000
Time
M(SD) 392.05(193.85) 230.27(90.43) 263.75(127.72) 351.14(194.31)
F 1498.53** 567.54**
N 1662 2000 1988 1674
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

Table 7-3
Success Rate and Time: Interaction Effect
Pro-self Pro-social Interaction
Low High Low High F
Success
M(SD) 98.80(10.89) 67.40(46.90) 100.00(.00) 100.00(.00) 425.33**
N 1000 1000 1000 1000
Time
M(SD) 330.34(145.39) 482.52(218.99) 197.95(53.30) 262.59(106.92) 92.50**
N 988 674 1000 1000
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

Figure 7-2 shows the time for groups with different motivations to reach
consensus. First, the graph shows the consistent main effect of social motivation and
epistemic motivation. Pro-social motivation and low epistemic motivation were
associated with less time to reach consensus. The results from participant groups with
pro-self-high epistemic motivation were somewhat abnormal, mainly due to small sample
size in the middle range, in which the success rate of the sample groups was very low.
Second, overall, it was found that participants with a high level of openness took less
time to reach consensus.

100

Figure 7-2. Time to Reach Consensus
Note. There were no successful groups in the category of pro-self-high epistemic
motivation in the lower range.

Both main and interaction effects of social and epistemic motivation on time to
reach consensus were statistically significant (Tables 7-2; 7-3). The result clearly shows
that pro-social motivation and low epistemic motivation, which are related to higher
success rate, are also related to less time required to reach consensus. Participants with
pro-social motivation took less time (230.27 iterations) than did those with pro-self
motivation (392.05 iterations). Participants with low epistemic motivation took less time
(263.75 iterations) than did those with high epistemic motivation (351.14 iterations). In
particular, the main effect of social motivation (F = 1498.53, p < .01) was larger than the
main effect of epistemic motivation (F = 567.54, p < .01) and the interaction effect was
relatively small (F = 92.50, p<.01).
In summary, when participants are motivated in pro-social and low epistemic
ways, they are more likely to reach consensus within a shorter time.

0
200
400
600
800
1000
Lowest Highest
pro-self-low pro-self-high
pro-social-low pro-social-high

101
Global Responsiveness of the Collective Decision
Figure 7-3 shows the level of global responsiveness achieved by groups with
different motivations. First, pro-social groups, on average, recorded a higher global
responsiveness than did pro-self groups. Second, from Figure 7-4, it is obvious that pro-
social motivation and high epistemic motivation were associated with high global
responsiveness at all levels of openness.

Figure 7-3. Global Responsiveness

-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Lowest Highest
pro-self-low pro-self-high
pro-social-low pro-social-high

102

Figure 7-4. Global responsiveness: Main effect

Regarding statistical analysis, the mean global responsiveness of the starting
alternative and the final decision, as well as the mean degree of improvement in global
responsiveness were considered (Tables 7-4; 7-5). Concerning the responsiveness of the
starting alternative, there was no statistically significant difference among the types of
motivation, which is consistent with the manipulation of the experiment.

Table 7-4
Global Responsiveness: Main Effect
Social Epistemic
Pro-self Pro-social Low High
Start
M(SD) .26(1.76) .30(1.78) .31(1.79) .25(1.76)
F .45 .94
End
M(SD) .17(1.82) .34(1.77) .21(1.82) .33(1.76)
F 5.47* 4.97*
Improve
M(SD) -.09(2.40) .05(2.38) -.10(2.39) .09(2.38)
F 1.57 5.72*
N 1662 2000 1988 1674
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Lowest Highest
pro-self pro-social
low epistemic high epistemic

103
Table 7-5
Global Responsiveness: Interaction Effect
Pro-self Pro-social Interaction
Low High Low High F
Start M(SD) .26(1.76) .25(1.78) .35(1.82) .24(1.74) .84
End M(SD) .03(1.83) .38(1.80) .39(1.80) .30(1.74) 13.66**
Improve M(SD) -.23(2.38) .13(2.41) .04(2.41) .06(2.36) 4.36*
N 988 674 1000 1000
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

Concerning the responsiveness of the final decision, the results should be read
with caution. First, the statistical significance of the main and interaction effects were
small (F
social
= 5.47, p < .05; F
epistemic
= 4.97, p < .05; F
interaction
= 13.66, p < .01). This
means that the type of motivation may not explain the difference of the responsiveness of
the final decision. Second, as seen in Figure 7-3, the global responsiveness of the final
decision in pro-self motivation groups was relatively sensitive to the level of openness.
One must be cautious in making a firm conclusion about the effect of motivation on the
global responsiveness of a final decision. When this researcher analyzed the results with
this caution, though, pro-social motivation groups (.34) and high epistemic motivation
groups (.33) recorded higher global responsiveness than did pro-self motivation groups
(.17) and low epistemic motivation groups (.21), respectively.
22

Finally, as for the degree of improvement in global responsiveness, two findings
are noticeable. One was that the difference of the degree of improvement across the

22
Note that the scale is the ratio of a constituent‟s level of satisfaction with the decision to the maximum
level of satisfaction (100). For example, the responsiveness of .21 indicates that the decision served
constituents positively by, on average, .21% of the sum of their weights. If the responsiveness measure was
-.21, it indicates that the decision was negatively responsive to general constituents‟ policy opinion. In the
situation that a policy opinion was defined as either 0 or 1, the expected level of global responsiveness was
0. Therefore, although the degree is trivial, the direction is theoretically meaningful.

104
types of motivation was even smaller than the difference of the global responsiveness of
the final decision across the types of motivation (F
social
= 1.57, p > .05; F
epistemic
= 5.72, p
< .05; F
interaction
= 4.36, p < .05). From this result, it cannot be said that the type of
motivation has a distinct effect on the improvement of global responsiveness through
deliberation. The other finding was that, although the difference was not significant, pro-
self motivation groups (-.09) and low epistemic motivation groups (-.10) recorded
decreased responsiveness. These results indicate the potential positive effect of pro-
social and high epistemic information-processing behaviors on global responsiveness of
the decision.
In summary, social and epistemic motivation had only limited effects on the level
of global responsiveness of the collective decision. With the limitation, pro-social and
high epistemic motivation seems to be associated with higher global responsiveness.
This result indicates that deliberation among participants with pro-self and low epistemic
motivation might not add value to the collective decision-making process.
Local Responsiveness of the Collective Decision
Another aspect of the responsiveness of the collective decision is local
responsiveness, which was defined as the mean responsiveness of the collective decision
to each party of constituents. In a party, all constituents had the same policy opinion in
the same focal dimensions. Therefore, the measure of local responsiveness focused not
on individual constituents or the public, but on individual parties characterized by a
unique, identical opinion. These parties are actually visible in collaborations in the real
world in the form of stakeholder groups. Figure 7-5 summarizes the average local

105
responsiveness across groups with different motivations at each level of openness. First,
Figure 7-5 demonstrates the consistent positive effect of pro-self motivation on local
responsiveness. Both pro-self-low epistemic motivation groups and pro-self-high
epistemic motivation groups recorded higher local responsiveness than did pro-social-low
or high epistemic motivation groups. Second, Figure 7-5 also demonstrates the consistent,
positive effect of high epistemic motivation on local responsiveness. Given the type of
social motivation, participants with high epistemic motivation recorded higher local
responsiveness than did participants with low epistemic motivation. Finally, in all
conditions, the relationship between local responsiveness of the collective decision and
the level of openness was negative.

Figure 7-5. Local Responsiveness

As with the analysis of global responsiveness, the statistical analysis of local
responsiveness included the mean local responsiveness of the starting alternative and the
0
10
20
30
40
50
Lowest Highest
pro-self-low pro-self-high
pro-social-low pro-social-high

106
final decision, and the mean degree of improvement in local responsiveness (Tables 7-6;
7-7). To begin, the result shows that, with one exception, there was no statistically
significant difference in the local responsiveness of the starting alternative, which is
consistent with the manipulation of the experiment. Note that the local responsiveness of
the starting alternative in high epistemic motivation groups was statistically different
from that in low epistemic motivation groups (F = 5.52, p < .05). However, this result
does not imply failure of the manipulation. Note that the samples in the analysis were
successful groups. Given the sample groups were successful in reaching consensus, this
result indicates that high epistemic motivation groups that successfully reached consensus
tended to start with an alternative with relatively high local responsiveness.

Table 7-6
Local Responsiveness: Main Effect
Social Epistemic
Pro-self Pro-social Low High
Start
M(SD) 3.30(10.41) 3.81(10.37) 3.18(10.11) 4.06(10.69)
F 1.66 5.52*
End
M(SD) 21.34(6.92) 17.13(8.46) 18.15(8.10) 20.11(7.93)
F 316.27** 96.30**
Improve
M(SD) 18.04(11.84) 13.32(12.23) 14.97(11.96) 16.05(12.62)
F 156.09** 18.05**
N 1662 2000 1988 1674
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

107
Table 7-7
Local Responsiveness: Interaction Effect
Pro-self Pro-social Interaction
Low High Low High F
Start M(SD) 3.07(10.23) 3.65(10.66) 3.28(9.99) 4.33(10.71) .46
End M(SD) 19.76(6.90) 23.66(6.28) 16.55(8.83) 17.71(8.03) 28.12**
Improve M(SD) 16.70(11.42) 20.01(12.16) 13.27(12.24) 13.38(12.22) 15.83**
N 988 674 1000 1000
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

Concerning the local responsiveness of the final decision, the results showed the
importance of social motivation (F = 316.27, p < .01). Higher local responsiveness was
associated with pro-self motivation (21.34) than with pro-social motivation (17.13).
23

The effect of epistemic motivation was also statistically significant (F = 96.30, p < .01),
with high epistemic motivation (20.11) associated with higher local responsiveness
compared to low epistemic motivation (18.15). Regarding the interaction effect, the
positive effect of high epistemic motivation was eminent among pro-self motivation
groups (F = 28.12, p < .01). The type of epistemic motivation made a larger difference
among pro-self motivation groups (pro-self-low: 19.76; pro-self-high: 23.66) than among
pro-social motivation groups (pro-social-low: 16.55; pro-social-high: 17.71).
Finally, concerning the degree of improvement in local responsiveness through
deliberation, the result showed the same pattern as the local responsiveness of the final
decision. That is, pro-self motivation (F = 156.09, p < .01) and high epistemic
motivation (F = 18.05, p < .01) had positive effects on the improvement of local

23
The scale used in measuring local responsiveness was the same as the scale used in measuring global
responsiveness. Because local responsiveness was measured on relatively homogenous constituents, it was
higher than global responsiveness measured on the whole, heterogeneous constituents.

108
responsiveness of the collective decision through deliberation. Contrary to the results on
global responsiveness, it should be noted that the deliberation process, simulated in this
study, contributed to the significant improvement of the local responsiveness of the
collective decision. In particular, social motivation appears to be an important factor in
determining the degree of improvement.
In summary, pro-self motivation and high epistemic motivation were associated
with high local responsiveness of the collective decision. In addition, the deliberation
process generally contributed to the improvement of the local responsiveness of the
collective decision under all types of motivation.
Learning
The fourth category of dependent variables is the degree of learning during
deliberation, which was measured using two indicators in this study. First, this researcher
measured the number of data points newly obtained from the deliberation process. This
measure indicated the quantitative aspect of learning. Second, this researcher measured
the number of focal dimensions whose binary value changed during the deliberation
process. This measure indicates the qualitative aspect of learning, since true learning
may cause a change in a participant‟s opinion. It was expected that as people are
motivated in pro-social manner and high epistemic manner, they would learn more in
terms of the quantity of new information obtained and the degree of change of their
opinion through the deliberation process.
The overall results of the simulation were consistent with this expectation with
some variations in the detail. Figure 7-6 illustrates the consistent positive main effect of

109
epistemic motivation and strong interaction effect between social motivation and
epistemic motivation. First, high epistemic motivation led to a higher degree of learning
in both pro-self and pro-social motivation. Second, the interaction effect showed that the
difference of the degree of learning increased when participants were motivated in a pro-
self manner compared to when they were motivated in a pro-social manner. Actually,
while pro-social groups were relatively homogeneous in the degree of learning,
regardless of the type of epistemic motivation, pro-self groups were clearly divided into
two groups by their epistemic motivation.

Figure 7-6. Newly Obtained Data Points through Information Sharing

The positive effect of pro-social motivation and high epistemic motivation was
confirmed again by statistical analysis (Tables 7-8; 7-9). Participants with pro-social
motivation increased the number of data points they knew by 10.47% point of the total
data points, while participants with pro-self motivation increased the number of data
0
5
10
15
20
25
30
Lowest Highest
pro-self-low pro-self-high
pro-social-low pro-social-high

110
points by 7.34% point (F = 266.46, p < .01). The effect of epistemic motivation on
learning was more obvious. Specifically, participants with high epistemic motivation
increased the number of data points by 13.05% point, while participants with low
epistemic motivation increased the number of data points only by 5.67% point (F =
4433.65, p < .01). The effect of epistemic motivation was magnified in pro-self
motivation groups, as indicated in the interaction effect between social and epistemic
motivation (F = 2078.63, p < .01). When participants were motivated in a pro-self way,
the difference between low epistemic motivation groups (2.05% point) and high
epistemic motivation groups (15.09% point) was quite large compared to the difference
between those two groups with pro-social motivation (9.25% point versus 11.69% point
respectively). As shown in Figure 7-6, these effects were consistent regardless of the
level of openness.

Table 7-8
Learning: Main Effect
Social Epistemic
Pro-self Pro-social Low High
Quantity
M(SD) 7.34(7.77) 10.47(2.72) 5.67(3.84) 13.05(5.19)
F 266.46** 4433.65**
Quality
M(SD) 22.74(8.53) 23.27(7.87) 24.83(7.69) 20.90(8.23)
F 36.37** 347.07**
N 1662 2000 1988 1674
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

111
Table 7-9
Learning: Interaction Effect
Pro-self Pro-social Interaction
Low High Low High F
Quantity M(SD) 2.05(.41) 15.09(6.89) 9.25(1.84) 11.69(2.92) 2078.63**
Quality M(SD) 26.92(7.24) 16.62(6.28) 22.75(7.57) 23.79(8.14) 518.76**
N 988 674 1000 1000
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

Another measure of learning is to what degree participants changed their policy
opinion as a result of group deliberation. From Figure 7-7, the main effects of social
motivation and epistemic motivation were not consistent. Around the middle level of
openness, the effect of motivation on learning was reversed. In the lower range of
openness, pro-social motivation and low epistemic motivation were associated with a
higher degree of a change of opinion. The higher range of openness, in contrast, is
characterized by the strong interaction effect. Specifically, epistemic motivation made a
larger difference among groups with pro-self motivation than among groups with pro-
social motivation.

Figure 7-7. Ratio of the Focal Dimensions Changed
0
20
40
60
80
100
Lowest Highest
pro-self-low pro-self-high
pro-social-low pro-social-high

112
An ANOVA test revealed that when participants were motivated in a pro-social
way (F = 36.37, p < .01) or a low epistemic way (F = 347.07, p < .01), the degree of
opinion change was higher. The main effects of social and epistemic motivation were
identified in the higher range of openness, as seen in Figure 7-7, and became clearer
following the statistical analysis. In addition, as shown in Figure 7-7, the strong
interaction effect (F = 518.76, p < .01) demonstrated that the effect of epistemic
motivation was contingent on social motivation. When participants were motivated in a
pro-social way, there was relatively small difference between low epistemic motivation
groups (22.75% of the focal dimensions changed) and high epistemic motivation groups
(23.79% changed). In contrast, when participants were motivated in a pro-self way, there
was relatively large difference between low (26.92% changed) and high (16.62%
changed) epistemic motivation groups.
In summary, pro-social motivation was associated with a higher degree of
learning. High epistemic motivation was associated with a higher degree of obtaining
new information. As for the degree of opinion change, low epistemic motivation was
associated with a higher degree of opinion change. The interaction effect was significant
in a manner that magnified the effect of epistemic motivation among groups with pro-self
motivation.
Satisfaction
Table 7-10 shows participants‟ average level of satisfaction, which was very
sensitive to the level of openness. Pro-self motivation and low epistemic motivation were
associated with a higher level of participant satisfaction. The main effect of the two types

113
of motivation was consistent at all levels of openness. A unique finding regarding the
level of satisfaction was the slope of increase in the level of satisfaction among the low
epistemic motivation groups. This result should be addressed cautiously. The measure
of participant satisfaction, developed in the computational model, reflected two dynamics
of the model. One was the changing degree of fit between each participant‟s policy
opinion and the current collective decision proposal. The other was the changing level of
confidence, that is, the size of the weight given to each data point, which was related to
the group dynamic known as mutual enhancement” (Wittenbaum et al., 1999). In the
case of low epistemic motivation groups, the level of satisfaction reflected the strong
effect of mutual enhancement. As specific information was shared among participants,
participants with the same opinion strengthened each other‟s confidence about the
validity and credibility of the information they shared. When specific information was
mentioned frequently, at some point it became the most important information to a
sufficient number of participants in terms of its large weight. When more and more
participants mentioned the information with a large weight, this resulted in adding even a
larger weight on the information. This loop of mutual enhancement could occur more
frequently when participants are motivated in a pro-self or low epistemic way, when only
a narrow range of information is shared.

114
Table 7-10
Level of Satisfaction by the Type of Motivation
Openness Lowest Highest
pro-Self-Low
46.74 190.89 534.15 2279.21 12774.49 110585.5 592278.9
pro-Self-High
55.98 79.88 111.65 125.08 140.93
pro-Social-Low
48.11 1122.74 1096.25 2310.55 865.87 16971.48 4305.7 6169.06
pro-Social-High
23.06 26.26 29.95 33.96 38.83 42.15 48.77 56.6

An ANOVA test revealed that the statistical significance of the effect of
motivation on the level of participant satisfaction was not large, even though the
differences appeared significantly large (F
social
= 4.75, p < .05; F
epistemic
= 4.95, p < .05;
F
interaction
= 4.75, p < .05) (Tables 7-11; 7-12). Table 7-11 shows that the level of
satisfaction, at the start, was successfully randomized. By the end of deliberation, pro-
self motivation and low epistemic motivation were associated with higher levels of
satisfaction and the interaction effect showed that the main effect of pro-self motivation
was magnified among the low epistemic motivation groups. As seen in Table 7-10, these
effects were consistent in all levels of openness. Additionally, as discussed above, these
results indicate that when participants reach consensus, those with pro-self motivation or
low epistemic motivation may leave with higher confidence and, by doing so, higher
satisfaction with their decision.

115
Table 7-11
Satisfaction: Main Effect
Social Epistemic
Pro-self Pro-social Low High
Start
M(SD)
.94
(2.87)
1.07
(2.83)
.93
(2.78)
1.11
(2.93)
F 1.49 3.10
End
M(SD)
352145.95
(5960834.86)
3112.83
(54463.82)
297455.05
(5451624.62)
90.55
(74.13)
F 4.75* 4.95*
N 1662 2000 1988 1674
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

Table 7-12
Satisfaction: Interaction Effect
Pro-self Pro-social
Interactio
n
Low High Low High F
Start M(SD)
.90
(2.80)
1.01
(2.97)
.96
(2.76)
1.18
(2.90) .31
End M(SD)
592278.93
(7723529.66)
140.93
(82.61)
6169.06
(76921.28)
56.60
(41.71) 4.75*
N 988 674 1000 1000
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; **p<.01

116
CHAPTER 8: EFFECT OF GROUP LEVEL VARIABLES ON THE
PERFORMANCE OF GROUP DELIBERATION

The purpose of the virtual experiment in this chapter was to examine the effect of
group-level variables on the performance of simulated collaborative governance systems.
The variables were grouped into two categories: the degree of interdependence of
stakeholder interest and modes of intervention. The former is usually an exogenous
condition to individual actors by the nature of the pertinent social issue. The latter is
something that leaders can manipulate to some degree. The degree of interdependence of
interest was manipulated by the number of focal dimensions (one, three, and five) that
each participant was concerned. That is, when the number of focal dimensions increased,
the degree of interdependence of interest among stakeholders also increased. In this case,
it could be said that the need for mutual adjustment of the stakeholder interests becomes
high. Therefore, the degree of overlap of interest and the degree of interdependence of
interest are used interchangeably hereafter. Modes of intervention included the rule of
determining a starting alternative (majority opinion, opinion opposite of the majority, and
random opinion), the rule of determining speaker turns (random selection, egalitarian,
and round-robin), and forum size (10, 30, 50). Consequently, a 3 x 3 x 3 x 3 factorial
experimental design was employed. Each participant‟s social and epistemic motivation
in a group was randomly assigned. That is, unlike the experiment discussed in Chapter 7,
a deliberation group, in this experiment, was composed of heterogeneous participants in
their information sharing behavior. This design was used to control for the effect of
motivation by randomizing the effect, in order to focus on the effect of group-level

117
variables. Further, it was considered that it would be more realistic to assume that a
collaborative forum is composed of heterogeneous actors in terms of their information-
processing motivation. As was the case in the statistical analysis of the experiment on
motivation, the level of openness of participants was set to the highest level. This
provided a sufficient number of successful groups to be included in the statistical analysis
in every factorial condition, which enhanced the reliability of the statistical results.
In each experimental condition, 200 sample groups were generated, which
resulted in the total sum of 16,200 sample groups for this experiment. Each system was
composed of 500 constituents. Each forum was given up to 1,000 speaking turns to share
information and make a decision. The dependent variables in the category of success,
global responsiveness, local responsiveness, and learning were used in this experiment.
Because the dependent variables are meaningful only in the case of groups that succeeded
in reaching consensus, only the successful groups were considered in the analysis.
The report of the analysis focused on the main effect of each independent variable.
Interaction effects between the variables are reported only when they were statistically
significant at the .01 significance level and added new insight to the main effect of each
variable. In particular, interactions were considered from two perspectives (1) how and
with what managerial intervention the degree of interdependence interacted and (2) what
managerial interventions interacted with each other.
Success
Table 8-1 summarizes the results of the ANOVA regarding the success rate and
time to reach consensus.

118
Table 8-1
Effect of Group Variables on Success Rate and Time to Reach Consensus
Interdependence Starting alternative Speaking turn Forum size
Success
rate
Low 89.07
(31.20)
majorit
y
91.09
(28.49)
random 96.11
(19.34)
small 99.83
(4.08)
Mediu
m
91.74
(27.53)
opposit
e
91.83
(27.39)
egalitaria
n
81.87
(38.53)
mediu
m
91.72
(27.56)
High 93.30
(25.01)
random 91.19
(28.35)
round 96.13
(19.29)
large 82.56
(37.95)
F

39.78**

1.42
590.58*
*

651.97**
N
Low
5400
majorit
y 5400
random
5400
small
5400
medium
5400
opposit
e 5400
egalitaria
n 5400
mediu
m 5400
high
5400
random
5400
round
5400
large
5400
time
Low
388.74
(249.55)
majorit
y
234.96
(189.42
)
random
245.06
(201.36)
small
122.52
(100.23)
medium
182.79
(94.51)
opposit
e
236.35
(193.64
)
egalitaria
n 215.85
(156.82)
mediu
m 281.27
(188.80)
high
140.43
(64.74)
random 234.59
(188.39
)
round
242.11
(203.80)
large
320.60
(209.14)
F
7845.07*
*

.45
100.15*
*
5085.51*
*
N
Low
4810
majorit
y 4919
random
5190
small
5391
medium
4954
opposit
e 4959
egalitaria
n 4421
mediu
m 4953
high
5038
random
4924
round
5191
large
4458
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; ** p<.01

Interdependence. An analysis on the success rate clearly shows that a higher
degree of interdependence was related to a higher success rate and a shorter time to reach
consensus. First, the success rates improved from 89.07% with low interdependence,
91.74% with moderate interdependence, and 93.30% with high interdependence. In
interpreting the result, the difference of the success rate between the levels of
interdependence may seem small. Nevertheless, the difference was statistically

119
significant (F = 39.78, p < .01). In addition, the author ran the computational model with
the medium level of openness to check the sensitivity of the results. This analysis
showed that the difference of the success rate was larger (i.e., 31.76% - 62.30% - 70.19%,
respectively) with a medium level of openness. Therefore, the results shown in Table 8-1
are conservative regarding the difference in the success rate caused by the degree of
interdependence.
The result also showed that the higher degree of interdependence was associated
with the shorter time to reach consensus. Not only did time to reach consensus reduce
from 388.74 (low degree of interdependence) to 182.79 (moderate) to 140.43 (high), but
so did the standard deviation of time (249.55 – 94.51 – 64.74, respectively). The
difference was statistically significant (F = 7845.07, p < .01).
Starting alternative. There was no significant difference in the success rate
among groups with different starting alternatives (F = 1.42, p > .10). In addition, there
was no difference in time to reach consensus among groups with different starting
alternatives (F = .45, p > .10).
Speaker turn. When collaborative forums followed the egalitarian rule of
speaker turn, the success rate was the lowest (81.87%), compared to that of the random
selection (96.11%) and round-robin rules (96.13%) (F = 590.58, p < .01). These results
reveal that it is difficult for groups that give priority to the minority‟s opinion to reach
consensus. However, time taken to reach consensus yielded an interesting result. The
random selection rule took 245.06 rounds, the round-robin rule took 242.11 rounds, and
the egalitarian rule took 215.85 rounds (F = 100.15, p < .01) to reach consensus. These

120
results suggest that, although success rate was low when following the egalitarian rule,
successful groups reached consensus relatively quickly.
Forum size. The simulation results show that there was a negative relationship
between the forum size and the success rate to reach consensus (F = 651.97, p < .01). As
the forum size increased from 10 to 30 to 50, the success rate decreased from 99.83% to
91.72% to 82.56%, respectively. In addition, as forum size increased, it took a longer
time for participants to reach consensus (F = 5085.51, p < .01).
Interaction effects. Noteworthy interactions between the group variables were
also found. First, as shown in Figure 8-1, an interaction effect between the
interdependence of interest and the rule of speaker turn was found (F = 200.57, p < .01).
This result indicates that when the degree of interdependence was moderate or high, the
adoption of the egalitarian rule of speaker turn had a negative effect on the probability to
reach consensus, which is consistent with the main effect of the egalitarian rule.
However, when the degree of interdependence was low, the effect was reversed. The
adoption of the egalitarian rule of speaker turn can be helpful in reaching consensus when
the degree of interdependence of interest among stakeholders is low.

121

Figure 8-1. Interaction between Interdependence and the Rule of Speaker Turn: Success
Rate

An interaction between the rule of speaker turn and forum size was also found (F
= 14.51, p < .01), in such a way that success rate decreased when large forums adopted
the egalitarian rule. The interaction shown in Figure 8-2 indicates that, when the
egalitarian rule of speaker turn is employed, the probability to reach consensus will
decrease to a significant degree as the forum size increases.

Figure 8-2. Interaction between the Rule of Speaker Turn and Forum Size: Success Rate

0
20
40
60
80
100
random egalitarian round
low
moderate
high
0
20
40
60
80
100
random egalitarian round
small
medium
large

122
Regarding time taken to reach consensus, Figure 8-3 shows an informative
interaction effect between the interdependence of interest and the rule of speaker turn (F
= 448.64, p < .01). Among the forums with a high degree of interdependence, the rule of
speaker turn did not make a significant difference. However, as the degree of
interdependence decreased, the egalitarian rule helped reduce time to reach consensus.
Similarly, as shown in Figure 8-4, a high degree of interdependence attenuated the
negative main effect of forum size (F = 944.11, p < .01). In short, a high degree of
interdependence neutralized the main effect of managerial intervention (speaker turn and
forum size). Further, there was no interaction between the interdependence of interest
and the rule of a starting alternative.

Figure 8-3. Interaction between Interdependence and the Rule of Speaker Turn: Time to
Reach Consensus

0
50
100
150
200
250
300
350
400
450
500
random egalitarian round
low
moderate
high

123

Figure 8-4. Interaction between Interdependence and Forum Size: Time to Reach
Consensus

Figure 8-5 illustrates another informative interaction between the rule of speaker
turn and forum size (F = 41.58, p < .01). Although the main effect of the egalitarian rule
was to reduce time to reach consensus, the effect was reversed when applied to large
forums. This result indicates that forum leaders should be aware of the effect of the
egalitarian rule when they form a large forum, for example, for political considerations.

Figure 8-5. Interaction between the Rule of Speaker Turn and Forum Size: Time to Reach
Consensus

0
100
200
300
400
500
600
small medium large
low
moderate
high
0
50
100
150
200
250
300
350
400
random egalitarian round
small
medium
large

124
Global Responsiveness
Table 8-2 summarizes the results of the ANOVA regarding global responsiveness.

Table 8-2
Effect of Group Variables on Global Responsiveness
Interdependence Starting alternative Speaking turn Forum size
Start
low
.07(1.97) majority .99(1.79) random .05(2.00) Small .05(1.85)
medium
.05(2.00) opposite -1.03(1.79) egalitarian .06(1.99) medium .06(2.04)
high
.05(1.98) random .22(1.82) round .06(1.96) Large .06(2.08)
F .47 1695.93** .02 .22
End
low
.13(1.79) majority .27(1.80) random .24(1.81) Small .19(1.79)
medium
.26(1.80) opposite .30(1.82) egalitarian .38(1.79) medium .39(1.81)
high
.48(1.82) random .31(1.80) round .28(1.82) Large .31(1.82)
F 51.61** .67 12.27** 17.69**
improve
low
.07(2.52) majority -.72(2.15) random .18(2.63) Small .14(2.42)
medium
.20(2.66) opposite 1.34(2.83) egalitarian .32(2.65) medium .33(2.72)
high
.43(2.69) random .09(2.45) round .22(2.62) Large .25(2.78)
F 32.59** 905.81** 6.57** 7.57**
N
low
4810 majority 4919 random 5190 Small 5391
medium
4954 opposite 4959 egalitarian 4421 medium 4953
high
5038 random 4924 round 5191 Large 4458
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; ** p<.01

Interdependence. The simulation results on global responsiveness of the
collective decision showed that a high degree of interdependence helps improve the
global responsiveness of the collective decision. First, the average responsiveness of
starting alternatives did not differ across three degrees of interdependence (F = .47, p
> .10). Second, there was a positive relationship between the degree of interdependence
and average global responsiveness of the final decisions (F = 51.61, p < .01). As a result,
there was also a positive relationship between the degree of interdependence and average
degree of improvement of global responsiveness (F = 32.59, p < .01). While groups with
a low degree of interdependence improved the responsiveness of their decision by .07

125
(from .07 to .13), groups with a moderate degree of interdependence did so by .20
(from .05 to .26), and groups with a high degree of interdependence did so by .43
(from .05 to .48).
Starting alternative. The different rules of setting a starting alternative resulted
in a difference in the average global responsiveness of the starting alternatives (F =
1695.93, p < .01). As was expected, the highest average for global responsiveness (.99)
at the start was recorded by forums that started with the majority opinion, while the
lowest average global responsiveness (-1.03) at the start was recorded by forums that
started with the opposite opinion. The random start recorded an average responsiveness
in the middle (.22). Second, in spite of the difference of starting point, there was no
significant difference in the average global responsiveness of the final decision across
three rules of determining a starting alternative (F = .67, p > .10). Finally, the average
degree of improvement of global responsiveness, as a result of deliberation, was different
across the three rules of determining a starting alternative (F = 905.81, p < .01). The
degree of improvement under each rule was noticeable. Starting with the majority
alternative was negatively related to improvement, with decreased global responsiveness
by -.72. In contrast, starting with the opposite alternative was positively related to
improvement, with increased global responsiveness by 1.34. As a comparison, groups
starting with a random alternative slightly improved global responsiveness (.09). In short,
the simulation results showed that with whatever alternative collaborative forums started
deliberation, the global responsiveness of their decision converged into an average level.

126
Speaker turn. While starting with no difference in global responsiveness as was
manipulated (F = .02, p > .10), forums with the egalitarian rule recorded a higher average
global responsiveness at the end of deliberation (.38 compared to .24 for the random rule
and .28 for the round-robin rule), which was statistically significant (F = 12.27, p < .01).
Forums with the egalitarian rule also succeeded in improving the most for local
responsiveness of their decision significantly more so than did groups with the other rules
(F = 6.57, p < .01). Forums with the round-robin rule were the next well-performing
groups, followed by forums with the random rule.
Forum size. Forum size did not make difference at the starting line (F = .22, p
> .10). The global responsiveness of the final decision showed a curvilinear relationship
with forum size. The average global responsiveness was highest among the medium-size
forums (.39), followed by the large-size forums (.31), and the small-size forums (.19) (F
= 17.69, p < .01). Consequently, the degree of improvement was also largest among the
medium-size forums (.33), followed by the large-size forums (.25), and the small-size
forums (.14) (F = 7.57, p < .01).
Interaction effects. Important interaction effects, concerning the degree of
interdependence and the modes of intervention, were found. There was no significant
interaction among the modes of intervention.
Interdependence interacted with the rule of setting starting alternatives, as shown
in Figure 8-6 (F = 4.58, p<.01). When the degree of interdependence was low, starting
deliberation with the opposite opinion produced a negative result to global
responsiveness of the final decision. In contrast, when the degree of interdependence was

127
high, starting deliberation with the opposite opinion produced a positive result. Therefore,
even though the main effect of staring with the opposite opinion was positive to global
responsiveness, one cannot expect the positive effect in the forums that have a low degree
of interdependence.

Figure 8-6. Interaction between Interdependence and the Rule of Starting Alternative:
Global Responsiveness

Second, a similar interaction effect was found between the degree of
interdependence and the rule of speaker turn (F = 4.92, p < .01). As shown in Figure 8-7,
the overall positive effect of the egalitarian rule to global responsiveness was negated
when the degree of interdependence was low.

-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
majority opposite random
low
moderate
high

128

Figure 8-7. Interaction between Interdependence and the Rule of Speaker Turn: Global
Responsiveness

Local Responsiveness
Interdependence. Table 8-3 summarizes the results of the ANOVA regarding
local responsiveness. The result in Table 8-3 shows that the average local responsiveness
of the final decision and degree of interdependence were negatively related (F = 6518.27,
p < .01). While the average local responsiveness of the final decision among forums with
a low degree of interdependence was as high as 41.56, it decreased to 20.64 with a
moderate degree of interdependence, and to 13.95 with a high degree of interdependence.
Since there was no difference in the average local responsiveness at the start, across
degrees of interdependence (F = 1.32, p > .10), the degree of improvement in local
responsiveness through deliberation showed the same negative relationship with the
degree of interdependence (F = 2520.91, p < .01).

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
random egalitarian round
low
moderate
high

129
Table 8-3
Effect of Group Variables on Local Responsiveness
Interdependence Starting alternative Speaking turn Forum size
Start
low 1.98
(26.87)
majority
17.01
(16.39)
random
1.66
(21.07)
small
3.27
(29.27)
medium 1.55
(19.25)
opposite
-16.78
(15.95)
egalitarian
2.22
(21.96)
medium
1.01
(16.86)
high 1.71
(17.19)
random
5.15
(16.29)
round
1.42
(21.32)
large
.71
(12.98)
F 1.32 5774.90** .86 41.13**
End
low 41.56
(22.33)
majority
25.30
(20.50)
random
25.34
(19.91)
small
39.22
(22.29)
medium 20.64
(14.00)
opposite
25.26
(20.09)
egalitarian
24.45
(20.75)
medium
19.79
(14.52)
high 13.95
(11.47)
random
24.92
(20.11)
round
25.58
(20.10)
large
14.11
(11.44)
F 6518.27** 1.82 142.42** 5935.28**
impr
ove
low 39.58
(33.88)
majority
8.29
(20.05)
random
23.68
(28.21)
small
35.96
(36.03)
medium 19.09
(22.97)
opposite
42.04
(30.44)
egalitarian
22.23
(28.94)
medium
18.78
(22.11)
high 12.24
(20.15)
random
19.77
(23.30)
round
24.16
(28.84)
large
13.40
(17.34)
F 2520.91** 3755.81** 68.04** 1897.04**
N
low
4810 majority 4919 random 5190 small 5391
medium
4954 opposite 4959 egalitarian 4421 medium 4953
high
5038 random 4924 round 5191 large 4458
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05;**p<.01

The absolute level of local responsiveness was noticeable. When the degree of
interdependence was low, participants could form a collective decision that was
responsive to their own parties and was, on average, as high as 41.56. This means that
the decision met 41.56% of the ideal satisfaction level of constituents in a stakeholder
group. Considering that 0 meant that the party was neither satisfied nor dissatisfied with
the collective decision, the level of responsiveness of 41.56 achieved by the collective
decision, along with the degree of improvement of 39.58, was a meaningful sign of the
virtue of deliberation.

130
Starting alternative. As manipulated, the average local responsiveness in forums
with three rules to determine starting alternatives was statistically different (F = 5774.90,
p < .01). Starting with the majority alternative was best for local responsiveness (17.01).
However, starting with the opposite alternative, due to its unpopularity, recorded the
lowest average local responsiveness (-16.78). Further, starting with a random alternative,
as a comparison, recorded local responsiveness in the middle (5.15). Second, there was
no difference among the three rules regarding the local responsiveness of the final
decision (F = 1.82, p > .10). Finally, though with different degrees, all forums improved
their local responsiveness through deliberation regardless of their rule of starting
alternative (F = 3755.81, p < .01). Specifically, forums starting with the opposite
alternative were most successful concerning improvement (42.04). Forums that started
with the majority alternative improved the least concerning local responsiveness of their
decision (8.29). Forums that started with a random alternative moderately improved
concerning local responsiveness of their decision (19.77). Unlike the case of global
responsiveness, local responsiveness could be improved through deliberation even when
they started deliberation with the majority opinion.
Speaker turn. The average local responsiveness at the start of deliberation was
not different across the three rules of setting speaker turns, as was manipulated (F = .86,
p > .10). The different rules for setting speaker turns had a statistically significant, but
practically small, difference in the average local responsiveness of the final decision (F =
142.42, p < .01) and degree of improvement (F = 68.04, p < .01). The results
demonstrated the inferior performance of forums with the egalitarian rule. Forums with

131
the egalitarian rule recorded the lowest average local responsiveness (24.45) compared to
forums with the random rule (25.34) and the round-robin rule (25.58). This result,
however, should be analyzed with caution. A sensitivity test regarding this result showed
that in the condition of a moderate level of openness, sample groups with the egalitarian
rule achieved the highest average local responsiveness (38.87) compared to groups with
the random rule (26.43) and the round-robin rule (25.19). So the performance of the
egalitarian rule on the local responsiveness of the decision was contingent on how open
participants were to others‟ opinion.
Forum size. Local responsiveness of starting alternatives was negatively related
to forum size (F = 41.13, p < .01). Local responsiveness of the final decision was also
negatively related to forum size (F = 5935.28, p < .01). While local responsiveness of
the final decisions among the smallest forums was as high as 39.22, among the largest
forums it was as low as 14.11. These results indicate that, given the number of policy
dimensions (i.e., eight in this study), the participants‟ diverse opinions in a large forum
could not be well reflected in the deliberation process. Accordingly, the average degree
of improvement was the highest among the smallest forums (35.96) and the lowest
among the largest forums (13.40) (F = 1897.04, p < .01).
Interaction effects. Interactions between the degree of interdependence and the
rule of speaker turn (F = 6.74, p < .01), as well as between the degree of interdependence
and forum size (F = 262.54, p < .01) were found to be significant in a way that magnified
the positive main effect of the low degree of interdependence on local responsiveness of

132
the final decision. There was no interaction between the degree of interdependence and
the rule of starting alternative, or among the modes of intervention.
Learning
Table 8-4 summarizes the results of the ANOVA regarding the degree of learning
measured in two ways: the amount of newly obtained data through deliberation (quantity)
and the degree of opinion change through deliberation (quality).

Table 8-4
Effect of Group Variables on the Degree of Learning
Interdependence Starting alternative Speaking turn Forum size
Quantity
low 13.86
(7.21)
majority
8.96
(5.92)
random
8.36
(5.64)
small
6.10
(4.22)
medium 7.38
(3.67)
opposite
9.00
(6.05)
egalitarian
10.61
(6.24)
medium
10.54
(5.90)
high 5.90
(2.54)
random
8.98
(5.88)
round
8.22
(5.73)
large
10.74
(6.44)
F 6320.68** .12 588.62** 2982.20**
Quality
low 29.22
(11.17)
majority
19.09
(12.80)
random
20.08
(12.24)
small
9.33
(10.47)
medium 16.91
(11.40)
opposite
19.04
(12.84)
egalitarian
17.35
(13.94)
medium
22.20
(10.39)
high 11.75
(8.98)
random
19.34
(12.86)
round
19.76
(12.27)
large
27.66
(9.74)
F 9300.12** 2.96 228.40** 9587.13**
N
low
4810 majority 4919 random 5190 small 5391
medium
4954 opposite 4959 egalitarian 4421 medium 4953
high
5038 random 4924 round 5191 large 4458
Note. Numbers are means. Numbers in parenthesis are standard deviations.
* p<.05; ** p<.01

Interdependence. The result concerning the amount of newly obtained data
points showed that the degree of learning and degree of interdependence were negatively
associated. Participants with a low degree of interdependence obtained the largest
number of new data points (13.86% of the total data points), followed by participants

133
with a moderate degree of interdependence (7.38%) and a high degree of interdependence
(5.90%) (F = 6320.68, p < .01). Another aspect of learning was the degree of change of
focal policy opinions that showed a negative relationship between the two variables (F =
9300.12, p < .01). That is, the low degree of interdependence was associated with the
high degree of opinion change. This result may seem counterintuitive. First, this result
can be analyzed from another aspect; for example, the absolute degree of opinion change.
The actual number of focal dimensions that changed during deliberation was the largest
in the high interdependence group. That is, while in the high interdependence
participants changed, on average, .59 focal dimensions (5 focal dimensions * 11.75% of
average change), participants in the moderate interdependence changed, on average, .51
focal dimensions (3 focal dimension * 16.91% of average change), and participants in the
low interdependence changed, on average, .29 focal dimensions (1 focal dimension *
29.22% of average change). This result indicates that when interdependence is high,
participants are more likely to change their opinions, which is not in a relative term but in
an absolute term. Second, the result on the degree of learning was also associated with
the time taken to reach consensus. Since forums with a high degree of interdependence
of interest among participants reached consensus more quickly than other forums (Table
8-1), they had fewer opportunities to exchange data points throughout the deliberation
process.
Starting alternative. The different ways of setting a starting alternative did not
significantly differ in the degree of learning either in terms of the amount of newly

134
obtained data points (F = .12, p > .10) or in terms of the degree of policy opinion change
(F = 2.96, p > .10).
Speaker turn. In terms of quantity of learning, groups with the egalitarian rule
learned the most (10.61%) compared to 8.36% for groups with the random rule, and 8.22%
for groups with the round-robin rule) (F = 588.62, p < .01). The high degree of learning
among groups with the egalitarian rule, despite the shortest time to reach consensus,
highlights the positive effect of the egalitarian rule in information sharing. A sensitivity
test of the effect, however, showed that in the condition of the medium level of openness,
participants with the egalitarian rule obtained the fewest new data points (10.80%
compared to 13.31% of the random rule and 14.27% of the round-robin rule). Therefore,
a firm conclusion on the relationship between the quantity of learning and the egalitarian
rule cannot be made.
In terms of the degree of change of policy opinion, the results show that groups
with the egalitarian rule changed the least (17.35%) compared to the other groups (20.08%
with the random rule and 19.76% with the round-robin rule) (F = 228.40, p < .01).
Unlike the case of the quantity of learning, a sensitivity test demonstrated consistency in
this relationship.
Forum size. As small forums reached consensus more quickly, learning in terms
of both quantity and quality occurred to a smaller degree in the smaller forums than in the
larger forums. First, participants in the small forums obtained the fewest number of new
data points (6.10%), compared to medium-size forums (10.54%) and large-size forums
(10.74%; F = 2982.20, p < .01). Second, participants in the small forums changed only

135
9.33% of their focal dimensions, while those in the large forums changed 27.66% of their
focal dimensions (F = 9587.13, p < .01).
Interaction effects. Regarding the quantity of newly obtained data points
through deliberation, interactions between interdependence and forum size are
noteworthy. As shown in Figure 8-8, a low degree of interdependence made a difference
when forum size was large (F = 638.06, p < .01). In other words, when the degree of
interdependence was moderate or high, forum size did not make difference in learning.
These results indicate that as long as a high degree of interdependence exists among
stakeholders, forum size is not critical to the degree of learning. Given a low degree of
interdependence in large forums, however, the degree of learning can be significant,
which may indicate the difficulty in building a common cognitive ground among
participants with divergent interests and in reaching consensus.

Figure 8-8. Interaction between Interdependence and Forum Size: Newly Obtained Data
Points

0
5
10
15
20
small medium large
low
moderate
high

136
The egalitarian rule of speaker turn had an effect to reduce the degree of opinion
change (Table 8-4). The effect was reversed, however, when the degree of
interdependence was low (F = 48.17, p < .01). As shown in Figure 8-9, a low degree of
interdependence negated the difference of opinion change among three rules of speaker
turn, even to the point of increasing the degree of opinion change under the egalitarian
rule.

Figure 8-9. Interaction between Interdependence and the Rule of Starting Alternative:
Change of Policy Opinion

0
5
10
15
20
25
30
35
random egalitarian round
low
moderate
high

137
CHAPTER 9: DISCUSSION

This study modeled the theoretical relationship between motivated information-
processing behavior and outcomes of motivated behavior in deliberative decision-making
processes in the context of collaborative governance. This study also investigated the
effect of interdependence of stakeholder interests and managerial intervention in the
deliberation process among participants. The relationships among variables were
simulated via the internal dynamics of the agent-based model developed in this study.
Using a computer simulation, virtual data on the direction and the strength of the
relationships were generated and analyzed. As described in the results chapters, results
are rather complex. Thus, it would not be helpful to discuss implications of all the details
of these simulation results. Rather, this researcher approaches the simulation results from
a broad theoretical interest discussed in the collaborative governance literature and
discusses results in light of generating theoretical propositions. Reflection from real
world case studies is applied to vitalize implications of the simulation results and assess
their practical relevance.
Two virtual experiments were performed at two levels. One experiment was
performed to investigate the effect of social and epistemic motivation at the individual
level on the outcomes of deliberation. The other experiment was performed to
investigate the effect of interdependence of stakeholder interest and managerial
interventions at the group level on the outcomes of deliberation. In the following
sections, theoretical implications of the results are discussed and propositions are
developed based on the results from these two virtual experiments. First, before

138
discussing the simulation results and propositions, three exemplary collaboration cases
are introduced. Second, propositions regarding the effect of motivated information-
processing are discussed. Finally, propositions regarding the effect of group level
variables are discussed.
Exemplary Cases for Analysis
To better understand and support the simulation results and propositions, three
collaborative governance cases are discussed below according to the findings and
implications of the simulation results. The cases include the Sacramento Area Water
Forum, the Westlake Joint-Use Collaborative in Los Angeles, and the Bay Area Alliance
for Sustainable Communities.
The Sacramento Area Water Forum. The Sacramento Area Water Forum
(Water Forum) was selected as a successful collaborative governance case. Not only is
the case regarded exceptionally successful by researchers, but also are there detailed
reports available on this case (Connick, 2006; Innes & Booher, 2010). In particular,
Innes and Booher (2010) described the Water Forum case from the perspective of their
diversity, interdependence, and authentic dialogue (DIAD) theory, which focuses more
on the deliberative process than on institutional backgrounds. Therefore, the case report
provides useful information regarding the theoretical focus of this study.
Beginning in 1993 and continuing for approximately six years, the Water Forum
was established to solve water resource usage and preservation issues around the lower
American River in Northern California. Stakeholders, including the City of Sacramento,
the County of Sacramento, the East Bay Municipal Utility District, water purveyors,

139
business developers, environmentalists, and taxpayers, gathered together to
collaboratively solve the water issue. The Water Forum followed five phases: planning,
organization, education, negotiation and resolution of issues, and implementation. In the
organization phase, the Water Forum was organized as an overarching Working Group
involving all stakeholders, caucuses, and four teams (Surface Water Team, Ground Water
Team, Demand Conservation Team, and Habitat Management Team). At the end of the
Forum, after six years and nearly $10 million, the MOU was signed, more than 41 entities
developed, and the Successor Effort was established to monitor the implementation phase.
Encouraged by the success of the Forum, many other public forums in the area were
established.
The Westlake Joint-Use Collaborative. The Westlake Joint-Use Collaborative
in Los Angeles was selected as another successful case. In 2006, the researcher
participated in a community-based research project that aimed to evaluate the process and
the result of the collaborative effort via qualitative methods including archival data
analysis and semi-structured interviews with 11 interviewees (Choi, Walsh, & Weiner,
2006). Therefore, the researcher is in a good position for deeper understanding of the
case using first-hand qualitative data. Because the specific interview data were to be
used only for reporting to the client organization (New Schools Better Neighborhoods),
the description about the case in this section is restricted to information from the open
archival data and some general information from the interviews.
The Westlake Joint-Use Collaborative was a collaborative forum convened by the
nonprofit organization, New Schools Better Neighborhoods (NSBN). Beginning in late

140
2002, the purpose of the forum was to solve the conflict between the Los Angeles Unified
School District (LAUSD) and the nonprofit housing developer, A Community of Friends
(ACOF), both of which designated a site near downtown Los Angeles to build a new
school or affordable housing, respectively. The Westlake area was a Latino community
with high density and poverty rate; therefore, both a new elementary school and
affordable housing were in urgent need. The Mayor of the City of Los Angeles and a Los
Angeles City Councilmember (District 1) realized that the issue should be solved
collaboratively and decided to invite NSBN, which was an expert in joint-use projects.
Diverse stakeholders, including LAUSD, ACOF, Los Angeles City, Del Sol Group, and
Council District 1 participated in the process. The forum continued for about a year with
a consensus that ACOF build affordable housing at the original site and the LAUSD build
a new Primary Center at the adjacent site, attaching an Early Education Center,
multipurpose room, and public restrooms.
The Bay Area Alliance for Sustainable Communities. The Bay Area Alliance
for Sustainable Communities (BAASC), reported by Innes and Booher (2010), was not a
successful collaborative forum. Therefore, this case was selected for comparison
purposes. The forum began in 1997 by civic leaders and nonprofit organizations in the
San Francisco Bay region, with the goal to make the Bay Area more sustainable. The
organization was composed of a Steering Committee and four caucuses. A final decision
was made after the caucuses developed a proposal and the Steering Committee approved
the proposal. According to Innes and Booher (2010), the process did not so much
resemble authentic dialogue as it did find a least common denominator among the

141
stakeholders. Specifically, they developed a document entitled “Compact” that contained
an agreement among the stakeholders. Because key actors, like local governments, were
not included in the forum, the Compact was not implemented.
Propositions Regarding the Effect of Motivated Information-Processing
Theoretical relevance of the research on motivated information-processing to
collaborative governance is obvious. Although theoretical interest is primarily on the
institutional level in the study of collaborative governance (Ansell & Gash, 2008; Tang &
Mazmanian, 2009), researchers have recognized the practical importance of face-to-face
dialogue (Innes & Booher, 2010), not in the dyadic but in the collective sense.
Specifically, if dialogue is a fundamental aspect of social learning and collaboration
(Fung & Wright, 2003), small groups or collaborative forums are the most appropriate
locus where face-to-face dialogue is possible (Innes & Booher, 2010; Senge, 1990).
From the results of the current study, theoretical propositions are developed here. First,
the discussion focuses on the effect of social and epistemic motivation. Then, the issue
of responsiveness in collaborative decision-making is discussed. Finally, a contingency
view on the virtue of authentic dialogue is discussed.
Effect of social motivation. In the current study, social motivation was
manifested by two motivated information-processing behaviors: pro-self and pro-social
information-processing. In their theoretical argument, De Dreu et al. (2008)
distinguished pro-self and pro-social motivation in terms of group members‟ pursuit of
self or collective-interest. From an information-processing perspective, this study
distinguished pro-self and pro-social motivation in terms of behavioral biases that each

142
type of motivation brings to the mutual learning process. Theoretical interest of
collaborative governance concerns whether information-processing with pro-social
motivation can yield better outcomes in the deliberation process compared to
information-processing with pro-self motivation. In fact, the behaviors of agents with
pro-social motivation are close to the idealized attitude of people; specifically, those
agents do not share information only to serve their own interest, and they are open to
others‟ opinion.
The computational model has operationalized specific behaviors in successful
information-processing with the discount rate in accepting others‟ opinions and produced
results consistent with what scholars have practically and normatively argued. From the
simulation results of the current study, we can say that pro-social information-sharing
behaviors are related to a higher probability of reaching consensus as well as a quicker
consensus. These behaviors are also associated with increased learning in terms of
knowledge acquisition and opinion change. Finally, these behaviors are positively
related to global responsiveness and negatively related to local responsiveness of the
collective decision.
The implication of the simulation results is that, even when we simplify the public
deliberation process as an information-processing process, the model predicts that
participants engaged in behaviors that resemble open dialogue will produce better
outcomes than participants engaged in the behaviors that do not resemble open dialogue.
These results are consistent with arguments raised from social psychology and
collaborative governance literature. For example, Parks et al. (1996) argued that trusting

143
negotiators hold the belief on their counterparts‟ sincerity and accuracy in information-
processing, and motivation to reach a collective consensus. This kind of trusted
information sharing leads to higher level of information exchange and better decision-
making (Thompson et al., 2010). Innes and Booher (2010) argued that inaccurate, biased,
self-serving, or inappropriate information that is more related to pro-self information
sharing than pro-social information sharing, leads to waste of time and failure of the
system.
The Water Forum case illustrates the positive effect of pro-social motivation on
reaching consensus.
24
Although stakeholders involved in the collaboration did not seem
to have begun with very high pro-social motivation, they ended with high trust among
each other, which facilitated pro-social information-processing behaviors among them.
While participating in the collaboration process, some stakeholders also acted on other
alternatives, such as lobbying and lawsuits, against the interests of other stakeholders
involved in the process (Innes & Booher, 2010). However, it was reported that
stakeholders were gradually engaged in routinized communication of information sharing.
In particular, the purpose of the education phase was to ensure that participants
understood the policy issues, not just from their own perspective, but also from the other
stakeholders‟ perspectives, which was aimed to facilitate pro-social information-
processing.

24
There is no data explicitly describing the social motivation of stakeholders. So here their motivated
information processing behavior is inferred from their changed behavior in other aspects.

144
Compared to the Water Forum case, the BAASC case illustrates that, when
deliberation focuses exclusively on individual participants‟ interest, the result might be to
agree on a least common denominator. Innes and Booher (2010) reported that, in the
BAASC case, participants wanted their interests reflected in the Compact, while not
interested in others‟ concerns. Ironically, consensus on the Compact was made because it
apparently reflected each stakeholder‟s interest well. In terms of local responsiveness,
the Compact was a positive result. However, the Compact was not implemented because
it did not reflect the deep interdependence of stakeholders‟ interests and creative
solutions.
The Westlake case demonstrated that pro-social information-processing could
lubricate the deliberation process. This case started in a deadlock; the LAUSD needed a
place to build a new primary center (K-2) and the site the LAUSD thought to be the best
was already planned to be a housing site that would be developed by a non-profit
developer, A Community of Friends. The deadlock was resolved in a creative way by
incorporating many stakeholders who brought with them ideas that supported both the
LAUSD and the developers. For example, by incorporating the idea from other
participants that an Early Education Center could be built by stacking it and the Primary
Center together using the slope of the site, the LAUSD was able to save a significant
amount of money. The idea, of course, came from the pro-social information-processing
among participants. The final decision may not necessarily be the best for each
stakeholder in terms of maximizing its self-interests, but the decision yielded a generally
good plan for the neighbors. The results and discussion led to the following propositions:

145
Proposition 1-1. As participants in the deliberation process within a collaborative
governance system are engaged in pro-social information-processing, they are more
likely to reach consensus with speed.
Proposition 1-2. As participants in the deliberation process within a collaborative
governance system are engaged in pro-social information-processing, they are more
likely to be responsive to the public and less likely to be responsive to specific
stakeholder groups.
Effect of epistemic motivation. Although pro-social motivation is a crucial part
of authentic dialogue in deliberate processes, it only tells about the departure from being
selfish to being open. Pro-social motivation does not automatically guarantee the quality
of information that participants share. They could be so nice to each other that they reach
a quick consensus with a handful of information that may not accurately reflect the
public‟s interest (Argyris, 1999). As social problems become more complex (Weber &
Khademian, 2008), a deliberate process should accompany intellectual exploration as
well as open information sharing. De Dreu et al. (2008) made a clear argument that we
have to explicitly consider “the fact that people can and will choose among a shallow and
heuristic versus a deep and deliberate information search-and-processing strategy” (p. 24).
De Dreu et al. (2008) suggested their own definition of epistemic motivation as
“the willingness to expend effort to achieve a thorough, rich, and accurate understanding
of the world, including the group task or decision problem at hand” (p.23). Brodbeck et
al. (2007) approached the issue from the perspective of symmetric and asymmetric
information-processing. They argued that factors, such as consent, among individual

146
decision preferences at the beginning of deliberation causes group members to reduce the
time and effort that will be spent for the full range of information sharing, which results
in a low degree of mutual learning. Given the norm of negotiation focus in a group as an
information processor (Brodbeck et al., 2007; Postmes et al., 2001), members tend not to
fully use the information available to them to make a better decision for the sake of
smooth consensus-building. From the perspective of collaborative governance, Innes and
Booher (2010) stated the negative the effect of this low epistemic information-processing
as follows:
People around the table often censor themselves, not saying what they are
thinking because they fear it would offend somebody or they think it is too
radical or peculiar… Moreover, some participants assume that since
collaboration is about finding common ground, they should not bring up
anything controversial. But without sincerity and without the questioning
of given knowledge and assumptions, a dialogue cannot be collaboratively
rational. (p. 100)

In the current study, epistemic motivation affected the range of information shared
among participants as well as the discriminating behavior between conforming and
refuting information. The simulation results concerning the effect of epistemic
motivation highlight some aspects of authentic dialogue that have not received sufficient
attention in the literature.
First, the measure of success showed that high epistemic motivation is associated
with lower probability to reach consensus and a longer time to reach consensus compared
to low epistemic motivation. This result is consistent with the theoretical expectation of
the effect of high epistemic motivation. On one hand, a high probability to reach
consensus among participants with low epistemic motivation reflects the agent behavior

147
that is inclined to negotiation focus. When the agent behavior followed the rule of
negotiation focus, the success rate was high. On the other hand, it might seem natural
that, when participants share a wide range of information, it takes a longer time to reach
consensus. In the real world, there could be various causes that prolong the time to reach
consensus when participants share a wide range of information. For example, the
behavior may gradually invoke new, related policy issues and enlarge the view of
participants, which could lead to a perceptual need for more information (De Dreu et al.,
2008). The behavior may also require more extensive and intensive intellectual
information-processing that consumes psychological efforts. Finally, high demand of
information without discerning the importance of a piece of information may have the
negative effect of delaying the decision-making, which was demonstrated by the
computational model. A longer time taken to reach consensus among participants with
high epistemic motivation in the current simulation was partly due to the built-in
ignorance of importance of the information. In the Water Forum case, participants
overcame this problem by employing experts who helped them share and learn relevant
information to make their decision.
Second, the measure of learning indicates that participants with high epistemic
motivation obtained a higher volume of new information and made fewer changes to their
policy opinions. The former result is quite straightforward, while the latter result needs
some explanation. The latter result is not necessarily a negative sign. As long as a group
of participants reaches consensus, less change of a policy opinion means that participants

148
could remain loyal to their constituents, which is an important factor for successful
collaborative governance.
Finally, the measure of responsiveness showed that when participants
demonstrated high epistemic information-processing behaviors, they made a more
globally and locally responsive decision than did participants with low epistemic
information-processing behaviors. This result confirms the normative claim that
authentic dialogue can result in a win-win solution. According to this result, however,
the winners are not exclusively the participants; the public, which is actually an aggregate
of specific stakeholder groups, can also benefit from the result.
The Water Forum case appropriately illustrates the importance of epistemic
motivation in the deliberation process. The deliberation process designed by the Water
Forum has been characterized by its education phase, in which “stakeholders met to
review information regarding water-supply issues and the Lower American River, and
educated one another about their perspectives on the issues” (Innes & Booher, 2010, p.
46). As previously mentioned, the purpose of the education phase was to build a
common cognitive ground regarding the policy issue among the stakeholder groups
(Connick, 2006; Klimoski & Mohammed, 1994). This education process was also
facilitated by hiring experts in pertinent areas. From the simulation perspective, all
deliberation processes helped participants come to possess a similar set of information,
which represented a common cognitive ground among them (Klimoski & Mohammed,
1994). Sometimes, the process even entailed a significant change of perspective that was
simulated in the computational model by a change of weight and the binary form of

149
policy opinion. For example, from a legal report during the process, the purveyors
realized that their right to use water could be restricted. This information changed the
perspective of the purveyors, which was among the main reasons why the purveyors
remained in the Forum. The Water Forum also demonstrated a prolonged time to reach
consensus due to stakeholders‟ high epistemic information-processing behaviors. Not
only did it take about six years for the Forum to wrap up at the macro level, but also at
the micro level, staff often prolonged the process to ensure that all participants had a clear
understanding what was going on (Innes & Booher, 2010).
The Water Forum also showed another important point that the deliberation
process, characterized by high epistemic motivation, could reconcile global and local
responsiveness. The success of the Water Forum meant that each stakeholder group
received what it wanted and the residents of the area generally benefited from the
orchestrated water plan. Compared to the BAASC case, in which only the least common
denominator was identified (Innes & Booher, 2010), the Water Forum ensured global
responsiveness of the decision while not sacrificing individual stakeholder groups‟
interests by developing a common cognitive ground in which their interests might be
changed or incorporated in a creative way.
The following propositions summarize the results of the simulation and the
discussion.
Proposition 2-1. As participants in the deliberation process within a collaborative
governance system are driven by high epistemic motivations, they are less likely to reach
consensus and do so more slowly.

150
Proposition 2-2. As participants in the deliberation process within a collaborative
governance system are driven by high epistemic motivation, they are more likely to be
responsive to both the general public and specific stakeholder groups.
Global and local responsiveness. It has been suspected that there is a dilemma
between self-interest and collective interest in group decision-making processes
(Thompson et al., 2010). In particular, the literature of collaborative governance has
focused on this dilemma when investigating the issue of responsiveness (Vigoda, 2002),
accountability (Bryson et al., 2006), recognition of interdependence (Thomson & Perry,
2006), and social learning process (Daniels & Walker, 1996; Muro & Jeffrey, 2008).
Although many studies have taken for granted that there is a dilemma between self-
interest and collective interest at the individual level, little is available concerning
whether the dilemma is inevitable or how strongly it affects the deliberation process at
the group level.
The results of the current simulation lead this researcher to some theoretical
propositions on the effect of motivated information-processing behavior at the individual
level concerning the responsiveness of the collective decision. Here, five propositions
regarding global and local responsiveness of the collective decision are suggested. The
effects of motivated information-processing on global and local responsiveness are
discussed respectively, followed by the discussion of the trade-off between global and
local responsiveness. While the Propositions 1 and 2 focus exclusively on either social or
epistemic motivation, here, both of the motivated information-processing behaviors are
considered.

151
First, the simulation results showed a negative effect of pro-self motivation on the
global responsiveness of a collective decision. Although the effect was not statistically
significant, pro-self motivation was associated with lower global responsiveness of the
final decision and degeneration of global responsiveness through the deliberations (see
Table 7-4). From this result, it can be said that, when participants remain pro-self, a
deliberation process among them may add no value to the collective decision. Likewise,
the simulation results yielded a negative effect of low epistemic motivation on the global
responsiveness of the collective decision. Participants were quick to reach consensus
with the least mutual learning. Further, they not only came to a decision of low global
responsiveness but also demonstrated degeneration of global responsiveness (see Table 7-
4).
The negative effect of low epistemic motivation on global responsiveness also
highlights the need for the balance between exploration and exploitation in learning.
According to March (1991), the speed of learning is associated to the width of learning
and eventually the quality of the decision. The implication of March‟s simulation results
is that when people are too sensitive to outside information, in terms of the speed and
degree of accepting it, they tend to lose the opportunity to make a full use of new and
critical information before making a decision. This lost opportunity results in a
degeneration of the quality of the decision. In other words, there could be a trade-off
between the speed of learning and the quality of the decision from learning. The results
of the simulation, concerning epistemic motivation, are equivalent to the theoretical

152
argument developed by March (1991). The results and discussion lead us to the following
proposition:
Proposition 3. The deliberation process may deteriorate the responsiveness of the
collective decision to the public when participants are self-oriented, narrow-informed, or
consensus-oriented in information-processing.
Second, the simulation results showed a positive effect of the deliberation process
in general on local responsiveness of the collective decision. Unlike the results of global
responsiveness, the results of local responsiveness demonstrate that, regardless of the
type of social and epistemic motivation, the deliberation process significantly improved
the local responsiveness of the collective decision (see Table 7-6). Therefore, it can be
said that the simulation demonstrates a process in which participants form a collectively
superior and mutually satisfactory solution through deliberation. To repeat, participants
started with a policy proposal, which was not highly responsive to all individuals.
However, the simulation found that, through the deliberation process, they could reach a
decision that would be more responsive to all participants, on average. These results lead
to the following proposition.
Proposition 4. The deliberation process will, in general, improve the average
responsiveness of the consensus-oriented decision to specific stakeholder groups.
The trade-off between global and local responsiveness and the dilemma
participants in a collaborative system face have been acknowledged in many ways. For
example, Innes and Booher (2010) referred to the dilemma of collaboration as meeting
one‟s self-interest and attaining the common good. Thomson and Perry (2006, p. 26)

153
used the term dual identity, which refers to maintaining participants‟ own distinct
identities and organizational authority separate from (though simultaneously with) the
collaborative identity. At the group level, Toma and Butera (2009) pointed to the
contrast between cooperation and competition among group members, which affects the
quality of the group decision. Lax and Sebenius (1986) framed the issue as balancing the
“twin tasks – creating value and claiming value.”
The dual identity perspective and the twin tasks perspective refresh an often
underestimated point in collaboration; Being selfish is not something to be eradicated.
Instead, selfishness is an integral part of the identity of participants and should be
appropriately incorporated in the theoretical framework of collaborative governance. I n
the absence of value-claiming, there is no guarantee that the collectively created value
appropriately serves, as shown in low local responsiveness of the decision among pro-
social participants. Participants need to endeavor to meet the needs of each interest
(Innes and Booher, 2010), as well as to act on behalf of the public.
The Water Forum case illustrates that a successful collaborative governance
system begins with correctly identifying each stakeholder‟s interest. For example, the
Forum was organized into caucuses and four teams (Surface Water Team, Ground Water
Team, Demand Conservation Team, and Habitat Management Team) to ensure that all
stakeholders‟ unique interests were fully expressed and discussed in sufficiently small
and relevant sub-forums. In addition, the education phase provided stakeholders with the
opportunity to incorporate their interests in the big picture. In the BAASC case,
stakeholders reached consensus on the Compact because the document at least reflected

154
their separate interests. One problem in the Westlake case was that the interests of the
LAUSD – building new schools as soon as possible to meet the educational demand from
the Los Angeles communities – were not well reflected in the deliberation process, which
made the LAUSD reluctant to be fully engaged. Participants in the Water Forum were
not an exception concerning their interest of being considered in the Forum, as in the case
of environmentalists (Connick, 2006).
The need to respect self-interest is also related to the accountability issue.
Researchers acknowledge that there is pressure regarding accountability of each
participant and the forum as a whole (Bryson et al., 2006; Lerner & Tetlock, 1999;
Provan & Milward, 2001). On one hand, Lerner and Tetlock (1999) argued that process
accountability works for participants to be engaged in appropriate information sharing
because they feel pressure from others who observe and evaluate whether group members
properly process information prior to making an important decision. This kind of process
accountability is imposed on the group as a whole. Unlike with groups within an
organization which possess enough distributed expertise regarding their projects, with
groups in collaborative governance systems, process accountability as a whole is often
ensured by participation of professionals who retain professional and technical
knowledge about the pertinent social issue, as shown in the case of the Water Forum and
the Westlake. In the former, a professional facilitator from the California State
University at Sacramento was hired. In the latter, a nonprofit organization, New Schools
Better Neighborhoods dedicated to consulting and developing joint-use projects, was
hired.

155
The concept of process accountability, however, is not as clear in the context of
collaborative governance as in the group settings where group members are more
cohesive and the collective goal is clearer. In the context of collaborative governance,
the pressure toward process accountability could come both from the specific stakeholder
groups and from the public, with emphasis on the former.
25
Thus, process accountability
is often ascribed to each representative so that the representative should be accountable to
his or her own constituents for what is delegated. In the Water Forum case, stakeholder
groups were asked to name their representatives, who were usually a high level staff in
the organizations, and to obtain board approval (Innes & Booher, 2010). For the
accountability concern, the representatives even reserved the right to leave the forum at
any time.
Representatives in a collaborative governance system, however, experience the
notion of dual accountability. In fact, an important issue in the Water Forum was to
ensure that the forum participants communicated with their constituents regarding
decisions made in the forum: “Once the Forum had a draft agreement, stakeholders
brought it to the boards of the groups they represented. This was a critical step to assure
that the agencies and advocacy groups would be prepared to implement it” (Innes &
Booher, 2010, p. 48). Representatives also had to persuade their own constituents to
understand and accept the forum decision. In the case of the Westlake project,
participants were concerned about persuading their organizations and outside supporters,

25
The visibility of the public is an issue in collaborative governance. The role of a participant in the
Westlake Collaborative was to organize the interest of Latino communities in the area so that it became
visible.

156
including political figures and funders, who had diverse goals, interests, and time
schedules not necessarily consistent with the collaborative forum‟s direction.
The results and the discussion lead to the following proposition.
Proposition 5-1. Collaborative governance will have a better opportunity to
succeed when the accountability pressure on participants is properly recognized by one
another in the deliberation process.
A relevant question was what would be the consequence of behavior sensitivity or
insensitivity to the accountability pressure. In the computational model, pro-self
motivation reflected the former and pro-social motivation reflected the latter. The
simulation result showed that when participants were motivated in a pro-self manner, that
is, were sensitive to the pressure from their own constituents, they could ensure the
responsiveness of the collective decision to their own constituents with the sacrifice of
responsiveness to the public and the probability to reach consensus. Toma and Butera
(2009) expected similar results based on their empirical study on group decision-making
and argued that
competition leads group members to engage in strategic behavior more than
cooperation and that this should be reflected in the withholding of information
and the unwillingness to put into question initial solutions, with the result of
reducing the quality of group decisions. (p. 794).

Participants could also be more accountable to their own constituents if they changed less
of their initial opinion delegated to claim at the forum.
In contrast, participants with pro-social motivation, succeeded in improving
global responsiveness with the sacrifice of local responsiveness. However, it is important

157
to note that these individuals still improved the local responsiveness of the decision to
some degree through deliberation. In other words, a collaborative forum composed of
participants with pro-social motivation may be somewhat less responsive to each specific
interest group than would a forum with pro-self participants. Nevertheless, this does not
mean that the former fails to improve local responsiveness of the decision through
deliberation: It actually does improve. In the case of the Westlake project, the
compromised plan would seem less responsive to each stakeholder group. However,
many participants agreed that the final plan served the urgent need of the neighborhood
well: They needed both affordable housing and a new school, not to mention green
spaces and playgrounds. The final plan was also locally responsive to an acceptable
degree.
From the result and the discussion, the following proposition is suggested:
Proposition 5-2. The accountability pressure on participants to their own
constituents will result in pro-self information-processing, reducing the probability to
reach consensus, and sacrificing the responsiveness to the public for the sake of the
responsiveness to their own constituents.
It should be noted that the computational model does not imply that participants
should be altruistic to reach consensus or enhance and balance the local and global
responsiveness. Pro-social participants in the simulation were still agents that pursued
satisfaction with a policy proposal of their own position. By design, global and even
local responsiveness are not of their explicit concern. This is because they are not bound
to argue for their own constituents‟ original policy opinion; they started with it but were

158
not bound to it; therefore, they learned from others and gradually changed their positions.
Pro-social participants succeeded in improving global responsiveness of their decision
without intention. However, they failed to reconcile between global and local
responsiveness. Consequently, another dimension of information-processing motivation
is warranted to solve the dilemma: epistemic motivation.
The simulation showed an instructive result regarding the effect of epistemic
motivation. Participants with high epistemic motivation succeeded in reconciling global
and local responsiveness such that it enhanced both simultaneously through deliberation.
In contrast, the deliberation process among participants with low epistemic motivation
resulted in degenerated global responsiveness and less improvement of local
responsiveness than the deliberation process among their counterparts. This result
supports the normative argument for social learning (Bull, Petts, & Evans, 2008; van
Buuren, 2009), in which stakeholder groups are engaged in a comprehensive collective
learning via information sharing and collective experiences, as shown in the education
phase of the Water Forum case.
The result of the simulation implies that a quicker decision will be made when
participants are biased to negotiation focus (Brodbeck et al., 2007) or consensus-oriented
norm (Postemes et al., 2001), as characterized by low epistemic motivation in the current
study. As shown in the case of BAASC, the process may seem smooth and conflict-free
(Innes & Booher, 2010). As discussed above regarding exploration and exploitation
(March, 1991), however, a quick decision made by participants with low epistemic
motivation does not guarantee that they fully use information available to correctly

159
recognize the need of general constituents, which would result in lower quality decisions.
In short, it is necessary in a deliberation process that participants have sufficiently high
motivation to make a decision with a wide range of information that tells more about
constituents‟ needs. The implication of the simulation results is that if participants are
engaged in active collective learning, the two conflicting responsiveness and the request
for accountability can be reconciled in a creative way. The results and the discussion
lead to the following proposition:
Proposition 5-3. When participants are engaged in high epistemic information-
processing, they will reach consensus in a way that reconciles a trade-off between the
responsiveness to the public and the responsiveness to participants‟ own constituents.
Authentic dialogue. Authentic dialogue is regarded as an integral part in
collaborative processes (e.g., Roberts, 2002). Symmetric information-processing,
discussed in the social psychology literature, can be bridged to the collaborative
governance literature via the concept of authentic dialogue. The researcher assumes that
the motivated information-processing model enriches the discussion of authentic dialogue.
First, the interpersonal aspect of authentic dialogue can be clarified. The motivated
information-processing model provides a framework with which to analyze interpersonal
dialogue. Second, the discussion about authentic dialogue needs to go further than
normative claims. The motivated information-processing model also provides a
framework with which to describe an individual‟s information-processing behavior, and
what kind of behavior should be encouraged. What is suggested in this section is a
contingency view of the virtue of authentic dialogue. In other words, an overarching

160
proposition is that, based on the simulation results, authentic dialogue is not always the
best for every aspect of the performance of group decision-making process.
The importance of dialogue has been emphasized in the collaboration literature.
Specifically, the recent theoretical interest on dialogue in the collaborative governance
literature stems partly from the classical concept of genuine dialogue and the “I-thou”
relationship developed by Buber (1958), the argument model developed by Toulmin
(1958) and his successor (e.g., Dunn, 1981), the work of Habermas (1981), who
developed the theory of communicative rationality, and from deliberative democracy
(Dryzek, 1990). The influence of Giddens who invocated the importance of public
forums in societal decision-making is also acknowledged (Bryson & Crosby, 1993;
Connick & Innes, 2003; Giddens, 1984). More recently, Yankelovich (1999) provided a
practical approach to authentic dialogue. Through the influence of the earlier thoughts,
dialogue, or deliberation, has emerged as the major issue in collaborative governance
(Dryzek, 2000; Fung & Wright, 2003; Healey, 2006; Innes & Booher, 2003; Roberts,
2002; Vigoda, 2002). Yankelovich (1999) emphasized dialogue as a process of seeking
mutual understanding, which makes it distinct from debate, conversation, or discussion,
all of which assume argument for each group‟s position (Fisher & Ury, 1981). Similarly,
Innes and Booher (2010) stated the uniqueness of authentic dialogue as following:
Authentic dialogue is not just talk, and it does not emerge naturally in large
groups, at least in industrialized countries. The norm is a formalized, almost ritual,
“debate” …, where interests are concealed rather than shared and where
compromise, logrolling, and tradeoffs behind the scenes are the modes of
reaching agreement (p. 97). … [P]articipants internalize the views of others to
enhance their mutual understanding. (p. 119)

161
From the information-processing perspective, authentic dialogue can be translated into
symmetric information-processing, where participants are free from the preference bias,
information sampling bias, and negotiation bias (Brodbeck et al., 2007). Brodbeck et al.
(2007) specified two conditions for groups to outperform individuals and voting schemes
in decision quality: the asymmetric information distribution among participants and the
symmetric information-processing by participants. Under the condition of symmetric
information distribution, the head of one may outperform or perform as good as the heads
of many because the heads are not mutually complementary (Fraidin, 2004; Maier, 1970).
Under the condition of asymmetric information-processing, the heads of many, even
when they are mutually complementary, may not be fully integrated and used for
collective intelligence. Debate or discussion, in the absence of mutual understanding, can
be featured as the asymmetric information-processing among participants. Authentic
dialogue, in contrast, entails the symmetric information-processing as its core processing
aspect. The education phase and negotiation phase of the Water Forum case illustrate a
good example of purposeful effort to encourage authentic dialogue among participants.
In short, the value of collaborative governance that makes it more attractive than majority
voting is in dialogue the system incorporates.
Unfortunately, case studies have demonstrated that authentic dialogue occurs less
often (Innes & Booher, 2010). Even successful collaborative systems in the end suffered
a terrible beginning in terms of the quality of dialogue; such was the case of the Westlake
Collaborative and the BAASC. From the information-processing perspective, this
finding is not surprising, even considering the muddy political nature of the deliberation

162
process. De Dreu et al. (2008) revitalized Kelley and Stahelski‟s (1970) research on
social dilemma that found the negative contagion effect of pro-self behavior in group
decision-making. That is, when a pro-social group member works with another pro-
social member, he or she continues to be cooperative. In contrast, when he or she faces a
pro-self member, he or she switches to a noncooperation stance. Here, one can simply be
reminded of the tit-for-tat strategy (Axelrod, 1981). What is worse is that pro-self
members consistently maintain their non-cooperative attitude and are not sensitive to
their counterpart‟s attitude. The implication of this finding is clear: Pro-self attitude in a
group will easily lead to the other (pro-social) members turn to pro-self attitude, which
can result in a serious stymie of collaborative process.
26

However, the findings have two limitations to apply to collaborative governance.
First, unlike participants in the experiments, participants in a collaborative governance
system may have no option but to cooperate, or their best strategy is to collaborate, as
shown in the case of the Water Forum (Connick, 2006). Second, successful collaborative
governance systems usually benefits from facilitative leadership (Ansell & Gash, 2008;
Bryson et al., 2006). For those reasons, a collaborative forum may have more
opportunities to develop a group norm that facilitates pro-social attitude.
As for epistemic motivation, information from minorities has been emphasized.
From the information-processing perspective, De Dreu et al. (2008) found that even when
minority factions in a deliberation group actually fail to convert the majority opinion,
their existence proves to “stimulate divergent thinking and innovation, reduce

26
This is the point where the role of the facilitative leadership appears critical. This issue is discussed later.

163
confirmatory information search, reduce group polarization, prevent groupthink, and
reduce conformity” (p. 28). This finding highlights the importance of epistemic
motivation in collaborative forums. Contrary to practical concerns that minority opinion
is usually ignored or at least underrepresented, the finding shows that group information-
processing unfolds in a way that allows the minority opinion to affect the group decision
in various positive aspects. More actively, recent social learning literature acknowledges
the importance of diversity of knowledge sources preexisting in collaborative forums.
For example, Feldman and Khademian (2007), in discussing the concept of inclusive
knowledge, emphasized the effort to invite stakeholders with various sources of
knowledge as an important factor for successful collaborative governance.
In summary, the literature of collaborative governance that has focused on
authentic dialogue tends to emphasize pro-social and high epistemic information-
processing behavior as the crucial facets of authentic dialogue at the personal and
interpersonal level. “Facilitation, small group work, egalitarian atmosphere, repeated
meetings, opportunities to influence the process, open communication, diverse
participation, unrestrained thinking, multiple sources of knowledge” (Muro & Jeffrey,
2008, p. 332) are all features of successful collaborative forums.
One theoretical gap in collaborative governance literature is that few explicitly
consider the negative sides of dialogue. The current study measured the performance of
virtual forums of participants with different types of information-processing behavior
from diverse aspects through simulation. The simulation results show that the effect of a
specific information-processing behavior on the performance of deliberation is contingent

164
on how performance is measured. As summarized in Table 9-1, each type of information-
processing behavior, except for the pro-self-low epistemic motivation, demonstrated its
relative home domain in the level of performance.

Table 9-1
Contingency of the Performance of Motivated Information-Processing on Measurement
Type of motivation

pro-self-low
epistemic
pro-self-high
epistemic
pro-social-low
epistemic
pro-social-high
epistemic
success rate medium low low high high
Time medium low low high medium high
final global
responsiveness
Low medium high high medium low
final local
responsiveness
medium high high low medium low
newly obtained
data points
Low high medium low medium high
opinion not
changed
Low high medium high medium low

Concerning pro-social-high epistemic motivation groups, which would best
reflect the condition of authentic dialogue, the simulation results demonstrated that
deliberation processes characterized by authentic dialogue did not always perform best
among the four types of motivated information-processing behavior. First, regarding
success in reaching consensus, pro-social-high epistemic motivation groups performed no
better than pro-social-low epistemic groups. The result indicates that for a quick
consensus, pro-social and low epistemic information-processing behavior is helpful.
Second, pro-social and low epistemic motivation groups initiated more global
responsiveness than self-interest. These groups actually recorded the best global

165
responsiveness among the four types of motivation groups. On the other hand,
improvement of global responsiveness via deliberation was highest among pro-self-high
epistemic motivation groups, followed by pro-social-high epistemic motivation groups.
Although the results about global responsiveness should be interpreted with caution
because of low statistical significance, the relational pattern demonstrated that pro-social-
high epistemic motivation is not the best for effective global responsiveness. Third,
regarding local responsiveness, pro-self-high epistemic motivation groups performed best,
showing that pro-self motivation is highly associated with local responsiveness and high
epistemic motivation had an influence, to a lesser degree, but still positively. Pro-social-
high epistemic motivation groups ranked the third regarding local responsiveness.
Finally, regarding the degree of learning, although pro-self-high epistemic
motivation groups are not likely to learn the most, the group members eventually learned
about the greatest number of data points because of the longer time taken to reach
consensus. What is interesting is that, despite the result that they obtained the most new
data points, they changed the least of their policy opinion. The result implies that
participants with this type of information-processing behavior could actually remain
accountable to their own constituents. Given that constituents did not change their
opinion during the deliberation process, less change in their representative‟s policy
opinion during the deliberation process indicates that their representative was loyal to
what it was originally delegated. Pro-social-high epistemic motivation groups were
ranked second in obtaining new data points, and third in their members‟ keeping their

166
original opinion, which could be risky when the constituents recede their support to the
representative.
Overall, instructively enough, except for a relatively low success rate and long
time to reach consensus, information-processing behavior with pro-self-high epistemic
motivation performed generally well. For consensus formation purposes, information-
processing behavior with pro-social-low epistemic motivation was best, but the
performance regarding improvement in global responsiveness and local responsiveness
was moderate. Information-processing behavior with pro-social-high epistemic
motivation performed moderate in all the dimensions of performance. Finally,
information-processing behavior with pro-self-low epistemic motivation performed worst.
These results emphasize the importance of claiming value as well as creating value. They
also emphasize the need for high epistemic motivation.
From the results and the discussion, the following propositions are developed:
Proposition 6-1. The generally discussed authentic dialogue is not always an ideal
for collaborative governance. Attitudes that combine pro-social and high epistemic
information-processing do not produce the best performance in reaching consensus, being
responsive to the general public and specific stakeholders, and learning.
Proposition 6-2. Attitudes that comprise authentic dialogue produce the moderate
and balanced performance between success and responsiveness.
Proposition 6-3. Except for the low likelihood of reaching consensus and a long
time to reach consensus, successful groups with pro-self and high epistemic motivation
will be more responsive to the public and to specific stakeholders simultaneously.

167
The Water Forum case can be regarded as a successful collaborative process. The
success was so exceptional that similar collaborative processes began in the Sacramento
area for transportation and land use (Connick, 2006). Although participants in the Forum
were engaged in deliberation close to authentic dialogue, it might not be necessarily pro-
social-high epistemic information-processing. The organization of the Forum implies
that the facilitator assumed that participants would be engaged in pro-self information-
processing, meaning that they will protect their interests in the deliberation process. As
discussed above, the participant attitude is reasonable from the accountability perspective
and even desirable from the information-processing perspective; their self-orientation
was the basis on which to build a shared mental model. What the simulation result
highlights is that pro-self information-processing should be accompanied by high
epistemic motivation. Driven by the facilitator and accelerated by growing mutual trust,
participants spent significant time sharing a wide range of information and developing a
collective idea on the policy issue. Although it took six years and more than $10 million
to reach a final decision, the collaboration process generated both such a globally and
locally responsive decision. Pro-social attitude was not a resource available at the
beginning, but a cultivated resource at the end. In short, the Water Forum case well
reflects the propositions developed in this section.
Propositions Regarding Group-Level Variables
In the previous section, the effect of the type of motivation in information-
processing on group deliberation process was discussed. In this section, the effect of
group-level variables on group deliberation process is discussed.

168
It is not easy for public leaders to change participants‟ motivations in information
sharing. Sometimes, it reflects a participant‟s personal disposition (Fiske, 1992) and
sometimes it reflects how much pressure the participant feels from his or her constituents.
What public leaders could do is to design the collaborative forum to encourage desirable
information-processing behavior and countervail the negative effect of undesirable
information-processing behavior. Therefore, the purpose of this section is to discuss the
theoretical relationship between group level variables and the performance of the
deliberation process. Practical implications of the theoretical discussion are also
suggested at pertinent places.
Interdependence. Interdependence of interests among stakeholders is the crucial
feature of collaborative governance; in the absence of conflict of interests among
stakeholders or in the absence of a path to resolve conflict, collaborative governance does
not stand (Muro & Jeffrey, 2008; Thomson & Perry, 2006). This study investigated the
effect of the degree of interdependence of interests among participants in virtual
collaborative forums. The simulation results in the current study revealed that a high
degree of interdependence was associated with high success rate, quick consensus, high
global responsiveness, low local responsiveness, and low degree of learning. The results
also describe a scenario in which participants with high interdependent interests gather,
share information, and adjust their policy opinions on a few dimensions. In doing so,
they quickly and rather easily reached consensus; the adjustment of their policy opinions
yielded good results to the general public, but not so good results to specific stakeholder
groups. Two questions can be discussed regarding the theoretical implication of the

169
results. First, how could participants with higher interdependence reach consensus more
often and more quickly? Second, what does the trade-off between global and local
responsiveness mean?
Regarding the question of how participants with higher interdependence reach
consensus more often and quicker, it should be clear that the simulation design highlights
two aspects of interdependence of interests among stakeholders. One is that a higher
degree of interdependence leads to a higher degree of conflict of interests among
stakeholders. Overlap of interest areas does not necessarily mean conflict of interest
because two participants could have the same policy opinion in the overlapping area(s).
However, with the same probability to have either 0 or 1 in their opinions, higher overlap
of interest with other participants generally results in conflict of interests with more
participants. The other aspect is that a higher degree of interdependence allows
flexibility for participants. For example, if a participant comes to the forum with interest
in one policy dimension, he or she should obtain a satisfactory collective proposal at this
dimension. This will not be easy because there will be counterparts who are in the same
inflexible situation. If a participant comes to the forum with interest in five policy
dimensions, in contrast, he or she can get more from a couple of dimensions while giving
up some from another couple of dimensions through negotiation.
27
For example, in the
Water Forum, when there was a conflict of interest and it was not easy to find a mutually
beneficial solution specific to that conflict area, they discussed whether one participant

27
Of course, the participant could be designed to have an urgent need to be satisfied in all five dimensions.
In terms of modeling this attitude, it is logically the same as that of participants with one dimension of
interest. The point is whether they have an alternative enough to be flexible in interest adjustment or not.

170
was willing to yield at that area when his or her interest there was not critical (Connick,
2006). This type of adjustment is possible only when the yielding participant‟s interest
can be satisfied in another area. This adjustment process may not exclude some political
pressure, which often works in collaboration cases such as with the Westlake project, in
which external political support for a collaborative solution played a central role for
participants to remain at the table. However, what is important is whether participants
can be flexible enough to yield at an area of interest and to be compensated at another
area of interest. Connick (2006) reports that “what they were able to agree on, …, was
that there were a number of other issues that they needed to address and on which they
anticipated they would be able to come to agreement” (p. 39).
While the degree of interdependence is discussed in a static view, it can be quite
dynamic in reality, specifically, concerning successful collaborative forums in which
participants gradually articulate their diverse, true interests. The literature on
collaborative governance emphasizes not only the importance of interdependence for a
collaborative forum to be launched (Chen & Graddy, 2005; Thomson & Perry, 2006;
Wood & Gray, 1991), but also the social learning process for participants to recognize a
broader range of interdependence of their interests (Connick & Innes, 2003; Innes &
Booher, 2003). Positive effects of a high degree of interdependence can be summarized
to three steps to build a positive relationship among stakeholders. First, a high degree of
interdependence encourages stakeholders with low trust to bond (Ansell & Gash, 2008).
As shown in the case of the Water Forum, stakeholder groups with no initial
collaborative experience and trust gathered together because they realized that there was

171
no way but to collaborate due to the extremely high degree of interdependence (Connick,
2006). Driven by this bond, a high degree of interdependence keeps the stakeholders in
the forum (Connick & Innes, 2003), which makes a room for them to build trust (Ansell
& Gash, 2008; Bull et al., 2008; Ostrom, 1998; Thomson & Perry, 2006). In particular,
remaining for a while in the forum is important to overcome the negative contagion effect
between pro-self and pro-social actors and strategic information-processing behavior
found in social dilemma research and motivated information-processing research (De
Dreu et al., 2008; Kelley & Stahelski, 1970; Toma & Butera, 2009). Research
demonstrates that pro-social behavior of group members in information sharing and
decision-making is vulnerable to their counterparts‟ pro-self or opportunistic behavior.
The collaborative governance literature also found the behavioral tendency in the form of
“I will if you will” mentality (Thomson & Perry, 2006, p. 27). Here, to overcome the
deadlock, the interaction should be continued (Axelrod, 1997). As trust accumulates,
participants start sharing true information, leading to the recognition of higher
interdependence among their interests (Feldman & Khademian, 2007; van Buuren, 2009).
In fact, recent reviews of case studies report a typical lifecycle of collaborative
governance, which is the beginning of the process with low trust and bad recognition of
interdependence that gradually moves into high trust and recognition of a broader range
of interdependence (Ansell & Gash, 2008; Innes & Booher, 2010).
In summary, the recognition of a high degree of interdependence accompanied by
the trust-building process is the critical point in collaborative process. Like the design of
the computational model, a high degree of interdependence does not exclusively mean

172
high level of conflict of interests among stakeholders in the real world. Rather, it is the
process, driven by conflict, of realizing a bigger picture about where the stakeholders‟
interests could be adjusted in a mutually beneficial way. A quicker formation of
consensus can be understood in this context. The simulation results imply that groups
with a high degree of interdependence among their participants benefit from two
mechanisms of interest adjustment at the individual level: Each participant adjusted the
weight of his or her opinion in each dimension according to the incoming information
and, as a result, adjusted the relative weights of his or her opinion across all focal
dimensions, both of which facilitated formation of consensus. Participants with a low
degree of interdependence lacked the second mechanism of adjustment. From the
simulation results and the discussion above, the following proposition is developed.
28

Proposition 7-1. Participants with a high degree of interdependence of interests
are more likely to reach consensus as long as they remain flexible about in which policy
area they want their desire to be realized.
The computational model showed the positive effect of a high degree of
interdependence on global responsiveness and the negative effect on local responsiveness.
The trade-off between global and local responsiveness in relation to interdependence
reflects three dynamics. First, a high degree of interdependence facilitates (or force in
some sense) participants in a collaborative forum to discuss and modify a wide range of
policy dimensions, which leads to high global responsiveness of the decision due to

28
This study did not consider increasing need for intellectual power in accordance with increasing
interdependence. So the result should be understood under the condition that the need for intellectual
power is controlled.

173
opinion adjustment based on more information sharing. Second, it is not easy to tailor
the collective decision to fit all specific stakeholder groups‟ desires. The results reflect
the difficulty of the task to reconcile among interdependent, incompatible interests.
Finally, participants with a high degree of interdependence, when combined with low
epistemic motivation, may reach consensus too quickly before using all information
available. Therefore, even though there could be another way to develop a better
proposal for both the general public and specific stakeholder groups, participants accept
the current proposal as satisfactory. In short, whatever their information-processing
motivation is, a high degree of interdependence can lead the deliberation process similar
to one where participants are motivated in a pro-social way.
The simulation results and the discussion lead to the following proposition.
Proposition 7-2. Participants with a high degree of interdependence of interests
are less likely to serve for each of participants‟ specific constituents than those with a low
degree of interdependence.
Majority vs. minority. Collaborative governance is often challenged by
alternative ways of decision-making through which conflicting parts that can make a
resolution without costly process (Muro & Jeffrey, 2008; Przeworski, 2009). Thomson
and Perry (2006) point out that collaboration should not be made for collaboration‟s sake
because it is costly. To investigate which condition the collaborative process can add the
most value to collective decision-making, the computational model manipulated agenda
setting at the start of deliberation. In particular, the focus was on the difference between
starting with the majority opinion and starting with the opposite opinion.

174
Majority rule may be a better decision-making rule than consensus when
information is symmetrically distributed among constituents or participants or if the
information-processing among group members is expected to be asymmetric (cf.,
Brodbeck et al., 2007). Innes and Booher (2010) echoed the proposition when they said
that collaborative planning process might not make a point when the conditions are met:
The issues are well understood, and there is considerable consensus around the solutions.
Those conditions correspond to symmetric information distribution.
Nevertheless, the relevance of collaborative governance in policy-making is not to
be underestimated in recent public decision-making environments. First, majority rule is
not an apt decision-making rule when stable consensus among stakeholders should be
formed. Innes and Booher (2010) stated:
The model of decision-making by majority rule not only may not take into
account key perspectives, it results in winners and losers, often creating
enemies… Decisions can be unstable, not only because 51 percent can easily be
shifted to 49 percent, but also because players are not bound together through
reciprocity. (p. 94)

What is paradoxical in this claim is that the easy conversion from 51% to 49% is exactly
the reason why majority voting is preferred: Majority voting is claimed to be the best and
most efficient way to exactly reflect the subtle change (Downs, 1961; Przeworski, 2009).
In addition, Innes and Booher did not consider the adaptation of the actors‟ practice using
a majority rule through a long-term repeated-game situation, such as the logrolling
behavior among stakeholders not to make any stakeholder a constant loser or enemy
through majority voting.

175
The key to assess the utility of decision-making rules is reciprocity (Ostrom, 1998;
Thomson & Perry, 2006; Wood & Gray, 1991) and the recent wicked policy environment
(Weber & Khademian, 2008) in which reciprocity and collaboration among stakeholders
becomes even more important. Reciprocity does not separate between winners and losers
among constituents: All could be winners with a creative solution or none could be.
Moreover, the complex web of social problems not only deepens the degree of reciprocity
among stakeholders but also increases uncertainty of knowledge developed in the past.
Through the investigation of group research, Kerr and Tindale (2004) concluded that
majority processes are a second-best solution compared to deliberation when information
based on which to make a decision is not accurate. Majority rule is apt to simple and
stable environments in which information asymmetries are of no significant problem.
The empirical findings from group information-processing and decision-making imply
that the recent wicked policy problems require more endeavoring decision-making
schemes than simple majority rule.
Why, then, do researchers not integrating these two ways? In other words, what
would it be like to find the majority opinion and take it as a starting alternative? The
simulation results of the computational model of collaborative governance illustrate that
the story is not as easygoing as expected. On one hand, the computer simulation results
showed the superiority of the majority opinion as a starting point to the random or the
opposite opinion on both global and local responsiveness. On the other hand, because the
global and the local responsiveness of the final decisions were not different across three
rules of starting alternatives, the improvement of global and local responsiveness, via

176
deliberation, was highest when participants started deliberation with the opposite opinion.
Simply put, starting deliberation with the majority opinion would not result in a better
decision because the majority opinion already reflects sufficiently responsive policy,
making the following deliberation process redundant or even misleading. On the
contrary, starting deliberation with the opposite opinion would result in a better decision
simply because there is a larger room for improvement.
It would be hasty to conclude from the results that a deliberation process is
inferior to majority rule. As discussed above, majority rule has its own significant
limitations in the recent policy environment. More importantly, a relevant theoretical
issue is whether or how we can identify the majority opinion. Note, the majority opinion
as identified by the computational model is an ideal one, which was possible only in the
virtual world. The opinion exactly recognized the sub-dimensions of a policy, and the
voting scheme exactly reflected the participants‟ preference. In the real world, this kind
of ideal starting proposal may not be available. For example, in the Water Forum case,
the County of Sacramento and EBMUD (East Bay Municipal Utility District) were
engaged in a lawsuit of nearly 20 years. This kind of all or nothing decision did not help
either group to receive what they wanted. The same thing occured in the Westlake case.
The LAUSD could have used the eminent domain power on the site they wanted to build
a new school on, but surely, the solution was not the best for the community as a whole.
Moreover, the issue was not whether the LAUSD or the ACOF should take the land.
Nevertheless, in a usual simple majority scheme, voters may be asked to vote for either

177
LAUSD or ACOF. A decision-making rule that reduces the policy issue into a narrow
agenda and produces winners and losers did not fit in these cases.
Furthermore, participants often come into discussion in the absence of clear
recognition of what the problem is and what they really want, which is the reason social
learning is emphasized in recent years (van Buuren, 2009). In addition, a collective
decision-making is not only a process to identify the stakeholders‟ preference, but also a
process to determine what policy area to consider (Bachrach & Baratz, 1962). In reality,
the degree of inclusiveness of the sub-dimensions of a policy that require a vote may not
be as high as the ideal one simulated in the model. Noticeably, in the Water Forum, what
the facilitators did first was identify the detailed interest areas. They carefully identified
four categories of stakeholders: water, development and business, environment, and the
larger public (Innes & Booher, 2010). In the Westlake case, through the deliberation
process, many stakeholders could be involved including First 5 L.A., L.A. City, Trust for
Public Land, and even neighborhoods in other areas of the city, which was a way of
identifying relevant interest areas. Finally, even though all sub-dimensions were
included, the decision of some dimensions via majority rule may be fixed due to technical,
economic, legal, or social feasibility. By directly involving stakeholders, the collective
deliberation process can make a creative solution that can overcome the obstacles from
the detail. The Westlake case well illustrates the effectiveness of going in detail in
collective decision-making process. Through this deliberation, memebers were able to
devise creative solutions in addition to affordable housing and the Primary Center such as

178
the Early Education Center, Boys and Girls Club, and the multipurpose room and
restrooms available to the neighbors.
In conclusion, the simulation result that showed improvement of global and local
responsiveness when participants started deliberation with the opposite or a random
proposal is theoretically more important than it seems. In fact, the preceding discussion
suggests that the ideal majority opinion is rarely available; the simulation showed an
alternative process that an ideal plan was not the starting point of deliberation, rather a
result of deliberation. Thus, the value of deliberation found through the simulation
implies the virtue of collaborative process. Collaborative deliberation process adds value
to collective decision-making when stakeholders start deliberation with, for political,
technical, or intellectual reason, an alternative that is far from an ideal. The simulation
results and the discussion lead to the following proposition.
Proposition 8. Under the condition that an ideal majority opinion cannot be
identified a priori, a collaborative deliberation process will add the most value in terms
of the responsiveness of the final decision to the general public and specific stakeholder
groups.
Speaker turn. Equality of collective deliberation is a crucial feature of
collaborative governance. However, it is one thing to say that collaborative governance
should be egalitarian and it is another to specify what constitutes the egalitarian quality of
collaborative governance. To summarize, a collective deliberation process is egalitarian
when the following conditions are met. First, participants have substantively equal
opportunities to provide the information they think is important, including minority

179
stakeholders. Second, information available by participants in a forum should be shared
as much as possible. As shown in Larson et al. (1996), unique information held by
minority members is often critical. In addition, as argued with the information
asymmetries model (Brodbeck et al., 2007), disparate information among stakeholders is
the main cause of conflict in a policy issue, which has been revealed in a study that
showed sharing information and experience through social learning resulted in reduced
conflict and a shared mental model on a social issue (Berkes, 2008). Finally, participants
should be open enough to listen to others‟ interests and adjust their interests in a mutually
beneficial way. Among the conditions, the virtual experiment on speaker turns focused
on the first. The others were reflected in individual information-processing behaviors.
An egalitarian deliberate process is important for the quality of consensus on the
collective decision. When an apparent consensus is made without incorporating
unrevealed opinions, which often occurs among the groups with low epistemic
motivation (Postmes et al., 2001), the consensus is likely thin and fragile (Innes &
Booher, 2010) and stakeholder groups with the suppressed opinion may find another
venue such as lawsuit to protect their interest. Innes and Booher (2010) pointed out that
when voices are heard and acknowledged during the deliberation process, opposition can
be reduced. The last point also implies that implementation of the agreed action plan will
be smoother when an egalitarian deliberate process works.
29
For example, the Water
Forum established the Successor Effort, whose members were a subset of the

29
Of course, with the egalitarian rule, opposition can be eminently visible and debate can increase. The
low success rate in reaching consensus among the groups with the rule may indicate the negative side of the
rule.

180
stakeholders in the Forum, to assess and monitor implementation of the decision
(Connick, 2006). Since the quality of consensus was good, participants in the Forum
implemented some of the decisions immediately (Connick, 2006).
An egalitarian deliberate process is theoretically tied to high epistemic motivation
in information-processing and to the role of facilitative leadership. Larson et al.‟s (1996)
empirical study on the information sharing behavior among medical residents is
instructive. When residents were engaged in a kind of low epistemic information-
processing behavior, that is, mentioning and repeating shared information only, the
quality of their decision was low. However, the facilitative leaders of the experimental
groups encouraged sharing of unique information held by minor members. This kind of
leadership supports the epistemic motivation of group members, often called
transformational leadership (Sosik, Kahai, & Avolio, 1998), which has been found to be
important in group creativity. A discussion from social psychology echoes the notion of
facilitative leadership that leads group members to be engaged in active information
sharing in the collaborative governance literature.
The computer simulation showed mixed results regarding the effect of egalitarian
managerial intervention. First, the egalitarian rule of speaker turn turned out to be a two-
edge sword. Forums with the rule were less likely to reach consensus. However,
successful forums reached consensus quickly compared to forums with the random or
round-robin rule. Under the egalitarian rule, the least satisfied participant always has a
chance to speak. The simulation demonstrated this by allowing the minor opinion to be
eminently expressed rather than to be suppressed. Consequently, the simulated forums

181
were pushed from easy consensus-building to continuous readjustment of their interests.
At the same time, however, letting the minor opinion be expressed helped them to pay
attention to what should have been coordinated immediately.
The round-robin rule provides a good comparison to the egalitarian rule. In one
sense, the round-robin rule is most egalitarian since it allocates the speaking turn equally
to all participants. If the egalitarian rule aims at equality in the result, then the round-
robin rule aims at equality in the process. In the case of the round-robin rule, a
participant with the minor opinion should wait for his or her turn. This has two meanings.
One is that the most urgent need from the egalitarian viewpoint could not be expressed
until it is the participant‟s turn. The other is that the participant might change his or her
opinion after obtaining new information while waiting to speak. Therefore, less satisfied
participants may come to accept the current proposal through learning.
It is difficult to find a study that focused on the speaker turns in the collaborative
governance literature. Actually, no collaborative forum would employ a strict rule like
those simulated in the computational model. Further, experimental research has found
that as group size increases, the information sharing process tends to be dominated by a
few people (Stasser, 2000). This phenomenon may not be significantly different in the
context of collaborative governance. Case studies indicate that dialogue is led not only
by powerful actors but also by facilitators or experts. From an information-processing
perspective, it is not a political issue but an issue of symmetric and asymmetric
information-processing. The Water Forum is an example that illustrates the role of a
facilitator and experts in organizing and supporting the deliberation process (Connick,

182
2006). The Westlake case shows the active role of a facilitator in sharing information
and proposing new alternatives, such as master plans drawn by reflecting the ongoing
discussion.
The results concerning global and local responsiveness of the final decision are
positive toward the egalitarian rule. Global responsiveness was highest with the
egalitarian rule of speaker turn. Local responsiveness was sensitive to the overall level of
openness. However, in general, local responsiveness of the decision under the egalitarian
rule was at least as high as that of the other rules. It is not surprising that group decision
with the egalitarian rule was responsive to specific stakeholder groups because the rule
would push the bottom line of local responsiveness up by giving priority to the least
satisfied. What is interesting is the relationship between the egalitarian rule and high
global responsiveness. Specifically, it was found in a previous analysis that high global
responsiveness is related to pro-social and high epistemic information-processing.
Therefore, from this result it can be inferred that, using the egalitarian rule, whatever the
information-processing motivation of the participants are, more diverse information can
be shared in the deliberation process.
The implication of these results is critical to the quality of consensus. High local
responsiveness, demonstrated by the simulation results, is an indicator that each
stakeholder group is well satisfied and the consensus is thick and robust, which leads to a
higher probability of being implemented (Ansell & Gash, 2008). In addition, participants
with the egalitarian rule changed their initial policy opinion the least, which means that
they remained more accountable to what they were delegated. Finally, the results imply

183
that the voice of the minority, rather than misleading the deliberation process to an
extreme, can contribute to global responsiveness of the collective decision.
In summary, except for the relatively low success rate, the performance of forums
with the egalitarian rule of speaker turn was better compared to the other two rules of
speaker turn. The simulation results support the normative claim for equal power and the
open or egalitarian dialogue condition in collaborative governance. From an information-
processing perspective, not only is the rule desirable in terms of political idea, but it will
also produce desirable outcomes in terms of responsiveness of the collective decision and
time taken to reach consensus. This further implies that conflict is not something to be
avoided but something to be revealed in collaborative governance. The results and the
discussion lead to the following propositions.
Proposition 9-1. When the deliberation process is egalitarian in terms of the
substantive opportunity to express interests, the likelihood of reaching consensus will be
low but the speed of reaching consensus will be high.
Proposition 9-2. When the deliberation process is egalitarian in terms of the
substantive opportunity to express interests, the process will produce a high quality
consensus with high global and local responsiveness.
Forum size. The computer simulation results demonstrate a clear trade-off
between small and large forum sizes in the performance of collaborative governance.
Large forums were less likely to reach consensus and, when they did, took longer
compared to small forums. Because of this longer time for discussion, participants in
large forums obtained newer data points and changed more of their policy opinion, which

184
eroded the accountability of the delegation. Global responsiveness of the final decision
showed a curvilinear relationship with forum size, with the best global responsiveness
among the middle-size forums. Finally, local responsiveness of the final decision showed
a negative relationship with forum size. Overall, a large forum size demonstrated more
disadvantages than advantages.
An irony here is that there are reasons for public leaders to enlarge a forum in the
practice of collaborative governance. First, there is a need for a wide range of knowledge
as the policy problems become wickeder, manifested by concepts such as inclusive
knowledge (Feldman & Khademian, 2007), joint fact-finding (Feldman & Khademian,
2007), adaptive co-management (Folke et al., 2005; Olsson, Folke, & Hahn, 2004), and
collaborative rationality (Innes & Booher, 2010). These concepts include as conceptual
elements the use of local knowledge, integration between local knowledge and
professional knowledge, unlimited information and preference sharing among
stakeholders, and authentic dialogue. To ensure that a forum collects all necessary
knowledge to yield a responsive decision, size could become inevitably large. From an
information-processing perspective, participants in a large forum may benefit from
diversity of information shared during the deliberation process. This benefit is even more
important when participants are pro-self because the diversity of participants can
countervail the negative effect of pro-self information-processing behavior. Case studies
have reported that forum size is usually no less than 30, although the size is usually
smaller at the initial stage. For example, when beginning in 1993, the Water Forum
consisted of representatives from 15 organizations but the size soon grew to include

185
other stakeholders. In the Westlake case, participants identified from the archival data
were nearly 50. Two collaborative city planning forums with more than a hundred
participants were also reported (Feldman & Khademian, 2007).
Second, inclusiveness is not only a matter of knowledge, but also a matter of
representation of diverse interests. Inclusiveness of diverse interests is at the core of the
normative value of collaborative governance. Stakeholders who would be affected by the
decision and who have the power to facilitate or block the implementation of the decision
should be included in the process (Innes & Booher, 2010). The BAASC case is a good
example. Unlike the case of the Water Forum and the Westlake, because some key
players, such as local governments, were not included in the BAASC case, the Compact
failed to be implemented (Innes & Booher, 2010). Thus, the role of conveners or
boundary-spanners is important in assuring that all relevant stakeholders are included in
the forum (Bryson et al., 2006; Gray, 1985; van Buuren, 2009).
From the simulation results and the discussion, it may be that an optimal forum
size is needed for inclusiveness, which was met at the minimum level. The curvilinear
relationship between forum size and global responsiveness of the final decision implies
the existence of an optimal forum size and indicates there could be redundancy of
participation beyond the optimal size. The Water Forum case illustrates how to manage
the forum size. For example, the staff divided the Forum into four teams according to the
specific interests identified. Then the final decision-making was made by the overarching
team. Simultaneously, the Westlake Collaborative remained relatively small throughout

186
the meetings. The point is that increasing the forum size should be conducted cautiously.
The results and the discussion lead to the following propositions.
Proposition 10-1. Larger forums will suffer a low likelihood to reach consensus,
prolonged discussion, and low responsiveness of the collective decision to specific
stakeholder groups.
Proposition 10-2. An optimal forum size will be determined at the smallest level
after considering the inclusiveness of the forum in terms of both interest and knowledge.

187
CHAPTER 10: CONCLUSION

The propositions developed in the Discussion chapter comprise an information-
processing perspective on collaborative governance. In this chapter, the researcher
provides the prospect of an information-processing perspective on collaborative
governance to investigate the mutual learning and deliberation process at the behavioral
level. Specifically, the researcher discusses answers from an information-processing
perspective to critical questions about collaborative governance identified by previous
researchers. This dissertation concludes with the limitations of this research and future
directions.
Information-Processing Perspective on Collaborative Governance
In this dissertation, the researcher approached collaborative governance from an
information-processing perspective. By combining the motivated information-processing
perspective from social psychology literature (e.g., Brodbeck et al., 2007; De Dreu et al.,
2008) and the collective learning perspective from collaborative governance literature
(e.g., Fung & Wright, 2001; Innes & Booher, 2010; Roberts, 2002), the computational
model described the collaborative process as an information-processing and decision-
making process. In this sense, the dynamics of collaborative governance has been
understood in terms of information sharing, learning, and collective decision-making.
Specifically, participants enter into a collaborative process with their own policy opinions,
supporting information, and type of motivation for information-processing. Driven by
their information-processing behaviors and led by group-level interventions, participants
in a collaborative system reach consensus regarding their collective policy opinions.

188
By taking this perspective and developing a computational model of collaborative
governance, this researcher developed a set of theoretical propositions. These
propositions provide a theoretical perspective from which to approach questions raised by
researchers and practitioners who face the new practice of collaborative governance. For
example, Mandell and Keast (2008), based on the discussion from the International
Research Society for Public Management conference on collaborative network in
Potsdam, Germany, 2007, summed up nine relevant questions regarding collaborative
governance. To summarize the discussion in this dissertation and evaluate the
contribution of the propositions in this study, answers to the questions informed by the
information-processing perspective and computational model are briefly addressed here.
The nature of collaborative governance. The basic question on collaborative
governance concerns the definition of collaborative governance (Mandell & Keast, 2008).
The most common perception on collaborative governance is that it is a political process
and no one would deny the overarching effect of political factors. Collaborative
governance is also an institutional arrangement, which is a prescription of collective
behavior. Collaborative governance is also compatible with the notion of collaborative
networking in which complex problems are collaboratively addressed by participants who
cannot solve them on their own (Mandell & Keast, 2008). These discussions on the
nature that collaborative governance is distinct from other forms of public governance
continue. Simultaneously, questions on the processing side of collaborative governance
have increased (Ansell & Gash, 2008; Hopkins, 2010).

189
In the current study, this researcher attempted to examine the veins of information
flow and their role in determining the performance of collaborative governance.
Collaborative governance has been characterized as a deliberative, consensus-oriented
decision-making process. This conceptualization emphasizes the information-processing
aspect of mutual interactions among stakeholders and implies an adaptive nature of
collaborative governance. It is not a rigid, mechanical decision-making system but a
flexible, adaptive one. In this sense, a collaborative governance system can be seen as a
complex adaptive system composed of intelligent actors (Miller & Page, 2007).
Specifically, there is no mechanism of central control and the formation of an action plan
is the result of a gradual, self-regulated process among intelligent actors. In this study, it
was shown that a collaborative governance system, as a whole, could be modeled as a
learning system to be responsive to constituents.
The computational model demonstrated that the performance of a deliberative
decision-making system is contingent on the type of motivated information-processing
behavior. Information-processing, driven by high epistemic motivation, was related to
high global and local responsiveness. In addition, information-processing, driven by pro-
social motivation, was related to high global responsiveness and low local responsiveness.
The computational model also demonstrated that different modes of managerial
intervention can be used to leverage the deliberation process. These results show that
system performance at the global level can be explained by motivation at the individual
level and by regulating rules at the group level (Arrow et al., 2000).

190
In summary, from an information-processing perspective, collaborative
governance is a system where intelligent actors interact locally to pursue their own and
collective goals. Strong political factors, including power imbalance, political pressure
from outside, or politics to pool resources, may override the effects of information-
processing behaviors. However, in line with the normative expectation on authentic
dialogue, the rising scholarly expectation is that appropriately designed institutions for
information-processing among participants will also overcome negative political factors
that might determine the success or failure of collaborative governance (Dryzek, 1990;
Feldman & Khademian, 2007; Gray, 1985; Habermas, 1981; Innes & Booher, 2010).
The meaning of effectiveness in collaborative governance. Another critical
question on collaborative governance is how to understand and measure effectiveness in
collaborative governance. Mandell and Keast (2008) summarized diverse criteria on
which to understand effectiveness of a collaborative network, which included (1)
changing perceptions, attitudes, and values, (2) changing relationships, (3) the degree to
which there is buy in by outside stakeholders, (4) achieving goals, and (5) formal
agreements. From an information-processing perspective, such criteria as changing
perceptions, attitudes, and values certainly reflect the outcomes of mutual learning.
Achieving formal agreements was also explicitly considered in the computational model.
In addition, from an information-processing perspective, the degree of mutual learning is
also counted as an independent, important criterion for measuring effectiveness. In fact,
from an information-processing perspective, formal agreements and changing perceptions
can be regarded as a derivation of mutual learning rather than a result of politics or

191
negotiation. In summary, an information-processing perspective highlights the degree of
mutual learning as the fundamental measure. Even when participants fail to reach
consensus, which is not surprising considering the difficulty of devising a mutually
beneficial solution under social complexities (Conklin, 2005), the legacy of mutual
learning can be a source of future collaboration (Bull et al., 2008; Connick & Innes,
2003).
Effectiveness as mutual learning can be perceived not only from individual actors‟
perspective but also from the system as a whole. From an information-processing
perspective, a dimension of effectiveness concerns to what degree available information
is learned and shared by actors through the deliberation process. Learning itself may not
be an explicit goal of individual actors when participating in a collaborative process.
However, knowledge-based decision-making is inevitable in a society when the goal is to
solve complex social problems. Therefore, at the system level, the degree of learning is
an important aspect of system effectiveness that determines the quality of the collective
decision. The result from the computational model is compatible with the argument.
When participants were motivated in a highly epistemic way, both the global and the
local responsiveness of the collective decision were improved. Furthermore, the
computational model showed that too quick a consensus could be detrimental to the
responsiveness of the decision because the deliberation process ends without fully using
the information available in the forum.
In conclusion, from an information-processing perspective on collaborative
governance, the degree of mutual learning is a critical measure of the effectiveness of a

192
collaborative governance system. Even when visible effectiveness, such as goal
achievement, is negative, the potential value of mutual learning remains significant.
Facilitator or broker in the network. Another question raised by Mandell and
Keast (2008) concerns the effect of having a facilitator or broker in the network and the
effectiveness of the collaborative network. Although the focus of interest is different,
this dissertation investigated the effect of leadership in the form of managerial
intervention on the deliberation process among stakeholders. Different modes of
managerial interventions, including agenda setting, speaking turns, and forum size, are
expected to have different effects on the collective decision. For example, the
computational model showed that a collaborative deliberation process, regardless of what
policy alternative a group begins with, could add value to public decision-making by
improving the responsiveness of the decision to the public and specific stakeholder
groups. The current model also demonstrates an advantage of the egalitarian leadership
in the deliberation process, which resulted in high global and local responsiveness and
quick consensus formation, when successful. Finally, the model suggests that the smaller
the collaborative forum, the more effective it will be, with a caution raised by some
curvilinear effects favoring medium sizes. Furthermore, these effects of managerial
interventions were contingent on the degree of interdependent interest among
stakeholders. When considering diverse effectiveness measures, there is no magic bullet
for successful governance. In other words, it is not easy to judge a collaborative
governance system as successful with a unitary dimension and leadership may be
effective in one aspect and ineffective in another aspect. The implication of an

193
information-processing perspective is that facilitators can be strategic in managing
collaboration by leveraging the deliberation processes among stakeholders.
Length of time a collaborative network lasts. The durability of a collaborative
governance system was another question raised by Mandell and Keast (2008). The
duration of a collaborative governance system may reveal diverse aspects about the
effectiveness of the system. The current study showed, via simulations, that a shorter
time to reach consensus is not necessarily a sign of a highly effective system.
Participants may reach consensus quickly as they are motivated pro-socially and are
epistemically low. This does not lead to critical thinking and may reduce effectiveness of
the system in terms of learning and responsiveness. In contrast, taking a long time to
reach consensus, while it may be burdensome to some participants whose interests are
time-sensitive, is not necessarily a sign of low effectiveness. From an information-
processing perspective, it will take a longer time for a sufficient level of social learning to
occur; it takes time for people to change their perceptions, attitudes, and values. It also
takes time for individuals to build a shared mental model about the pertinent social issue.
Simulations, conducted in this study, actually emulated the process. Considering factors
such as the complexity of the social problem, the variance in diverse actors‟ intellectual
needs, and inertia in changing their perceptions (Termeer, 2009), a long lifespan for a
collaborative governance system as a measure of effectiveness at the decision-making
stage should be considered more or less contingent on an issue by issue basis and
evaluated in light of the quality of the decision.

194
In conclusion, this study contributes to the collaborative governance literature by
highlighting an information-processing aspect of collaborative governance and by
providing a computational model of collective decision-making processes. Specifically,
this study developed a behavioral model of information-processing and learning among
participants in collaborative governance and linked the model with performance variables
relevant to collaborative governance. Based on conceptually and formally defined
measures as well as on a computational model that formally defined the complex
relationships among behavioral factors and institutional factors, this study suggests
various theoretical propositions. The propositions developed in this study comprise a
contingency framework of collaborative governance from an information-processing
perspective.
Limitations and Future Research
The limitations of this research are twofold in the broadest sense (1) the
inclusiveness or reality of the computational model and (2) its external validity. These
limitations also provide a prospect of future research. In this section, the limitations and
suggestions for future research to fill the gap are discussed. Future research can be
conducted by manipulating the aforementioned parameters that were set as constants in
this study. Rather than repeating them, two unmentioned, but important, factors to be
considered in future research for significant theory development are discussed here: trust
and a broad range of social learning.
Trust. Trust is an important by-product of a deliberation process. A high level of
trust will encourage stakeholders to be more pro-social. The process of accumulating

195
mutual trust among participants in a collaborative forum is an important part of the
overall process for the success of collaborative governance (Ansell & Gash, 2008).
Nevertheless, a collaborative governance system does not always result in enhanced
mutual trust among participants. On one hand, it has been reported that unsuccessful
collaborative governance systems can contribute to accumulating mutual trust (Innes &
Booher, 2010). In other words, sometimes, the major purpose of a collaborative forum is
not consensus-building but social networking and trust-building. On the other hand, even
a successful collaborative governance system, in terms of achieving its substantive goal,
may not contribute to the accumulation of long-term mutual trust, as shown in the case of
the Westlake project. In conclusion, the achievement of the goal of collaborative
governance may not necessarily accompany the accumulation of mutual trust and vice
versa.
The process of trust accumulation can follow three different scenarios. First,
when we consider a successful incidence of accumulating mutual trust, we can expect a
model characterized by a positive circle of trust accumulation. That is, the trust of actor
A for actor B affects positively the trust of actor B for actor A and vice versa. This
positive circle is expected to be found frequently in successful collaborative governance
systems. In this positive circle, participants will end with an ideal consequence of high
mutual trust, which will be available for consecutive implementation processes or future
collaborations. A second scenario is group polarization (Burnstein & Vinokur, 1977). In
this scenario, participants in a collaborative governance system are divided into a few,
usually two or three, subgroups. Because groups are polarized in their positions on the

196
pertinent issue, there will be a severe conflict among subgroups and a relatively tight
bond within each subgroup. In this case, trust building will occur only within the
subgroups, which is not beneficial to the whole group. This scenario occurs when
ideologically motivated actors are involved. A final scenario is a complete fragmentation
among participants. In this scenario, there is no trust building among participants. This
scenario can be realized when participants only need to solve their conflicts on a narrow
issue, when they do not expect their relationship to continue, or when external political
pressure drives them to remain in the forum against their willingness.
The current study did not consider the dynamics of trust building. During the
iterations, participants did not change their pro-self or pro-social attitude. Thus, the static
nature of the model is a limitation. A future study will consider a model of trust building
in a deliberative decision-making system and the model will consider diverse scenarios
of trust building mentioned above. As participants realize one of the three scenarios, a
researcher may be able to measure success to reach consensus and responsiveness of their
decision.
Social learning on a broad range. In the simulation, mutual learning occurred
only among participants in a forum. However, learning by different actors can occur
simultaneously. First, constituents are active learners; they learn from their own
experiences, from their neighbors, and from the media. Accordingly, they may change
their policy opinions over time. This means that the basis on which to determine the
level of responsiveness of the collective decision can also change over time. Second,
constituents and their representative learn from each other. In the current model, the

197
knowledge constituents possessed was the source of the knowledge their representatives
possessed. Considering the importance of local knowledge in environmental decision-
making (Feldman & Khademian, 2007), learning by representatives from their
constituents is an important step in social learning processes. Simultaneously,
constituents learn from their representatives. As representatives obtain new information
in the forum, they convey this information to their constituents. Gladwell (2000) called
representatives who play the role of information delivers as connectors. The connectors
not only push local knowledge possessed by local constituents to a forum of broader
interests, but also push back global knowledge generated by the forum to their local
stakeholders. What is necessary for the forum to reach consensus is the latter process of
learning (Bull et al., 2008). That is, the idea and information developed and shared in the
forum should spread throughout constituents and affect their policy opinions to form a
social consensus.
This broader scale of social learning can be a focus of future research as the
model may consider diverse aspects of a theoretical framework of social learning. First,
the model should consider empirical findings on the willingness and effectiveness of
representatives in transferring ideas and knowledge obtained from the collaborative
forum to their own constituents. For example, in the Water Forum case,
environmentalists were concerned that the collective decision would be unacceptable to
their supporters (Connick, 2006), in which case they could lose their legitimacy as
representatives and be claimed as unaccountable. This issue is related to the political

198
concept of representativeness. Thus, a future model may be informed by the political
theory of representativeness as well as the managerial theory of accountability.
Second, a future model may consider independent learning by constituents. This
feature involves theoretical concepts including environmental turbulence and time gap of
a policy decision into the model. When their environments, other than feedback from
their representatives, affect constituents, the learning process will work as a moderating
factor on the effectiveness of mutual learning between constituents and representatives.
Here, both the degree and the speed of learning and opinion change are relevant
theoretical issues.
Finally, a future model that considers all social learning processes may describe a
collaborative governance system as a complex adaptive system (Miller & Page, 2007).
Future studies may examine the effectiveness of institutional factors and processes of a
social learning system on recognizing the environment and solving the problem.
Agent-based modeling and the issue of external validity. This study employed
agent-based modeling to develop theoretical propositions regarding deliberative decision-
making processes in collaborative governance. Via an agent-based model, the real world
is simplified and data are artificially generated. However, there are two general
limitations of this method. First, although simplification of the real world is often not a
shortcoming but a virtue of a theory, an agent-based model cannot fully avoid bias in the
simplification process. Second, the simulation results do not verify the external validity
of the propositions developed in this study. The simulation results show that the
propositions can logically be derived from the assumptions adopted into the agent-based

199
model. Thus, to ensure the external validity of the propositions, empirical data that
support the propositions are necessary.
Following are some future research directions to overcome the current limitations.
First, controlled experiments can be employed to validate the propositions regarding
information-processing behaviors. This venue will be an expansion of research that has
been conducted in social psychology literature (see Wittenbaum et al., 2004). For
example, subjects may be manipulated to be motivated in pro-self or pro-social and low
or high epistemic ways with specific payoff structures and then be provided a task to
make a collective, consensus-oriented decision. Since experimental procedures are well-
developed in the field, the design of future research will have to modify these procedures
to fit with the propositions developed in the current study. Second, the effect of
managerial interventions can be examined by both controlled experiments and
participatory observation. For example, an observer may convene subjects of a decision-
making group with a certain managerial intervention. It is also possible that a researcher
participate in real world collaborative forums, observe the deliberation process, and
analyze the type of managerial intervention used in the forums. Multiple observers might
also evaluate leadership, combined with self-evaluation by the leader. Finally, using
extant data (for example, case studies); the computational model developed in the current
study could be modified to more accurately reflect real world conditions. For example, a
computational model could simulate the actual two-tiered group rule applied in the
Sacramento Water Forum case. In addition, a computational model could adopt a new
design of the task as the task design in this study was rather universal. For example, a

200
task design characterized by a high degree of interdependence between sub-dimensions
can be employed.
As collaborative governance becomes more popular in solving social problems in
public spheres, knowledge about the deliberation process in collaborative governance
becomes more critical for better practice of innovative public management. Therefore,
we need keen understanding about the effects of individuals‟ motivated information-
processing behaviors during deliberation and the strengths and weaknesses of different
modes of managerial interventions in the deliberation process. Understanding
collaborative governance should be based on theories that weave together a consistent
perspective on the complex process of collaborative governance. Finally, an appropriate
research method should accompany research that investigates these complex processes.
This dissertation is an effort to respond to the scholarly need to better understand
collaborative governance. By combining collaborative governance literature with
behavioral models of information-processing in social psychology and by using agent-
based modeling to integrate such models, this dissertation proposed an information-
processing perspective on collaborative governance that has received less attention but
will gain relevance in the social learning era as being the key to solving complex social
problems.

201
BIBLIOGRAPHY

Abelson, R. P. (1986). Beliefs are like possessions. Journal for the Theory of Social
Behaviour, 16(3), 223-230.
Adler, P. S. (2001). Market, hierarchy, and trust: The knowledge economy and the future
of capitalism. Organization Science, 12(2), 215-234.
Ansell, C. & Gash, A. (2008). Collaborative governance in theory and practice. Journal
of Public Administration Research and Theory, 18(4), 543-571.
Argote, L. (1999). Organizational learning: Creating, retaining and transferring
knowledge. Boston: Kluwer Academic Publishers.
Argyris, C. (1999). On organizational learning. Malden, MA: Blackwell Publishers.
Arrow, H., McGrath, J. E., & Berdahl, J. L. (2000). Small groups as complex systems:
Formation, coordination, development and adaptation. Thousand Oaks, CA: Sage
Publications.
Axelrod, R. M. (1997). The complexity of cooperation: Agent-based models of
competition and collaboration. Princeton, NJ: Princeton University Press.
Axelrod, R. & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211(4489),
1390-1396.
Axtell, R., Axelrod, R., Epstein, J. M., & Cohen, M. D. (1996). Aligning simulation
models: A case study and results. Computational & Mathematical Organization
Theory, 1(2), 123-141.
Bachrach, P. & Baratz, M. S. (1962). Two faces of power. The American Political
Science Review, 56(4), 947-952.
Beierle, T. C. & Konisky, D. M. (2001). What are we gaining from stakeholder
involvement? Observations from environmental planning in the Great Lakes.
Environment and Planning C, 19(4), 515-528.
Berkes, F. (2008). Evolution of co-management: Role of knowledge generation, bridging
organizations and social learning. Journal of Environmental Management, 90(5),
1692-1702.
Blatner, K. A., Carroll, M. S., Daniels, S. E., & Walker, G. B. (2001). Evaluating the
application of collaborative learning to the Wenatchee Fire Recovery Planning
effort. Environmental Impact Assessment Review, 21(3), 241-270.

202
Booher, D. E. (2004). Collaborative governance practices and democracy. National Civic
Review, 93(4), 32-46.
Booher, D. E. & Innes, J. E. (2002). Network power in collaborative planning. Journal of
Planning Education and Research, 21(3), 221-236.
Bouwen, R. & Taillieu, T. (2004). Multi-party collaboration as social learning for
interdependence: Developing relational knowing for sustainable natural resource
management. Journal of Community & Applied Social Psychology, 14(3), 137-
153.
Brodbeck, F. C., Kerschreiter, R., Mojzisch, A., Frey, D., & Schulz-Hardt, S. (2002). The
dissemination of critical, unshared information in decision-making groups: The
effects of pre-discussion dissent. European Journal of Social Psychology, 32(1),
35-56.
Brodbeck, F. C., Kerschreiter, R., Mojzisch, A., & Schulz-Hardt, S. (2007). Group
decision making under conditions of distributed knowledge: The information
asymmetries model. Academy of Management Review, 32(2), 459-479.
Bryson, J. M., Crosby, B. C., & Stone, M. M. (2006). The design and implementation of
cross-sector collaborations: Propositions from the literature. Public
Administration Review, 66, 44-55.
Buber, M. (1958). I and thou. New York: Scribner.
Bull, R., Petts, J., & Evans, J. (2008). Social learning from public engagement: Dreaming
the impossible? Journal of Environmental Planning and Management, 51(5), 701-
716.
Burnstein, E. & Vinokur, A. D. (1977). Persuasive argumentation and social comparison
as determinants of attitude polarization. Journal of Experimental Social
Psychology, 13(4), 315-332.
Burton, R. M. (2003). Computational laboratories for organization science: Questions,
validity and docking. Computational & Mathematical Organization Theory, 9(2),
91-108.
Buuren, A. van. (2009). Knowledge for governance, governance of knowledge: Inclusive
knowledge management in collaborative governance processes. International
Public Management Journal, 12(2), 208-235.
Carley, K. M. (1991). Designing organizational structures to cope with communication
breakdowns: A simulation model. Organization Environment, 5(1), 19-57.

203
Carley, K. M. (1992). Organizational learning and personnel turnover. Organization
Science, 3(1), 20-46.
Carley, K. M. (1995). Computational and mathematical organization theory: Perspective
and directions. Computational & Mathematical Organization Theory, 1(1), 39-56.
Carley, K. M. (1996). Validating computational models. Working paper: Social and
Decision Sciences. Pittsburgh, PA: Carnegie Mellon University.
Carley, K. M. (2002). Computational organization science: A new frontier. Proceedings
of the National Academy of Sciences of the United States of America, 99(Suppl 3),
7257-7262.
Carley, K. M. & Gasser, Les. (1999). Computational organization theory. In G. Weiss
(Ed.), Multiagent systems: A modern approach to distributed artificial
intelligence (pp. 299-330). Cambridge, MA: The MIT Press.
Carley, K. M. & Newell, A. (1994). The nature of the social agent. The Journal of
Mathematical Sociology, 19(4), 221-262.
Carley, K. M. & Prietula, M. J. (1994). Computational organization theory. Lawrence
Erlbaum Associates.
Chen, B. & Graddy, E. A. (2005). Inter-organizational collaborations for public service
delivery: A framework of preconditions, processes, and perceived outcomes.
Paper presented at the 2005 ARNOVA conference, November 17-19, Washington,
DC.
Choi, T., Walsh, T., & Weiner, J. (2006). Finding a third way for community: An
evaluation of the Westlake Joint-Use Collaborative. Paper presented to New
Schools Better Neighborhoods, May 2006, Los Angeles, CA.
Cohen, M. D., March, J. G., & Olsen, J. P. (1972). A garbage can model of organizational
choice. Administrative Science Quarterly, 17(1), 1-25.
Cohen, S. G., Ledford Jr, G. E., & Spreitzer, G. M. (1996). A predictive model of self-
managing work team effectiveness. Human Relations, 49(5), 643-676.
Conklin, E. J. (2006). Dialogue mapping: Building shared understanding of wicked
problems. Hoboken, NJ: John Wiley & Sons Inc.
Connick, S. (2006). Sacramento Area Water Forum: A case study. Institute of Urban &
Regional Development, IURD working paper series 2006-06: University of
California, Berkeley.

204
Connick, S. & Innes, J. E. (2003). Outcomes of collaborative water policy making:
Applying complexity thinking to evaluation. Journal of Environmental Planning
and Management, 46(2), 177-197.
Cooper, T. L., Bryer, T. A., & Meek, J. W. (2006). Citizen-centered collaborative public
management. Public Administration Review, 66, 76-88.
Daniels, S. E. & Walker, G. B. (1996). Collaborative learning: Improving public
deliberation in ecosystem-based management. Environmental Impact Assessment
Review, 16(2), 71-102.
De Dreu, C. K. W., Nijstad, B. A., & Knippenberg, D. van. (2008). Motivated
information processing in group judgment and decision making. Personality and
Social Psychology Review, 12(1), 22-49.
Donahue, J. D. & Nye, J. S. (2002). Market-based governance: Supply side, demand side,
upside, and downside. Washington, DC: Brookings Institution Press.
Downs, A. (1961). Problems of majority voting: In defense of majority voting. The
Journal of Political Economy, 69(2), 192-199.
Dryzek, J. S. (1990). Discursive democracy. New York: Cambridge University Press.
Dryzek, J. S. (2000). Deliberative democracy and beyond: Liberals, critics, contestations.
New York: Oxford University Press.
Dunn, W. (1981). Public policy analysis: An introduction. Englewood Cliffs, N.J.:
Prentice-Hall.
Edmondson, A. C., Dillon, J. R., & Roloff, K. S. (2007). Three perspectives on team
learning: Outcome improvement, task mastery, and group process. In J. P. Walsh
& A. P. Brief (Eds.), The Academy of Management Annals (Chapter 6).
Psychology Press.
Edwards, K. & Smith, E. E. (1996). A disconfirmation bias in the evaluation of
arguments. Journal of Personality and Social Psychology, 71, 5-24.
Epstein, J. M. (1999). Agent-based computational models and generative social science.
Complexity, 4(5), 41-60.
Epstein, J. M. (2006). Generative social science: Studies in agent-based computational
modeling. Princeton, NJ: Princeton University Press.
Epstein, J. M. & Axtell, R. (1996). Growing artificial societies: Social science from the
bottom up. Washington, D.C.: Brookings Institution Press.

205
Feldman, M. S. & Khademian, A. M. (2007). The role of the public manager in inclusion:
Creating communities of participation. Governance, 20(2), 305-324.
Fisher, R., Ury, W., and Patton, B. (1981). Getting to yes: Negotiating agreement without
giving in. New York: Penguin Group.
Fiske, A. P. (1992). The four elementary forms of sociality: Framework for a unified
theory of social relations. Psychological Review, 99, 689-723.
Folke, C., Hahn, T., Olsson, P., & Norberg, J. (2005). Adaptive governance of social-
ecological systems. Annual Review of Environment and Resources, 30, 441-473.
Fossett, J. & Thompson, F. J. (2005). Administrative responsiveness to the disadvantaged:
The case of children's health insurance. Journal of Public Administration
Research and Theory, 16(3), 369-392.
Fraidin, S. N. (2004). When is one head better than two? Interdependent information in
group decision making. Organizational Behavior and Human Decision Processes,
93(2), 102-113.
Fung, A. & Wright, E. O. (2001). Deepening democracy: Innovations in empowered local
governance. Politics and Society, 29(1), 5-41.
Fung, A. & Wright, E. O. (2003). Deepening democracy: Institutional innovations in
empowered participatory governance. London: Verso.
Giddens, A. (1984). The constitution of society: Outline of the theory of structuration.
Berkeley, CA: University of California Press.
Gladwell, M. (2000). The tipping point: How little things can make a big difference.
Little, Brown and Company.
Gray, B. (1985). Conditions facilitating interorganizational collaboration. Human
Relations, 38(10), 911-936.
Greitemeyer, T. & Schulz-Hardt, S. (2003). Preference-consistent evaluation of
information in the hidden profile paradigm: Beyond group-level explanations for
the dominance of shared information in group decisions. Journal of Personality
and Social Psychology, 84(2), 322-339.
Greitemeyer, T., Schulz-Hardt, S., Brodbeck, F. C., & Frey, D. (2006). Information
sampling and group decision making: The effects of an advocacy decision
procedure and task experience. Journal of Experimental Psychology Applied,
12(1), 31-42.

206
Habermas, J. (1981). The theory of communicative action: Reason and the rationalization
of society. Boston, MA: Beacon Press.
Hahn, T., Olsson, P., Folke, C., & Johansson, K. (2006). Trust-building, knowledge
generation and organizational innovations: The role of a bridging organization for
adaptive comanagement of a wetland landscape around Kristianstad, Sweden.
Human Ecology, 34(4), 573-592.
Halley, J. D. & Winkler, D. A. (2008). Classification of emergence and its relation to
self-organization. Complexity, 13(5), 10-15.
Harrison, J. R., Lin, Z., Carroll, G. R., & Carley, K. M. (2007). Simulation modeling in
organizational and management research. Academy of Management Review, 32(4),
1229-1245.
Hastie, R. & Stasser, G. (2000). Computer simulation methods for social psychology. In
Harry T. Reis & Charles M. Judd (Eds.), Handbook of research methods in social
and personality psychology (pp. 85–114). New York: Cambridge University Press.
Healey, P. (1996). Consensus-building across difficult divisions: New approaches to
collaborative strategy making. Planning Practice and Research, 11(2), 207-216.
Healey, P. (2006). Collaborative planning: Shaping places in fragmented societies.
Basingstoke, UK: Palgrave Macmillan.
Himmelman, A. T. (1996). On the theory and practice of transformational collaboration:
From social service to social justice. In Chris Huxham (Ed.), Creating
collaborative advantage (pp. 19-43). Thousand Oaks, CA: Sage Publications.
Hinsz, V. B., Tindale, R. S., & Vollrath, D. A. (1997). The emerging conceptualization of
groups as information processors. Psychological Bulletin, 121, 43-64.
Holland, J. H. (1995). Hidden order: How adaptation builds complexity. Cambridge, MA:
Perseus Books.
Holland, J. H. (1998). Emergence: From chaos to order. Reading, MA: Helix Books.
Holland, J. H. & Miller, J. H. (1991). Artificial adaptive agents in economic theory. The
American Economic Review, 81(2), 365-370.
Hollenbeck, J. R., Ilgen, D. R., Sego, D. J. , Hedlund, J., Major, D. A., & Phillips, J.
(1995). Multilevel theory of team decision making: Decision performance in
teams incorporating distributed expertise. Journal of Applied Psychology, 80(2),
292-316.

207
Hollingshead, A. B. (1996). The rank-order effect in group decision making.
Organizational Behavior and Human Decision Processes, 68(3), 181-193.
Hopkins, D. (2010). Planning a city through 'dialogue': Deliberative policy-making in
action in Western Australia. Urban Policy and Research, 28(3), 261-276.
Huber, G. P. (1991). Organizational learning: The contributing processes and the
literatures. Organization Science, 2(1), 88-115.
Innes, J. E. & Booher, D. E. (2003). Collaborative policymaking: Governance through
dialogue. In Maarten A. Hajer & Hendrik Wagenaar (Eds.), Deliberative policy
analysis: Understanding governance in the network society (pp. 33–59). New
York: Cambridge University Press.
Innes, J. E. & Booher, D. E. (2010). Planning with complexity: An introduction to
collaborative rationality for public policy. New York: Routledge.
Katzenbach, J. R. & Smith, D. K. (2003). The wisdom of teams: Creating the high-
performance organization. New York: HarperCollins.
Kauffman, S. A. (1993). The origins of order: Self-organization and selection in
evolution. New York: Oxford University Press.
Kelley, H. H. & Stahelski, A. J. (1970). Social interaction basis of cooperators' and
competitors' beliefs about others. Journal of Personality and Social Psychology,
16(1), 66-91.
Kerr, N. L. & Tindale, R. S. (2004). Group performance and decision making. Annual
Review of Psychology, 55, 623-55.
Kim, D. H. (1993). The link between individual and organizational learning. In David A.
Klein (Ed.), The Strategic management of intellectual capital (pp. 41-62).
Woburn, MA: Butterworth-Heinemann.
Klimoski, R. & Mohammed, S. (1994). Team mental model: Construct or metaphor?
Journal of Management, 20(2), 403-437.
Larson, J. R., Christensen, C., Abbott, A. S., & Franz, T. M. (1996). Diagnosing groups:
Charting the flow of information in medical decision-making teams. Journal of
Personality and Social Psychology, 71, 315-330.
Larson, J. R., Christensen, C., Franz, T. M., & Abbott, A. S. (1998a). Diagnosing groups:
The pooling, management, and impact of shared and unshared case information in
team-baeed medical decision making. Journal of Personality and Social
Psychology, 75, 93-108.

208
Larson, J. R., Foster-Fishman, P. G., & Franz, T. M. (1998b). Leadership style and the
discussion of shared and unshared information in decision-making groups.
Personality and Social Psychology Bulletin, 24(5), 482-495.
Larson, J. R., Foster-Fishman, P. G., & Keys, C. B. (1994). Discussion of shared and
unshared information in decision-making groups. Journal of Personality and
Social Psychology, 67, 446-461.
Lawler, E. E., Mohrman, S. A., & Ledford, G. E. (1992). Employee involvement and total
quality management: Practices and results in fortune 1000 companies. San
Francisco: Jossey-Bass.
Lax, D. A. & Sebenius, J. K. (1986). Interests: The measure of negotiation. Negotiation
Journal Process Dispute Settlement, 2(1), 73-92.
Leach, W. D. & Pelkey, N. W. (2001). Making watershed partnerships work: A review of
the empirical literature. Journal of Water Resources Planning and Management,
127, 378-385.
Lerner, J. S. & Tetlock, P. E. (1999). Accounting for the effects of accountability.
Psychological Bulletin, 125(2), 255-275.
Levine, J. M. & Moreland, R. L. (1994). Group socialization: Theory and research.
European review of social psychology, 5(1), 305-336.
Levitt, R. E. (2004). Computational modeling of organizations comes of age.
Computational & Mathematical Organization Theory, 10(2), 127-145.
Lewis, K. (2004). Knowledge and performance in knowledge-worker teams: A
longitudinal study of transactive memory systems. Management Science, 50(11),
1519-1533.
Lomi, A. & Larsen, E. R. (2001). Dynamics of organizations: Computational modeling
and organization theories. Cambridge, MA: MIT Press.
Maier, N. R. F. (1970). Problem solving and creativity in individuals and groups.
Monterey, CA: Brooks/Cole.
Mandell, M. P. & Keast, R. (2008). Introduction. Public Management Review, 10(6),
687-698.
March, J. G. (1991). Exploration and exploitation in organizational learning.
Organization Science, 2(1), 71-87.

209
March, J. G. (2001). Foreword. In Alessandro Lomi & Erik R. Larsen (Eds.), Dynamics
of organizations: Computational modeling and organization theories (pp. ix-xvii).
Menlo Park, CA: AAAI Press.
McGuire, M. (2006). Collaborative public management: Assessing what we know and
how we know it. Public Administration Review, 66(s1), 33-43.
Mezias, S. J. & Lant, T. K. (1994). Mimetic learning and the evolution of organizational
populations. In Joel A. C. Baum & Jitendra V. Singh (Eds.), Evolutionary
dynamics of organizations (pp. 179-198). New York: Oxford University Press.
Miller, J. H. & Page, S. E. (2007). Complex adaptive systems. Princeton, NJ: Princeton
University Press.
Moreland, R. L. (1999). Transactive memory: Learning who knows what in work groups
and organizations. In L. L. Thompson, J. M. Levine & D. M. Messick (Eds.),
Shared cognition in organizations: The management of knowledge (pp. 3-31).
Mahwah, NJ: Lawrence Erlbaum Associates.
Muro, M. & Jeffrey, P. (2008). A critical review of the theory and application of social
learning in participatory natural resource management processes. Journal of
Environmental Planning and Management, 51(3), 325-344.
O'Leary, R., Gerard, C., & Bingham, L. B. (2006). Introduction to the symposium on
collaborative public management. Public Administration Review, 66(s1), 6-9.
O'Toole, R. & O'Toole, A. W. (1981). Negotiating interorganizational orders. The
Sociological Quarterly, 22(1), 29-41.
Olsson, P., Folke, C., & Hahn, T. (2004). Social-ecological transformation for ecosystem
management: The development of adaptive co-management of a wetland
landscape in Southern Sweden. Ecology and Society, 9(4), 2-27.
Ostrom, E. (1998). A behavioral approach to the rational choice theory of collective
action: Presidential address, American Political Science Association, 1997. The
American Political Science Review, 92(1), 1-22.
Ostrom, E. (2005). Understanding institutional diversity. Princeton, NJ: Princeton
University Press.
Parker, K. C. H. (1988). Speaking turns in small group interaction: A context-sensitive
event sequence model. Journal of Personality and Social Psychology, 54(6), 965-
971.

210
Parks, C. D., Henager, R. F., & Scamahorn, S. D. (1996). Trust and messages of intent in
social dilemmas. Journal of Conflict Resolution, 40, 134-51.
Pfeffer, J. & Salancik, G. R. (1978). The external control of organizations: A resource
dependence perspective. Stanford, CA: Stanford University Press.
Postmes, T., Spears, R., & Cihangir, S. (2001). Quality of decision making and group
norms. Journal of Personality and Social Psychology, 80(6), 918-930.
Prietula, M. J., Carley, K. M., & Gasser, L. (1998). Simulating organizations:
Computational models of institutions and groups. Cambridge, MA: MIT Press.
Provan, K. G. & Milward, H. B. (2001). Do networks really work? A framework for
evaluating public-sector organizational networks. Public Administration Review,
61(4), 414-423.
Przeworski, A. (2009). Self-government in our times. Annual Review of Political Science,
12, 71-92.
Rawls, J. (1958). Justice as fairness. The Philosophical Review, 67(2), 164-194.
Roberts, N. C. (2002). Keeping public officials accountable through dialogue: Resolving
the accountability paradox. Public Administration Review, 62(6), 658-669.
Rourke, F. E. (1992). Responsiveness and neutral competence in American bureaucracy.
Public Administration Review, 52(6), 539-546.
Saltzstein, G. H. (1992). Bureaucratic responsiveness: Conceptual issues and current
research. Journal of Public Administration Research and Theory, 2(1), 63-88.
Schein, E. H. (1999). Process consultation revisited: Building the helping relationship.
MA: Addison-Wesley Reading.
Schulz-Hardt, S., Brodbeck, F. C., Mojzisch, A., Kerschreiter, R., & Frey, D. (2006).
Group decision making in hidden profile situations: Dissent as a facilitator for
decision quality. Journal of Personality and Social Psychology, 91(6), 1080-1093.
Semler, R. (1989). Managing without managers. Harvard Business Review, 67(5), 76-84.
Senge, P. M. (1990). The fifth discipline. New York: Doubleday.
Sosik, J. J., Kahai, S. S., & Avolio, B. J. (1998). Transformational leadership and
dimensions of creativity: Motivating idea generation in computer-mediated
groups. Creativity Research Journal, 11(2), 111-121.

211
Sowa, J. E. & Selden, S. C. (2003). Administrative discretion and active representation:
An expansion of the theory of representative bureaucracy. Public Administration
Review, 63(6), 700-710.
Srivastava, A., Bartol, K. M., & Locke, E. A. (2006). Empowering leadership in
management teams: Effects on knowledge sharing, efficacy, and performance.
The Academy of Management Journal, 49(6), 1239-1251.
Stasser, G. (1988). Computer simulation as a research tool: The discuss model of group
decision making. Journal of Experimental Social Psychology, 24(5), 393-422.
Stasser, G. (2000). Information distribution, participation, and group decision:
Explorations with the discuss and speak models. In Daniel R. Ilgen and Charles L.
Hulin (Eds.), Computational modeling of behavior in organizations: The third
scientific discipline (pp. 135-156). Washington, DC: American Psychological
Association.
Stasser, G. & Stewart, D. (1992). Discovery of hidden profiles by decision-making
groups: Solving a problem versus making a judgment. Journal of Personality and
Social Psychology, 63(3), 426-434.
Stasser, G., Stewart, D. D., & Wittenbaum, G. M. (1995). Expert roles and information
exchange during discussion: The importance of knowing who knows what.
Journal of Experimental Social Psychology, 31(3), 244-265.
Stasser, G. & Taylor, L. A. (1991). Speaking turns in face-to-face discussions. Journal of
Personality and Social Psychology, 60(5), 675-684.
Stasser, G. & Titus, W. (1985). Pooling of unshared information in group decision
making: Biased information sampling during discussion. Journal of Personality
and Social Psychology, 48(6), 1467-1478.
Stasser, G. & Titus, W. (1987). Effects of information load and percentage of shared
information on the dissemination of unshared information during group
discussion. Journal of Personality and Social Psychology, 53(1), 81-93.
Stasser, G. & Titus, W. (2003). Hidden profiles: A brief history. Psychological Inquiry,
14(3&4), 304-313.
Stasser, G., Vaughan, S. I., & Stewart, D. D. (2000). Pooling unshared information: The
benefits of knowing how access to information is distributed among group
members. Organizational Behavior and Human Decision Processes, 82(1), 102-
116.

212
Stewart, D. D. & Stasser, G. (1995). Expert role assignment and information sampling
during collective recall and decision making. Journal of Personality and Social
Psychology, 69, 619-628.
Stivers, C. (1994). The listening bureaucrat: Responsiveness in public administration.
Public Administration Review, 54(4), 364-369.
Tang, S. Y. & Mazmanian, D. A. (2009). Collaborative governance approached through
theory. Retrieved from SSRN: http://ssrn.com/abstract=1516851.
Termeer, C. J. A. M. (2009). Barriers to new modes of horizontal governance. Public
Management Review, 11(3), 299-316.
Thompson, L. L., Wang, J., & Gunia, B. C. (2010). Negotiation. Annual Review of
Psychology, 61, 491-515.
Thomson, A. M. & Perry, J. L. (2006). Collaboration processes: Inside the black box.
Public Administration Review, 66, 20-32.
Thomson, A. M., Perry, J. L., & Miller, T. K. (2009). Conceptualizing and measuring
collaboration. Journal of Public Administration Research and Theory, 19(1), 23-
56.
Tjosvold, D. & Field, R. H. G. (1983). Effects of social context on consensus and
majority vote decision making. Academy of Management Journal, 26(3), 500-506.
Toma, C. & Butera, F.. (2009). Hidden profiles and concealed information: Strategic
information sharing and use in group decision making. Personality and Social
Psychology Bulletin, 35(6), 793-806.
Toulmin, S. E. (1958). The uses of argument. Cambridge, England: Cambridge Univ
Press.
Van Swol, L. M. & Ludutsky, C. L. (2007). Tell me something I don't know: Decision
makers' preference for advisors with unshared information. Communication
Research, 34(3), 297-312.
Van Swol, L. M., Savadori, L., & Sniezek, J. A. (2003). Factors that may affect the
difficulty of uncovering hidden profiles. Group Processes & Intergroup Relations,
6(3), 285-304.
Vigoda, E. (2000). Are you being served? The responsiveness of public administration to
citizens' demands: An empirical examination in Israel. Public Administration,
78(1), 165-191.

213
Vigoda, E. (2002). From responsiveness to collaboration: Governance, citizens, and the
next generation of public administration. Public Administration Review, 62(5),
527-540.
Vroom, V. H. & Jago, A. G. (1998). The new leadership: Managing participation in
organizations. Englewood Cliffs, NJ: Prentice Hall.
Wade, R. (1987). The management of common property resources: Collective action as
an alternative to privatisation or state regulation. Cambridge Journal of
Economics, 11(2), 95-106.
Waldrop, M. M. (1992). Complexity: The emerging science at the edge of order and
chaos. New York: Simon & Schuster Paperbacks
Weber, E. P. & Khademian, A. M. (2008). Wicked problems, knowledge challenges, and
collaborative capacity builders in network settings. Public Administration Review,
68(2), 334-349.
Wegner, D. M. (1987). Transactive memory: A contemporary analysis of the group mind.
In B. Mullen & G. Goethals (Eds.), Theories of group behavior (pp. 185-208).
New York: Springer-Verlag.
Weiss, G. (1999). Multiagent systems: A modern approach to distributed artificial
intelligence. Cambridge, MA: MIT Press.
West, M. A. & Anderson, N. R. (1996). Innovation in top management teams. Journal of
Applied Psychology, 81(6), 680-693.
Wittenbaum, G. M. (1998). Information sampling in decision-making groups: The impact
of members' task-relevant status. Small Group Research, 29(1), 57-84.
Wittenbaum, G. M. (2000). The bias toward discussing shared information: Why are
high-status group members immune. Communication Research, 27(3), 379-401.
Wittenbaum, G. M., Hollingshead, A. B., & Botero, I. C. (2004). From cooperative to
motivated information sharing in groups: Moving beyond the hidden profile
paradigm. Communication Monographs, 71(3), 286-310.
Wittenbaum, G. M., Hubbell, A. P., & Zuckerman, C. (1999). Mutual enhancement:
Toward an understanding of the collective preference for shared information.
Journal of Personality and Social Psychology, 77, 967-978.
Wittenbaum, G. M. & Park, E. S. (2001). The collective preference for shared
information. Current Directions in Psychological Science, 10(2), 70-73.

214
Wood, D. J. & Gray, B. (1991). Toward a comprehensive theory of collaboration. The
Journal of Applied Behavioral Science, 27(2), 139-162.
Yankelovich, D. (1999). The magic of dialogue: Transforming conflict into cooperation.
New York: Simon & Schuster.

Abstract (if available)

Abstract In recent decades, there has been an increase in theoretical attention to collaborative governance as a deliberative decision-making process among stakeholders. Meanwhile, relatively less attention has been paid by scholars to behavioral and procedural aspects of the decision-making process in collaborative governance. The following study was based on group information-processing and decision-making literature in social psychology to develop a decision-making model of collaborative governance. This model conceptualizes collaborative governance as a collective, egalitarian, deliberate, and consensus-oriented decision-making process. Further, this model considers different types of human motivation and biases in information-processing and group decision-making. This researcher employed an agent-based modeling method to design a computational model of the deliberation process in the context of collaborative governance to identify relationships among actors’ social and epistemic motivations

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Elements of a successful multi-sectoral collaborative from a local government perspective: a framework for collaborative governance – dimensions shared by award winning multi-sectoral partnerships

PDF

The effects of interlocal collaboration on local economic performance: investigation of Korean cases

PDF

How does collaborative governance work? The experience of collaborative community-building practices in Korea

PDF

Homegrown violent extremism: designing a community-based model to reduce the risk of recruitment and radicalization

PDF

A robust model for disaster risk and decision analysis

PDF

Governing regional collaboratives: institutional design, management and leadership

PDF

A public sector organizational change model: prioritizing a community-focused, inclusive, and collaborative approach to strategic planning

PDF

Governance and urban design: Omaha by design

PDF

Resilience by anticipating change: simple and robust decision making for coastal adaptation planners, communities and elected officials

PDF

Does collaborative R&D policy work? The effect of collaborative R&D policy on innovative activities of firms in Korea

PDF

Governance networks: internal governance structure, boundary setting and effectiveness

PDF

A two-level analysis of foreign policy decision making: an empirical investigation of the case of China-Taiwan

PDF

County governance reform in California: introduction of the council‐executive model and the elected county‐executive

PDF

A multiattribute decision model for the selection of radioisotope and nuclear detection devices

PDF

A multi-regional computable general equilibrium model in the Haritage of A. Anas and I. Kim

PDF

Global health diplomacy: a new era of health in U.S. foreign policy

PDF

Resilient and equitable urbanism by design: insights from the collaborative process to reimagine the SF Bay Area

PDF

Water security, national security and MCIWest: a grounded theory for operationalizing risk management

PDF

Choice neighborhoods: a spatial and exploratory analysis of Housing Authority City of Los Angeles public housing

PDF

Collaborating to provide microfinance to caregivers of orphans and vulnerable children in Ethiopia

Asset Metadata

Creator Choi, Taehyon (author)

Core Title Information sharing, deliberation, and collective decision-making: A computational model of collaborative governance

School School of Policy, Planning, and Development

Degree Doctor of Philosophy

Degree Program Policy, Planning, and Development

Publication Date 03/11/2011

Defense Date 01/20/2011

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag collaborative governance,deliberation, decision-making, agent-based modeling, learning,information-processing perspective,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Robertson, Peter J. (committee chair), Fulk, Janet (committee member), Heikkila, Eric J. (committee member)

Creator Email taehyon.choi@gmail.com,taehyonc@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m3688

Unique identifier UC1452477

Identifier etd-Choi-4391 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-446996 (legacy record id),usctheses-m3688 (legacy record id)

Legacy Identifier etd-Choi-4391.pdf

Dmrecord 446996

Document Type Dissertation

Rights Choi, Taehyon

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu