University of Southern California Dissertations and Theses

Ecology and network evolution in online innovation contest crowdsourcing

(USC Thesis Other)

Ecology and network evolution in online innovation contest crowdsourcing

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

ECOLOGY AND NETWORK EVOLUTION IN ONLINE INNOVATION
CONTEST CROWDSOURCING
by

Yiqi Li

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
(COMMUNICATION)

August 2021

Copyright 2021 Yiqi Li

ii

Acknowledgments
This dissertation is a fruit of the support, guidance, and love I have received over the past
five years. I am deeply grateful to the best dissertation committee I could ever wish for: Dr. Peter
Monge, Dr. Janet Fulk, and Dr. Aimei Yang. Their constant guidance and encouragement have
made writing this dissertation possible. I would like to express my special thanks to my advisor,
Dr. Peter Monge, who has been my solid support, and whose work has enlightened and will
continue to inspire my research. His invaluable suggestions have motivated me to work hard to
become a diligent and rigorous scholar. I would also like to thank Dr. Janet Fulk, who has always
encouraged me to think deeply and critically. In addition, I am sincerely grateful to Dr. Aimei
Yang for offering tremendous encouragement in my pursuit of academic passion.
My sincere gratitude goes to the other mentors in the Annenberg School of
Communication (ASC). I wish to gratefully acknowledge Dr. Dmitri Williams and Dr. Patricia
Riley, my qualifying exam committee members, who have inspired and helped me in so many
ways. Thank you to the other brilliant mentors who have offered countless insightful
suggestions: Dr. Emilio Ferrara, Dr. Lindsay Young, Dr. Marlon Twyman II, and Dr. Margaret
L. McLaughlin. I am also grateful for all the administrative support provided by Anne Marie
Campian and Sarah Holterman. I wish to express my sincere thanks to Dr. Michelle Shumate and
Dr. Richard Kolsky, who encouraged me to pursue my academic dream. Lastly, I am grateful to
Dr. Jiawei Sophia Fu, who has always been my role model and influenced every step I have
taken on the academic path.
I am so lucky to have met a group of brilliant colleagues at ASC and the Annenberg
Networks Network Lab. My deep gratitude goes to Dr. Bei Yan, Dr. Emily Sidnam-Mauch, Dr.
Grace Yuehan Wang, Hye Min Kim, Dr. Ignacio Cruz, Jingyi Sun, Ke Maddie Huang-

iii

Isherwood, Dr. Larry Zhiming Xu, Liyuan Wang, Lichen Zhen, Mingxuan Liu, Dr. Yao Sun, Dr.
Yu Xu, and many others. Thank you for filling my life at USC with so much color and
happiness.
I am most indebted to my family, who have provided me with unconditional love and
support. Special thanks to my mom and dad for always believing, understanding, and
encouraging me. My sincere appreciation also goes to my parents-in-law for their kindness and
love. I also want to express my appreciation to my amazing husband, Siyuan Yao, who has been
so patient, loving, resourceful, and understanding. Moreover, I wish to thank my dear friends—
Dr. Chuqing Dong, Mingyang Li, Wenjie Yu, Wenting Yin, Xiao Lian, Xiuhui Li, Dr. Yafei
Zhang, Yiyi Zhang, Dr. Yu Hou, Yuanzhi Zeng, and many others—for their guidance,
friendship, and encouragement. Finally, thank you, little Offer, my dear Chinchilla friend, for his
company.

iv

Table of Contents
Acknowledgments........................................................................................................................... ii
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Abstract ........................................................................................................................................ viii
Chapter 1: Introduction ................................................................................................................... 1
Chapter 2: Literature Review and Hypotheses ............................................................................... 7
2.1. Crowdsourcing Innovation Contests .................................................................................... 7
2.2. Research Site ...................................................................................................................... 12
2.3. Two Theoretical Perspectives: Ecological and Evolutionary Perspectives ....................... 16
2.3.1. Ecological Theories ..................................................................................................... 17
2.3.2. Evolutionary Theories ................................................................................................. 41
2.4. Crowdsourcing Innovation Contests as Communities of Practice ..................................... 55
Chapter 3: Methods ....................................................................................................................... 61
3.1. Data Collection ................................................................................................................... 61
3.2. Network Construction ........................................................................................................ 63
3.3. Analytical Procedures ........................................................................................................ 64
3.4. Measures............................................................................................................................. 74
Chapter 4: Results ......................................................................................................................... 77
4.1. Ecological Perspective ....................................................................................................... 77
4.1.1. One-mode ERGM Results (H1) .................................................................................. 78
4.1.2. Multilevel ERGM Results (H2a and H2b) .................................................................. 86
4.2. Evolutionary Perspective.................................................................................................... 96
4.3. REM Results (Community of Practice) ........................................................................... 114
Chapter 5: Discussion and Conclusion ....................................................................................... 123
5.1. Findings Discussion ......................................................................................................... 123
5.1.1. Multilevel Ecology of Categories and Social Networks ........................................... 123
5.1.2. Network Evolution of Interdependent Populations ................................................... 127
5.1.3. Community of Practice .............................................................................................. 132
5.2. Theoretical Contributions ................................................................................................. 136
5.2.1. Crowdsourcing Literature .......................................................................................... 136
5.2.2. Developing Ecology Theory ...................................................................................... 137
5.2.3. Developing Network Evolution Theory .................................................................... 140
5.2.4. Developing CoP......................................................................................................... 141
5.3. Practical Contributions ..................................................................................................... 142

v

5.4. Limitations ....................................................................................................................... 144
5.5. Future Direction ............................................................................................................... 145
5.6. Conclusion ........................................................................................................................ 151
References ................................................................................................................................... 154
Appendices .................................................................................................................................. 181

vi

List of Tables
Table 1. OpenIDEO contest phases .............................................................................................. 14
Table 2. Population types .............................................................................................................. 25
Table 3. Eight relations in the community ecology ...................................................................... 49
Table 4. One-mode ERGM results for the review network (H1) ................................................. 79
Table 5. Multilevel ERGM results for the multilevel ideation network ....................................... 87
Table 6. Multilevel ERGM results for the multilevel review network ......................................... 91
Table 7. One-mode ERGM Results for the Ideation Network (H3 and H4a)............................... 98
Table 8. One-mode ERGM results for the review network (H4b and H4c) ............................... 106
Table 9. Ordinal time REM results on comment ties during the ideation phase ........................ 116
Table 10. Ordinal time REM results on comment ties during review ........................................ 118
Table 11. A summary of hypotheses, populations relevant to the hypotheses, and results ........ 122

vii

List of Figures
Figure 1. Hypothetical partially and fully multidimensional networks ........................................ 28
Figure 2. A hypothetical multilevel network ................................................................................ 30
Figure 3. A multilevel network illustration for the OpenIDEO ideator population ...................... 32
Figure 4. An example of opportunity areas under an OpenIDEO innovation contest .................. 36
Figure 5. Population types in crowdsourcing contests and interpopulation relations .................. 50
Figure 6. An illustration of the three reciprocity patterns in Poquet and Dawson (2018) ............ 60
Figure 7. The network signature being tested for H1 ................................................................... 65
Figure 8. Cross-level closure network signatures being tested for H2a and H2b ......................... 67
Figure 9. Visualizations that help distinguish the mechanisms of H2a and H2b ......................... 68
Figure 10. Network signatures being tested for H3, H4a, H4b, and H4c ..................................... 69
Figure 11. Network signatures for H5a and H5b (REM Model) .................................................. 72
Figure 12. Network signatures for RQ1 (REM Model) ................................................................ 73
Figure 13. Goodness of Fit plots for Models 1 and 2 ................................................................... 82
Figure 14. Goodness of Fit plots for Model 3 ............................................................................... 90
Figure 15. Goodness of Fit plots for Models 4 and 5 ................................................................... 93
Figure 16. Goodness of Fit plots for Models 6 and 7 ................................................................. 100
Figure 17. Goodness of Fit plots for Models 8 and 9 ................................................................. 109

viii

Abstract
Research attention to crowdsourcing contest communities has been increasing. However,
community members’ patterns and evolution of communication interaction have been largely
unexplored. Communication networks channel knowledge and information exchange, as well as
reflect the organizing of the community. First, this research applies ecology theory to understand
how the categorized ecological niches influence patterns of social interaction. Then, the
variation-selection-retention (V-S-R) framework is adopted to studying the evolution patterns of
communication networks. Finally, the innovation contest community is conceptualized as a
community of practice organized by evolving communication networks to understand voluntary
knowledge contribution of the ideators (i.e., the contest participants). Using single-level,
multilevel and longitudinal network analyses, this research discovered important findings. For
example, ecological niches were found to influence social interaction patterns significantly. This
study also discovered that new entrants and existing community members did not prefer to share
communication ties. However, interactions between new entrants and existing members were
proven to be beneficial for achieving a better network position in the future. Finally, findings
also suggested that despite being a contest community, the ideator network was cohesive and
cooperative.
Overall, this research has a few notable contributions. First, it contributes to the
development of ecology theory by arguing that platform categories can be used to conceptualize
resource niches in online communities, and that categorized niches significantly influence social
interactions in online communities. Second, this research contributes to the development of
network evolution theory by including population interdependencies when analyzing the sources
and directions of network evolution. Finally, this dissertation argues that multiple types of

ix

community members (e.g., ideators and visitors) should be considered when analyzing
communities of practice. Practical implications for platform managers, contest organizers, and
individual ideators were also discussed.

Keywords: networks, crowdsourcing contests, ecology theory, evolutionary theory, network
evolution, niches, populations, online communities, multilevel, community of practice

1

Ecology and Network Evolution in Online Innovation Contest Crowdsourcing
Chapter 1: Introduction
Internet technologies “enable scalability, interactivity, management of flexibility,
branding, and customization in a networked business world” (Castells, 2002, p. 76). The wide
adoption of Internet technologies has brought changes to organizations’ innovation and
knowledge creation activities. Organizations of various types have adopted crowdsourcing tools.
Such organizations include firms (e.g., Boudreau & Lakhani, 2013; Pénin & Helmchen, 2011),
governments (e.g., Brabham, 2012; Brunswicker et al., 2017), and nonprofit organizations (e.g.,
Fuger et al., 2017; Wang, 2016). Many organizations have extended traditional closed-
innovation efforts that “develop creative ideas internally” within the organizational boundaries
(Sun & Majchrzak, 2020, p. 1) and adopted a crowdsourcing approach to tap into the large
communities of users for their knowledge and expertise (Baldwin & Von Hippel, 2011). For
example, InnoCentive was a pharmaceutical company that faced challenges in designing a new
product. The company built a new “matchmaking” system which allowed external experts to
solve their research and development questions and rewarded problem-solvers with monetary
incentives (Albors et al., 2008, p. 197).
The mechanisms of crowdsourcing communities’ organization and interaction have
ignited interest among many scholars (e.g., Bullinger et al., 2010; Lindberg et al., 2016). This
study focuses on crowdsourcing innovation contests, a common type of crowdsourcing form. In
such contests, organizations define problems they aim to solve and then broadcast and organize
these contests on intermediary platforms such as InnoCentive and OpenIDEO. Communication is
vital on these platforms (Lindberg et al., 2016). They often feature communication and
collaboration tools, which are essential in coordinating interactions among participants and
maintaining a community memory (Lindberg et al., 2016). Therefore, analyzing the mechanisms

2

behind communication and collaboration networks is critical in unveiling the organizing logics
behind contest-based crowdsourcing communities.
In this study, a specific type of crowdsourcing innovation contest is investigated:
coopetitive contests. In these contests, participants engage in cooperative activities such as
supporting and commenting on each other’s ideas to improve innovative outcomes. In addition,
ideators (i.e., the contest participants) are also competitors in the contests. A monetary prize is
often offered, and only a few ideas can win (Sun & Majzrak, 2020).
Although a growing number of scholars have focused on studying crowdsourcing
contests (Dahlander & Frederiksen, 2012), how participants interact for innovative purposes is
largely unexplored (Sun & Majchrzak, 2020, p. 10). Ecological and evolutionary theories help
provide theoretical perspectives on how different types of community members communicate
and collaborate, as well as how their interactions evolve. Inspired by Hawley (1950), academic
interest has been paid to applying ecological and evolutionary perspectives to examing human
activities. Organizational ecology and evolutionary approach emerged later, drawing inspiration
from Campbell’s (1965) work. Ecology theory helps communication scholars understand
interactions and interdependencies within and across organization populations, and evolution
theories tackle the mechanisms of entities’ emergence, transformation, and demise (Carroll,
1985; Freeman & Hannan, 1989; Hannan & Freeman, 1977). Evolutionary theories have also
been applied to studying mechanisms of network evolution by Monge et al. (2008).
Traditional ecologists have focused mainly on population dynamics. In other words, their
unit of analysis has been populations (Abbott et al., 2016), a concept that describes collections of
entities sharing similar forms and overlapping resources. These ecologists focus on aggregated
changes in populations of various forms and density (i.e., numbers of entities) (Abbott et al.,

3

2016), including both individuals (e.g., Xu, 2019; Davis, 2017) and organizations (e.g., Lowrey,
2017; Weber, 2012). Entities, in this research, will be used to indicate individuals or
organizations in an ecological community.
More recently, ecological studies have started to place more emphasis on entity-level
analysis, and on an exploration of organizational changes and strategic choices (e.g., Bigelow et
al., 2019; Wang et al., 2018; Wang et al., 2019; Zhao et al., 2018). Most of these studies focused
on organizations, but a few exceptions proved that the ecological and evolutionary perspective is
also useful in studying individuals (e.g., Davis, 2017; Shen et al., 2014a; Xu, 2019). For
example, Davis (2017) studied ex-combatants during civil wars and found defection was most
likely to be observed among individuals who were “demographically atypical” or were in the
niche overlap of different organizations in battles (p. 115). These theories have been increasingly
applied to studying individual dynamics in online communities (e.g., Shen et al., 2014a; Xu,
2019). For example, Shen et al. (2014) applied evolutionary theories to analyzing social ties.
They found social ties in online gaming evolve through the processes of “formation,
maintenance, and demise” (p. 1). Treating a crowdsourcing community as a multidimensional
and multilevel complex system which consisted of both humans and affiliated categories, Xu
(2019) applied ecological and evolutionary theories to studying a crowdsourcing community and
found they are useful frameworks in helping to understand cross-level network interactions
between individual artists and their affiliated platform categories. For example, he found that
category densities encourage individuals’ entry rate.
Network analysis has gained popularity among ecologists, especially among
communication scholars (e.g., Fu & Shumate, 2015; Lee et al., 2007; Weber, 2012; Yang, 2020),
who have emphasized the importance of studying social network structures using the ecological

4

framework (e.g., Margolin et al., 2015; O’Brien et al., 2019; Shumate & Lipp, 2008; Weber,
2012). For ecologists, the value in observing social networks is that niches, resources at which
entities survive and reproduce (Hannan & Freeman, 1977), and social networks are interrelated.
Niches shape network evolution, which in turn, affects network nodes’ resource access (Yang,
2020). Analyzing the evolution of social networks allows a better understanding of
communication and interaction patterns and helps scholars make accurate predictions (Margolin
et al., 2015).
Most ecological research studied unidimensional networks, which means they only
examined one type of relation and a single type of network node. With multidimensional
networks, multiple types of nodes and relations can be studied simultaneously (Contractor et al.,
2011). Multilevel networks refer to networks that explore not only within-level interactions such
as employees’ communication ties, but also across-level relations such as employees’ affiliation
to companies (Lomi et al., 2016). A few studies that adopted multidimensional and multilevel
network analysis have suggested that examining multiple types of nodes and relations
simultaneously offers novel and nuanced angles (e.g., Amati et al., 2019; Lee & Monge, 2011).
For example, Amati and colleagues (2019) found that interorganizational communication
structures are affected by organizations’ internal activities. Lee and Monge (2011) found that
multiple types of network relationships are closely associated. In fact, ecological theories are
inherently multilevel. Forces that change networks and populations of entities exist on several
possible levels. Attributes on the organizational level, population level, and even the
environment level (e.g., features of resources in the environment) can all affect the ecology and
evolution of a community. For instance, Kim et al. (2006) proposed a multilevel model to
explain network evolution based on structural inertia theory of organizational ecology (Freeman

5

& Hannan, 1989; Hannan & Freeman, 1984). In their proposed model, characteristics from
multiple levels—within the organizations, on the dyadic level, on the population network level,
as well as from the external environment—all influence network changes or the lack of changes.
Therefore, this study will apply a multilevel network analysis (or a fully multidimensional
network analysis; detailed definitions and distinctions can be found in Section 2.3.1.2), a type of
analysis that examines multiple types of network ties and network entities simultaneously, and
studies how forces on different levels or dimensions affect the patterns of communication
interaction in crowdsourcing innovation contests.
In addition to the ecological and evolutionary frameworks that shed light on how
ecological factors shape the communication patterns of the ideators (the competitors in the
contests), this study also examines whether community-oriented knowledge exchange exists
among contest participants. The contest community is conceptualized as a community of practice
(CoP), which describes that members active in contests form communities by sharing practices
such as innovating ideas and exchanging feedback to improve one another’s innovation
outcomes (Wenger et al., 2002). Communication networks are critical in their community
organizing (Wang et al., 2019).
In this dissertation, the ecology of crowdsourcing innovation contests will be studied
from three angles. First, this study adopts a multilevel network analysis angle to examine how
categories designed by problem-defining organizations enable or constrain communication of the
ideators. This angle is critical because it will provide insights on how social architecture shapes
the communication structures of the ideators whose idea submissions are affiliated with platform
categories. This study also has practical implications for platform managers and contest

6

organizers as they should be aware of the impact of platform designs on communication
interactions.
Second, this study also examines how communication networks evolve. Specifically,
within- and inter-population dynamics are included to study network evolution patterns. Two
types of populations and their networks are examined: ideators (i.e., the contest participants), and
visitors (i.e., those who do not participate in contests but engage in the innovation processes by
providing comments to ideators). Within the ideator population, this research studies two types
of crowdsourcing ideators: existing members, the experienced contest participants who are
familiar with the rules and norms of the platform, and new entrants, who are less experienced but
may potentially bring in diverse and novel information into the community. This angle helps
tease out potential interdependencies among different types of community members and how
these interdependencies affect network evolution. Findings contribute to the theoretical
development of the variation-selection-retention (V-S-R) framework that will be introduced in
Section 2.3.2. The role of interpopulation connections in network evolution will be better
understood. Moreover, the results of this study also provide strategic suggestions for ideators to
perform better on these platforms and occupy more advantageous network positions.
Lastly, in addition to the ecological and evolutionary patterns, whether the ideators share
community-oriented communication patterns is examined applying the CoP framework.
Interesting reciprocity patterns and factors motivating contest participants’ knowledge
contribution are explored and discussed. Previous studies on CoPs mainly focus on community
members’ communication within a single type of CoP group (e.g., Rivera & Cos, 2016; Gilbert,
2016; Wang et al., 2019), which may risk overlooking other interconnected groups’ potential

7

influence. Therefore, the ideator group will not be studied in isolation in this research. Instead,
whether ideators’ CoP networks are affected by the visitors will be examined.
Chapter 2: Literature Review and Hypotheses
2.1. Crowdsourcing Innovation Contests
The advance in media technologies has made it possible for a large crowd of individuals
to engage in solving problems collectively, communicatively, and simultaneously on the internet
(Brabham, 2013). Organizations have become increasingly interested in crowdsourcing, a
practice of tapping into collective intelligence through an open call on the internet. Through an
open-call format, problem-solving tasks are distributed on the internet, and are undertaken by
loosely connected individuals around the world (Howe, 2008; Kietzmann, 2017). The
assumption of crowdsourcing is that a large number of online crowd members possess collective
wisdom and intelligence (Lévy & Bononno, 1997; Surowiecki, 2004), the composition of which
is diverse and offers various heuristics (Jonassen, 2004). In addition to improving organizations’
performance by offering a knowledge and idea repertoire that is hard to find internally,
crowdsourcing reduces their innovation costs (Pénin & Helmchen, 2011).
More recently, some scholars have suggested that the benefits of crowdsourcing go
beyond reducing costs and improving the results of organizational innovation. It also leads to
favorable outcomes for crowdsourcing participants (Sánchez et al., 2015). Empowerment-
oriented crowdsourcing is defined as an “online collaborative and non-profit process open to the
participation of diverse citizenry in order to perform tasks whose final result brings social
benefits to the political, social, economic and/or environmental field, in which the knowledge
generated can be accessed, shared and reused freely” (Sánchez et al., 2015, p. 6). Projects on
OpenIDEO, the research site for this study, satisfy this definition (see Section 2.2). Projects like

8

these empower citizens in various ways, such as secured information access, inclusion,
participation, and the monitoring of the institutions’ or organizations’ accountability (Sánchez et
al., 2015). Among various forms of crowdsourcing, this study focuses on innovation contests, in
which problems are defined by organizations who broadcast and organize them on intermediary
platforms, such as OpenIDEO and InnoCentive (Howe, 2008; Liang et al., 2018; Sun &
Majchrzak, 2020).
As mentioned in the definition above, crowdsourcing contests involve three types of
parties: problem-defining organizations, intermediary platforms, and large crowds of problem-
solvers, who are often-times heterogenous and anonymous (Pénin & Helmchen, 2011). These
platforms are often characterized by a hierarchical structure under the management and guidance
of the problem-defining organization, coupled with the rules and social architecture enforced by
the intermediary platforms. They also feature “a community of contributors with self-organizing
social structures” (Blohm et al., 2013, p. 208). A vibrant and self-sustained community is crucial
for attracting a large crowd of contributors, saving resources for “platform management,”
creating community values and norms (e.g., trust and empathy), and developing “a self-
energizing cycle” of a healthy amount of contributions, evaluations, and comments (Blohm et al.,
2013, p. 208). Moreover, the ideas that are perceived as the best are selected based on both the
organizations’ criteria and the community members’ votes (Brabham, 2012; Howe, 2006, 2008;
Pénin & Helmchen, 2011). How participants communicate and collaborate within an organizing
form like this is understudied in the crowdsourcing literature, offering an exciting research
opportunity. Therefore, this research aims to understand the impact of the hierarchical platform
structure by studying how communication can be enabled or constrained by the social

9

infrastructure. Then, this research also aims to identify community members’ interaction patterns
over time as contests evolve over different stages.
There are three types of innovation contests: competitive, collaborative, and coopetitive
(Sun & Majchrzak, 2020). Specifically, competitive contests such as Kaggle and Topcoder (King
& Lakhani, 2013; Lakhani et al., 2010) rely on creating a highly competitive environment and an
extremely large crowd to encourage high-quality solutions. Collaborative innovative contests, in
comparison, encourage participants’ co-design and communication (Majchrzak et al., 2013).
Tasks are voluntarily taken by participants with different knowledge and expertise.
This study will focus on the third type of crowdsourcing contest: coopetitive innovation
contests, in which participants collaborate and compete simultaneously (Hutter et al., 2011). In
these contests, problems are defined broadly, and participants are encouraged to exchange
knowledge and information while competing over a prize (Füller et al., 2012; Leimeister et al.,
2009). Ideators submit their ideas for competition, while at the same time, they also
communicatively refine and improve each other’s ideas (Tucci et al., 2018). Interactions among
participants are highly encouraged, while constraints are also placed upon these interactions by
the underlying competitive forces. Communication features on these platforms allow knowledge
collaboration and information exchange, thereby creating interdependencies among participating
crowds of ideators and visitors (Faraj et al., 2011; Stohl, 2014; Yan et al., 2020). In
crowdsourcing contests, members can communicate by commenting on one another’s designs
and forming teams to co-design ideas. Such a phenomenon poses the question of how people
communicate on platforms like these, with mixed intentions to collaborate and compete.
The effect of communication among crowds can be both positive and negative to
collective judgment, depending on the network structure and composition of the crowds (Yan et

10

al., 2020). Decentralized communication with diverse and complementary expertise improves the
aggregated judgment of a crowd (Becker et al., 2017; Noriega-Campero et al., 2018). Such an
effect of communication is also called the collective learning effect (Yan et al., 2020). However,
communication can generate adverse effects through social influence bias, which means under
the social influence, individuals’ decisions are likely to be swayed by other people in the crowd
(Muchnik et al., 2013; Salganik, 2006).
In summary, the importance of communication is evident in crowdsourcing platforms,
and the mixed outcomes of communication make it important to fully understand complex
communication mechanisms in crowd-based platforms. In this study, communication network
structure will be investigated to achieve a good understanding of the social interaction and
communication outcomes in crowdsourcing innovation contests.
The relationship between communication and idea generation has been “long pondered in
the disciplines of sociotechnical systems, architecture, and urban planning” (Guth & Brabham,
2017, p. 4). In the field of communication, scholars have only recently started to analyze the
relationship between communication and idea generation in crowdsourcing (Guth & Brabham,
2017). Few past studies have used network analysis to examine crowdsourcing communication
within the field of communication. Shaw (2012), for instance, investigated social hierarchies in
an online participatory political blog community by studying how individual characteristics such
as experience, activity level, and status influence the number of recommendations a person
receives from peers. He found that in online open communities, users’ experiences and levels of
activity positively associate with the number of recommendation comments received. He
suggested the participatory blog community is both decentralized and centralized. It is
decentralized because there was no evidence to suggest that elite members shared a higher

11

reputation than other members. It is still centralized because there existed social closure among
elites and leaders. His study shows that analyzing the community’s social structure is crucial to
an understanding of the organizing form of online communities. However, while Shaw examined
structural centralization or decentralization, he did not investigate specific network structures.
Renard and Davis (2019) clustered users based on their social behaviors into four categories:
coopetitors, competitors, cooperators, and supporters. They emphasized the importance of
studying social interdependencies in crowdsourcing communities. They also analyzed how a
mixture of competition and cooperation positively contributes to creative results because it
allows participants to receive useful feedback without experiencing the negative impacts of the
competitive pressure. However, like Shaw (2012), although they emphasized the importance of
studying structural interdependencies of online communities, they did not study specific network
structures or mechanisms behind the question of how communication ties are formed, thus
overlooking nuanced communication patterns.
Moreover, many previous studies analyzing social behaviors in crowdsourcing
communities have investigated how individuals’ characteristics affect behavioral outcomes (e.g.,
Brabham, 2012; Shen et al., 2014). For example, Brabham (2012) found that crowdsourcing
participants have various motivations for participating beyond merely winning the contests and
earning monetary rewards. Ideators, the contest participants, are often motivated by “the
opportunity to advance their careers, to have fun, to express themselves, to contribute to a
collaborative effort, to gain peer recognition, and to learn new skills and knowledge” (p. 323) .
Shen et al. (2014) found that both altruistic and selfish intentions motivate online social
behaviors in online communities.

12

The aforementioned literature shares an individual-centric focus that informs us of some
important antecedents of people’s behaviors in online communities, but this focus runs the risk
of overlooking how environment-level factors, such as resource niches and population affiliation,
fit in the equation. Thus, applying ecological theories addresses the gap and sheds light on the
roles of environmental factors in online communities. Furthermore, although an increasing
number of social network studies have been conducted on open innovation platforms, few have
adopted ecological and evolutionary frameworks. Situated on OpenIDEO, an innovation contest
platform selected as the research site of this study, the general goal of the first section of the
research is to examine: (1) the structure and dynamics of communication networks of coopetitive
crowdsourcing communities; and (2) determine whether the ecological factors affect
communication network structure and dynamics, and if so, how. Understanding the relationship
between communication structure and environmental factors is valuable for two reasons. First, it
has the potential to reveal the dual processes of cooperative and competitive communication
patterns behind the ideation process in crowdsourcing contest platforms. Second, as discussed,
the results of this study could offer useful suggestions to the designs of these crowdsourcing
contest platforms and strategic management of the crowds. The research site will be introduced
in the next section (Section 2.2), and ecological and evolutionary theories will be reviewed
(Section 2.3).
2.2. Research Site
In this section, the research site is introduced. OpenIDEO, a well-known design
company, was established in 2010 to solve social issues. It is a representative innovation contest
platform (Sun & Majchrzak, 2020). Its principal activities revolve around contests. Each contest

13

lasts approximately three to five months, allowing community members to collaborate and
compete. At the final stage of the contests, winning solutions/submissions are selected.
OpenIDEO develops contests with sponsor organizations that are interested in supporting
different social causes. Together, they develop a question centered around pressing social issues
to attract ideas and discussions (“OpenIDEO FAQ,” n.d.). Every user who has an OpenIDEO
account can comment on the posts to exchange ideas. In this current study, community members
who have participated in contests are identified: ideators and visitors. Ideators are idea
contributors who enter contests to compete by submitting ideas, and can be identified by looking
at the “ideas” page of each contest. Visitors, in comparison, participate by contributing materials
for ideators for reference and providing feedback to others’ contributions in the comment section
without competing in the contest. Final winners will be selected among the ideators, not visitors.
On OpenIDEO, challenges often go through three phases: ideation, review, and
evaluation (or sometimes called refinement; see Table 1). Ideation is a stage when people
brainstorm new ideas collectively and build off others’ insights. All ideators submit the first draft
of their ideas by the end of the ideation stage. At the beginning of the ideation stage, some
contest organizations may decide to post opportunity areas, categories that are used to guide
directions of the idea development and categorize ideas (Sun & Majchrzak, 2020). Ideators
decide which category or categories to submit their ideas. During the review stage, community
voters anonymously vote on the ideas by clicking the “like” button. During this stage, comments
are also the most crucial channel for ideators to receive feedback and anonymous applauds,
which resemble the likes function on social media platforms such as Twitter. After the
conclusion of the review stage, the evaluation stage begins with the selection of finalist ideas.
The rest of the ideas lose the opportunity to win the contest. These finalist ideas will receive

14

more rounds of comments from the community to improve. In the end, the top ideas will be
selected and announced by the competition organizers, but their decisions may be affected by
community members’ comments on the posts, the number of anonymous applauds the idea
received, or other unknown factors that are not recorded on the website. In some contests,
winning ideas are also implemented offline. Commenting is an essential communication channel
throughout all stages of OpenIDEO. Users are allowed to follow comment threads and even
subscribe to receive email updates on selected comment threads to stay tuned to conversations
they find interesting (Wang, 2016).

Table 1. OpenIDEO contest phases
Contest Phases Tasks
Before Each
Contest
OpenIDEO develops contests with sponsor organizations
that define a question aiming to solve pressing social
issues.
During
Each
Contest
Ideation People collectively brainstorm and build off each other’s
ideas. By the end of the ideation stage, all ideators have
completed submitting the first draft of their ideas. Ideators
also decide to submit ideas in one or more “opportunity
areas.”
Review Ideators and visitors provide comments and feedback to the
ideas submitted during the ideation phase.
Evaluation
(Refinement)
A few ideas are selected into finalist group to receive an
additional round of comments and feedback. The rest of the
ideas are eliminated. Finally, winning ideas are selected by
OpenIDEO and sponsoring organizations.

In this research, because communication among the whole contest community is the
focus, only the first two stages (i.e., the ideation and review stages) will be studied when all
ideators actively compete in the contest. The evaluation stage will not be studied because this is
when the finalists are selected. While submissions that enter the finalist group continue to

15

receive more feedback for refinement, the role of the rest of the ideators resembles visitors who
only participate by commenting or they may drop out of the contest entirely.
To better illustrate what types of communication are involved in each stage, a contest
tackling the food waste social issue is described here as an example. The name of the contest is
“empowering caregivers in immunization innovation challenges.” This contest is sponsored and
created by the Bill & Melinda Gates Foundation. During the ideation stage of the competition,
ideators were encouraged to brainstorm and submit ideas. The organizers raised seven categories
to help guide people’s ideation towards their interested directions. In total, 99 ideas were
submitted to the ideation stage. Sylvanus Okumu and Anne Riitho collaboratively completed a
submission proposing to build a community network of volunteers to increase immunization
awareness. They submitted the idea in the opportunity area called “enhancing service quality and
accountability” to train frontline workers and enhance community engagement. During the
ideation stage, community members exchanged comments as well as brainstormed and built off
each other’s ideas. During the review stage, all ideators received a round of feedback through
comments. After the review stage, 33 ideas were shortlisted to receive another round of
evaluation (refinement), and finally, 9 ideas were selected to be the finalists. This project by
Sylvanus Okumu and Anne Riitho was shortlisted but did not win.
OpenIDEO is an emerging research site for social science and social network scholars.
For instance, Wang’s (2016) research was based on OpenIDEO. She examined the antecedents
for the collaboration ties and found that experts were preferred collaborators. Moreover, she also
found that diverse team composition led to better performance than teams with less diverse
configurations. Similarly, Fuger et al. (2017) analyzed how the “constellation of teams” affects
teams’ success (p. 439). They classified users and found four different types of community

16

members. Those who were active in commenting and low in contributing were called
collaborators. Those who were high on both commenting and contributing were categorized as
contributors. The third type of user was called allrounder, representing those who have moderate
comments and contributions. The rest were passive users who were inactive in commenting as
well as submission. They did not find evidence that diverse teams led to better performance, but
they identified collaborators as essential team members that boosted idea quality. In conclusion,
OpenIDEO is a rich platform suitable for social network applications targeted at observing
communication and collaboration behaviors. In the following section (Section 2.3), ecological
and evolutionary perspectives are introduced and hypotheses are raised.
2.3. Two Theoretical Perspectives: Ecological and Evolutionary Perspectives
Ecological theories study interactions and interdependencies within and across
populations and communities of organizations in their ecosystems, whereas evolutionary theories
focus on mechanisms of their emergence, transformation, and demise (Carroll, 1984; Freeman &
Hannan, 1989; Hannan & Freeman, 1977). Both theories share an origin in the biological theory
of natural selection (Darwin, 1859). Inspired by Campbell (1965) and Hawley (1944, 1986),
many scholars have applied ecological and evolutionary theories to studying organizations. Both
theories have received broader empirical support ever since. They have started to be adopted in
studies of group dynamics (e.g., Lai, 2014), gamers in gaming communities (e.g., Shen et al.,
2014a), and individuals in online crowdsourcing communities (e.g., Xu, 2019). Both theoretical
perspectives offer useful explanations for how populations interact and co-evolve. For example,
Lai (2014) studied mixed-mode groups that operate both online and offline, and she found that
the population density increases individual group’s competition and lowers survival rates.
External network ties offer groups legitimacy and resource access and increases their survival

17

chance. Xu examined intra-organizational ecology of the International Communication
Association (Xu et al., 2021), and found that membership density of interest groups decreases the
likelihood of members leaving the groups, because denser groups are perceived to be more
legitimate than other groups.
2.3.1. Ecological Theories
The concepts of niches and populations are at the core of ecological theories. Niches are
defined as “all those combinations of resource levels at which the population can survive and
reproduce itself” (Hannan & Freeman, 1977, p. 947). A niche is a multi-dimensional resource
space (Hannan & Freeman, 1977). The distribution of niche resources affects entities’ changes
(Aldrich & Wiedenmayer, 2019) and social networks (Monge et al., 2008). For example, niche
affects founding rates of organization populations. As the result of some organizations’ failure,
new niche spaces opened up, creating room for new organizations to emerge (Aldrich &
Wiedenmayer, 2019). Monge et al. (2008) raised that just as a niche has limited carrying capacity
of the number of entities operating within it, it also has limited carrying capacity for network
ties. The reason is that network ties require resources (e.g., time, energy, attention) to build and
maintain, without which network links tend to dissolve (Monge et al., 2008). Populations of
organizations rely on similar niches, and therefore often respond to similar environmental forces
(Hawley, 1950; Scott, 2004). For example, “the ratification of the United Nations Convention on
the Rights of the Child (UNCRC)”, an important event, has affected the entire community of
children’s rights nongovernmental organizations (NGOs; Margolin et al., 2015, p. 30). Through
the UNCRC, clear community norms were articulated, and NGOs’ legitimacy practices shifted.
The tendency to seek information from more experienced organizations reduced because

18

younger organizations no longer needed to seek information regarding legitimate practices from
older organizations (Margolin et al., 2015).
For online communities, membership, participation, and voluntary contribution are the
primary niche resources that maintain the running of online communities (e.g., Wang et al.,
2013; Shen et al., 2014a). For members of online communities, “people come together with
others to converse, exchange information or other resources, learn, play, or just be with each
other” (Kraut & Resnick, 2012, p. 1). The resources sought by different populations also differ.
In coopetitive crowdsourcing contest platforms, for ideators who participate in the contests, there
are two essential resources that improve their chance of winning: (1) knowledge and information,
and (2) understanding of norms and rules about the platforms. Communication is the main
activity of community members, and it is the channel of knowledge and information exchange.
For visitors, their motivation can be different. They are not competing for any prizes. Examples
of visitors’ motivation can include entertainment, societal accountability, and self-identities
(Nonnecke & Preece, 2001).
A population describes a group of entities with shared forms, which can be characteristics
such as structural configurations, functions, behaviors, or developmental processes (McKelvey,
1982). Population categorization is a crucial task to identify meaningful structure in the ecology.
Ecologists have not reached a consensus on how to operationalize populations in empirical
studies (Hsu & Hannan, 2005). Therefore, the strategy of empirically classifying entities varies
from study to study, depending on the research focus and aims. Four methods are used by
researchers.
The first method is studying populations using industrial categories, such as the
semiconductor industry (Podolny et al., 1996), the newspaper industry (Dobrev, 2001), and

19

breweries and telephone companies (Barnett, 1997). Industrial categories are examples of pre-
existing fundamental niches that do not overlap with one another. They are often stable and
widely acknowledged or even institutionalized categories (Freeman & Hannan, 1989; Hannan et
al., 2003; Podolny et al., 1996). Studies applying this strategy often focus on legitimation and
institutionalization of a market industry (e.g., Dobrev, 2001), or changes or characteristics of
specific market segments (e.g., Podolny et al., 1996).
The second population-defining method is grouping organizations with similar
organizational structures, routines, and identities (Aldrich et al., 2019) into the same population.
Lee and Monge (2011), for instance, studied organizations of the same type (i.e., “governmental
organizations, international governmental organizations, NGOs, and for-profit corporations” as
population members who occupy homophilous niche spaces (p. 766). Studies applying this
method often seek to understand cross-sectoral or within-sectoral relations among organizations
(e.g., Atouba & Shumate, 2010; Lee & Monge, 2011). For these studies, forces that shape
organizational behaviors often come from different organizational structures and different
institutional environments in which organizations operate (Fu & Shumate, 2015).
The third method is proposed by Hilbert and colleagues (2016). They presented the
possibility to partition socially connected individuals into a population and proved it as an
effective population-defining method of identifying selection forces in the environment. Lastly, a
group of scholars has adopted the idea that social identities are useful in defining populations and
organizational forms (e.g., Carroll & Swaminathan, 2000; Durand et al., 2017; Hannan et al.,
2019; Hsu & Hannan, 2005).
2.3.1.1. A Categorization Framework to the Ecology of Crowdsourcing Innovation
Contests. Categorization research is an extension of community ecology. Hannan and colleagues

20

suggested that social codes or identities should be “at the forefront of ecological analysis” when
defining populations (2007, p. 21). The definition of categories is the “groupings of entities that
simplify our apprehension of what surrounds us, focusing attention on a limited number of
dimensions or features, enabling recognition and action.” (Durand et al., 2017, p. 4). Their
population-defining approach is similar to other community ecologists in that they still retain “a
focus on features (defined broadly to include relations)” of the entities being studied (Hannan et
al., p. 31). However, this framework slightly differs in that it also considers “the social
construction of categories, forms, and populations” (Hannan et al., 2007, p. 31). They believe
that categories, a formal language being adopted, clearly define the ecology of fluid
communities. Unlike traditional ecological studies, it allows partial membership and category-
spanning. Because of the social nature of the category construction, entities can also shape the
ecology by socially proposing new categories (Hannan et al., 2019).
An example of that in OpenIDEO is the “other” category allowed in the contests. Ideators
can choose to submit ideas according to existing categories created by the contest organizers, or
propose their own and make an argument for it by submitting to the “other” category. Over time,
if the new categories are successful and recognized, they may be upgraded to a formal category
in the ecology if the organizer allows (Hannan et al., 2019). Once the categories become
institutionalized, certain norms, practices, and assumptions are expected to be associated with the
entities located in the given categorization (Durand et al., 2017). A social issue (e.g.,
humanitarian, environmental issue) can be an example of an institutionalized category that
denotes NGO members’ identities, because it “outlines what an NGO does and does not do and
forms the basis for external audience expectations” (Yang, 2020, p. 8). Categories also have
cognitive meanings because they influence audiences’ judgments of the entities being

21

categorized. Each established category has cognitive and “taken-for-granted” meaning to it and
attracts distinct types of interested audiences (Durand et al., 2017). For example, in the research
by Granqvist et al., (2013), company executives strategically adopt market labels that appeal to
specific audiences so that their companies are positioned in desirable market positions.
This line of research has been extended to studying individuals, instead of mainly
organizations which have been the focus of traditional ecology studies. Ecologists started asking
questions such as how producers of cultural products like music and films, or how individual
business sellers strategically occupy categories to achieve an optimal balance of being novel and
legitimate (e.g., Askin & Mauskapf, 2017; Zhao et al., 2018). Scholars interested in the tradeoff
of occupying a focused and specialized niche category versus spanning multiple categories drew
inspiration from niche width theory discussed below (Zuckerman et al., 2003). The reason is
individuals’ behaviors, like behaviors of organizations, are affected by the norms and taken-for-
grantedness of category niches.
Xu (2019) has also applied an adapted density dependence theory to studying how
categories’ density affects designers’ category entry. He found that the increases in a category’s
density, measured by the total grade of membership of the members, positively influence the rate
of entry of the category. The reason is that densely occupied categories tend to be perceived as
legitimate and tend to attract external members to entry. Xu and colleagues (2020) proposed that
although interorganizational ecology is well-studied (e.g., Bryant & Monge, 2008; Margolin et
al., 2015; Shumate & Lipp, 2008), “what is less understood is the ecological dynamics at the
intra-organizational level” (p. 5). They proposed that “an investigation of the evolution of
individual membership in subgroups” within organizations “can capture the ecological dynamics
at the intra-organizational level” (p. 5). Similarly, Xu (2019) studied an online crowdsourcing

22

community and found that the density of categories (i.e., niches for individuals) affects
individuals’ entry rate, which means individuals sense the overall structure of the community’s
environment and strategically manage their resource affiliation.
However, what has been understudied, is how the categories, or niches in online
communities, affect individuals’ social interactions. To fill the understudied gap, this study tests
whether categories can be studied as ecological forces that influence individuals’ communication
network ties, and if they do, what influences can be observed. Therefore, in innovation
crowdsourcing contests, populations can also be defined by the categories.
For OpenIDEO specifically, organizers identify several opportunity areas to help clearly
define the problems and guide the direction of idea generation (Sun & Majchrzak, 2020). For
online innovation contests, using opportunity areas to cluster ideators with similar interests is an
appropriate population-defining strategy that teases out ecological forces coming from contest-
designing organizations. In OpenIDEO, all submissions, based on their focus, are also
categorized into specific opportunity areas. Ideas that do not belong to any of the opportunity
areas are categorized as “other.” Submissions centered around each opportunity area address the
same problem from different perspectives. Pre-existing opportunity area categorization can be
used to group online community members into different areas of interest.
Members submitting to each category occupy different resource niches. For each interest
area, the focus is on a slightly different perspective of the problem, and submitters’ toolboxes of
knowledge domains or tactics (Wang & Soule, 2016) are also different. Audiences, consisting of
fellow ideators and visitors, also differ within each interest area (Hannan et al., 2019). Categories
defined by the platform are also “institutionally consolidated identities,” reflecting submitters’
niche identities (Hsu & Hannan, 2005, p. 478). Therefore, it is reasonable to use predefined

23

categories based on the content of each submission to define population boundaries, and
opportunity areas represent resource niches of each contest-based community. Using existing
categories to define population boundaries helps identify fundamental niches in the resource
space, matching the rationale for the first population-defining strategy mentioned above.
Hsu and Hannan (2005) propose to go beyond explicit categories by considering the
concept of identities, in which “codes represent default expectations held by audiences about the
organizational properties and constraints over properties” (p. 475). These identity categories can
be clearly stated or can be latent but widely recognized. The “implicit categories” are defined as
social codes that “capture the unarticulated norms of style and content,” which shape the
evaluation of the audiences in the online marketplace, “even when these categories are not
explicitly defined” (Sargent et al., 2020, p. 6).
In OpenIDEO, although the categorization of ideators and visitors is not clearly
announced, their roles and presences are very different in this community. Ideators, or contest
participants, are the main contributors to the contests who submit ideas to solve contests. Their
idea submission is also on their profile pages, emphasizing their ideator roles. In comparison,
visitors do not participate in the contest but leave their comments and suggestions on ideators’
posts. On visitors’ profile pages, they do not have idea submissions to specific contests, but their
recent comments are also present, reflecting their visitor identities. Therefore, it is very
straightforward for community members to distinguish the identities of the “ideators” and
“visitors” in each OpenIDEO contest. This will be another population-defining strategy applied
by this dissertation to studying the ecology of OpenIDEO. Note that their identities are not
permanent. An ideator in one contest can be a visitor in another contest and vice versa.

24

Within the contest-organized communities, in the current research, population-defining
strategies are used based on two research angles. First, opportunity areas (i.e., categories) are
used to define populations in this current section (Section 2.3.1) to explore the relationship
between category niches and social interactions. The second strategy is grouping ideators into
new entrants and existing members to study the network evolution of interdependent populations
of different evolutionary stages. This strategy is applied in Section 2.3.2. to study network
evolution and co-evolution of two populations of different evolutionary stages. Finally, in
Section 2.4, ideators and visitors’ interdependencies are studied. See Table 2 for population types
identified in this research.

25

Table 2. Population types
Level 1 Population A: Ideators
Population
B: Visitors
Level 2
Ecological Perspective Evolutionary Perspective
Population
B is not
subdivided
at this level
Population 1: Opportunity
Area 1
Population 2: Opportunity
Area 2
…
Population X: Opportunity
Area X
Population X+1: Undefined
Opportunity Area (Details of
this category can be found in
Section 2.3.1.3)
Population 1: New entrants
Population 2: Existing members

Finally, some scholars also study realized niches, “subspace of the fundamental niche”
(Carroll, 1985, p. 1267). Most scholars taking this path often apply niche width theory and
resource partitioning theory, two commonly applied ecological theories, to study generalists and
specialists (Aldrich et al., 2020). Generalists describe entities that occupy “a wide range of
environmental resources,” whereas specialists rely on a narrow niche space for survival (Carroll,
1985, p. 1266). For example, Carroll and Swaminathan (2000) found that the density of specialty
breweries increased in the presence of dominating generalist competitors. Instead of categorizing
populations into generalists and specialists, some other scholars have grouped them into entrants
and incumbents. Giustiziero et al. (2019), for instance, studied the competition and learning
between entrants and incumbents. They suggested that incumbents approach new entrants to
learn their technologies, but paradoxically, their investments in the new technologies make new
entrants retreat. Similarly, Weber (2012) examined how new entrants and existing organizations
affect each other’s performance and hyperlink network evolution in the news media industry. He
found that hyperlinking with the new entrants helps increase the page views of existing news

26

organizations. Unlike the previous strategy that focuses on industries, such a strategy allows
researchers to investigate resource partitioning within a resource space. In OpenIDEO, ideators
are also categorized as new entrants and existing members, which will be clearly defined in later
sections (see Section 2.3.2).
2.3.1.2. A Multilevel Extension of Ecology Theories. Driven by the availability of
digitalized data on human behaviors, and the increasing computational power to analyze more
complicated social structures, a growing number of studies have applied a multidimensional
network approach over the past decade (Contractor, 2009; Contractor et al., 2011; Ognyanova &
Monge, 2013; Su & Contractor, 2011; Wang et al., 2016). Multidimensional networks feature
more than one dimension by being multimodal or multiplex, or both (Contractor et al., 2011, p.
682).
Multimodal means networks can include multiple types of nodes. Such nodes are not
limited to humans but can also be non-human entities (Contractor et al., 2011). Contractor and
colleagues listed technologies, policies, and routines as possible examples of non-human entities
(2011). The increasing availability of cyberinfrastructure technologies such as categories,
hashtags, and rating systems in digital traces data has provided more research opportunities for
researchers to understand network tie formation patterns not only among humans, but also
between human and non-human agents (Contractor, 2009; Su & Contractor, 2011). The
multiplex criterion specifies that several different types of network linkages can be studied
simultaneously in multidimensional networks, and therefore, this feature can sometimes be
referred to as multirelational (Contractor et al., 2011; Ognyanova & Monge, 2013).
In multidimensional networks, multiple networks can be studied simultaneously among
multiple types of nodes, including human and non-human nodes. Human nodes form

27

interpersonal relations such as collaboration and friendship. Linkages among non-human entities
can be interorganizational collaboration ties, compatibility among technologies, similarity among
genres, etc. Relations can also form between human and non-human entities. Examples of such
network linages are affiliation ties between individuals and their employer organizations,
between individuals and technologies they adopt, or between individuals and music genres they
enjoy, etc. Contractor and colleagues (2011) also pointed out the existence of partially and fully
multidimensional networks. Partially multidimensional networks include: (1) multimodal uniplex
networks featuring multiple types of nodes but single type of relations, (2) unimodal multiplex
networks that include only one type of nodes connected by multiple types of relations, and (3)
multimodal and multiplex networks that have at least two types of nodes and relations. Fully
multidimensional networks need to satisfy stricter criteria by containing “multiple sets of nodes
and multiple sets of relations with relations both within sets of nodes and among all sets of
nodes” (Contractor et al., 2011, p. 695). To better illustrate different types of multidimensional
networks, Figure 1 displays hypothetical partially and fully multidimensional network with two
types of nodes: twitter users and hashtags they apply in their posts.

28

Figure 1. Hypothetical partially and fully multidimensional networks
Note. The orange linkages indicate their follower-followee ties. Because such a network consists of nodes of the
same level, it is also referred to as a within-level network. #MeToo and #WomenShould are both commonly used in
feminist social movements, while #savetheplanet advocates solving the environmental social issue. Therefore,
#MeToo and #WomenShould share a similarity tie, which is colored green in the visualization. Similarly, because it
consists of nodes that belong to the same level, it is also referred to as a within-level network. The blue links connect
the users and the hashtags they use. This network is called cross-level network because it indicates relationships
across the two types of nodes on two levels. Such ties are undirected and colored in blue and orange.

29

Multilevel network is a related but distinct concept that is easily confused with
multidimensional network. The idea of level in “multilevel” networks refers to “a set of actors,
or agents, and the levels are interdependent with respect to the conditions for action and/or
outcomes” (Lazega & Snijders, 2016, p. 4). Therefore, multilevel networks require the
identification of more than one level (e.g., individual employees and companies), and networks
within levels (e.g., network ties among employees, or inter-company networks) and across levels
(e.g., the affiliation ties between employees and companies) (Lomi et al., 2016). Lazega and
colleagues (2004, 2006, 2008) analyzed a multilevel network dataset on French cancer
researchers and their affiliated laboratories as an example to introduce this term (Lazega et al.,
2004, 2006, 2008). See Figure 2 for a hypothetical visualization of this network. This multilevel
network comprises lower-level agents (i.e., individual researchers), higher-level nodes (i.e.,
labs), and middle-level affiliation ties between researchers and labs. The lower-level network ties
are colored blue, representing researchers’ advisor-advisee relations. The higher-level network
(colored in green) features inter-lab collaboration ties. The orange links in between are the
affiliation ties indicating individuals’ lab membership, which can also be referred to as a middle-
level (or cross-level) network.

30

Figure 2. A hypothetical multilevel network

Networks tend to be embedded in “a hierarchical or nested structure” (Harary & Batell,
1981, p. 30). In other words, each network node is “in fact, a graph itself” (Moliterno & Mahony,
2011, p. 446). For example, in an interorganizational network, each organization is also a
network of intra-organizational “network of groups, departments, or divisions” (Moliterno &
Mahony, 2011, p. 446). This nested relationship between network levels, or cross-network
interaction, is essential when studying multilevel networks. Ignoring it risks overlooking
interdependencies across levels, as well as oversimplifying social systems (Lomi et al., 2016;
Paruchuri et al., 2019). In multilevel networks, the lower and the higher levels share a nested
relationship because lower-level nodes are embedded in higher-level network nodes such as
groups, organizations, or other collectives. Such cross-level relations can be complex in that they
are not simple “aggregation of ties” between the two levels (Lomi et al., 2016, p. 267). Such

31

nested relation can be whole or partial, which means individuals can be affiliated with one
collective (e.g., a group, an organization, or a lab in our example) or co-employed by several
collectives at the same time. The nesting relation also has a clear direction, because one cannot
say that groups, organizations, or other collectives are nested in, or affiliated to, individuals.
Multilevel network analysis is distinct from multilevel or hierarchical statistical models
(e.g., hierarchical linear modeling). For multilevel statistical models, the random effects in the
individual level and the group level are considered independent. For multilevel networks,
however, the nested relationship in multilevel networks is referred to as a cross-level network
relationship, such as affiliation, and the two network levels are, in fact, interdependent (Lomi et
al., 2016). Multilevel network analysis is also different from meta network analysis which uses
ERGM parameters estimated for each class and then assesses the overall within-class friendship
pattern in the given school (Lubbers & Snijders, 2007). Driven mainly by the development of the
multilevel ERGM approach (Wang et al., 2013), multilevel network analysis is “the analysis of a
specific multilevel network, directed to the study of a multilevel network data structure, with ties
between nodes at more than one level, and often ties between levels” (Lomi et al., 2016, p. 267).
Therefore, based on their different definitions, multidimensional networks include not
only multilevel networks that incorporates both within- and cross-level network (also called fully
multidimensional network) but also other partially multidimensional network types. In this study,
in order to understand individual crowdsourcing users’ interdependencies with one another and
with platform categories, a multilevel (or a fully multidimensional) network framework is
adopted. Both within-level social interactions among individuals, as well as cross-level
individual-population affiliation, are examined.

32

In summary, multilevel and multidimensional networks are frameworks that examine
networks from multiple levels or dimensions. They are extensions of traditional unidimensional
networks that oversimplify “the rich complexity that exists in most social networks” (Contractor
et al., 2011, p. 685). These extensions have the potential to enhance the development of many
theoretical streams (Contractor, 2009), including social network theory (Monge & Contractor,
2003), network society (Castells, 2011), actor-network theory (Latour, 2005), and more. In the
current study, a fully multidimensional/multilevel application will be conducted to examine the
ecology of online innovation contests (see Figure 3 for a network illustration for this research).
For simplicity, “multilevel” will hereafter be used to describe the multilevel and fully
multidimensional perspective.

Figure 3. A multilevel network illustration for the OpenIDEO ideator population

For ecologists, ecological forces from the environment can be studied at two different
levels: on the community level among populations (e.g., Audia et al., 2006) and on the

33

population level among members (e.g., Lai, 2014). Most ecologists have chosen to study only a
single level (e.g., Carroll & Wade, 1991; Doerfel et al., 2013; Hannan & Freeman, 1987).
However, processes on different levels simultaneously affect entity-level change, and hence
more comprehensive research designs are needed (Aldrich & Wiedenmayer, 2019). The more
recent applications of ecological theories (e.g., Bodin & Tengö, 2012; Hollway et al., 2017) have
suggested that ecological factors shape social mechanisms in online communities. For example,
Bodin and Tengö proposed a socio-ecological framework and found that humans sharing the
same ecological resources (e.g., land, forest resources) tend to be socially connected. Hollway
and colleagues (2017) identified that organizations’ specialty affiliation affects their social
relations. Using multilevel network analysis, they found that generalist hospitals were more
likely to initiate collaboration network ties than chance alone. Moreover, when collaboration is
reciprocated, hospitals with the same specialties share collaboration ties. However, network
effects across levels are not well-understood for online communities between humans and virtual
ecological resources. Therefore, questions such as how human interactions are affected by non-
human factors, how humans interact with the non-human factors in online systems, and how
social interactions shape resource distribution on these communities await to be answered.
Applying the multilevel framework to study online ecology helps answer these questions.
Although ecological forces can be studied at several levels, among ecological studies, a
multilevel application is novel and seldomly adopted, with a few notable exceptions. The first
exception is a multilevel ecological study conducted by Amati and colleagues (2019). Amati et
al. (2019) studied the network evolution of multilevel networks and analyzed within-organization
activities and inter-organizational network ties. Their findings advanced the understanding of
resource partition (Carroll et al., 2002). They indicated that generalist organizations were likely

34

to have more network partners, and organizations with fewer partners were more likely to
expand their niche portfolios. Their multilevel analysis provided empirical evidence that social
communication networks on one level and affiliation networks on the other level co-evolve,
which indicates that niche-width co-evolves with social networks (Amati et al., 2019). Xu (2019)
has also studied ecological theories from a multidimensional angle, contributing to the
understanding of networks with both human and non-human entities. He found that category
density increases artists’ rate of entry to that category. In conclusion, ecological forces shape the
evolution of social networks and should not be overlooked. Instead of only analyzing within-
level social networks among individuals, this study uses a multilevel perspective that also
considers the effect of ecological factors.
A multilevel and ecological framework is appropriate for studying online communities
for a few reasons. First, online communities such as crowdsourcing platforms are often
characterized by boundaries and category-spanning activities (Faraj & Shimizu, 2018).
Ecological theories have the potential to offer useful theoretical explanations of community
members’ relationships with online environmental resources. Second, the networks of many
online communities are inherently multilevel (Stephen et al., 2016). Moliterno and Mahony
(2011) raised a multilevel network theory of organizations, in which they propose that many
formal organizations are multilevel in nature. In organizations, network ties may exist between
people, teams, departments within organizations (Zappa & Robins, 2016). Similarly,
crowdsourcing platform users not only interact with one another but also with artifacts such as
categories (Xu, 2019). For OpenIDEO (i.e., the research site), similarly, two levels can be
identified: users and categories. Therefore, a multilevel approach is appropriate. Conducting a
multidimensional and ecological study on an online crowdsourcing community makes significant

35

contributions. Instead of merely examining social networks on a single level, observing multiple
levels simultaneously provides a more realistic picture. This approach is valuable because it is
able to observe mechanisms that are hard for traditional single-level network analysis to identify.
For example, Zappa and Robins (2016) used the multilevel network framework to study
organizational learning. They examined both individuals and their affiliated units, and found that
within-unit knowledge transfer is more likely to happen than by chance alone, and that cross-unit
knowledge transfer is likely to happen when the two units are already connected by work-flow
network ties. Such results are much harder to surface using single-level network approach. In this
study, a multilevel approach is adopted to studying the ecology of an online crowdsourcing
platform.
2.3.1.3. Relationship Within and Across Populations. Early application of ecology
theory was limited to single populations of organizations (e.g., Hannan & Freeman, 1977;
Hawley, 1986). Astley (1985) then suggested populations should not be examined in isolation.
Other adjacent or related populations should also be considered, and the entire community of
organizations should be viewed as a whole, thus inspiring the development of community
ecology. Community ecology is a theory that examines populations of entities interacting with
one another within a community environment (Aldrich & Ruef, 2020). In this study, community
ecology is applied, and both dynamics within and across populations are considered.
On the research site, during the review stage of a contest, submitters leave comments on
one another’s submissions to help improve these designs. As mentioned above, populations in a
crowdsourcing contest can be defined by pre-existing opportunity areas and an undefined area
called “other.” Figure 4 provides an example of opportunity areas.

36

Figure 4. An example of opportunity areas under an OpenIDEO innovation contest

Note. The overall problem being addressed in the contest was, “how might we as people on the move and neighbors
build bridges to a shared future of stability and promise.”

Ideators submit to the “other” category when their ideas do not fall within the predefined
categories. They are the “outliers” that are emergent and novel (Niederman & Lukyanenko,
2018, p. 149). Novel categories require substantial collective actions to acquire legitimation
(Alexy & George, 2013). Similarly, for ideators submitted to a new category that is not
predefined by the contest organizer, they have to make extra efforts to argue why their
submissions deserve to be considered equally as important as ideas submitted to predefined
categories. Members of emerging categories often suffer from loss of audiences’ recognition

37

because of the lack of legitimacy. For example, Alexy and George (2013) found that software
companies affiliated with emergent categories experienced lower stock prices than those
affiliated with existing categories. Previous research has also suggested that less legitimate
populations, such as young entrepreneurial organizations, are more likely to fail (e.g., Hunt,
2017; Singh & Lumsden, 1990).
The legitimation of emergent organizing forms (i.e., the undefined categories in this
study context) often calls for collective behaviors (Carroll & Khessina, 2006, p. 191). For
example, to establish new organizational forms, entrepreneurs often adopt collective actions such
as associations (McKendrick et al., 2003), campaigns (Hannan et al., 1995), and conferences
(Carroll & Swaminathan, 2000). Esparza and colleagues (2014) found that in an emerging food
truck industry, trade associations were formed to increase the legitimacy of the industry.
Collective activities also help educate the audiences (Swaminathan & Wade, 2001). The “social
movement” analogy is important for theorizing collective legitimation behaviors because it helps
interpret the collective and often irrational actions of members of a novel form (Tilly, 1978;
Carroll & Khessina, 2006). Members of the undefined population, therefore, are motivated to
pursue legitimation processes to gain recognition for their population forms (Hannan & Carroll,
1992; Weber et al., 2016). Here, a hypothesis can be posed that in online crowdsourcing
communities, it is likely that facing the pressure of legitimation, members of the emerging
category also tend to adopt similar collective behaviors. Because for OpenIDEO, comments are
one of the most important communication channels (as discussed in Section 2.2), here, it is
reasonable to hypothesize that such collectively oriented behavior of the undefined category can
be observed in comments. Comments on this platform, as mentioned, are often either endorsing
the ideas or sharing opinions to help improve the ideas. Comments that clearly announce

38

competition are rare. The review stage is when ideas are fully submitted and evaluated by
community members. Therefore, if comment ties are shared within populations during this
competitive phase, it is a legitimation. To examine whether the legitimation behaviors are
commonly seen among the “other” category members, the following hypothesis is proposed:
Hypothesis 1 (H1). During the review stage of the contest, within the undefined
population, ideators are more likely to share comment network ties than by chance alone.
2.3.1.4. Multilevel Closure. In this section, the investigation will be extended to a
multilevel angle by analyzing both interactions among ideators as well as ideators’ linkages to
opportunity areas. Multilevel closure describes multilevel clustering mechanisms (Amati et al.,
2019). Amati and colleagues’ study (2019) pointed out that two mechanisms potentially
contribute to multilevel closure: focal closure and membership closure. This is similar to the
long-standing debate of homophily versus influence (or sometimes labeled as selection versus
contagion) for single-level network studies (Leenders, 1995; Ma et al., 2014). For single-level
studies, this debate states that two opposing mechanisms—selection and influence—may help
explain why people who are close in social networks share similar traits. These two mechanisms
lead to the same outcomes, and scholars should carefully separate the two to identify which
affects the network evolution (Steglich et al., 2010). Selection, or homophily effect, states that
social entities tend to choose those who share similar traits to establish social connections.
Influence, or contagion, describes an opposite mechanism, in which the connected network
entities grow alike because ideas, characteristics, and practices travel through ties, and as a
result, network neighbors influence one another (Friedkin, 2001). Past literature studying the two
mechanisms on single-level networks has suggested that the mechanism varies per different
settings. For instance, Wang and Soule (2012) found the collaboration network of social

39

movement organizations was driven by selection rather than by influence. Powell et al. (2005)
also identified that the homophily effect drove the network evolution among organizations in the
biotechnology industry. Checkley et al. (2014), however, found that instead of selection,
influence effect drove venture capital firms’ network evolution in UK.
For multidimensional networks, a similar debate has not received much empirical
attention. Two similar mechanisms are: focal closure and membership closure. Focal closure
describes the situation in which that entities sharing similar network identities in one level are
likely to be socially connected on another level, implying a homophily effect. Apart from the
homophily effect, “being located in similar resource spaces would allow organizations to engage
in repeated and less costly interactions” (Lee & Monge, 2011, p. 764). In comparison,
membership closure explores the tendency for socially connected entities to share similar
affiliation actions on another level, which indicates an influence effect taking place. Burshell and
Mitchell (2017) analyzed firms’ decisions to enter an emerging niche. They pointed out that
when a large number of the focal firm’s socially proximate peers enter a new niche, it increases
the probability for the focal firm to enter. The reason is that peers’ entries signal to the focal
firms that the new niche may contain substantial opportunities. Peers’ entries also lower firms’
costs of adapting to the new niche, allowing them more access to information and strategies of
adaptation. In Burshell and Mitchell’s study, the authors used two different measures to
operationalize social proximity. They treated firms in the same market segments and
geographical proximate areas as being socially proximate. However, these measures are only
proxies. They did not analyze social networks, a direct indicator of social proximity, offering
research opportunities for network scholars.

40

As discussed, these two competing ideas have been tested in populations of organizations
and their internal activity portfolios (Amati et al., 2019; Burshell & Mitchell, 2017). In this
study, the two competing dynamics are tested in the innovation contest community. The goal, in
the context of the crowdsourcing contest community, is to answer whether it is the social
proximity that contributes to ideators’ similar problem-solving strategies (i.e., occupying the
same opportunity area), or it is the opportunity area affiliation that contributes to ideators’ social
proximity. Understanding which mechanism (i.e., focal closure and membership closure)
dominates the multilevel mechanism allows us to understand how ideators’ two decisions—(1)
which opportunity area to concentrate on, and (2) who to communicate—are connected. This
investigation is also meaningful in that it may provide insights on how ideators deal with
uncertainty: whether they address uncertainty by connecting with structurally equivalent similar
others or by imitating their network neighbors (Amati et al., 2019; DiMaggio & Powell, 1983).
Communication of two stages of a contest is studied: the ideation stage and the review stage.
In OpenIDEO, all categories are decided by the contest organizers and are formally
posted on the platform with detailed descriptions. During the ideation phase, ideators can choose
which category to submit their ideas to, and this is when ideators’ category affiliation becomes
clear. The ideation phase is followed by the review stage, when the ideators have completed the
submission, and community members (both ideators and visitors) start to review the ideas and
provide comments. Communication networks on this platform evolve as the contest stages
unravel. As suggested in categorization literature, categories are meant for audiences, such as
consumers, investors, and suppliers (Koçak et al., 2014). In this platform, categories are meant
for organization reviewers, community peers, or the public who read and evaluate these ideas.
Categories may shape the expectations of these audiences.

41

The focal closure effect is tested by the network evolution mechanism that ideators
submitting to the same categories are more likely to talk to each other than by chance alone. In
other words, it means when cross-level closure occurs in the community, comment ties are
formed after the idea submission ties. This mechanism indicates that idea submission drives
social interaction, and that homophily is a mechanism driving the multilevel network
configuration.
Hypothesis 2a (H2a). Individual ideators with the same category affiliation are more
likely to share comment ties than by chance alone in the ideation and review stage.
Membership closure is also tested, which reflects the mechanism that ideators who
exchanged interactions during the ideation stage are likely to be affiliated to the same category.
In other words, when cross-level closure exists, at least one idea submission tie occurs after
comment ties, which indicates that social interactions drive idea submission behaviors.
Therefore, membership closure is a different mechanism in comparison to focal closure
described in H2a. This hypothesis is only posted for the ideation stage because in the review
stage, category submission has completed, and the formation of multilevel closure can only be
driven by focal closure, not membership closure. The next hypothesis is:
Hypothesis 2b (H2b). Individual ideators sharing comment ties are more likely to share
the same category affiliation than by chance alone in the ideation stage.
2.3.2. Evolutionary Theories
Evolution occurs through three processes: variation, selection, and retention (abbreviated
hereafter as V-S-R) (Aldrich & Ruef, 2006; Campbell, 1965). The conceptualization of these
processes provides researchers with tools to explain both “dynamics and outcomes” (Aldrich et
al., 2020, p. 19). Potential directions of evolutionary changes arise from the variation processes,

42

during which entities develop new routines and activities in response to change. The opposite of
variation is inertia, meaning resistance to change (Hannan & Freeman, 1977, 1984, as cited in
Monge & Contractor, 2003). During selection, certain variations are chosen over others.
“Selection criteria are set through the operation of market forces, competitive pressures, the logic
of internal structuring, conformity to institutionalized norms, and other forces” (Aldrich & Ruef,
1999, p. 21). Finally, the retention process describes that the selected variations that are
sustained. This is when new variations over-write existing rules, and new norms are established.
Monge et al. (2008) promulgated the idea that ecological and evolutionary theories
should consider not only entities of organizations or individual human communicators, but also
the networks among them. They suggest that just as an environmental resource niche has a
carrying capacity, which means the number of organizations it can support is limited,
communication networks also have a carrying capacity, with “a limited number of links” they are
capable of supporting (Monge et al., 2008, p. 457). Monge et al. (2008) suggest that like
populations, networks also go through V-S-R processes. During variation, entities experiment
with links, through which they acquire resources such as information, knowledge, and advice.
Over time, network entities form preferences for with whom they communicate, a process
considered as selection. These preferences are later institutionalized as norms, and some network
connections become formalized and retained. Therefore, retention is also defined as “the process
of institutionalizing a selected variation” through “biases and habits via formal rules or
procedures” (Monge et al., 2008, pp. 460–461). This theory offers theoretical explanations of the
formation, selection, sustaining, and decay of network ties.
Tie fitness is at the center of this theory. While network links can generate resources such
as access to valuable and novel knowledge and information, their formation and maintenance

43

also require resources, such as time and effort. Because of the costs of each tie formation and
maintenance, network ties also have carrying capacities. Such a concept is called relational
carrying capacity. Network entities selectively form ties, and only some of these ties are chosen
to be long-term connections. Each network link also varies in fitness. Network fitness describes
“the propensity for a relationship to sustain itself, that is, to survive or to reproduce itself”
(Monge et al., 2008, p. 462). Network ties with better fitness have a higher probability of being
selected and retained. Network fitness can be determined by the extent to which network nodes
depend on resources provided by the given connections, whether the connections are easy to
maintain, whether they hold potential future benefits, or are decided merely by the strength of the
ties (Monge et al., 2008). Moreover, fit ties tend to be copied in the future. For example,
although by the end of an academic project, the collaboration between two researchers may
terminate temporally, this termination does not indicate poor network fitness. If this
collaboration is reproduced in future projects, it is one of several possible indicators that the
given network tie between the researchers is fit.
Most previous studies applying V-S-R only studied the network evolution of a single
population and overlooked the influence of other populations. For example, Margolin et al.
(2015) examined the network evolution of a population of children’s rights NGOs. Similarly, Fu
(2019) studied the tie evolution of environmental NGOs. Shen et al. (2014a) analyzed the
network evolution of game players. Although they classified the community of players into new
entrants and existing players, they studied them as two independent populations without
analyzing specific inter-population influences.
However, populations co-exist in a community, and they co-evolve through relationships
such as symbiosis and commensalism (Aldrich & Ruef, 2006). Carroll and Wade (1991)

44

identified the interdependencies between urban and rural microbreweries without directly
running network analysis. They found that initially, as urban breweries grew in number, the
founding rate of the rural breweries increased, and the mortality rate dropped, indicating that
they shared a mutualist relationship. Similarly, Walker and colleagues (2011) studied
membership and nonmembership advocacy organizations and found that the density of
membership organizations increased the founding of nonmembership organizations, which also
reflected the mutualist relationship between the two types of organizations. These studies pointed
out that the populations are interdependent. However, such interdependencies are not relational
as they did not directly analyze network connections.
Only a few studies studied population interdependencies by directly observing networks.
Meyskens et al. (2010), for example, conducted exploratory research on social ventures and
found their symbiosis network with other organizational populations such as corporations and
governments provide them with adequate resource access. Interdependencies such as symbiosis
and mutualism also affect their network evolution. A seminal research study conducted by
Powell and colleagues (2005), for example, examined the network evolution of the
biotechnology community over 12 years. Using a multidimensional and longitudinal design, they
identified that micro network structural changes are closely related to macro-dynamics, including
market shifts and institutional pressures. Their study was multiplex as they examined several
types of network ties (i.e., financial ties, R&D collaboration, commercialization, and licensing)
in the analysis. It was also multimodal by examining five different types of organizations (i.e.,
biotechnology firms, government institutes, medical corporations, universities, and venture
capital). They analogized organizations’ network partner selection as choosing dancing partners
in a dance hall. When the music changes, people switch dance partners. For organizations, when

45

the environment shifts, their network partners change as well. Such network shifts also shaped
the overall organizational field. Their results reflected that the community also experienced
comparable network evolution similar to V-S-R. For example, they found that initially,
organizations sought partners with diverse portfolios and central entities that were well-
connected. This was a phase similar to network variation in which entities experimented with
diverse network partners. As the environment shifted, organizations selected partners of different
functionalities. For instance, in 1988 and 1989, motivated by venture capital organizations’
preference for research and development (R&D), biotechnology companies changed their
preference for commercialization partners and established ties with more R&D partners. Such
shifts assimilate the selection process in network evolution. Eventually, leading entities formed
dominating and cohesive network clusters, which is like the retention process. In summary, they
found that organizational networks went through evolutionary processes that resembled V-S-R,
and populations of organizations in different sectors were interdependent, and their network co-
evolved. More recently, Weber (2012) also found that establishing symbiotic ties with emergent
populations allowed media organizations to access new information and technology, and over
time, they were more likely to be selected as network partners.
In the following section, different populations in an innovation contest community are
defined, and hypotheses regarding their V-S-R processes will be raised to study how networks in
the two populations are related and how social networks of populations co-evolve.
Two Contest-Based Populations. Two types of members are present in most online
communities: community inhabitants who participate in community activities frequently and are
familiar with the community, as well as visiting members who barely know any community
norms and rules. Scholars studying public goods have separated these two types of community

46

members as maintainers and visitors (Bighash et al., 2018), or as core and periphery members
(Safadi et al., 2018). In innovation contest platforms, similar categorization applies. Contest
participants serve more as the maintainers of the platform who participate in the contests and
provide ideas to the platforms. In contrast, some people serve as a visitor role. These people are
registered users, meaning that they have community memberships, which sets them apart from
the hidden audiences who merely read posts on these platforms without contributing any
comments. Visitors are registered users who participate not by submitting ideas to compete in
contests but by commenting on ideators’ ideas, which is also important for maintaining the
community. They communicate with the ideators by commenting on their posts. Therefore, in
this section, two populations with diverse community roles are studied: ideators and visitors.
Within the ideator population, there are two types of members. Most online innovation
contest platforms have some degrees of temporariness, in contrast to other formally defined
organizations whose structures are characterized by stable staffs, strong norms, task
modularization, hierarchy, or formal leadership (Majchrzak & Malhotra, 2016; Malhotra &
Majchrzak, 2014). They often feature temporary crowds, the definition of which is “groups of
strangers coming together for a short period of time for particular purposes, disbanding once the
period ends or the purposes are achieved, and with minimal normative effects created by a
shadow of the future” (Majchrzak & Malhotra, 2016, p. 686). In innovation contest platforms
such as OpenIDEO, each contest is independent in that it revolves around a distinctive problem
proposed by a different organization. Contests also share some degrees of interdependency
because they are presented and organized on the same platform with consistent rules, norms, and
procedures. Based on this logic, apart from temporary crowds who participate in only one contest
briefly, a crowd of ideators who are more persistent in the community should also be given

47

research attention. By identifying these two types of ideators, this study observes the organizing
structure of innovation contests that is a mixture of persistent norms and traditions, as well as the
flexible and temporary nature of each contest. Moreover, past studies on innovation contests
have either focused on serial ideators (Bayus, 2013) or temporary ideators (Majchrzak &
Malhotra, 2016), which ignores their possible influence on one another. This study will fill this
gap by analyzing the patterns and dynamics of communication interaction of both temporary and
serial ideators.
In ecological terms, these two types of community members can be referred to as existing
members and new entrants (e.g., Aldrich et al., 2020; Weber, 2012). In an OpenIDEO contest,
specifically, ideators who submitted to past contests are members of the existing population.
These are serial ideators who are experienced with the underlying norms and traditions of the
platform. In contrast, all the new ideators are new entrants who have not previously participated
in any other contests. They bring diverse external knowledge and ideas to the community
(Majchrzak & Malhotra, 2016). They are unfamiliar with the platform norms, at least upon entry.
The boundary of the two types of ideators may become increasingly permeable as the contest
unfolds. The reason is that as new entrants learn by doing, they may become more familiarized
with the platform norms. Previous research has found that new entrants of a community have less
experiences, and as a result, they are more explorative (March, 1991), and they are more active
in building social networks than community members with longer tenure (Shen et al., 2014a; Xu,
2019). In comparison, even though existing members have more experiences, their creativity
may be constrained by previous experiences (Bayus, 2013).
Populations do not exist in isolation, and there are eight types of inter-population
relationships (Aldrich et al., 2020; see Table 3 for definitions and examples). The first six types

48

are commensalist relationships. Commensalism is defined as “the existence in the population of
common interests or similar tasks that can be pursued more effectively when two or more like-
acting units pool their energies” (Hawley, 1986, p. 36). Commensalist relationships between
entities sharing similar or overlapping resource niches range from competition to mutualism: full
competition, partial competition, predatory competition, neutrality, partial mutualism, and full
mutualism. (1) Full competition between two groups describes a competition relationship that is
harmful to both parties. (2) Partial competition harms one group of entities, but it does not affect
the other group. (3) Predatory competition, in comparison, is a relationship that is beneficial to
one group while being harmful to the other group. (4) Neutrality means both groups stay neutral
to each other. (5) Partial mutualism refers to a relationship that is beneficial to one group and
neutral to the other group. (6) Full mutualism describes full cooperation between two groups that
are beneficial to both parties. The second type of interpopulation relationship is symbiosis. (7)
Symbiosis is described as “mutual dependence based on functional differences” (Hawley, 1986,
p. 36). In ecological terms, it represents a supportive relationship between two species, like an
anemone and clownfish (Bryant & Monge, 2008). Entities from two distinct populations that
occupy different resource niches can exchange resources and show support through symbiotic
relationships that benefit both parties (Brittain, 1994). (8) The third type is dominance, which
indicates that one population has complete control over the other population.

49

Table 3. Eight relations in the community ecology (Adapted from Aldrich et al., 2020, p. 251)
I. Commensalism
(−, −) Full competition: growth in each population detracts from growth in the other;
for example, competition between voluntary associations for members from the same
socio-demographic groups (McPherson, 1983)
(−, 0) Partial competition: relations are asymmetric, with only one harming the other;
for example, right-wing newspapers increased the failure rate of centrist papers in
interwar Vienna (Barnett and Woywode, 2004)
(+, −) Predatory competition: one party expands at the expense of the other; for
example, sharecropping and share tenancy arrangements developed at the expense of
plantations in the postbellum South (Ruef, 2004)
(0, 0) Neutrality: populations do not affect each other; for example, medical marijuana
dispensaries thrive despite the arrival of recreational cannabis outlets (Hsu, Koçak, and
Kovács, 2018).
(+, 0) Partial mutualism: relations are asymmetric, with only one population benefiting
from the presence of the other; for example, lesbian and gay advocacy groups
experience more success in counties with businesses oriented toward lesbian and gay
consumers, but advocacy success does not lead to more businesses (Negro, Perretti,
and Carroll, 2013).
(+, +) Full mutualism: two populations in overlapping niches benefit from the presence
of the other; for example, small and large railroads and telephone companies benefited
from the other’s presence (Barnett, 1995; Dobbin, 1994)
II. Symbiosis
(+, +) Symbiosis: two populations are in different niches and benefit from the presence
of the other; for example, venture capitalists make profits by investing in high-
technology firms, thereby enabling both populations to grow (Brittain, 1994)
III. Dominance
A dominant population controls the flow of resources to other populations (Hawley,
1950); effects depend on the outcome of commensalistic and symbiotic relations
Note. Signs in parentheses refer to the effect of one population, A, on a second population, B:
+ positive effect
0 no effect
− negative effect

As discussed earlier, in crowdsourcing contest communities, there are two types of
populations (see Figure 5). Ideators are the first population type, consisting of existing
participants and new entrants, who are both contest participants competing in the contests. Their
resource niches are similar. The types of knowledge, expertise, information, and requirements
that are needed for winning a contest are similar. They also compete for the same awards of the

50

contests. Therefore, these two types of players share a commensalist relationship because of their
shared resource dependence.
In comparison, visitors do not rely on the same resources. As mentioned, visitors’
contributions are driven by entertainment, societal accountability, and self-identities (Nonnecke
& Preece, 2001). Visitors bring in an important outsider perspective to the innovation in the
contest, and they provide unbiased comments that are not affected by the competition because
they are not competing for prizes. Therefore, the relationship between visitors and ideators is
symbiosis. Ideators provide learning and entertaining materials for the visitors, and visitors
provide comments to help maintainers improve their ideas.

Figure 5. Population types in crowdsourcing contests and interpopulation relations

51

This study aims to tease out how networks of interdependent populations co-evolve.
After clearly defining the two populations, hypotheses will be proposed in the following
sections.
Network Variation. Network variation, by definition, is experimenting with a large
number of connections or new and different network ties (Monge et al., 2008). Previous
empirical research measured variation as a large network size (Shen et al., 2014a) and as diverse
network connections outside the ego’s population (Doerfel et al., 2013). They found that
variation is more evident when significant shifts happen in a community that alter the norms in
the environment and create uncertainty (Bryant & Monge, 2008). “Entrance and exit of
populations (are also) the main factors in the increase or decrease of this possibility” (Bryant and
Monge, 2008, p. 169). Moreover, Doerfel and colleagues (2010) have observed that
organizations tend to experience network variation after a disaster.
In a contest-based community, although, the platform has norms that persist across
contests, each contest is still organized by different organizations and is centered around an
entirely different question defined in distinctive ways (Sun & Majchrzak, 2020). Therefore, it is
expected that in the first stage of the contest, namely the ideation stage of each contest, ideators
experiment with network ties by interacting with a large variety of members to get familiarized
with the contest and other ideators. Therefore, network variation may exist in the ideation stage.
The first phase of each challenge is when ideators expose themselves to diverse ideas, get
inspired, and locate potential collaborators (Sun & Majchrzak, 2020). It can also be deemed as a
phase when network variation activities are high for innovative contest ideators.
Shen and colleagues (2014) have also found that new entrants’ networking behaviors
follow the pattern of network variation. Upon entering a new environment, in comparison to

52

existing entities, newcomers tend to experiment with a large number of ties to familiarize
themselves with the environment. More experienced entities are more selective of whom to
communicate with, and they prefer to maintain a smaller network. Similarly, Doerfel et al.
(2013) found that organizations with low resilience or challenging resource availability tend to
experiment with ties after disasters. Newcomers’ resource availability is more challenging
because they do not have as much knowledge about the platform in comparison to the existing
populations. They know fewer people to provide information, feedback, and support. Therefore,
they are more motivated to seek network variation in comparison to existing members.
Hypothesis 3 (H3). During the ideation stage, the first stage of a contest, new entrants are
more likely to be senders of comment ties (in general, including both within- and cross-
populations) in comparison to existing members.
During the review stage, the stage after idea submission, is when the ideas are evaluated.
Competition intensifies in this stage because this is when ideas receive feedback from the
community users, and by the end of the review stage, some ideas are shortlisted while the others
fail (“OpenIDEO FAQ,” n.d.). The contest stages that are specified and enforced on the platform
present an excellent opportunity to observe how environment shifts may affect network
evolution.
At the beginning stage of the contest, new entrants are still very unfamiliar with the
community. They may seek mutualist relationships with existing members to help them gain tacit
knowledge about the platform, learn the norms, and increase their legitimacy. Younger entities
also tend to connect with more experienced ones to “gain knowledge of what is legitimate
through the imitation of older, more established partners” (Margolin et al., 2015, p. 37). For
existing members, mutualist relationships with the emergent population provide novel

53

information and diverse knowledge they are not likely to be exposed to when they communicate
with other existing members (Majchrzak & Malhotra, 2016). Therefore, the following hypothesis
is:
Hypothesis 4a (H4a). During the ideation stage, comment ties are more likely to exist
between existing members and new entrants than by chance alone.
Network Selection. Most previous research applying the network V-S-R framework
(Monge et al., 2008) focused on analyzing mechanisms of tie selection. Both endogenous and
exogenous factors contribute to the selection process of network ties (Wang et al., 2016).
Endogenous factors describe the “inherent structure of the network itself, the properties of the
relations that comprise the network, such as mutuality, transitivity, and clustering” (Monge et al.,
2008, p. 464). Take transitivity as an example. Previous scholars found that network transitivity
is a stable structure that helps foster norms (Stephens et al., 2005), and is often formed when
entities in the networks seek stability and legitimacy (Margolin et al., 2015). Preferential
attachment is also a common signature tested in network evolution (Margolin et al., 2015; Wang
et al., 2016). It describes the phenomenon in which highly connected network entities tend to
receive more connections in the future (Merton, 1968).
Environmental factors affect the direction of network evolution. For instance, network
ties are more likely to form when resources are abundant in the environment (Wang et al., 2016;
Yang, 2020). The underlying explanation is that ties need resources to build, and when resources
are abundant, carrying capacity is higher. Margolin et al. (2015) compared NGOs’ network
patterns before and after changes of the institutional logic of their environment – the codification
of standards. They found as the uncertainties reduced in the environment, the gaps between
experienced and less experienced organizations decreased. They found, as a result, the tendency

54

for the older organizations to be favored in the network lowered. Similarly, as crowdsourcing
contests unravel, and new entrants are increasingly familiar with the contest community during
the review stage, they become less dependent on existing members for knowledge and
experiences. Also, as the review stage unravels, ideators start to receive more feedback on the
ideas, and they become gradually more certain about the community. Review stage, the second
stage of the contest is selected to observe network selection. The reason is that this stage is when
ideators get increasingly familiarized with the environment, and selection process may start to
happen when they decide to select some ties over others. Therefore, the following hypothesis is
as follows:
Hypothesis 4b (H4b). During the review stage, new entrants no longer maintain more
comment ties with the existing members than by chance alone.
Because mutualist ties between existing members and new entrants satisfy both parties’
informational needs, they may also benefit their performances. Weber (2012) found that
symbiotic relationships positively predict the number of future network connections of the focal
news organization because more information access is bridged between the two populations.
However, to the best of the author’s knowledge, the effect of mutualist network ties on future
network selection has never been studied. Following Weber’s (2012) logic, links connecting
existing members and the new entrants give them better access to knowledge and diverse
information (p. 191). These ties help enhance the profiles of the focal entities, which makes them
preferred network partners and attracts network connections in the future (Weber, 2012).
Therefore, the following hypothesis is:

55

Hypothesis 4c (H4c). Ideators’ increases in the number of mutualist ties exchanged
between the new entrants and existing members in the ideation phase lead to increases in the
likelihood of receiving comments in the review stage.
Retention. As discussed, retention is “the process of institutionalizing a selected
variation” (Monge et al., 2008, pp. 460–461). Unlike many other platforms in which long-lasting
relationships (e.g., friendship, follower-followee relationship, mentor-mentee relationship) can
be easily observed, OpenIDEO does not have an infrastructure that allows users to establish
formal and lasting relationships. In addition, as discussed, each contest is temporary and only
lasts about three to five months. Different contests also feature distinct themes and center around
diverse social issues, so they attract different crowds of participants. Therefore, network
retention is very hard to observe in OpenIDEO, and this process is not examined in this current
research.
2.4. Crowdsourcing Innovation Contests as Communities of Practice
For coopetitive challenges, “both competition and cooperation play critical roles in
generating innovation by establishing a balanced approach that encourages participants to not
only work for competitive outcomes, but also transfer their knowledge and learn from each other
in order to ultimately evolve collective knowledge” (Sun & Majchrzak, 2020, p. 9). The
ecological and evolutionary approach adopted in the previous section explores ideators’ network
behaviors that “work for competitive outcomes” (Sun & Majchrzak, 2020, p. 9), or in other
words, are relatively self-interested. In this section, to get a more comprehensive view of the
crowdsourcing contest community, ideators’ collective learning and knowledge creation will be
explored. In this section, this study explores: (1) do ideators’ interactions indicate a form of a

56

community of practice? And if so, (2) what are the factors motivating contest participants’
knowledge sharing with competitors?
Community of practice (CoP) describes “a group of people who share a common set of
problems, or a passion about a topic, and who deepen their knowledge and expertise in the area
by interaction on an ongoing basis” (Wenger et al., 2002, p. 5). A CoP is organized by shared
practices, through which “ways of doing things, views, values, power relations, ways of talking”
are developed over time (Eckert, 2006, p. 2). It has three main elements: domain, community,
and practice (Wenger et al., 2002). Domain describes the situation in which community members
have a shared commitment and interest that evolves. Both members and relationships among
them constitute the community. Practice describes the situation in which members of a CoP are
joined informally by doing things, and practice leads to the development of shared norms and
meaning systems (Wang et al., 2019). This concept was initially developed to study collective
learning in informal settings, such as engineers learning from each other in a water drilling task,
and a group of consultants exchanging marketing strategies (Wenger & Snyder, 2000). Wenger
and Snyder (2000) suggested that managers in companies should identify CoPs, provide
infrastructure for community-building, and assess the value of CoPs. CoPs can help enhance
companies’ strategic capabilities by generating knowledge and renewing such knowledge. They
argued that maintaining a CoP is like raising a “golden goose” that lays golden eggs (Wenger &
Synder, 2000, p. 143).
More recently, scholars have extended the theoretical scope to analyzing online
communities such as online newsrooms (Weiss & Domingo, 2010), a cluster of wineries (Wang
et al., 2020), social movements (Wang et al., 2019), For example, Weiss and Domingo (2010)
studied digital newsrooms as CoPs. They identified that the innovation of digital news depends

57

on the “complex interactions between the members” who “enter and leave, participate and
observe, learn and teach, communicate and negotiate” (p. 1168). Wang and colleagues (2020)
pointed out that current CoP studies are “primarily descriptive and exploratory” (p. 619). They
studied the wine industry in Ontario and found that a winery’s CoP engagement positively
contributes to their performance and that geographically remote wineries perform not as well as
their peers, but CoP engagement improves their performance. Wang et al. (2019) analyzed
network dynamics in a Twitter social movement. They found that Twitter social movement
grows more connected over time, increasingly resembling CoPs, and that serial participants
function as brokers in the network. This line of research often takes a longitudinal perspective
and studies how a CoP is organized and evolved (e.g., Eckert, 2006; Rivera & Cox, 2016). The
contest community shares the features of CoP in a few ways. First, in each contest, individuals
interested in addressing the same question posed by the contest organizers join together to share
ideas and exchange feedback. Therefore, they share the same clearly defined domain: their
commitments to solving the problems posted in an innovation contest. Second, it is organized by
people’s voluntary practices such as idea submission as well as knowledge sharing and
endorsement through comments. Finally, ideators also share experience in and commitment to
the relevant social issue, just as many other CoPs (Eckert, 2006).
However, crowdsourcing innovation contests are also different from the typical CoPs in
several ways, pointing towards important gaps in the CoP studies. Traditional CoP studies have
largely focused on communities with similar motivations and collective understandings (Rivera
& Cox, 2016). However, interactions across different types of groups in CoPs have been largely
ignored in previous research. Online crowdsourcing contest communities, for instance, feature
multiple different interconnected populations of individuals holding diverse aims. Most CoP

58

studies focus only on the evolution of one group of individuals, which ignores the influence of
other interrelated groups operating in the same ecology. In fact, as Gilbert (2016) has rightly put
it: “as CoP moves online, there is increased potential for more expansive and diverse
membership” (p. 1214). Therefore, this study identifies two populations of community members
and examines their interdependencies.
Specifically, in crowdsourcing contest communities, a CoP consists of both ideators who
compete in the contests and aim to win the prizes, as well as visitors who do not compete in the
contest but leave comments and feedback to the contest participants. They have different
identities and motivations. The contest participants are more deeply involved in the community,
and they need to play by the rules of the contests. They often participate either because they wish
to win, learn, or entertain. In comparison, visitors may only participate with the goals of learning
or entertaining, and therefore, they are less competitive. Including them both in the analysis
speaks to Wenger et al. (2002)’s argument that a CoP is often maintained by core and peripheral
network members. This research, therefore, examines a CoP consisting of both ideators and
visitors, and aims to understand how the existence of the visitor population influences the
ideators’ knowledge contribution.
Wasko and Faraj (2005) took a step further and theorized virtual CoP as a “network of
practice” to describe large, loosely connected, and geographically dispersed individuals’
practices. They defined a “network of practice” as “a self-organizing, open activity system
focused on a shared practice that exists primarily through computer-mediated communication”
(p. 37), which also resembles the visitor populations on OpenIDEO. This concept highlights the
essential role of studying networks in CoPs.

59

Wasko et al. (2009) studied knowledge exchange in an online community. They found a
reversed pattern which showed that CoP network structure features generalized knowledge
exchange, which means “one’s giving is not reciprocated by the recipient, but by a third party”
(p. 258). Generalized reciprocity contributes to public goods maintenance (Wasko et al., 2009).
They argue that individuals do not know one another in online communities, and contribution
behaviors are voluntary (Wasko et al., 2009). Poquet and Dawson (2018) studied the structural
patterns of a Massive Open Online Course platform and found different networks feature diver
network patterns. They suggested that when individuals of the community share significant
direct reciprocity, they tend to behave out of self-interest. In comparison, communities that
feature indirect reciprocity are often solid, altruistic, and contain “pay it forward” norms (Poquet
& Dawson, 2018, p. 52). Finally, communities that feature significant triadic closure tend to be
cohesive (see Figure 6). Understanding which exchange pattern exists in the community is
crucial because it allows us to understand how the community is organized, evolved, and
sustained (Wasko et al., 2009). To fully understand the organizing form and norms of knowledge
exchange among contest participants, as well as how such norms shift over time on coopetitive
platforms, the first research question explores:
Research Question 1 (RQ1). As each contest phase unravels, how do the reciprocity
patterns of the ideators population shift?

60

Figure 6. An illustration of the three reciprocity patterns in Poquet and Dawson (2018)

Wasko and Faraj (2005) found that individuals with high knowledge exchange activities,
measured by high network centrality, were more likely than others to share in the network of
practice due to their “habit of cooperation” (Wasko & Faraj, 2005, p. 41). First, to replicate
Wasko and Faraj’s finding within a community of individuals who share similar identities (i.e.,
lawyers that exchange legal advice in their study), a hypothesis is posted below:

61

Hypothesis 5a (H5a). Over time, ideators who are more central in the ideator comment
network are more likely to send out comment ties to other ideators in the future.
To address the aforementioned theoretical gap in the CoP literature, another hypothesis is
posed to study the influence of the visitor group on the ideator group. Similar logic applies.
Cross-group interactions may also be beneficial for ideators to foster a cooperative habit.
Moreover, visitors are not competitors in the crowdsourcing contest, and therefore as mentioned
in Section 2.3.2, they share a symbiosis relationship with the ideators. Being embedded in the
visitor network can expose ideators to cooperative norms, and as a result, they are more likely to
comply with cooperative norms (Rogers & Kincaid, 1981; Wasko & Faraj, 2005).
Hypothesis 5b (H5b). Ideators who have exchanged (both sent and received) more
comments with visitors in the ideation phase are more likely to send comment ties to other
ideators in the future.
Chapter 3: Methods
3.1. Data Collection
The research site is OpenIDEO, a contest platform that aims at solving social issues such
as education, human rights, poverty, and health concerns. On this site, NGOs, big companies,
and the platform organize social issue-based challenges. A recent example is a challenge that
was inspired by the ongoing COVID-19 pandemic. The platform organized it to answer the
following question: “how might we rapidly inform and empower communities worldwide to stay
safe and healthy during the COVID-19 outbreak”. All community members can comment on
each other’s designs. Finally, based on the organizers’ selection, top ideas are selected into the
finalist stage. Ideators’ and visitors’ profiles contain rich information to indicate their
performance. They are evaluated based on their expertise, winning history, submission history,

62

teamwork, and other criteria. There are two types of challenges: one that aims at connecting
NGOs and funders, in which designs are mostly closed to the general public, and one that targets
general audiences (Gaggioli, 2019). Due to that the interest specifically lies in open innovation,
this study only collected data on challenges tailored for the public because both ideators and
visitors are allowed to join public contests.
A challenge with predefined opportunity areas needed to be selected for analysis. Out of
the 60 completed challenges, 35 public contests have opportunity areas specified by the contest
organizers. Some profiles are marked as “deleted users,” indicating that given users are no longer
active on the platform and their profiles are deleted. Behaviors of delete users are impossible to
record because all delete users’ profiles are empty. Moreover, older challenges often have more
delete users who would create more errors for analysis. Therefore, the most recent challenge with
predefined opportunity areas out of the 35 challenges was selected for analysis. Through
preliminary estimation, out of 430 ideators, only one ideator’s profile was deleted. The following
challenge was analyzed: “how might we as people on the move and neighbors build bridges to a
shared future of stability and promise?” The contest was organized by the OpenIDEO platform in
collaboration with GHR Foundation, a philanthropic NGO working on global development,
education, and Alzheimer’s prevention. Three opportunity areas were predefined in this
challenge to address post-disaster communities suffering from poverty. The three areas were:
“the full human journey”, “beyond survival to potential,” “equipping diverse communities,” and
“others” (see Figure 4 for detailed descriptions of each category). The first opportunity area
specifically focused on designing stages of migrants’ movement, and helping people make
informed decisions. The second area focused on providing education so that immigrants have
opportunities to learn and improve themselves. The third area focused on improving community

63

solidarity, hospitality, and diversity. The “other” contained ideas that are not defined by existing
categories. All submission pages, the opportunity area each submission was affiliated with, all
comments during the ideation and review stage, and comments in reply to other comments were
collected using a Python-based package called “Selenium.” It is a powerful package which
controls a web driver that mimics users’ web-scrolling actions and allows automatic extraction of
long web materials spanning several pages on the research site.
3.2. Network Construction
Comment Networks in the Ideation and Review Stages. All comments shared in the
ideation and review stage were collected. Then, ideators were identified. Among all ideators
(Nideators = 429), during each contest stage, if they shared at least one comment with one other
ideator, they were included in the network. Two directed networks were constructed: a comment
network during ideation stage, and a network during the review stage. If ideator i commented on
ideator j’s submission, there was a comment communication tie sent from i to j. The ideation
network contained 277 ideators who shared at least one comment tie with another ideator, and
the review network consisted of 284 ideators.
Two Multidimensional Networks (H2a, H2b). In this study, two network dimensions
were identified. In multilevel network studies, the lower level network is often referred to as
Network A, the higher level network as Network B, and the cross-level network between two
levels is labelled as Network X (e.g., Wang et al., 2013). In this multidimensional network,
similar labels are used. A multidimensional network was constructed (see Figure 3 for the
network composition). The one-mode network of ideators and comment ties are referred to as
Network A. There were two one-mode networks that contained comment ties during the ideation
stage and the review stage. These two networks were the same as the “comment networks in the

64

ideation and review stages” mentioned above. The bipartite cross-level network that contained
the links between ideators and opportunity areas are referred to as Network X. When an entity
submitted an idea to a challenge addressing an opportunity area, there was a tie between the
ideator and the opportunity area. Note that in this setting, there was no connection among
opportunity areas, and therefore, Network B did not exist. Both networks used to test H2a and
H2b followed the same cross-level closure structure (see Figure 8). The network signatures used
to test H2a and H2b are presented in Figure 8. Although in this network, opportunity areas were
treated as independent, and they do not share ties, this network is still a multidimensional
network because it satisfies the criteria of being both multiplex and multimodal.
3.3. Analytical Procedures
Ecological Perspective (H1, H2a, & H2b). H1 explored whether members of undefined
populations were more likely to give each other comment ties than by chance alone (see Figure
7). ERGM was used to test H1. ERGM is a widely adopted network analysis method that
captures the interdependencies among network nodes (Pattison & Wasserman, 1999; Robins et
al., 1999; Wasserman & Pattison, 1996). ERGMs calculate estimates of the likelihood of various
network configurations, also called network signatures, using Markov chain Monte Carlo
(MCMC) maximum likelihood estimation (Shumate & Palazzolo, 2010). This method also
allows observation of the effect of exogenous variables. Each ideator’s population affiliation was
included as the exogenous variable in the ideator network to test H1. It will be referred to as one-
mode ERGM, which describes ERGM that examines structures of unimodal and uniplex
networks. Using this term also helps distinguish this method from the multilevel ERGM

65

introduced in the following paragraph. To test H1, the review network (Nideators = 284) was
examined using one-mode ERGM.

Figure 7. The network signature being tested for H1

Note. Blue squares represent network nodes. Squares with orange shades represent members of the undefined
population. This network structure tests that ideators in the undefined population are more likely to share a tie than
by chance alone.

Multilevel Exponential Random Graph Modeling (ERGM) was used to test H2a and H2b.
Multilevel ERGM is an extension of one-mode ERGM that models the structure of multiple
networks simultaneously so that both within- and across-network structures can be examined
(Wang et al., 2009). H2a argued that ideators sharing the same opportunity areas were more
likely to share comment ties. H2b predicted that ideators sharing comment ties were likely to
submit to the same categories. The network signature included in the model to test the cross-level
closure pattern described in H2a and H2b is shown in Figure 8. During the ideation stage, if the
cross-level closure structure existed more than by chance alone, the sequence of comment ties
and cross-level idea submission ties of all cross-level closure structures should be further
examined. If cross-level affiliation ties preceded comment ties, it means focal closure was at play
(H2a). If affiliation ties were formed before communication ties, it means the membership
closure was present (H2b). Then, a chi-square test was adopted to compare whether the
frequencies of observing focal closure and membership closure were significantly different.

66

During the review stage, all ideators’ submissions were completed, and all comment ties were
formed after the idea submission. Therefore, if cross-level closure existed more than would be
expected than by chance in the review stage, focal closure would be present during the review
stage (H2a would be supported during the review stage). See Figure 9 for visualization of the
two mechanisms (i.e., focal closure and membership closure). MPnet software (Version 1.04)
was used to fit the multilevel ERGM modeling (Wang et al., 2014).

67

Figure 8. Cross-level closure network signatures being tested for H2a and H2b

68

Figure 9. Visualizations that help distinguish the mechanisms of H2a and H2b

Note. Blue squares represent ideators and yellow circles represent categories (i.e., opportunity areas). Ties
connection ideators and categories are idea submission ties, and those connecting ideators represent their comment
ties. Solid lines occur before the dotted lines.

Evolutionary Perspective (H3, H4a, H4b, and H4c). H3, H4a, H4b, and H4c were
examined using one-mode ERGM, which is the same method applied to test H1 in the previous
section. H3 inquired whether new entrants were more active than existing members in the
network. H4a hypothesized that comment ties were more likely to exist between existing
members and new entrants than by chance alone during the ideation stage, and H4b predicted
that comment ties were less likely to exist between existing members and new entrants during
the review stage than by chance alone. To test H3, H4a, and H4b, each ideator’s identity (i.e.,
existing member, new entrants) was added as a categorical nodal covariate. H4c predicted that
the number of mutualist ties exchanged in the ideation stage would be positively associated with
the likelihood of receiving comment ties in the review stage. The number of mutualist ties was

69

added in the model as a continuous nodal covariate. The visualization of the signatures that
tested H3, H4a, H4b, and H4c were included in Figure 10. H4a and H4b can both be tested using
the mismatch signature shown in Figure 10. To test H3 and H4a, one-mode ERGM was applied
to the network for comments during the ideation stage (Nideators = 277). To test H4b and H4c, the
review network was studied (Nideators = 284).

Figure 10. Network signatures being tested for H3, H4a, H4b, and H4c

Note. Blue squares represent network nodes. Squares with orange and green shades describe new entrants and
existing members, respectively. The large node size describes nodes that exchanged a high number of mutualist ties
between the new entrants and existing members. Squares with dashed borders indicate that the attribute (i.e., how
many mutualist ties were exchanged across populations) of the given node is not important.

70

Relational Event Modeling and Community of Practice (RQ1, H5a, and H5b). H5a
and H5b examine how comment networks evolve over time, and they were tested using
Relational Event Modeling (REM). The reason for using REM is that communication ties are
short-lived events instead of relationship states such as friendship or collaboration ties, which
violates the assumptions of the Stochastic Actor-Oriented Model (SAOM; Snijders et al., 2010).
REM is a powerful and flexible longitudinal network modeling tool capable of recording each
relational event between every pair of people and the specific time when each event happened or
the order of each event (Brandes et al., 2009; Butts, 2008). The model is based on a rate function
(Schecter & Contractor, 2017) which indicates that “the rate of an event represents its pace over
time; more frequent events have a higher likelihood of occurring, relative to events with a lower
rate” (p. 4). Individuals’ exogenous traits, time-varying, or static variables, can also be added
into the model to consider whether a sender effect or a receiver effect exists that influences the
odds of the relational event.
Two directed comment networks connecting commenters and post owners during the
ideation and review stage were constructed. As described in Section 3.2, ideators who shared at
least one comment tie were included in the ideation (Nideators = 277) and review (Nideators = 284)
networks. Relational events were directed comment ties. Each event was both the dependent
variable of all previous events, and the independent variable predicting all future events (Welles
et al., 2014). For REMs, the R package relevent was used for analysis. The sequence of each
comment event was recorded and ordered for this model. H5a hypothesized that ideators central
in the ideator network during the inspiration stage were more likely to send out comment ties in
the future. Because the network was directed, centrality can be tested by indegree and outdegree
centrality. Therefore, this hypothesis was tested using the “NIDSnd” and “NODSnd” in the

71

“relevent” package, which capture normalized indegree and outdegree centrality of each node to
predict the possibility for future sending rate (see Figure 11 for the visualization). H5b indicated
that the total number of comment ties exchanged with the visitor increases the possibility of the
visitors sending out comment ties to other ideators. The number of ties exchanged with visitors
(both sent and received) of each node was added as a covariate to test the nodes’ future sender
possibilities (see Figure 11 for visualization).

72

Figure 11. Network signatures for H5a and H5b (REM Model)

Note. The bold line(s) indicates the first event(s), while the dotted line(s) represents the following event(s). The
large size of the node describes that the focal node had received a high number of comments from the visitors.
Squares of dashed borders were used here to indicate that attributes of the given nodes were not important.

73

RQ1 asked which type of reciprocity pattern existed in this community. Several
reciprocity patterns (i.e., triadic closure, direct reciprocity, and indirect reciprocity) were tested
by adding the network structures shown in Figure 12.

Figure 12. Network signatures for RQ1 (REM Model)

Note. The bold line shows the first event, and the dotted line(s) indicates the following event(s).

74

3.4. Measures
Ideators and visitors. To prepare the analysis of all hypotheses and a research question,
it is crucial to first differentiate community members who have participated in the contest as
ideators and visitors. Ideators were defined as authors of the submitted ideas in each contest.
Visitors, in comparison, were the rest of the community members who did not submit ideas but
offered suggestions through comments. In total, 429 ideators were identified from the contest. A
total of 303 visitors shared comments during the ideation stage, and 366 visitors commented
during the review stage.
Population membership based on submission content categorization (H1). The
categorization of each idea can be extracted from the platform. Based on the categorization
information of the ideas, ideators were grouped into different populations defined by opportunity
areas (see Section 2.3.1 for the reasoning of this population-defining method). Among the
opportunity areas, one theme was called “other,” which is an undefined category, featuring ideas
that are out of the framework defined by the contest organizers. A categorical exogenous nodal
attribute was created, with members belonging to the undefined population marked as 1, and
members of other populations marked as 2. Among the 429 ideators, 164 submitted to the
“Beyond Survival to Potential” category, 109 were affiliated with the “Equipping Diverse
Communities” category, 52 were affiliated with “the Full Human Journey” category, and 44
submitted to the “other” category, leaving 111 who did not submit to any of the categories.
Among the ideators affiliated with categories, 51 submitted their ideas to more than one
category.
New entrants and existing members (H3). Ideators were coded as new entrants if they
had not participated in any previous contests. Otherwise, they were coded as existing members.

75

This is a categorical variable. Among the 429 ideators, 330 were new entrants, and 99 were
existing members.
The number of mutualist ties exchanged between the new entrants and existing
members in the ideation phase (H4c). This variable was added in the one-mode ERGM as a
nodal covariate to test H4c, which predicted that the number of ties exchanged across the two
populations during the inspiration stage increased the chances of receiving comment ties during
the review stage. For each network node, the number of mutualist ties exchanged across
populations during the inspiration stage was calculated. For example, if the focal node was a new
entrant, the total number of comment ties with an existing member was calculated, and vice
versa. Human coders verified that these comments were not negative critiques to rule out
possible comments that were not mutualist. Comments mentioning what can be improved in the
idea did not count as negative critiques. In total, 697 messages that were exchanged between
existing members and new entrants during the ideation phase were identified. To test intercoder
reliability, 209 messages (30% of the 697 messages) were coded by the author and another coder
into three categories: mutualist, negative, and unknown. Among the 209 messages, 1 message
was coded by the two coders as unknown because it was short and unclear. Two coders agreed
that 207 messages were mutualist. Because of the disproportionate percentage of the mutualist
messages, the expected disagreement would be extremely low. Traditional intercoder reliability
methods (e.g., Krippendorff’s alpha, Cohen’s Kappa) are not meaningful in this case. As a result,
percentage agreement is reported instead. Two coders’ percentage of agreement was very high
(99.52%). Finally, 686 mutualist messages were identified. This variable was calculated for 284
ideators (Mean = 4.37, SD = 23.67) who shared at least one comment tie during the review stage.

76

Total number of comment ties exchanged with visitors during the ideation stage
(H6b). The total number of ties exchanged with visitors (both sent and received) of each ideator
during the ideation stage was calculated (Mean = 41.44, SD = 262.71) and added to the model as
a nodal covariate. This variable was standardized so the strength of the effect can be compared
with other variables.
Control Variables
Previous interactions. Previous empirical research points out that past network structure
predicts future network structure (e.g., Pilny et al., 2016; Quintane et al., 2014). Therefore, prior
relational ties need to be considered in model fitting. This control variable of comment ties
during the ideation stage was added in the review stage models.
Tenure. Network members’ tenure is closely related to their level of familiarity with the
environment (Ohly et al., 2010). Ideators who had been on this platform longer were more
familiarized with the rules and norms in comparison to the new ideators. Knowledge of
community norms and rules are also important resources. Research has shown that individuals
with longer tenure tend to be favored network partners because they may possess greater human
capital (Chipidza & Tripp, 2018). Ideators’ tenure was measured by the number of days since the
day they joined the OpenIDEO community. Tenure was calculated for 275 ideators who shared
at least one comment tie during the ideation stage (Mean = 288.11, SD = 530.91), and 284
ideators (Mean = 286.59, SD = 516.74) who shared at least one comment tie during the review
stage. This measure was standardized so that the strength of the effect can be compared with
other variables.
Experience. Similarly, ideators with higher experience values have participated in more
activities such as submitting ideas, commenting, and collaborating on the platform. This variable

77

also indicates ideators’ familiarity with the platform, and it was added to all models. It was
calculated for 275 ideators (Mean = 62.24, SD = 183.28) who shared at least one comment tie
during the ideation stage, and 284 ideators who shared at least one comment tie during the
review stage (Mean = 60.29, SD = 179.42). This measure was also standardized so that its effect
can be compared with other variables.
Chapter 4: Results
The goal of this dissertation is to understand the network structure and evolution of
OpenIDEO contest communities. To achieve this goal, the dissertation examines the platform
from three theoretical perspectives: the ecological, evolutionary, and community of practice
theories. To start with, this research aims to examine and demonstrate how ecological factors
affect social interactions is understudied in terms of using a multilevel approach. Such an
approach allows individuals and category niches to be studied simultaneously in order to
examine how platform categories affect individual-level interaction. Detailed theoretical
contributions and implications will be elaborated in the discussion section.
4.1. Ecological Perspective
Three hypotheses (i.e., H1, H2a, and H2b) were developed to study the platform from an
ecological perspective. H1 was conceptualized to examine the interaction patterns of the “other”
population, and H2a and H2b were proposed to inquire how population affiliation and
communication patterns are associated by studying two cross-level closure mechanisms. H1 can
be tested by treating ideators’ population affiliation to be “other” or not as an exogenous binary
nodal covariate in the ideator network. Therefore, H1 was tested using a one-mode ERGM
model. H2a and H2b were tested using multilevel ERGMs because these two hypotheses needed
to be examined using a cross-level network structure (see Figure 8).

78

4.1.1. One-mode ERGM Results (H1)
Model Fitting & Goodness of Fits (GoFs). Because H1 hypothesized that ideators
submitted to the “other” category were likely to share ties during the review stage, one-mode
ERGM models were fit to study the structure of the review network. A forward selection
approach was adopted for the model fitting. First, only variables that test hypotheses and the
control variables were added to the model (see Table 4, Model 1). Then, more network structures
were added to improve the model fit. The best model was reported as the final model for
hypothesis testing (see Table 4, Model 2). In Table 4, all network structures were visualized. In
these visualizations, blue squares represent ideators. When these blue squares are shaded, it
indicates that categorical variables were added in the model. All yellow squares indicate that a
continuous variable was added. The size of the nodes represents the size of the continuous
attribute. Squares with dashed boundaries are nodes whose attributes were not important. To
explain the notation, the variable indegree(exp) and indegree(tenure) were testing whether nodes
with high experience values and tenure were likely to receive ties or not. Absolute differences
(exp) and (tenure) were added to study whether the magnitude of the differences between two
nodes’ attributes influence the likelihood of them sharing a tie. Orange ties indicate that edge
covariates were added in the model.

79

Table 4. One-mode ERGM results for the review network (H1)
Model 1
(Hypothesized and
Control Variables
Only)
Model 2
(Final Model)
Notes
Structural features
Edges

-3.47 (0.04)***
Outdegree 0

8.35 (0.28)***
Indegree 2

0.87 (0.17)***
Edge covariates
Past network
(ideation stage)

0.61 (0.16)*** 1.17 (0.15)***
Nodal covariates
Node match (“other”
opportunity area;
H1)

-2.96 (0.59)*** -1.00 (0.52) H1 not
supported
Control
Indegree (exp)

6.71 (0.08)*** 0.09 (0.04)*
Indegree (tenure)

0.88 (0.03)*** -0.03 (0.03)
Absolute differences
(exp)

-9.10 (0.11)*** -0.06 (0.04)
Absolute differences
(tenure)

-1.55 (0.04)*** 0.06 (0.02)***
Null Deviance: 111419
on 80372 degrees of
freedom
Residual Deviance:
20719 on 80366 degrees
of freedom
AIC: 20731
BIC: 20787
Null Deviance: 11419 on
80372 degrees of freedom
Residual Deviance:
10646 on 80363 degrees
of freedom
AIC: 10664
BIC: 10748

Note. p* < .05, p** < .01, p*** < .001. Values in the parentheses are standard errors. Blue squares are ideators.
Shaded blue squares indicate that categorical exogenous variables were added, while yellow squares show that
exogenous continuous variables were added. The large node size describes that the continuous attribute to be large,
and the small node size shows it is small. Attributes of squares with dashed outlines were not important. Orange ties
were ties with edge covariates.

80

The GoFs of the two models were compared. As shown in Table 4, the null deviance of
an empty model was 111419 on 80372 degrees of freedom. In Model 1, the residual deviance
was 20719 on 80366 degrees of freedom. The difference between the null and residual deviance
of Model 1 was 90700 on 6 degrees of freedom, and based on the χ
2
distribution, p < .001. Model
1 had a better model fit than the null model. Model 2 had the residual deviance of 10646 on
80363 degrees of freedom. The differences between the residual deviance of Model 1 and 2 was
10073 on 3 degrees of freedom, and p < 0.001. AIC and BIC statistics are GoF statistics that
penalize the number of predictors, and they are often used in model comparison. The smaller the
AIC and BIC, the better the model is. Even though Model 2 was fit with more variables than
Model 1, it had smaller AIC and BIC than Model 1. Therefore, Model 2 had a better model fit
than Model 1. Finally, the explained amount of variance in the outcome variable was 100773
with 9 predictors in Model 2.
Other GoF statistics of the two models were reported in Figure 13 and Appendix A.
Using the converged model, the ergm software simulated 1000 random graphs and provided a
comparison between the observed and simulated statistics. The logic for both the GoF plots and
tables in Appendices is the same, which is to provide comparison between the observed and
simulated values. GoF plots used boxplots to show the distribution of values for simulated
networks, and solid lines to indicate the value for the observed network. If the solid line is not
extreme in the boxplot, it means the model provides a decent representation of the observed
network. For the GoF tables, using the model estimates, the mean values of the simulated
networks are calculated and reported, and the observed values are reported. This table provides
the comparison between these two values. The closer the mean of the simulated statistics and the
observed statistics, the better the model is in describing observed networks.

81

The discrepancy between the observed and simulated graph statistics of model statistics
as shown in Appendix A was small for both Model 1 and 2, so they both had adequate
estimations for the variables included in the model. This can be further confirmed by the model
statistics plots in Figure 13. Figure 13 contains GoF graphs of model estimates and the
distribution of simulated graphs. We would expect that the summary for the features in our
observed data, represented by the solid lines in the plot, would not be extreme in the distribution
of all graphs, presented as box plots. If all observed values were within the interquartile range of
the simulated graph distribution, the model estimates were fit well. Apart from the model
statistics, GoF plots (see Figure 13) also show how well the model captured the global network
features, including degree distribution (i.e., indegree and outdegree) and shared partner
distributions. In comparison, Model 2 described the indegree and outdegree distribution better
than Model 1 because the solid lines for the GoF graphs of the Model 2 matched the pattern of
the distribution of simulated networks much better than Model 1. GoFs on edgewise shared
partner distribution was quite similar for Models 1 and 2, but Model 1 performed slightly better
than Model 2. Detailed statistics of the observed and the simulated network features (i.e.,
indegree, outdegree, shared partner distribution) can also be found in Appendix A. Considering
all the GoFs criteria, Model 2 was a better model than Model 1, so Model 2 was reported here as
the final model. Estimates from Model 2 were reported for the hypothesis testing.

82

Figure 13. Goodness of Fit plots for Models 1 and 2
Panel 1: Goodness of Fit Plots for Model 1 with Hypothesized and Control Variables Only

83

84

Panel 2: Goodness of Fit plots for Model 2, the Final Model.

85

86

Model Results (H1 and Post-Hoc Results). H1 predicted that within the undefined
population, members were more likely to share comment ties than by chance alone during the
review stage. The estimate (Estimate = -1.00, p >.05; see Model 2) for node match of the “other”
opportunity area was not significant. H1 was not supported.
Apart from results for H1, interesting post-hoc results also emerged. The comment
network from the ideation stage significantly predicted the network structure in the review stage
(Estimate =1.17, p < .001; see Model 2). During the review stage, ideators who had high
experience values were more likely to receive comments (Estimate = 0.09, p < .05; see Model 2).
4.1.2. Multilevel ERGM Results (H2a and H2b)
Model Fitting & GoFs. Multilevel ERGMs were fit on the structure of two networks: the
multilevel network in the ideation stage and the review stage. A forward selection approach was
also adopted in the multilevel ERGM model fitting procedure. To start with, the model with only
hypothesized and control variables was fit, which was followed by models with other network
structures being added to improve model fit. Models with the best model fit were reported as the
final models. For all multilevel ERGM models, the network density of the ideator network
(Network A) and the category affiliation network (Network X) were fixed because model
convergence was otherwise challenging, and this approach was suggested as a solution in the
MPnet Manual (Wang et al., 2014). The manual also advised that when density is fixed for
networks, the signature “edges” should not be included in the model (Wang et al., 2014).
Therefore, “edges” was not included in the models.
Model results for the multilevel ideation network can be found in Table 5 where graphs
of the variables were added. Squares represent ideators, and circles describe platform categories.
Network ties in the ideation stage were colored green. Affiliation ties between ideators and

87

categories were colored blue. Yellow squares indicated that continuous categories were tested in
the model. Node size is proportional to the size of the continuous variables. Blue squares with
dashed outlines were not important nodes.

Table 5. Multilevel ERGM results for the multilevel ideation network
Note. Squares represent ideators, and circles represent platform categories. Green ties are comment ties in the
ideation stage, and blue ties are affiliation ties connecting ideators and categories. Yellow squares show that
exogenous continuous variables were added. The large node size describes that the continuous attribute has a high
value, and the small node size indicates the attribute value is small. The blue squares with dashed outlines represent
nodes whose attributes were not important. * indicates that the absolute value of a parameter is greater than twice
the standard errors, and the parameter is significant, which is equivalent to p < .05 (Wang et al., 2014).

Model 3
(Hypothesized Variables and Controls
Only; Final Model)
Parameters SE t-ratios
Structural parameters
TriangleXAX (H2a and H2b)

0.54* 0.06 -0.04
Controls
Standard tenure activity

-0.05 0.05 -0.03
Standard exp. activity

-0.11 0.15 0.02
Standardized tenure difference

-0.05 0.07 -0.05
Standardized exp. difference

0.08 0.15 0.01

88

Model 3 contains the hypothesized and control variables only. It was also the final model
because adding any of the other network structures impaired model fit. Model 3 was the model
with the best fit. GoF statistics and plots can be found in Figure 14 and Appendix B. Similar to
one-mode ERGM models, GoF statistics for multilevel models were calculated by comparing
1000 simulated graphs based on converged models and the observed graph. The goal of the GoF
plots and tables (in Appendix B) for multilevel ERGMs is to provide comparisons between
simulated network and observed network statistics. The interpretation for GoF figures is the
same for multilevel ERGM and one-mode ERGM. In GoF figures, if the solid lines were not
extreme in the boxplots, it means the model fit is good. In the tables, the means of the simulated
graphs and observed values can be compared. Different from the one-mode GoF tables, MPnet
provides standard deviation of degree distribution and t-ratios for multilevel ERGMs (Wang et
al., 2013). The t-ratios are heuristic GoF statistics, and the absolute value of t-ratio is considered
adequate when it is smaller than 2.0 (Snijders, 2002; Wang et al., 2013). For those signatures
included in the model, suggested t-ratios should be smaller than 0.1 in absolute value to
reconfirm model convergence (Wang et al., 2014).
In Figure 14, statistics of the variables included in the model and other global network
structural configurations of the simulated models were reported as boxplots. If the observed
model statistics, represented as the orange line, fell within the interquartile range of the boxplots,
the models are fit well. As shown in Figure 14, all model statistics fell within the interquartile
range of the simulated graph distribution, which means the model estimates were in general
accurate. For the other global network configurations, overall, the solid line matched the boxplot
distribution. However, the model fell short in capturing the standard deviation or the skewness of
the degree distribution of the ideatior network (i.e., Network A), the skewness of the interaction

89

between the affiliation network (i.e., Network X) and the ideation network (i.e., Network A). In
Appendix B, observed statistics were compared with the mean statistics of the simulated graphs.
According to Appendix B, for all variables included in the model, the absolute t-ratios were
below 0.1. The global network statistics not included in the model were higher than 2.0 in
absolute value, and therefore, the model did not capture the global network features well. The
Mahalanobis distance was reported to show the overall fit of the model, where a smaller
Mahalanobis distance indicates a better fit. It is an indicative measure to compare models with
the same configurations and cannot be tested with χ
2
statistics (Wang et al., 2014). Because
Model 3 was the best model, and all variables included in the model were well-fit and, therefore,
overall accurate, hypotheses were reported using Model 3.

90

Figure 14. Goodness of Fit plots for Model 3

91

For the multilevel review network, model results can be found in Table 6. Model 4
presented the model with only hypothesized variables and control variables. Using Model 4 as a
baseline model, additional network structures were added to improve its model fit. Finally, the
best model was reported in Model 5.

Table 6. Multilevel ERGM results for the multilevel review network
Note. Squares represent ideators, and circles represent platform categories. Purple ties are comment ties in the
review stage, and blue ties are affiliation ties connecting ideators and categories. Yellow squares show that
exogenous continuous variables were added. The large node size describes that the continuous attribute has a high
value, and the small node size indicates the attribute value is small. The blue squares with dashed outline represent
those attributes of these nodes are not important. * indicates that the absolute value of a parameter is greater than
twice the standard errors, and the parameter is significant, which is equivalent to p < .05 (Wang et al., 2014).

Model 4
(Hypothesized Variables and
Controls Only)
Model 5
(Final Model)
Parameters SE t-ratios Parameter SE t-ratios
Structural parameters
TriangleXAX (H2a
and H2b)

0.13 0.07 0.03 0.50* 0.11 0.04
L3XAX

-0.19* 0.05 0.05
Controls
Standard tenure
activity

0.07 0.04 -0.004 0.07 0.04 -0.08
Standard exp.
Activity

0.13* 0.06 0.03 0.13* 0.06 0.02
Standardized tenure
difference

0.06 0.05 -0.04 0.06 0.05 -0.09
Standardized exp.
Difference

-0.23* 0.07 0.05 -0.23* 0.07 0.03

92

GoF plots and statistics can be found in Figure 15 and Appendix C. As shown in Figure
15, all model statistics for both Model 4 and Model 5 were fit well as the observed statistics were
not extreme in the simulated distribution. For the global network configurations, Model 5 was
better than Model 4. This can also find support in Appendix C, which shows that differences
between the observed and simulated graph statistics were overall smaller in Model 5 than Model
4. Based on Figure 15, Models 4 and 5 described the global structures well because the solid line
followed the patterns of the boxplots relatively well, but they were not within the interquartile
ranges. Moreover, according to Appendix C, although t-ratios for variables included in the model
had the absolute value smaller than 0.1, which indicates good GoF, the absolute values for t-
ratios were larger than 2 for global network structures. The global network structures were not fit
well. Although neither Model 4 nor Model 5 described the global network structures adequately
(i.e., the standard deviation, skewness, and clustering of the multilevel networks), the absolute t-
ratios were smaller for Model 5 than Model 4. Specifically, TriangleXAX, which was used to
test H2a and H2b, had a better fit in Model 5 than Model 4, as the t-ratio was much smaller in
Model 5 compared to Model 4. Moreover, the Mahalanobis distance in Model 5 was also
smaller. In conclusion, Model 5 had a better fit compared to Model 4. Considering all the
variables included in the model had good model fit, Model 5 was used for the result report.

93

Figure 15. Goodness of Fit plots for Models 4 and 5
Panel 1: Goodness of Fit Plots for Model 4

94

Panel 2: Goodness of Fit Plots for Model 5

95

Model Results (H2a, H2b, and Post-Hoc Results). H2a and H2b hypothesized that
cross-level closure was more likely to exist than by chance alone in the ideation and review
stages. A multilevel network signature (TriangleXAX) was included to test these hypotheses.
This cross-level closure structure was positive and significant (Estimate = 0.54, p < .05; see
Model 3) during the ideation stage, and in the review stage (Estimate = 0.50, p < .05; see Model
5). H2a tested the focal closure mechanism, and H2b tested the membership closure mechanism
(see Figure 9 for the distinction between the two). To further unpack the mechanism behind the
significant cross-level closure during the ideation stage, descriptive results are examined. There
were 439 pairs of ideators who shared the same category affiliation. Among them, 322 pairs of
ideators shared comments after both ideators completed submission (i.e., focal closure). Only
117 pairs formed membership closure (i.e., one ideator submitted to a category first and then
after sharing comment ties with another ideator, the other ideator followed suit by submitting to
the same category). Subsequently, a χ
2
test was run to see whether the number of focal closure
and membership closure differs from what would be expected by chance (χ
2
= 95.73, p < .001).
Therefore, both membership closure and social proximity existed during the ideation stage, but
the chance of observing focal closure was significantly higher than observing membership
closure. In other words, focal closure was the dominant phenomenon. In the review stage, all
ideators had completed their idea submission. As a result, all the cross-level closure during the
review stage were focal closure (H2a). In conclusion, focal closure dominated both the ideation
and review stage (H2a supported). This result indicates that members of the same category
tended to share communication ties. Membership closure is the opposite of focal closure. It
describes that individuals sharing communication ties tend to submit to the same category.
Membership closure (H2b) was also observed during the ideation stage, but it was not the main

96

force driving cross-level closure (H2b not supported). Note that in Appendix B, there were only
243 TriangleXAX, which was fewer than the 439 reported here. It was because the multilevel
model was fit on an unweighted network while in fact, the pairs of ideators sharing the same
category was 439.
Finally, an interesting post-hoc result was present here. The negative and significant
L3XAX (Estimate = -0.19, p < .05, See Model 5) in the review stage indicated that ideators did
not like to exchange comments with others outside their own categories during the review stage.
4.2. Evolutionary Perspective
An evolutionary perspective is applied to studying how communication networks on
OpenIDEO evolve through variation and selection processes. As discussed in the literature
review, retention is not studied in this current research because OpenIDEO does not have a
platform infrastructure to encourage long-lasting relationships, and therefore it is impossible to
observe network tie retention on this platform. Moreover, challenges on OpenIDEO are quite
independent because: (1) they are organized by different sponsors aimed at solving diverse social
issues, and (2) they are often participated by different ideators.
This research highlights the importance of interdependency among different populations,
and such interdependency should not be overlooked when analyzing network evolution.
Specifically, two types of populations (i.e., new entrants and existing members) are studied in
this research. Hypotheses in this section focus largely on how these two populations exchange
network connections over time and how their interpopulation connections influence the future
direction of the network evolution. Detailed contributions and implications will be discussed in
the discussion section.

97

4.2.1. One-Mode ERGMs on the Ideation Network
Model Fitting & GoFs. One-mode ERGMs were fit on two networks: the ideation and
review networks. To model the structure of the ideation network, a model with only
hypothesized and control variables was first fit (see Table 7, Model 6). Then, more network
structures were added with the goal of model improvement, and the model with the best model
fit is reported here (see Table 7, Model 7).

98

Table 7. One-mode ERGM Results for the Ideation Network (H3 and H4a)
Model 6
(Hypothesized
Variables and
Controls Only)
Model 7
(Final Model)
Notes
Structural features
Edges

-3.98 (0.08)***
Mutual

3.58 (0.13)***
Geometrically
weighted
dyadwise shared
partner (outgoing
two-path)

0.97 (0.06)***
Indegree 2

0.37 (0.15)*
Nodal covariates
Node mismatch
(new entrants and
existing members)
-1.35 (0.07)*** -0.16 (006)* H4a was
not
supported

Node outdegree
(new entrants)

-2.18 (0.06) -0.84 (0.07)*** H3 was
not
supported
Control
Indegree (exp)

1.94 (0.07)*** 0.25 (0.05)***
Indegree (tenure)

0.63 (0.03)*** 0.09(0.04)*
Absolute
differences (exp)

-2.48 (0.09)*** -0.22 (0.04)***
Absolute
differences
(tenure)

-1.65 (0.04)*** -0.24 (0.04)***
Null Deviance: 104457
with 75350 degrees of
freedom
Residual Deviance:
12182 with 75344
degrees of freedom
AIC: 12194
BIC: 12249
Null Deviance:
104457 with 75350
degrees of freedom
Residual Deviance:
8661 with 75340
degrees of freedom
AIC: 8680
BIC: 8772

Note. p* < .05, p** < .01, p*** < .001. Values in the parentheses are standard errors. In the signature visualizations,
blue squares represent ideators. Shaded blue squares indicate that categorical exogenous variables were added, while
yellow squares show that exogenous continuous variables were added. The large node size describes that the
continuous attribute had a high value, and the small node size indicates the attribute value was small. The blue
squares with dashed outline represent nodes whose attributes are not important.

99

In Table 7, the null deviance and residual deviance are reported to compare the model fit
of Model 6 and Model 7. The null deviance was 104457 with 75350 degrees of freedom. In
Model 6, the residual deviance was 12182 with 75344 degrees of freedom. The difference
between the null model and Model 6 was 92275 with 6 degrees of freedom, and based on the χ
2
distribution, p < .001. Model 6 was significantly improved in comparison to an empty model.
Model 7 had residual deviance of 8661 with 75340 degrees of freedom. Model 7 significantly
improved the fit of Model 6 as their difference in the residual deviances was 83614 with 4
degrees of freedom, and based on χ
2
distribution, p < .001. Moreover, the AIC and BIC of Model
7 were also smaller than those of Model 6. In summary, the amount of variance explained in
Model 7 was 95796, and the number of predictors was 10. The GoF statistics were reported in
Figure 16 and Appendix D. Figure 16 contains GoF graphs of the observed and simulated
graphs. As described, when the observed estimates (i.e., the solid lines in the plots) were not
extreme in the distribution of simulated graphs (i.e., the boxplots), the model had a good fit.
Figure 13 reports the GoFs of the model statistics and other network structures such as indegree
and outdegree distributions and shared partner distributions. As indicated in Figure 16, both
models had acceptable GoFs in model statistics as the solid lines fell within the boxplots. Model
7 captured the global network features, especially the indegree distributions, better than Model 6.
The two models had very similar edgewise shared partners and outdegree GoFs. The detailed
GoF statistics of observed and simulated network structures can be found in Appendix D, and as
discussed, the smaller the difference between the observed and simulated network values, the
better the model is. In Appendix D, for most network statistics, differences between observed
and simulated networks in Model 7 were smaller than in Model 6. In conclusion, Model 7 had a
better model fit than Model 6, and Model 7 was used to report final results.

100

Figure 16. Goodness of Fit plots for Models 6 and 7
Panel 1: Goodness of Fit Plots for Model 6 with Hypothesized and Control Variables Only

101

Note. The indegree GoF plot of Model 6 indicated that this model did not perform well in capturing the indegree
distribution of the observed network.

102

Panel 2: Goodness of Fit Plots for Model 7, the final model

103

104

Model Results (H3, H4a, and Post-Hoc Results). H3 tested whether new members of
the community were actively engaging in network variation. Specifically, it predicted that new
entrants were more likely to be active comment senders during ideation. H3 was not supported as
the estimate was negative and significant (Estimate = -0.84, p < .001; see Model 7), which
indicates that the opposite was true. The negative and significant estimate suggests that new
entrants were actually less likely to send out ties in general than existing members.
H4a and H4b studied network variation patterns across populations of new entrants and
existing members. H4a hypothesized that new entrants and existing members were likely to share
comment ties. Node mismatch of new entrants (Estimate = -0.16, p < .05; see Model 7) was
negative and significant, which means new entrants were less likely to share comments with
existing members. Therefore, H4a was not supported.
Apart from hypothesis testing, some interesting post-hoc results appeared. During the
ideation stage, ideators with high experience values (Estimate = 0.25, p < .001; see Model 7) and
long tenure (Estimate = 0.09, p < .05; see Model 7) were preferred receivers of comment ties.
Moreover, ideators were not likely to comment on others who had different experience values
(Estimate = -0.22, p < .001; see Model 7) and tenure (Estimate = -0.22, p < .001; see Model 7).
In other words, ideators who share similar experience values and tenure were likely to share
comment ties. The Edges variable was negative and significant (Estimate = -3.98, p < .001; see
Model 7), which indicates that this network was sparse. Ideators preferred to reciprocate
comment ties as the mutual structure was positive and significant (Estimate = 3.58, p < .001; see
Model 7). Moreover, the existence of outgoing two-path (Estimate = 0.97, p < .001; see Model 7)
and incoming two-path (Estimate = 0.37, p < .05; see Model 7), which was presented as indegree

105

2 in the model indicates that centralization existed in the ideation network, and some ideators
were active comment senders while some were popular comment receivers.
4.2.2. One-mode ERGMs on the Review Network
Model Fitting & GoFs. One-mode ERGMs were also fit on the review network. The
model fitting procedure remained the same. It started with a baseline model with hypothesized
and control variables (see Table 8, Model 8). Then, different additional network structures were
tried and if it improved the model fit, it was included. Otherwise, it was dropped. Finally, the
best model was reported (see Table 8, Model 9). Same as in the table above, network signatures
were visualized. In these structural visualizations, blue squares denote ideators, and shaded
squares indicate that categorical variables were added. Yellow squares stand for nodes with
exogenous continuous variables. For these yellow squares, the size denotes the magnitude of the
continuous attributes. Blue squares with dashed outline stand for nodes whose attributes were not
important. Orange ties illustrate edge covariates.

106

Table 8. One-mode ERGM results for the review network (H4b and H4c)
Model 8
(Hypothesized Variables
and Controls Only)
Model 9
(Final Model)
Notes
Structural features
Edges

-3.20 (0.05)***
Outdegree 0

8.11 (0.28)***
Indegree 2

0.75 (0.17)***
Indegree 3

-0.03 (0.18)
Edge covariate
Past network
(ideation stage)

0.31 (0.16)* 0.87 (0.14)***
Nodal covariates
Node mismatch (new
entrants and existing
members)

-2.34 (0.05)*** -0.53 (0.06)*** H4b was not
supported, given
H4a was not
supported
Indegree (No.
mutualist ties)

-0.002 (0.0006) 0.006 (0.0005)*** H4c was
supported
Control
Indegree (exp)

4.68 (0.07)*** 0.14 (0.05)**
Indegree (tenure)

0.61 (0.03)*** -0.004 (0.03)
Absolute differences
(exp)

-6.04 (0.10)*** 0.14 (0.05)*
Absolute differences
(tenure)

-1.43 (0.04)*** 0.003 (0.02)

107

Table 8. (continued)
Model 8
(Hypothesized Variables
and Controls Only)
Model 9
(Final Model)
Notes
Null Deviance: 111419 on
80372 degrees of freedom
Residual Deviance: 17946
on 80365 degrees of
freedom
AIC: 17960
BIC: 18025
Null Deviance: 111419
on 80372 degrees of
freedom
Residual Deviance:
10461 on 80361 degrees
of freedom
AIC: 10483
BIC: 10585

Note. p* < .05, p** < .01, p*** < .001. Values in the parentheses are standard errors. Blue squares represent
ideators. Shaded blue squares indicate that categorical exogenous variables were added, while yellow squares show
that exogenous continuous variables were added. The node size describes that the size of a continuous attribute. Blue
squares with dashed outline represent nodes whose attributes were not important. Orange ties describe an edge
covariate was added.

The null deviance was 111419 with 80372 degrees of freedom, and as shown in Table 8,
Model 8 had residual deviance of 17946 with 80365 degrees of freedom. Model 8 was
significantly better in comparison to the null model because their deviance difference was 93473
with 7 degrees of freedom, which, according to the χ
2
distribution, p < .001. The residual
deviance of Model 9 was 10461 with 80361 degrees of freedom, which reduced the deviance of
6585 with 4 degrees of freedom. Based on χ
2
distribution, p < .001. Model 9 explained the
amount of variance of 100958 with 11 variables. Based on these model fit statistics, Model 9 was
a better model than Model 8.
GoFs were included in Figure 17 and Appendix E. Figure 17 shows the GoF graphs.
When the solid lines fell inside the interquartile ranges of the boxplots, it means the model
captures the given network structures well in the observed data. Figure 16 contains GoF plots for
model statistics, indegree and outdegree distribution, and edgewise shared partners distribution.
Detailed GoF statistics were also reported in Appendix E. If the observed statistics are close to

108

the simulated means, the model has a good GoF. As shown in Figure 16, model statistics had
acceptable GoFs because the solid line falls within the interquartile range of the boxplots. Model
9 captured the indegree and outdegree distribution better than Model 8. The two models had
similar edgewise shared partners distribution. Therefore, Model 9 was the better fit modal when
compared to Model 8. Therefore, estimates from Model 9 were reported as the final model.

109

Figure 17. Goodness of Fit plots for Models 8 and 9
Panel 1: Goodness of Fit Plots for Model 8 with Hypothesized and Control Variables Only

110

Note. Model 8 did not capture the indegree and outdegree distribution of the observed network very well because in
the GoF plots, the solid lines did not went through the interquartile range of the boxplots.

111

Panel 2: Goodness of Fit Plots for Model 9, the Final Model

112

113

Model Results (H4b, H4c, and Post-Hoc Results). H4b predicted that new entrants and
existing members were not likely to share comment ties. Node mismatch was negative and
significant (Estimate = - 0.53, p < .001; see Model 9), which was congruent with the prediction
in H4b that new entrants and existing members were not likely to share comments during the
review stage. However, the preposition for H4b to be supported is for H4a to be supported,
because H4b predicted that these two populations no longer shared comments during the review
stage. Given that H4a was not supported, H4b was not supported. H4c hypothesized that the
number of mutualist ties shared in the ideation stage would positively predict the likelihood of
receiving ties in the review stage, and it was supported (Estimate = 0.006, p < .001; see Model
9). H4c was supported.
Interesting post-hoc results were also summarized from the model. Both Model 9 and
Model 2 tested the same network (i.e., the review network), so their post-hoc results were
compared. Results (Model 9 Estimate = 0.14, p < .001; Model 2 Estimate = 0.09, p < .05) from
Models 2 and 9 indicate that people of high experience values were preferred comment receivers.
However, in Model 9, results indicate that ideators preferred sharing ties with people whose
experience values were different from their own (Model 9 Estimate = 0.14, p < .05), but this
variable was not significant in Model 2 (Model 2 Estimate = -0.06, p >.05). Also, although
results from Model 2 suggest that ideators like to share comments with others who had very
different tenure (Model 2 Estimate = 0.06, p < .001), results from Model 9 did not signal a
significant pattern (Model 9 Estimate = 0.003, p > .05). Therefore, based on one-mode ERGM
results, people preferred sending ties to highly experienced ideators, and the other inconsistent
results were not conclusive.

114

4.3. REM Results (Community of Practice)
The last section of this research examines the ideator community from a community of
practice (CoP) perspective. Because the community features a combination of cooperation and
competition dynamics, the CoP perspective recognizes that community-oriented knowledge
exchange may exist among the ideators. The ecological and evolutionary perspectives applied in
the previous sections have limitations in identifying cooperative behaviors driven by altruistic
intentions. The CoP literature helps explain voluntary and cooperative communication among
contest competitors. The network is accentuated in the CoP literature as most computer-mediated
CoPs are connected and organized by networks (Wasko & Faraj, 2005). In this section, first,
RQ1 examined which types of reciprocity existed on this platform. Then, H5a and H5b explored
ideators’ motivations behind voluntary comment contribution behaviors. Specifically, it explored
how network position and the role of visitors’ communication affected voluntary commenting.
Theoretical contributions and implications will be further unpacked in the discussion section.
RQ1, H5a, and H5b were tested using ordinal time REMs on comment ties during the ideation
and review stage.
Model Fitting & GoFs. Like the procedures taken in the previous section, the model
with hypothesized and control variables was fit first, and then more structural configurations
were added to improve model fit. The models with the best model fit were reported.
To start with, network evolution during the ideation stage was analyzed. Model results
and GoF statistics are shown in Table 9. Network graphs were also presented in Table 9 to
illustrate the variables. In these graphs, bold lines represent first events, and dotted lines stand for
subsequent events. The null deviance was 31758.15 with 1414 degrees of freedom, and for
Model 10, the residual deviance was 26710.2 with 1407 degrees of freedom. The improved

115

deviance was 22952.05 with 7 degrees of freedom, and based on the χ
2
distribution, the
corresponding p value was < .001. Therefore, Model 10 exceeded the null model. As seen in
Model 11, residual deviance was 20950.43 and the degrees of freedom were 1406. In comparison
to Model 10, Model 11 decreased residual deviance by 24614.77 with 1 degree of freedom. The
χ
2
distribution demonstrated that the difference was significant at p < .001. Model 11 explained
the amount of 10807.72 in the variances of the outcome variable with 8 predictors.
Consequently, Model 11 is a better model fit than Model 10, and the results from Model 11 were
reported for hypothesis testing.

116

Table 9. Ordinal time REM results on comment ties during the ideation phase
Model 10
(Hypothesized and
Control Variables
Only)
Model 11 (Final Model) Note
Structural
features

NIDSnd

34.88 (1.61)*** 16.79 (2.30)*** H5a was
supported
NODSnd

9.57 (0.31)*** 5.88 (0.52)*** H5a was
supported
ITPSnd
(triadic
closures)

0.11 (0.009)*** 0.12 (0.009)*** RQ1
PSAB-BA
(direct
reciprocity)

4.09 (0.28)*** 6.83 (0.27)*** RQ1
PSAB-BY
(indirect
reciprocity)

-1.63 (0.32)*** 1.08 (0.31)*** RQ1
PSAB-AY

5.77 (0.07)***
Nodal covariates
(control)

Exp (receive) -0.05 (0.03) 0.04 (0.02).
Tenure
(receive)
0.03 (0.03) 0.03 (0.03)

117

Table 9. (continued)
Model 10 (Hypothesized
and Control Variables
Only)
Model 11 (Final Model) Note
Model Fit Null deviance: 31758.15 on
1414 degrees of freedom
Residual deviance: 26710.2
on 1407 degrees of freedom
AIC: 26724.2
BIC: 26760.98
Null Deviance: 31758.15
on 1414 degrees of
freedom
Residual Deviance:
20950.43 on 1406 degrees
of freedom
AIC: 20966.43
BIC: 21008.47

Note. The bold line shows the first event, and the dotted line(s) indicates the following event(s). p* < .05, p** < .01,
p*** < .001. Values in the parentheses are standard errors.

Table 10 presents the model results for the review network. Therefore, Model 12 contains
all hypothesized and control variables. Then, based on Model 12, additional network structures
were added to improve the model fit. Model 13 was the final model with the best model fit. As
shown in Table 10, the null deviance was 37633.01 with 1666 degrees of freedom. Model 12
(residual deviance of 34086.60 on 1657 degrees of freedom) improved the null model as the
difference in residual deviance was 3546.41 with 9 degrees of freedom, and p < .001. In
comparison to Model 12, the residual deviance of Model 13 was 26043.57 with 1656 degrees of
freedom, which significantly outperformed Model 12 by 8043.03 with only 1 degree of freedom,
p < .001. Model 13 explained 11589.44 in the total variances in the outcome variable with 10
predictors. Both AIC and BIC scores were smaller for Model 13 compared to Model 12, which
also suggested that Model 13 had a better fit than Model 12. Therefore, Model 13 was used as
the final model.

118

Table 10. Ordinal time REM results on comment ties during review
Model 12 (Hypothesized
& Control Variables
Only)
Model 13 (Final
Model)

Note
Structural features
NIDSnd

46.18 (2.52)*** 17.77 (2.73)*** H5a was
supported
NODSnd

12.45 (0.20)*** 7.93 (0.41)*** H5a was
supported
ITPSnd (triadic
closures)

0.21 (0.01)*** 0.20 (0.01)*** RQ1
PSAB-BA
(direct
reciprocity)

5.45 (0.26)*** 7.20 (0.26)*** RQ1
PSAB-BY
(indirect
reciprocity)

-0.25 (0.30) 1.44 (0.30)*** RQ1
PSAB-AY

5.99 (0.06)***
Edge covariate
Aggregated
ideation
network
1.48 (0.16)*** 0.49 (0.17)**
Nodal covariates
(control)

Number of ties
exchanged with
visitors during
ideation (send)
-0.0006 (0.00006)*** -0.06 (0.03)* H5b was
not
supported
Exp (receive) -0.06 (0.03)* -0.06 (0.03)*
Tenure
(receive)
0.04 (0.03) 0.04 (0.03)

119

Table 10. (continued)
Model 12 (Hypothesized
& Control Variables
Only)
Model 13 (Final Model)
Model Fit Null Deviance: 37633.01
on 1666 degrees of
freedom
Residual Deviance:
34086.6 on 1657 degrees
of freedom
AIC: 34104.60
BIC: 34153.36
Null Deviance: 37633.01 on
1666 degrees of freedom
Residual Deviance: 26 043.57
on 1656 degrees of freedom
AIC: 26063.57
BIC: 26117.75

Note. The bold line shows the first event, and the dotted line(s) indicates the following event(s). p* < .05, p** < .01,
p*** < .001. Values in the parentheses are standard errors.

Model Results (H5a, H5b, and RQ1). H5a hypothesized that ideators who were more
central were more likely to send out ties in the future. During the ideation stage, the positive and
significant NIDSnd (Estimate = 16.79, p < .001; see Model 11) indicated that indegree predicted
senders’ future tie sending rate. NODSnd (Estimate = 5.88, p < .001; see Model 11) was also
positive and significant, which means outdegree also predicted future tie sending rate. During the
review stage, NIDSnd was positive and significant (Estimate = 17.77, p < .001; see Model 13).
NODSnd was also positive and significant (Estimate =7.93, p < .001; see Model 13). Therefore,
H5a was supported in both the ideation and review stages.
H5b predicted that ideators would exchange (both sent and received) more comments
with visitors in the ideation phase and were more likely to send comment ties to other ideators in
the future. H5b was not supported in Model 13 (Estimate = -0.06, p < .05) because the estimate
was negative and significant, which was contrary to the hypothesis. This result indicated that the
more comments exchanged with the visitors, the less likely for the focal ideators to send future
comment ties to other ideators. RQ1 asked what reciprocity patterns existed among ideators.
Three types of reciprocity were examined: direct reciprocity, indirect reciprocity, and triadic

120

closures. During the ideation stage, direct reciprocity pattern (Estimate = 6.83, p < .001; see
Model 11), and indirect reciprocity (Estimate = 1.08, p < .001; see Model 11), and triadic
closures (Estimate = 0.12, p < .001; see Model 11) existed. Estimates for Model 15 were
reported here as the results were consistent with Model 13. During the review stage, direct
reciprocity pattern (Estimate = 7.20, p < .001), indirect reciprocity (Estimate = 1.44, p < .001),
and triadic closures (Estimate = 0.20, p < .001) also existed (see Model 13). Reciprocity patterns
were consistent across the two stages. This result shows that all three types of reciprocity were
present throughout the ideation and review stages, which means this is an overall cooperative
and cohesive community. Detailed discussion of these three structures can be found in the
discussion section.
Interesting post-hoc results can also be observed from the REMs in the ideation and
review stages (see Models 11 and 13). PSAB-AY was significant and positive throughout the
ideation (Estimate = 1.08, p < .001; see Model 11) and review stages (Estimate = 5.99, p < .001;
see Model 13). This result shows that sending to one ideator significantly led to future sending to
a different ideator, which suggested some active comment senders who sent ties with more than
various ideators in the community. Results indicated that during the ideation stage, tenure
(Estimate = 0.03, p > .05; see Model 11) and experience values (Estimate = 0.04, p > .05; see
Model 11) were not significant predictors for future tie receiving actions. During the review
stage, tenure was still not a significant predictor for tie receiving (Estimate = 0.04, p > .05; see
Model 13). However, experience values became a negative predictor for future receiving
(Estimate = -0.06, p < .05; see Model 13). That means, ideators with high experience values were
less likely to receive ties during the review stage. Longitudinal model results contradicted with
previous one-mode ERGM findings that highly experienced ideators were more likely to receive

121

ties (Model 9 Estimate = 0.14, p < .001; Model 2 Estimate = 0.09, p < .05) than low experience
values during the review stage. The longitudinal model (Estimate = 0.03, p > .05; see Model 11)
also failed to support one-mode ERGM results that ideators with longer tenure were preferred tie
receivers (Estimate = 0.09, p < .05; see Model 7) during the ideation stage. The inconsistency of
results generated from the cross-sectional and longitudinal models suggests that conclusions
cannot be drawn on control variables including tenure and experience values. Results from REM
(Estimate = 0.49, p < .01; see Model 13) and one-mode ERGM (Estimate = 0.87, p < .001, See
Model 9) agreed on the finding that network structures from the ideation stage significantly
predicted tie formation in the review stage. Finally, the summary hypotheses and their results can
be found in Table 11.

122

Table 11. A summary of hypotheses, populations relevant to the hypotheses, and results
Ecological Perspective Populations
Relevant to
the
Hypotheses
Notes
H1: During the review stage of the contest, within the
undefined population, ideators are more likely to share
comment network ties than by chance alone.
Ideators H1 was not
supported
H2a: Individual ideators with the same category affiliation
are more likely to share comment ties than by chance
alone in the ideation and review stage.
Ideators H2a was
supported
H2b: Individual ideators sharing comment ties are more
likely to share the same category affiliation than by
chance alone in the ideation stage.
Ideators H2b was not
supported
Evolutionary Perspective
H3: During the ideation stage, the first stage of a contest,
new entrants are more likely to be senders of comment
ties (in general, including both within- and cross-
populations) in comparison to existing members.
Ideators H3 was not
supported
H4a: During the ideation stage, comment ties are more
likely to exist between existing members and new entrants
than by chance alone.
Ideators H4a was not
supported
H4b: During the review stage, new entrants no longer
maintain more comment ties with the existing members
than by chance alone.
Ideators H4b was not
supported given
H4a was not
supported
H4c: Ideators’ increases in the number of mutualist ties
exchanged between the new entrants and existing
members in the ideation phase lead to increases in the
likelihood of receiving comments in the review stage.
Ideators H4c was
supported
Community of Practice
RQ1: As each contest phase unravels, how do the
reciprocity patterns of the ideators population shift?
Ideators NA
H5a: Over time, ideators who are more central in the
ideator comment network are more likely to send out
comment ties to other ideators in the future.
Ideators H5a was
supported
H5b: Over time, ideators who have exchanged (both sent
and received) more comments with visitors in the ideation
phase are more likely to send comment ties to other
ideators in the future.
Ideators and
visitors
H5b was not
supported

123

Chapter 5: Discussion and Conclusion
5.1. Findings Discussion
Drawing from ecology theory, network evolution theory, and the community of practice
(CoP) framework, this research analyzed the network dynamics in a coopetitive crowdsourcing
contest community. Several important findings are reported. First, this research identified
interesting patterns of how ecological categories influence social network structures. Second, this
research suggests that the scope of network evolution research should go beyond a single
population. Specifically, interpopulation interdependencies (Aldrich et al., 2020; Astley, 1985;
Powell et al., 2005) were considered in this dissertation when analyzing individual contest
participants’ network variations and selection patterns through two different contest stages:
ideation and review. Finally, to explore ideators’ knowledge exchange and voluntary sharing,
this dissertation studied the community as a CoP. Specifically, comment exchange dynamics of
the ideators, the contest competitors, were examined. Findings are summarized and discussed in
the following sections.
5.1.1. Multilevel Ecology of Categories and Social Networks
To start with, this research examined the ecology of a contest community from a
multilevel perspective. A multilevel approach presents a more realistic representation of the
community structure as it allows the influence of ecological factors to be included when
analyzing the social interaction patterns within the community. By conceptualizing populations
as groups of ideators defined by preexisting platform categories, this research suggests that
platform categories may shape community users’ social interaction patterns. Specific findings are
discussed below.

124

5.1.1.1. Population Emergence and the “Other” Category. Contrary to Hypothesis 1
(see Table 11 for the summary of hypotheses and results), results indicate that ideators in the
“other” category were not more likely to exchange comments than by chance alone during the
review stage. It can be inferred that legitimation was not observed within the “other” category,
and therefore, it is possible that the “other” category should not be recognized as a population.
This result indicates that the form of this category should be reconsidered in terms of its identity
and evolutionary stages. The arguments are presented as follows. First, identity-wise, the “other”
category contained ideators who chose not to conform with the predefined categories. The
“other” category has an open and undefined nature. In 1970s, living organisms were classified as
either animals or plants, but then scientists found new species that were not animals or plants,
and new categories— “fungi, protozoans and algae”—were proposed in addition (Gandhi, 2019,
para. 6). For example, Mesodinium chamaeleon was a creature that found to possess features of
both animals and plants that challenged scientists’ previous categorization. Therefore, the
categorization system has been also constantly modified, and the “other” category can be used as
an open-ended and comprehensive label for new possibilities to be proposed when they do not
belong to any of the existing categories. Therefore, contrary to what was proposed initially in the
literature review, these ideators’ identities may be too diverse to be a population that share
similar “core” attributes and common identities (Hannan et al., 2019, p. 3). The open and
undefined nature of the “other” category provides opportunities for ideators to think out of the
boxes, which resembles the variation concept in evolutionary theory that describes intentional or
random processes such as “mutations” or “trials” (Campbell, 1965, p. 306) that depart “from
routines or tradition” (Aldrich et al., 2020, p. 19). The ideas submitted to the “other” category are
diverse variations that diverge from ideas categorized by platform categories. Variations of ideas

125

can fuel community evolution. Some variations fail, while some are selected and retained that
give rise to new niches (Aldrich et al., 2020). Recognizing that the “other” category to be open
and potentially diverse, these ideators should not be recognized as one collective sharing the
same identity and niche space. Therefore, the collective legitimation efforts often observed in
novel categories (Alexy & George, 2013) and entrepreneurial industries (Esparza et al., 2014)
may not apply here.
Second, in terms of evolutionary status, the “other” category might have been in a
nascent stage, when routines are not established and focused identities have not yet formed
(Aldrich, 2020). Thus, it could have been a stage before population emergence. Before new
populations are founded and stabilized, “founders” tend to “implicitly compete to have their
approach taken for granted, appealing to potential customers, investors, and others to accept their
vision” (Aldrich et al., 2020, p. 199). This may have caused the “other” population to be less
cooperative with one another than in the rest of the categories. Over time, separate categories
may emerge from the “other” category and may even become institutionalized if the contest
organizers allow. Later on, when these new populations stabilize, within-population collectives
emerge to exchange information, establish norms and rules, and face challenges together
(Esparza et al., 2014).
5.1.1.2. Category Niches. Apart from the “other” category, cross-level closure was
observed in both the ideation and review stages. The result for Hypothesis 2a shows that
dominant mechanism in the crowdsourcing challenge community across different contest stages
was focal closure, which describes the tendency for ideators of the same category affiliation to be
socially connected. Membership closure, which represents the pattern when socially connected

126

nodes submit to the same category, also existed in the ideation stage but was not a dominating
mechanism based on results for Hypothesis 2b.
This finding extended the traditional research on homophily versus influence debates
grounded in single-level networks (e.g., Leenders, 1995; Ma et al., 2014) to multilevel networks
(Amati et al., 2019). Homophily and influence can drive network formation and evolution, but it
is challenging to distinguish these two processes (Xu, 2018). Amati et al. (2019) proposed a
longitudinal and multilevel solution to help separate these two mechanisms. The reason is that
multilevel network analysis reveals out not only network patterns among humans, but also the
interaction of multiple networks. In this situation, attributes can be included as endogenous to
networks so that interaction between a human network of social relations and a human-attribute
network can be examined. Although the multilevel ERGM applied in this research was not a
longitudinal model, the researcher collected longitudinal data and examined the sequence of the
social relations and attribute affiliation ties. This research applied the aforementioned approach
and found that homophily is a dominating process that drives the formation of multi-level social
networks on the crowdsourcing contest platform.
In summary, this research aimed to study how ecological niches affect social interactions.
The findings suggest that the concept of the “other” category should be reconceptualized because
legitimation was not observed among its members. Unlike members in the “other” category that
did not prefer to share social ties, intra-population (i.e., within-category) connections were likely
to be observed in general. More detailed theoretical contributions will be discussed in the next
section (See Section 5.2).

127

5.1.2. Network Evolution of Interdependent Populations
This research also examined the network evolution patterns across the ideation and
review stages of the OpenIDEO contest community. Across-population interaction patterns were
analyzed. Two types of ideators were identified: new entrants and existing members. They
competed in the same contest, and their relationship was identified as commensalism, which is
defined as the relationship shared by resource-dependent organizations and/or individuals
(Aldrich et al., 2020). Apart from ideators, there were also visitors who did not compete in the
contest but shared comments with the ideators to provide feedback. Visitors and ideators shared
a symbiotic relationship (Aldrich et al., 2020) because the visitors do not directly compete with
the ideators. Their relationship may bring resources to both parties. Ideators provide content to
entertain, inform or teach visitors, while visitors share feedback, knowledge, and information to
help ideators improve their ideas. These two types of interpopulation interdependencies (i.e.,
commensalism and symbiosis) were included in the current research to study network evolution
(Aldrich et al., 2020).
Complementing previous research (e.g., Fu, 2019; Margolin et al., 2015; Shen et al.,
2014a) that has examined the patterns of network evolution only on a single population, this
research applied network evolution theory (Monge & Poole, 2008) to analyzing the role of
interpopulation influences (Aldrich et al., 2020) on the evolving communication network
structures. This research found that mutualist ties between new entrants and existing members
enhance ideators’ network portfolios and affect directions for network selection.
5.1.2.1. Network Variation in the Ideation Stage. Findings show that new entrants
were less likely to share communication ties during the ideation stage than existing members,
contrary to Hypothesis 3. This result contradicts previous research findings that new entrants

128

actively seek network variations by experimenting with a large number of ties to familiarize
themselves with their surroundings (Shen et al., 2014a) and seek resources (Doerfel et al., 2013).
An explanation on why this finding contradicts previous research might be that this current study
was based on a different context. There are two notable features of crowdsourcing contest
communities. First, comment communication on crowdsourcing contest communities is visible to
the public. Second, contests on this type of community are temporary and short-lived as they
often only last several months. Finally, contests are also relatively independent of each other.
When each contest starts, the theme, organizers, and participants might be completely different
from the previous contests. Holding these three features in mind, this current research result is
compared against Shen and colleagues’ and Doerfel and colleagues’ findings. Shen et al. (2014a)
identified that new entrants are more active in network variation process in a gaming community.
These new players experiment with a “promiscuity of connections” to gain knowledge about the
platform (Shen et al., 2014a, p. 3). As they accumulate more experiences over time, they tend to
become more selective of their network partners (Shen et al., 2014a). The network ties of this
current research were different in nature from the ties described in Shen et al.’s (2014a) work.
The main difference between comment ties in a crowdsourcing contest community and
communication on gaming chats is that talking through gaming chats is low-stakes, fleeting, and
only visible to limited players in the same battle. Some battle-related group chats vanish after the
battle ends. In comparison, comment ties on crowdsourcing platforms last longer and can be seen
by anyone visiting the website. They resemble representational ties, which, although from a
cursory glance, are simply exchanged among network actors, also contain communications of
value to the general public (Shumate & Contractor, 2013). New entrants are very new to the
platform and might be intimidated by the stakes involved in sharing comments with others. They

129

may silently lurk and learn from others by reading others’ comments until they are familiar with
the community norms (Nonnecke & Preece, 2001).
Doerfel and colleagues (2013) studied interorganizational networks post Hurricane
Katrina. Their study analyzed a crisis condition where new entrants lacked resources and were
motivated to engage in network variation activities to seek resources. However, in contest
platforms on crowdsourcing contest platforms, each contest is overall temporary and
independent. They only last several months and often attract quite different crowds. During the
ideation stage, when a contest starts, although existing members have participated in contests
before and are more familiar with the platform’s norms and rules, they have to adapt to the new
contest just as the new entrants do. Moreover, new entrants may possess novel information and
knowledge from outside of the community that existing members may not have. Therefore,
resource gaps between new entrants and existing members may not be as significant as in an
uncertain crisis stage that puts new entrants in vulnerable situations (Doerfel et al., 2013). New
entrants’ lack of familiarity with the platform may be reduced because it is the start of a new
contest for everyone, and new entrants do not feel more motivated than other members to seek
connections.
5.1.2.1. Network Selection in the Review Stage. Contest stages are recognized as
possible sources of environmental shifts and were included in this longitudinal research design.
Hypothesis 4a posited that initially, during the ideation stage, network ties would be likely to
exist between new entrants and existing members (see Table 11). Past research has shown that
new entrants seek existing members for knowledge about the platform and legitimation
(Margolin et al., 2015), and existing members tend to connect with new entrants for diverse and
novel knowledge (Majchrzak & Malhotra, 2016). As the contest unfolds, new entrants are less

130

dependent on existing members for knowledge about the community during the review stage.
Therefore, this research hypothesized in Hypothesis 4b that new entrants and existing members
would no longer maintain more comment ties than by chance alone during the review stage. The
results show that throughout the ideation and review stages, a consistent pattern was observed
that new entrants and existing members were unlikely to share comment ties. Therefore,
communication patterns across new entrants and existing members were not affected by the
shifts of contest stages. Such results may reflect that these two stages share some overlap and
consistency, which contradicted Hypotheses 4a and 4b that asserted the shift between ideation
and review would be significant.
Previous literature has suggested that environmental shifts affect network selection (e.g.,
Margolin et al., 2015; Wang et al., 2016). The stable network patterns between new entrants and
existing members in this research may indicate that the magnitude of shifts from the ideation to
the review stage was not large enough to disrupt network evolution process between new
entrants and existing members. The post-hoc results also show that the existence of network ties
during the ideation stage was significantly associated with the existence of ties of the review
stage. In other words, network inertia was a driving mechanism in the crowdsourcing contest
community (Shen et al., 2014a), and the shift between the ideation and review stage was not
large.
As shown in the measures section (See Section 3.4.), comments in OpenIDEO were often
praise or instructive suggestions because the human coding results on comment ties
demonstrated that negative comments were almost unseen. The lack of interaction between the
new entrants and existing members, as shown in results for H4a and H4b (see Table 11), may
also suggest that existing members felt threatened by emerging members seeking to develop in

131

the same resource space (Aldrich et al., 2020). Feeling threatened, existing members may have
been reluctant to share praises that recognize the legitimacy for new entrants (Aldrich et al.,
2020) or suggestions that help emerging members improve. This finding also reflects the
likelihood that a clear population boundary existed between the new entrants and existing
members.
In addition, this research also confirmed that the relationship between new entrants and
existing members was mutualist, a type of commensalism relationship that describes populations
sharing similar resource needs benefitting from each other (Aldrich et al., 2020). New entrants
and existing members compete in the same contest, and the resources required for winning are
similar. Therefore, they share similar resource needs. In fact, results indicate that their
interpopulation relationship was beneficial. Specifically, in congruence with what was predicted
in Hypothesis 4c, ideators’ increase in the number of mutualist ties exchanged between new
entrants and existing members during ideation was positively associated with the likelihood of
receiving comment ties in the review stage. This finding confirms Weber’s (2012) logic that
links connecting new and existing members channel complementary resources that help them
acquire advantageous network positions and attract future network ties. Mutualist
communication shared between new entrants and existing members enhances their portfolios.
New entrants can share novel information and knowledge, while existing members can share
tacit knowledge and information accumulated from their experiences. This finding also advances
the evolutionary theory premise that population interdependency does influence network
evolution (e.g., Monge & Poole, 2008; Weber, 2012). Adding to Weber’s (2012) findings that
symbiotic relationships bridge two populations and help increase future connections, this
research found that mutualist relationships also channeled valuable resources that attracted future

132

network ties. However, as indicated in the results for Hypotheses 4a and 4b (see Table 11), new
entrants and existing members did not prefer sharing comment ties. Therefore, such mutualist
relationships were beneficial but rare.
In conclusion, this research contributes to our understanding of how environment
influence network evolution patterns (e.g., Wang et al., 2016; Yang, 2020). Findings indicate that
new entrants were hesitant to share comment ties in general during the ideation stage.
Throughout the contest, new entrants and existing members were unlikely to share comment ties.
These findings can be explained by the temporary and independent nature of the crowdsourcing
contest communities and the representational nature of the commenting ties discussed in
previous paragraphs. Another important contribution of this research is that this dissertation
studied the interdependencies of two different populations (i.e., new entrants and existing
members) and found that interpopulation interdependencies significantly influenced network
evolution. This finding adds to a body of literature that applies V-S-R to understanding network
evolution in a single population (e.g., Fu, 2019; Margolin et al., 2015). This study provides
evidence that the network evolution of multiple interrelated populations should be studied
simultaneously. More detailed theoretical contributions will be discussed in the next section as
well (See Section 5.2).
5.1.3. Community of Practice
Online crowdsourcing contests, as described in previous sections, are coopetitive, that is,
both cooperative and competitive. Contest participants compete for winning, but they are also
encouraged to exchange knowledge and feedback to help each other develop ideas (Sun &
Majchrzak, 2020). Complementing the ecological and evolutionary approaches that study
ideators’ interactions with the environment and across populations, the CoP approach touches

133

upon ideators’ collective community-oriented knowledge creation through commenting. This
research conceptualized contest-organized communities as evolving CoPs. The reason is that
people were drawn together by their interests and commitments to solving a common question
raised by the contest organizer. The organizing of such a community is through practices such as
idea sharing and voluntary commenting (Eckert, 2006). Based on this conceptualization, the
evolution of comment ties was studied. The questions explored in this section include (1)
whether the communication interactions for ideators resembled the structure of a CoP that is
reciprocal (Research Question 1), and (2) what factors motivated ideators’ knowledge sharing
with other ideators competing in the same contest (Hypotheses 5a and 5b; see Table 11).
The first question inquires about the reciprocity patterns in CoP (Research Question 1).
Building on Wasko and Faraj (2005), this research studied virtual CoPs from a network
perspective and highlighted the importance of communication networks in community
organizing. Several different types of reciprocity patterns were examined, aiming to understand
patterns of individuals’ voluntary contributions and knowledge sharing, which are essential for
public goods maintenance in online communities (Wasko et al., 2009). Results indicate that the
three types of reciprocity (Taylor & Nowak, 2007)—direct reciprocity, indirect reciprocity, and
triadic closure—were all present through the ideation and review contest stages. As mentioned,
direct reciprocity is linked with self-interest, and collaborative behaviors are contingent upon
benefits received. Direct reciprocity can also be described by the “I help you and you help me”
sentiment (Taylor & Nowak, 2007, p. 2283). Indirect reciprocity signals an altruistic and “pay it
forward” norm within the community (Dawson, 2019, p. 52). In communities with indirect
reciprocity, or generalized exchange, people’s voluntary contributions are often repaid by third
parties. Therefore, this study’s results indicate that ideators’ networking patterns followed the

134

patterns of both self-interested direct social exchanges and community-oriented and altruistic
contributions, which was beneficial for the creation and maintenance of community public goods
(Wasko et al., 2009). Because triadic closure significantly drove network evolution, this
community was dense and cohesive (Dawson, 2018). In online communities, triadic closure can
be conduits to cohesion, support, trust, homogeneous groups, and knowledge and opinion
reinforcement in the community, but excessive closure can also be bad for innovation because
network nodes may suffer from knowledge and information redundancy and lack of flexibility in
collaboration (Burt, 1992; Gargiulo & Benassi, 2000; Shen et al., 2014b). In conclusion, the
crowdsourcing contest community can be quite cooperative even though the setting is a
competition in nature. Ideators’ contributions are driven by not only direct reciprocity, but also
by the generalized exchange that is closely related to public goods production. It is reasonable to
assume that the community is also evolving towards the direction of being denser and more
cohesive and the triadic closure mechanism was driving network evolution. Contest competitors’
voluntary knowledge sharing, therefore, can be driven by direct social exchange, community-
oriented and altruistic contribution, as well as cohesive local clustering mechanisms.
The second question deals with factors motivating voluntary knowledge sharing among
contest participants. Findings support the idea that the maintenance of public knowledge goods
in a crowdsourcing contest CoP is closely related to individuals who are central in the
communication network (Wasko & Faraj, 2005). Specifically, both popular ideators, who have
received many ties in the past, and active ideators, who have been active sharing ties, are likely
to be cooperative and share comments with others in the future. This finding confirms previous
research (Baek & Bae, 2019; Wasko & Faraj, 2005) that network centrality is closely related to
voluntary knowledge contribution. This study found evidence in the results for Hypothesis 5a

135

(see Table 11) that individuals with high indegree and outdegree centrality can form a habit of
contribution. Moreover, this research studied the networks in a CoP using a longitudinal method,
which is a response to Wasko and Faraj’s (2005) saying that the “dynamic nature of network
structuring” should be considered (p. 53). Compared to previous cross-sectional research that
only identified that centrality is related to total contribution frequency (e.g., Baek & Bae, 2019;
Wasko & Faraj, 2005), the longitudinal model results indicate that centrality positively predicts
the frequency of future contributions, offering more nuances to the existing literature.
Another contribution of the current study to the CoP literature is also discussed here.
Previous studies have only identified CoPs formed by individuals with communal or collective
identities and shared practices (e.g., Li et al., 2009; Rivera & Cox, 2016). This study proposed
that the composition of CoPs can be more complicated. Specifically, contest-organized CoPs
feature two types of interconnected community members: ideators and visitors. Acknowledging
that populations are interdependent, and that online CoPs are increasingly diverse (Gilbert,
2016), studying CoPs that include multiple interconnected types of members is an inevitable
trend.
In the current study, contrary to Hypothesis 5b, ideators who exchanged more comments
with visitors were not more likely to share comments in the future. Therefore, exposure to the
cooperative visitor population did not make ideators more cooperative. Baek and Bae (2019)
argued that people with high centrality are motivated to share knowledge because they wish to
maintain advantageous network positions. Wang et al. (2019) studied a social media CoP
advocating for climate awareness and found that being central in the community of main
contributors’ network was rewarding, but being central in the entire social issue network was not.
They explained that these main contributors are elites and leaders in the community, and

136

connecting to them helps increases network visibility. In CoP crowdsourcing contests, ideators
are the main contributors as they generated ideas, and all comment communication is located
within these ideas. Therefore, being central in the visitor network (i.e., frequently interact with
visitors) may not be as rewarding as being central in the ideators’ network.
In summary, CoP was applied to studying crowdsourcing contest communities’
knowledge sharing patterns. This research contributes to the CoP literature by studying two types
of community members: ideators and visitors. Results found interesting results that occupying
central network positions in the ideator network motivated knowledge contributions, but
interacting with visitors was not related to the likelihood for sharing comments in the future. This
research also identified reciprocity patterns in the crowdsourcing contest community that
provides more insights into the network mechanisms that drive network formation and evolution.
Detailed theoretical contributions are discussed in the following section (See Section 5.2.).
5.2. Theoretical Contributions
5.2.1. Crowdsourcing Literature
This research adds to an important line of literature that has shown that analyzing online
communities from a network perspective is rewarding and important (e.g., Haythornthwaite,
2007; Lee & Lee, 2010; Shaw, 2012). Networks can be a useful lens to study the infrastructure
and organization of various platforms. This is because different platforms support different
patterns of information flow (Haythornthwaite, 2007). There are several main takeaways from
this current research. First, content homophily is an important force that drives network
formation patterns in a crowdsourcing contest community. In this virtual community of
“strangers” (Gupta & Kim, 2004, p. 2680), ideas are the faces of the ideators that shape

137

interactions, and ideas are classified by platform categories. As a result, platform categories are
useful in identifying categorized resource niches in crowdsourcing contest platforms.
Second, this research also helps communication scholars understand the organization of
online crowdsourcing contest communities. This research shows that despite the contest going
through stages that served the contest organizers’ various aims (e.g., brainstorming diverse ideas,
providing feedback, etc.), the network interaction patterns remained overall consistent. This
result may reflect that designs in contest stages are limited in shaping communication patterns in
crowdsourcing contest communities. Third, even though this community was a coopetitive
venture in which users shared both cooperative and competitive relationships (Hutter et al.,
2011), cooperation was still a dominating social norm in this community. This conclusion is
drawn from the results that within-population comments were frequently observed, and all three
types of reciprocity were significantly motivating network evolution. The reason why
cooperation was dominating might be that actively establishing social ties positively relates to
performance in crowdsourcing contests (Dissanayake et al., 2015). Also, the significant network
closure tendency that existed in contest communities may also risk knowledge and information
redundancy and the lack of flexibility in knowledge collaboration (Burt, 1992; Gargiulo &
Benassi, 2000; Shen et al., 2014a), that may not be very constructive for generating innovative
ideas.
5.2.2. Developing Ecology Theory
This research has opened up the possibility of applying ecological perspectives in
studying social interactions in online communities. There have been past attempts that applied
ecology theory to online communities (e.g., Lai, 2014; TeBlunthuis et al., 2020; Xu, 2019).
Specifically, Lai (2014) studied how network ties can affect the survival of mixed-mode groups

138

that operate both online and offline. TeBlunthuis et al. (2020) proposed extending ecology theory
to studying online collective action and found that the Internet is a competitive space for
collective actions. They found that collective actions in topical areas with high density are
competitive and therefore attract less participation. Xu (2019) analyzed the influence of
ecological factors on network structures and found that category density in an online
crowdsourcing community positively affects users’ rates of entry.
However, what remained unanswered in these studies is how ecological factors are
related to the structure of communication networks. Social networks take resources to form and
maintain (Monge et al., 2008), and networks also channel resource access (Doerfel et al., 2013).
Networks are often solutions for a lack of resources (Doerfel et al., 2013). Therefore, it is vital
for researchers to understand the relationship between resources and social network structures.
This research suggests that niches, conceptualized as platform categories, can potentially
influence their social network structures.
This research design also suggests that in online communities, where resources are less
tangible, categories are useful means to identify resource distributions. Platform categories are
important ecological resource niches of online communities for several reasons. They are
institutionalized by the platform (Hsu & Hannan, 2005), and therefore signal members’ identities
within a contest. Relatedly, they also have cognitive meanings. Supported by the results for
Hypotheses 2a and 2b, focal closure was the significant mechanism driving cross-level closure,
and membership closure was rare. Therefore, categories attract like-minded audiences (Durand,
2017). Resources contained in categories can include but are not limited to attention, audience,
knowledge, and information in each area. In conclusion, categories can be used to identify

139

populations in online communities effectively. Because membership closure was rare, category
entry was not significantly influenced by communication.
Driven by the inquiry of how ecological factors and social interactions are related, this
current study included categories as endogenous to networks so that individuals’ communication
and relationship with platform categories were examined simultaneously. This research also
contributes to an emerging field of research that applies a multilevel network approach to
studying ecology (e.g., Amati et al., 2019; Xu, 2019). The multilevel application opens up the
possibilities for including other ecological factors such as social issues (Yang, 2020), intra-
organizational activities (Amati et al., 2019), expertise (Fulk, 2016), categories (Xu, 2018), and
identities (Brickson, 2007).
With the multilevel network approach, the categorized resource space was better
understood, and knowledge exchange patterns within and across the populations were also
observed. For example, the significant focal closure pattern indicates that intrapopulation
communication is frequently observed in a crowdsourcing contest community, suggesting that
individuals tend to seek knowledge and information from others sharing similar interests and
areas of expertise (Aiello et al., 2012). Such homophily is also often observed in knowledge
communities (Zhang & van der Schaar, 2017), social media (Aiello et al., 2012), and online
collective action (Xu & Zhou, 2020), and this type of homophily can be summarized as content
homophily (Xu & Zhou, 2020). Content homophily describes the tendency for like-minded
individuals to maintain social ties because of shared interest in specific content. This finding
provides preliminary evidence of the importance of content in driving social interactions in an
ideation innovation community. Finally, a dark side of homophily should be made clear because

140

it is often linked with “segregation, inequality, and reduced inter-group information transfer”
when an echo chamber is formed among like-minded people (Passe et al., 2018, p. 262).
5.2.3. Developing Network Evolution Theory
This research applied the V-S-R framework to studying network evolution in
crowdsourcing contest communities (Monge et al., 2008). Specifically, variation and selection
were examined. There are two contributions. First, this research adds insights to understanding
the relationship between environment and the sources and directions of network variation (e.g.,
Wang et al., 2016; Yang, 2020). This research found that in online communities where
communication ties have representational value (Shumate & Contractor, 2013), communication
stakes can be high, and new entrants may feel intimidated and avoid network variation in the
form of offering comments to others’ ideas. Moreover, in contest crowdsourcing communities
that are often temporary and independent of previous contests, resource differences between the
new entrants and existing members may not be big enough to motivate communication between
them. These findings suggest that network evolution research should carefully consider the
characteristics of the community infrastructures.
Second, this research also contributes to a line of research that has applied V-S-R to
studying network evolution (e.g., Fu, 2019; Shen et al., 2014a; Margolin et al., 2015) by pointing
out that the interdependencies among populations should be included in network evolution
research. Drawing from Meysken et al. (2010) and Power et al. (2005), this research proposes
that population interdependencies can be observed directly through social networks. This
approach opens the possibilities of studying the co-evolution of the multiple networks of a whole
community, which pushes the development of network evolution theory from focusing on single

141

populations to analyzing an entire community of interdependent populations (Astley, 1985;
Meyskens et al., 2010; Powell et al., 2005).
5.2.4. Developing CoP
Finally, this research also contributes to the CoP literature (e.g., Rivera & Cox, 2016;
Wenger et al., 2002). The current research suggests that multiple interconnected groups with
different identities and resource needs should be studied simultaneously as a whole CoP. This
theoretical development speaks to Gilbert’s (2016) argument that online CoPs are large and more
diverse than offline counterparts. This research also studied community members’ voluntary
contributions to other members they were competing against using a longitudinal and network
approach (Wasko et al., 2009). Longitudinal network exchange patterns were examined. Direct
reciprocity, indirect reciprocity, and triadic closures (Poquet & Dawson, 2018) provided a more
nuanced understanding of the CoP structures than dyadic knowledge exchange research (Pan et
al., 2015). Although still rare in the CoP literature, studying longitudinal network structures has
been proven helpful in understanding the changes in norms and social selection patterns in CoPs
(Wang et al., 2019).
According to the previous literature, the reasons centrality is closely related to
contribution can be both altruistic and self-interested. Wasko and Faraj (2005) explained that
central individuals have formed a habit of contribution through either receiving or sending
comment ties, and therefore are more motivated to have high contribution activity. However,
Baek and Bae (2019) pointed out that people with high centrality are more motivated to keep
occupying advantageous network positions, and therefore are more active in knowledge
contribution. In the current research, degree centrality in the ideator network contributed to
knowledge contribution, but the frequency of interaction with visitors who were cooperative did

142

not predict knowledge contribution. This result indicates that Baek and Bae’s (2019) view is
more relevant to the crowdsourcing contest CoPs than Wasko and Faraj (2005)’s perspective.
Ideators are core CoP members as they directly contribute ideas that are being evaluated and
commented. Visitors are peripheral and temporary (Safadi et al., 2018) as they do not submit
ideas and only give feedback by commenting on other’s ideas, and many only give comments a
few times throughout the contests. Therefore, occupying a central ideator network, is more
rewarding than occupying a central position in the visitor network. This argument is also
supported in Wang et al.’s (2019) finding that connecting with serial members of CoPs helps
increase visibility and connections with broader networks. Therefore, ideators highly connected
with other ideators benefit more from the centrality and are motivated to maintain such
advantages. In comparison, interacting with visitors does not provide the same level of reward
and is less motivating for knowledge contribution. Finally, network inertia was also found to be
significant in network evolution. This might indicate that members in the CoP tend to resist
network changes regardless of the shifts in contest stages. Structural inertia can be described as
results of previous successful synergy (Kim et al., 2006). Future research should dive in and
analyze the cause of inertia and its influence on innovation.
5.3. Practical Contributions
This research analyzed the social network structures on contest crowdsourcing platforms.
Using this approach, mechanisms (e.g., homophily, three types of reciprocity) that drive the
formation and evolution of social networks have been identified. However, in the meantime, this
research also noted the lack of communication across populations that platform managers,
contest organizers, and ideators should pay special attention to.

143

This finding on focal closure indicates that ideators preferred to exchange comments with
others sharing the same categories, but the post hoc result also indicates that during the review
stage, ideators were not likely to exchange comments with other ideators outside their own
categories. As discussed, this finding suggests that platform categories have the power to define
populations in online communities, and the category boundaries are directly related to social
boundaries too. Therefore, categories can enable communication within populations and
discourage communication to other different populations. For knowledge communities like
online crowdsourcing contest communities, platform designers should be aware of the impact of
categories on social interactions and be aware of the possibility of communication siloes being
created by categories. Online communities have advantages in knowledge innovation because of
the diverse knowledge, background, and interests participants possess (Faraj et al., 2016).
Interactions among people of wide diversity can be beneficial for quality brainstorming
(Nickerson & Sakamoto, 2010) and crowd wisdom (Surowiecki, 2004). Platform organizers
should strategically “design the network” (Nickerson & Sakamoto, 2010; p. 1), and they should
encourage interactions among diverse individuals. Individual ideators should also be strategic
about category submission, as such behavior leads to exposure to a certain group of people with
the same category submission.
This research also found that new entrants and existing members were unlikely to share
comment ties. Still, this mutualist connection could benefit them in the long run because those
who exchanged many mutualist connections evolved to become preferred network partners. This
result suggests that a communication vacuum tends to exist between new and existing members.
Platform managers, contest organizers, and ideators should also be aware of the separation

144

between the two types of actors and discuss or implement ways to tear down the wall between
the two types of members.
Relatedly, this research found that new entrants were reluctant to engage in network
variation. The reason might be that they are unfamiliar with the platform and the communication
stake can be high on contest crowdsourcing communities. It may also indicate that existing
members are forming closed clusters, and it is hard for new entrants to enter. Future research
should employ methods such as field research, survey, or interview to understand whether it is
possible that the community can be unfriendly to new entrants. Platform organizers should
design programs to familiarize new entrants with the communities and encourage them to be
more active communicators.
5.4. Limitations
OpenIDEO, the focal platform being investigated in this research, is very structured in its
organizing style. Specifically, the platform has a clear timeline for each challenge, in which all
ideators are required to complete idea submission by a deadline. The existence of this deadline
may have directly led to more focal closure structures in comparison to membership closure. The
reason is that ideators were required to submit their ideas by a specific time, and they may not
have had enough time to socialize with others before idea submission was required and only
started doing so after submission. Deadlines may also directly lead to membership closure, the
influence mechanism, because ideators might feel stressed under the time pressure, and
individuals have the tendency to imitate others under stress (Buckert et al., 2017). In summary,
this current research did not disentangle the effect of the platform and contest designs in the
formation of the cross-level closure mechanism. Future research efforts could further untangle
such effects through interviews. More empirical research efforts could also be contributed to

145

studying different types of platforms with diverse designs. For example, future research could
replicate this research using a less strictly structured community and compare findings with this
current paper.
Network retention was not examined in the current research because, in crowdsourcing
contest communities, each contest-organized community is temporary, which only exists through
the duration of the contests. The platform also did not have an infrastructure that encouraged
long-term relationships (e.g., follower-followee relationships, friendships, mentor-mentee
relationships). Future research endeavors should include surveys or interviews to determine
whether some network ties can be retained as routines on crowdsourcing contest communities. If
so, what type of connections lead to retention? Moreover, hypotheses and research questions
were only tested in one contest. Future research should collect data on more contests to see
whether the result can be replicated across contests.
5.5. Future Direction
Based on Aldrich’s (2020) summary of the population emergence literature, several
population emergence processes might be relevant to the ideator populations. These processes
include (1) the emergence of key pioneering individuals diverging from existing forms, (2) the
announcement of a new population form through media or other public channels, (3) social
movements or collective actions that advocate a new form, (4) the reinforcement of new
regulations that apply to new populations, or (5) the publication of innovative ideas in the forms
of products, research articles, patents, and more. As discussed, the “other” category might be a
diverse space that contains various new forms, and it might also be a stage before population
emergence. Future research should identify whether the processes summarized by Aldrich (2020)
exist within this category. Note, these population emergence processes often require room for

146

new population emergence and time for emergence. Crowdsourcing contests tend to be highly
structured because platform categories are determined by organizers and therefore constrain the
room for new population emergence.
The nature of the “other” category also needs more careful research to understand. If it is
an undefined resource space filled with variations, it may be capable of new category emergence
(Durand & Khaire, 2017). Category emergence is still a under-researched area of study (Durand
& Khaire, 2017; Kennedy & Fiss, 2013). According to Durand and Khaire (2017), the emergence
of novel categories requires conditions such as (1) novel attributes that is different from
preexisting categories, (2) promoters of the new categories, (3) legitimacy of the new category.
Therefore, future research should first more carefully analyze the nature of the “other” category,
and then observe the possibility of new category emergence. Such category can be explicit or
latent.
Moreover, each challenge usually only lasts about three or four months, and each contest
is often based on entirely different themes and attracts a different crowd of participants and
audiences. The aforementioned population emergence processes often take time. Therefore, the
temporariness of contest crowdsourcing sites may pose challenges for population emergence.
Future research efforts can be devoted to understanding the influence of the temporariness
feature of crowdsourcing contest sites on population emergence and analyze whether populations
face extra challenges emerging. Future research can also study the “other” category in a less
structured community that lasts longer and examine the evolution direction of such a category.
Will new forms emerge out of such a category? Are these new forms similar or different? What
are the processes through which these new forms emerge?

147

Moreover, although within-population legitimation cannot be observed, emerging
populations with the “other” population may still employ legitimacy-seeking strategies to
increase their chances of success. Legitimacy can be cognitive or sociopolitical (Aldrich et al.,
2020). Cognitive legitimacy describes “the acceptance of a new kind of venture as a taken-for-
granted feature of the environment,” and sociopolitical legitimacy refers to the “acceptance by
key stakeholders” (Aldrich et al., 2020, p. 195). Cognitive legitimacy strategies include imitating
dominant ideas (David, 1985), strategic negotiation on the creation of categories and collective
identities (Durand & Khaire, 2017), collective action (Esparza et al., 2014), and knowledge
diffusion and raising of public recognition in the community (Delacroix & Rao, 1994), etc.
Strategies to achieve sociopolitical legitimacy include collective action, cooperation with other
populations, etc. (Aldrich et al., 2020). Future research could explore whether these within-
population strategies are adopted by emerging populations and what strategies are more effective
for temporary crowdsourcing contest communities.
This research only teases out how categories shape individuals’ interaction patterns.
Future studies should also consider the possibilities that users’ interactions may, in turn, shape
categories in the ecological space. For example, in platforms where users create their own
categories (Oh & Monge, 2013), such as Twitter, where users create hashtags to label their
communication, certain hashtags may become popular as a result of diffusion of community
members’ communication interactions. This idea echoes the recent development of
categorization theory, which argues that categories can be socially constructed (Hannan et al.,
2019). The role of social interaction is pronounced in the legitimation and institutionalization of
emerging categories (Durand et al., 2017). Specifically, categories’ meanings and boundaries can

148

be socially construed and negotiated by people with different goals and agendas (Durand et al.,
2017).
According to ecology theory, individuals occupying the same resource niches share niche
overlap as they have similar resource requirements (Baum & Singh, 1994). Individuals’ shared
niche overlap weight is positively related to competition intensity. An individual’s niche overlap
weights can be calculated by the overlapping niche over the total niches of an individual (Baum
& Singh, 1994). In the current research, the overall pattern of the community was that those
individuals in the same population sharing resource overlap tended to share communication ties.
The niche overlap weights, however, were not calculated for ideators.
Moreover, categories may not be completely independent of each other, as they may be
distributed across conceptual spaces where some are conceptually close to each other while
others may be distant (Hannan et al., 2019). Different categories may vary conceptually. For
example, Xu and Zhou (2020) found Twitter users adopting similar hashtags tend to share
communication ties, and different types of hashtags present various types of homophily patterns.
They found that #MeToo communities attract like-minded individuals while #MAGA alienates
outgroup users. More nuanced differences among categories may exist that could affect social
interactions among individuals. Future research should provide a fuller consideration of various
ecological factors such as niche overlap weights and interdependencies among categorized
niches. In addition to explicit categories that can help identify similar content, latent categories
can also identify content that draws similar people together (Conover et al., 2011). Therefore,
future research efforts could also be devoted to identifying the relationship between latent niches
and social interactions. Relatedly, the network evolution perspective in this research highlights
the importance of studying population interdependencies. Although this research identified the

149

interdependencies between the new entrants and existing members, future research should extend
the application of this idea to study the interdependencies of other types of populations, such as
populations identified by categories or latent categories.
This research also adds to a few ecological studies that have analyzed population
interdependencies directly, using social network analysis (e.g., Meyskens et al., 2010; Powell et
al., 2005; Weber, 2012). Communication networks contain important insights about the nature of
the relationships among network nodes. However, due to the culture of the platform examined in
this study, community members’ comments were mostly positive. It would be interesting for
future research to study other platforms where both positive and negative communication exists,
or to seek beyond positive comments and study the more profound nature of the relationship
through interviews or surveys.
Relatedly, research findings suggest that interpopulation interdependencies should be
included when studying network evolution. This argument also speaks to the classic theoretical
advancement from population ecology, which only studies a single population, to community
ecology (Astley, 1985; Powell et al., 2005). Similarly, network evolution should also be analyzed
from a community ecology perspective instead of a population one (Powell et al., 2005).
Community-level studies can tease out how “many populations are closely linked via their
orientation to a common technology, normative order, or legal-regulatory regime” (Aldrich, et
al., 2019, p. 257). As is evident in Powell and colleagues’ (2005) research, the network evolution
of interconnected populations of organizations (e.g., for-profit companies, government agencies,
venture capitalists) co-evolve in the biotechnology community. For example, they found that
National Institute of Health (NIH) establish ties with new entrants to help their product
development so that they became less constrained by big and dominating companies. The

150

collaboration between NIH and new entrants, in turn, attracted connections from venture
capitalists, which eventually drove many giant companies out of the market. Specifically for
online community studies, how different types of populations co-evolve in the same platform, or
crowds working for a similar cause interact over multiple platforms, can be studied using
community-level analysis. Future research could also explore the evolution and interplay of the
other types of interpopulation relationships listed in Aldrich (2020): the six types of
commensalism, symbiosis, and dominance (details can be found in Table 3). Analyzing the
structure and evolution of these different types of interpopulation connections allows us to
understand how populations gain resources from the existence of each other and/or compete with
each other (Powell et al., 2005), which provides a more nuanced understanding of the organizing
of virtual communities.
This current research did not find network variation to be related to ideators’ identity of
being new entrants or existing members. Future research should keep exploring what factors may
drive network variation or whether the intention of network variation even exists in
crowdsourcing contest communities, considering the duration of each contest is very short. As
summarized in Aldrich et al., (2020), sources of variation often come from organizations’
intentional nudges such as formal training and incentives, “everyday variation” caused by
activities such as “trial-and-error learning, luck, imitation, mistakes, passion, misunderstanding,
idle curiosity,” or unexpected environmental shifts (p. 20). For online crowdsourcing contest
platforms, network variations may be generated from networking programs or creativity training
programs organized on the platform, environmental shifts such as the renovation of the website,
or more. Whether these forms of variations exist can be researched in the future. Relatedly,
future research should also test whether being new entrants or existing members drives network

151

variation in other crowdsourcing communities that do not feature short-term contests and see
whether the hypotheses in this article can find support in communities that are more stable and
consistent than crowdsourcing contest communities.
Finally, it is important to note that resources required to survive and succeed on online
communities might be different from organizational populations that are often studied in
organizational ecology literature (e.g., Aldrich et al., 2020; Hannan & Freeman, 1987). The role
of material goods might not be very important for users in online communities, but other less
tangible resources such as identification, information, knowledge, and norms may become
important. Future research applying ecological and evolutionary perspectives to studying online
communities should find ways to understand the composition and nature of resources required in
online communities to succeed. It is also important to note that one assumption of information
goods is “jointness of supply” (Fulk et al., 1996, p. 60). This concept describes that information
goods in online communities cannot be reduced because of other people’s consumption.
Therefore, in online communities, users may not need to compete intensely for information
goods access, which may also explain the overall cooperative atmosphere of the community.
Therefore, ecological and evolutionary theories are very useful to study online communities, and
researchers should push the boundary of the theory to study more diverse online behaviors.
However, more research efforts should also be devoted to carefully theorizing the nature of the
community ecology.
5.6. Conclusion
In conclusion, this research applied ecology theory, network evolution theory, and CoP
literature to studying the network formation and evolution of an online crowdsourcing contest
community. First, this research suggests that platform categories can be used to conceptualize

152

resource niches of online platforms. Using a longitudinal and multilevel approach, this current
research distinguished the homophily and influence effects, and found that content homophily
drives network formation in the contest crowdsourcing community. This study contributes to
ecology theory by identifying how categorized resource niches are related to social interactions.
Second, this research noted that population interdependencies should be included when
analyzing network evolution (Aldrich et al., 2020; Astley, 1985; Powell et al., 2005), which
pushes future network evolution research to shift from focusing on a single population to
multiple interrelated populations. This research did not identify the tendency for new entrants
and existing members to connect but found such mutualist connections between the two
populations hold benefits as they predict future incoming network ties. Finally, this research also
suggests multiple different populations of platform users should be identified and studied as a
whole CoP. Specifically, this research argues that crowdsourcing contest CoPs should study not
only ideators but also visitors. A longitudinal network approach was adopted, and interesting
knowledge contribution patterns were identified.
Practical implications were also discussed. Specifically, this research helps understand
the organization of crowdsourcing contest communities. Communication vacuums existed
between new entrants and existing members, but considering such ties are beneficial, platform
managers and contest organizers should discuss ways to tear down the communication walls
between the two populations. Contest participants should strategically manage their
communication with people from outside of their own populations to achieve advantageous
network positions in the future. Finally, crowdsourcing contest communities are overall
cohesive, and the network formation and evolution were driven by content homophily, direct and
indirect reciprocity, and triadic closure. Platform managers and contest organizers should know

153

the benefits and drawbacks of these forces on knowledge innovation, and design and manage the
platform strategically. Limitations and future research directions were also discussed.

154

References
Abbott, K. W., Green, J. F., & Keohane, R. O. (2016). Organizational ecology and institutional
change in global governance. International Organization, 70(2), 247–277.
https://doi.org/10.1017/S0020818315000338
Aiello, L. M., Barrat, A., Schifanella, R., Markines, B., & Menczer, F. (2012). Friendship
prediction and homophily in social media. ACM Transactions on the Web, 6(2), 9–33.
https://doi.org/10.1145/2180861.2180866
Aldrich, H. E., Ruef, M., & Lippmann, S. (2020). Organizations evolving: (Third edition).
Edward Elgar Publishing.
Aldrich, H. E., & Wiedenmayer, G. (2019). From traits to rates: An ecological perspective on
organizational foundings. In J. A. Katz & A. C. Corbet (Eds.), Seminal Ideas for the Next
Twenty-Five Years of Advances (Vol. 21, pp. 61–97). Emerald Publishing Limited.
https://doi.org/10.1108/S1074-754020190000021010
Aldrich, H. E., & Ruef, M. (1999). Organizations evolving (2nd ed.). Sage.
Aldrich, H. E., & Ruef, M. (2006). Organizations evolving. SAGE Publications, Inc.
Alexy, O., & George, G. (2013). Category divergence, straddling, and currency: Open innovation
and the legitimation of illegitimate categories. Journal of Management Studies, 50(2),
173–203. https://doi.org/10.1111/joms.12000
Amati, V., Lomi, A., Mascia, D., & Pallotti, F. (2019). The co-evolution of organizational and
network structure: The role of multilevel mixing and closure mechanisms.
Organizational Research Methods, 109442811985746.
https://doi.org/10.1177/1094428119857469

155

Askin, N., & Mauskapf, M. (2017). What makes popular culture popular? Product features and
optimal differentiation in music. American Sociological Review, 82(5), 910–944.
https://doi.org/10.1177/0003122417728662
Astley, W. G. (1985). The two ecologies: Population and community perspectives on
organizational evolution. Administrative Science Quarterly, 30(2), 224–241.
https://doi.org/10.2307/2393106
Atouba, Y., & Shumate, M. (2010). Interorganizational networking patterns among development
organizations. Journal of Communication, 60(2), 293–317.
https://doi.org/10.1111/j.1460-2466.2010.01483.x
Audia, P. G., Freeman, J. H., & Reynolds, P. D. (2006). Organizational foundings in community
context: Instruments manufacturers and their interrelationship with other organizations.
Administrative Science Quarterly, 51(3), 381–419. https://doi.org/10.2189/asqu.51.3.381
Baek, S. I., & Bae, S. H. (2019). The effect of social network centrality on knowledge sharing.
Journal of Service Science Research, 11(2), 183–202. https://doi.org/10.1007/s12927-
019-0009-2
Baldwin, C., & Von Hippel, E. (2011). Modeling a paradigm shift: From producer innovation to
user and open collaborative innovation. Organization Science, 22(6), 1399–1417.
Barnett, W. P. (1997). The dynamics of competitive intensity. Administrative Science Quarterly,
42(1), 128–160. JSTOR. https://doi.org/10.2307/2393811
Baum, J. A. C., & Singh, J. V. (1994). Organizational niches and the dynamics of organizational
mortality. American Journal of Sociology, 100(2), 346–380.

156

Bayus, B. L. (2013). Crowdsourcing new product ideas over time: An analysis of the Dell
IdeaStorm community. Management Science, 59(1), 226–244.
https://doi.org/10.1287/mnsc.1120.1599
Becker, J., Brackbill, D., & Centola, D. (2017). Network dynamics of social influence in the
wisdom of crowds. Proceedings of the National Academy of Sciences, 201615978.
https://doi.org/10.1073/pnas.1615978114
Bigelow, L., Nickerson, J. A., & Park, W. Y. (2019). When and how to shift gears: Dynamic
trade-offs among adjustment, opportunity, and transaction costs in response to an
innovation shock. Strategic Management Journal, 40(3), 377–407.
https://doi.org/10.1002/smj.2996
Bighash, L., Oh, P., Fulk, J., & Monge, P. R. (2018). The value of questions in organizing:
reconceptualizing contributions to online public information goods. Communication
Theory, 28(1), 1–21. https://doi.org/10.1111/comt.12123
Blohm, I., Leimeister, J. M., & Krcmar, H. (2013). Crowdsourcing: How to benefit from (too)
many great ideas. MIS Quarterly Executive, 12(4), 199–211.
Bodin, Ö., & Tengö, M. (2012). Disentangling intangible social–ecological systems. Global
Environmental Change, 22(2), 430–439. https://doi.org/10.1016/j.gloenvcha.2012.01.005
Boudreau, K. J., & Lakhani, K. R. (2013). Using the crowd as an innovation partner. Harvard
Business Review, 91(4), 60–69.
Brabham, D. C. (2012). Motivations for participation in a crowdsourcing application to improve
public engagement in transit planning. Journal of Applied Communication Research,
40(3), 307–328. https://doi.org/10.1080/00909882.2012.693940

157

Brandes, U., Lerner, J., & Snijders, T. A. B. (2009). Networks evolving step by step: statistical
analysis of dyadic event data. 2009 International Conference on Advances in Social
Network Analysis and Mining, 200–205. https://doi.org/10.1109/ASONAM.2009.28
Brickson, S. L. (2007). Organizational identity orientation: The genesis of the role of the firm
and distinct forms of social value. Academy of Management Review, 32(3), 864–888.
Brittain, J. (1994). Density-independent selection and community evolution. In Evolutionary
dynamics of organizations (pp. 355–378). Academic Press.
Brunswicker, S., Bilgram, V., & Fueller, J. (2017). Taming wicked civic challenges with an
innovative crowd. Business Horizons, 60(2), 167–177.
https://doi.org/10.1016/j.bushor.2016.11.001
Bryant, A. J., & Monge, P. R. The evolution of the children’s television community, 1953-2003.
International Journal of Communication, 2 (2008), 160-192.
Buckert, M., Oechssler, J., & Schwieren, C. (2017). Imitation under stress. Journal of Economic
Behavior & Organization, 139, 252–266. https://doi.org/10.1016/j.jebo.2017.04.014
Bullinger, A. C., Neyer, A. K., Rass, M., & Moeslein, K. M. (2010). Community-based
innovation contests: where competition meets cooperation. Creativity and Innovation
Management, 19(3), 290–303. https://doi.org/10.1111/j.1467-8691.2010.00565.x
Burshell, J., & Mitchell, W. (2017). The social construction of market categories: How
proximate social space creates strategic incentives to be early claimants of the fiscal
sponsor label. In Emergence (Vol. 50, pp. 171–221). Emerald Publishing Limited.
https://doi.org/10.1108/S0733-558X20170000050006
Burt, R. S. (1992). Structural holes: The social structure of competition. Harvard university
press.

158

Butts, C. T. (2008). S relational event framework for social action. Sociological Methodology, 3,
155–200.
Campbell, D. T. (1965). Variation and selective retention in socio-cultural evolution. Social
Change in Developing Area.
Carroll, G. R. (1985). Concentration and specialization: Dynamics of niche width in populations
of organizations. American Journal of Sociology, 90(6), 1262–1283.
Carroll, G. R., Dobrev, S. D., & Swaminathan, A. (2002). Organizational processes of resource
partitioning. Research in Organizational Behavior, 24, 1–40.
https://doi.org/10.1016/S0191-3085(02)24002-2
Carroll, G., & Khessina, O. (2006). The ecology of entrepreneurship. https://doi.org/10.1007/0-
387-23622-8_9
Carroll, G. R., & Swaminathan, A. (2000). Why the microbrewery movement? Organizational
dynamics of resource partitioning in the U.S. Brewing industry. American Journal of
Sociology, 106(3), 715–762. https://doi.org/10.1086/318962
Carroll, G. R., & Wade, J. (1991). Density dependence in the organizational evolution of the
American brewing industry across different levels of analysis. Social Science Research,
20(3), 271–302.
Castells, M. (2002). The Internet galaxy: Reflections on the Internet, business, and society.
Oxford University Press on Demand.
Castells, M. (2011). The rise of the network society (Vol. 12). John wiley & sons.
Checkley, M., Steglich, C., Angwin, D., & Endersby, R. (2014). Firm performance and the
evolution of cooperative interfirm networks: UK venture capital syndication. Strategic
Change, 23(1–2), 107–118.

159

Conover, M. D., Goncalves, B., Ratkiewicz, J., Flammini, A., & Menczer, F. (2011). Predicting
the political alignment of Twitter users. 2011 IEEE Third International Conference on
Privacy, Security, Risk and Trust and 2011 IEEE Third International Conference on
Social Computing, 192–199. https://doi.org/10.1109/PASSAT/SocialCom.2011.34
Contractor, N. (2009). The emergence of multidimensional networks. Journal of Computer-
Mediated Communication, 14(3), 743–747. https://doi.org/10.1111/j.1083-
6101.2009.01465.x
Contractor, N. S., Monge, P. R., & Leonardi, P. M. (2011). Multidimensional networks and the
dynamics of sociomateriality: bringing technology inside the network. International
Journal of Communication, 39.
Dahlander, L., & Frederiksen, L. (2012). The core and cosmopolitans: a relational view of
innovation in user communities. Organization Science, 23(4), 988–1007.
Darwin, C., 1809-1882. (1859). On the origin of species by means of natural selection, or
preservation of favoured races in the struggle for life. London: John Murray, 1859.
David, P. A. (1985). Clio and the Economics of QWERTY. The American Economic Review,
75(2), 332–337.
Davis, A. P. (2017). A social ecology of civil conflict: Shifting allegiances in the conflict in
Sierra Leone. Social Science Research, 67, 115–128.
https://doi.org/10.1016/j.ssresearch.2017.06.002
Delacroix, J., & Rao, H. (1994). Externalities and ecological theory: Unbundling density
dependence. Evolutionary Dynamics of Organizations, 255, 268.

160

DiMaggio, P., & Powell, W. W. (1983). The iron cage revisited: Collective rationality and
institutional isomorphism in organizational fields. American Sociological Review, 48(2),
147–160.
Dissanayake, I., Zhang, J., & Gu, B. (2014). Virtual team performance in crowdsourcing
contests: A social network perspective completed research paper. In 35th International
Conference on Information Systems "Building a Better World Through Information
Systems", ICIS 2014 Association for Information Systems.
Dobrev, S. D. (2001). Revisiting organizational legitimation: cognitive diffusion and
sociopolitical factors in the evolution of bulgarian newspaper enterprises, 1846-1992.
Organization Studies, 22(3), 419–444. https://doi.org/10.1177/0170840601223002
Doerfel, M. L., Chewning, L. V., & Lai, C.H. (2013). The evolution of networks and the
resilience of interorganizational relationships after disaster. Communication Monographs,
80(4), 533–559. https://doi.org/10.1080/03637751.2013.828157
Doerfel, M. L., Lai, C. H., & Chewning, L. V. (2010). The evolutionary role of
interorganizational communication: modeling social capital in disaster contexts. Human
Communication Research, 36(2), 125–162. https://doi.org/10.1111/j.1468-
2958.2010.01371.x
Durand, R., Granqvist, N., & Tyllström, A. (2017). From categories to categorization: Studies in
sociology, organizations and strategy at the crossroads. Emerald Publishing Limited.
Durand, R., & Khaire, M. (2017). Where do market categories come from and how?
Distinguishing category creation from category emergence. Journal of Management,
43(1), 87–110.

161

Eckert, P. (2006). Communities of practice. Encyclopedia of language and linguistics, 2(2006),
683-685.
Esparza, N., Walker, E. T., & Rossman, G. (2014). Trade associations and the legitimation of
entrepreneurial movements: Collective action in the emerging gourmet food truck
industry. Nonprofit and Voluntary Sector Quarterly, 43(2_suppl), 143S-162S.
https://doi.org/10.1177/0899764013512723
Faraj, S., Jarvenpaa, S. L., & Majchrzak, A. (2011). Knowledge collaboration in online
communities. Organization Science, 22(5), 1224–1239.
https://doi.org/10.1287/orsc.1100.0614
Faraj, S., & Shimizu, T. (2018). Online communities and knowledge collaborations. Oxford
Research Encyclopedia of Business and Management.
https://doi.org/10.1093/acrefore/9780190224851.013.146
Faraj, S., von Krogh, G., Monteiro, E., & Lakhani, K. R. (2016). Special section introduction—
online community as space for knowledge flows. Information Systems Research, 27(4),
668–684. https://doi.org/10.1287/isre.2016.0682
Freeman, J., & Hannan, M. T. (1989). Setting the record straight on organizational ecology:
Rebuttal to young. American Journal of Sociology, 95(2), 425–439.
https://doi.org/10.1086/229275
Friedkin, N. E. (2001). Norm formation in social influence networks. Social Networks, 23(3),
167–189.
Fu, J. S. (2019). Unpacking the influence of informational, organizational, and structural factors
on the longitudinal change of the NPO follower-followee network on Twitter.
International Journal of Communication, 24.

162

Fu, J. S., & Shumate, M. (2015). Social media activity and hyperlink network analysis: a holistic
media ecology perspective. 2015 48th Hawaii International Conference on System
Sciences, 1808–1817. https://doi.org/10.1109/HICSS.2015.218
Fuger, S., Schimpf, R., Füller, J., & Hutter, K. (2017). User roles and team structures in a
crowdsourcing community for international development – a social network perspective.
Information Technology for Development, 23(3), 438–462.
https://doi.org/10.1080/02681102.2017.1353947
Fulk, J. (2016). Conceptualizing multilevel expertise. In Expertise, communication and
organizing. Oxford University Press.
Fulk, J., Flanagin, A. J., Kalman, M. E., Monge, P. R., & Ryan, T. (1996). Connective and
communal public goods in interactive communication systems. Communication Theory,
6(1), 60–87.
Füller, J., Hutter, K., & Fries, M. (2012). Crowdsourcing for goodness sake: Impact of incentive
preference on contribution behavior for social innovation. In Interdisciplinary
Approaches to Product Design, Innovation, & Branding in International Marketing.
Emerald Group Publishing Limited.
Gaggioli, A. (2019). Potential of open innovation platforms for solving social challenges.
Cyberpsychology, Behavior, and Social Networking, 22(6), 428–429.
https://doi.org/10.1089/cyber.2019.29154.csi
Gandhi, M. (2019, August 7). Several living organisms are neither animals nor plants, defy
traditional ways of classification.

163

Gargiulo, M., & Benassi, M. (2000). Trapped in your own net? Network cohesion, structural
holes, and the adaptation of social capital. Organization Science, 11(2), 183–196.
https://doi.org/10.1287/orsc.11.2.183.12514
Gilbert, S. (2016). Learning in a Twitter-based community of practice: an exploration of
knowledge exchange as a motivation for participation in# hcsmca. Information,
Communication & Society, 19(9), 1214-1232.
Giustiziero, G., Kaul, A., & Wu, B. (2019). The dynamics of learning and competition in
schumpeterian environments. Organization Science.
https://doi.org/10.1287/orsc.2018.1264
Granqvist, N., Grodal, S., & Woolley, J. L. (2013). Hedging your bets: explaining executives’
market labeling strategies in Nanotechnology. Organization Science, 24(2), 395–413.
https://doi.org/10.1287/orsc.1120.0748
Gupta, S., & Kim, H.-W. (2004). Virtual community: Concepts, implications, and future
research directions. 10, 2679–2687.
Guth, K. L., & Brabham, D. C. (2017). Finding the diamond in the rough: Exploring
communication and platform in crowdsourcing performance. Communication
Monographs, 84(4), 510–533.
Hannan, M. T., & Carroll, G. R. (1992). Dynamics of organizational populations: Density,
legitimation, and competition. Oxford University Press.
Hannan, M. T., Carroll, G. R., & Polos, L. (2003). The organizational niche. Sociological
Theory, 21(4), 309–340. https://doi.org/10.1046/j.1467-9558.2003.00192.x
Hannan, M. T., & Freeman, J. (1977). The population ecology of organizations. American
Journal of Sociology, 82(5), 929–964. https://doi.org/10.1086/226424

164

Hannan, M. T., & Freeman, J. (1984). Structural inertia and organizational change. American
Sociological Review, 149–164.
Hannan, M. T., & Freeman, J. (1987). The ecology of organizational founding: American Labor
Unions, 1836-1985. American Journal of Sociology, 92(4), 910–943.
https://doi.org/10.1086/228587
Hannan, M. T., Mens, G. L., Hsu, G., Kovács, B., Negro, G., Pólos, L., Pontikes, E., & Sharkey,
A. J. (2019). Concepts and categories: foundations for sociological and cultural analysis.
Columbia University Press.
Harary, F., & Batell, M. F. (1981). What is a system? Social Networks, 3(1), 29–40.
Hawley, A. H. (1944). Ecology and human ecology. Social Forces, 398–405.
Hawley, A. H. (1986). Human ecology: A theoretical essay. University of Chicago Press.
Haythornthwaite, C. (2007). Social networks and online community. The Oxford Handbook of
Internet Psychology, 121–137.
Hilbert, M., Oh, P., & Monge, P. R. (2016). Evolution of what? A network approach for the
detection of evolutionary forces. Social Networks, 47, 38–46.
https://doi.org/10.1016/j.socnet.2016.04.003
Hollway, J., Lomi, A., Pallotti, F., & Stadtfeld, C. (2017). Multilevel social spaces: The network
dynamics of organizational fields. Network Science, 5(2), 187–212.
https://doi.org/10.1017/nws.2017.8
Howe, J. (2006). The rise of crowdsourcing. Wired Magazine, 14(6), 1–4.
Howe, J. (2008). Why the power of the crowd is driving the future of business. The International
Achievement Institute.

165

Hsu, G., & Hannan, M. T. (2005). Identities, genres, and organizational forms. Organization
Science, 16(5), 474–490. https://doi.org/10.1287/orsc.1050.0151
Hunt, R. A. (2017). The oxpecker and the rhino: the positive effects of symbiotic mutualism on
organizational survival.
Hutter, K., Hautz, J., Füller, J., Mueller, J., & Matzler, K. (2011). Communitition: The tension
between competition and collaboration in community‐based design contests. Creativity
and Innovation Management, 20(1), 3–21.
Jonassen, D. H. (2004). Learning to solve problems: An instructional design guide (Vol. 6). John
Wiley & Sons.
Kennedy, M. T., & Fiss, P. C. (2013). An ontological turn in categories research: from standards
of legitimacy to evidence of actuality. Journal of Management Studies, 50(6), 1138–
1154. https://doi.org/10.1111/joms.12031
Kietzmann, J. H. (2017). Crowdsourcing: A revised definition and introduction to new research.
Business Horizons, 60(2), 151–153. https://doi.org/10.1016/j.bushor.2016.10.001
Kim, T.Y., Oh, H., & Swaminathan, A. (2006). Framing interorganizational network change: A
network inertia perspective. Academy of Management Review, 31(3), 704–720.
King, A., & Lakhani, K. R. (2013). Using open innovation to identify the best ideas. MIT Sloan
Management Review, 55(1), 41.
Koçak, Ö., Hannan, M. T., & Hsu, G. (2014). Emergence of market orders: Audience interaction
and vanguard influence. Organization Studies, 35(5), 765–790.
https://doi.org/10.1177/0170840613511751
Kraut, R. E., & Resnick, P. (2012). Building successful online communities: Evidence-based
social design. MIT Press.

166

Lai, C. H. (2014). Can our group survive? An investigation of the evolution of mixed-mode
groups. Journal of Computer-Mediated Communication, 19(4), 839–854.
https://doi.org/10.1111/jcc4.12075
Lakhani, K. R., Garvin, D. A., & Lonstein, E. (2010). Topcoder (a): Developing software
through crowdsourcing. Harvard Business School General Management Unit Case, 610–
032.
Latour, B. (2005). Reassembling the social: An introduction to actor-network-theory. Oxford
university press.
Lazega, E., Jourda, M. T., Mounier, L., & Stofer, R. (2008). Catching up with big fish in the big
pond? Multi-level network analysis through linked design. Social Networks, 30(2), 159–
176.
Lazega, E., Mounier, L., Jourda, M. T., & Stofer, R. (2006). Organizational vs. personal social
capital in scientists’ performance: A multi-level network study of elite French cancer
researchers (1996-1998). Scientometrics, 67(1), 27–44.
https://doi.org/10.1556/Scient.67.2006.1.3
Lazega, E., Mounier, L., Stofer, R., & Tripier, A. (2004). Discipline scientifique et discipline
sociale: Réseaux de conseil, apprentissage collectif et innovation dans la recherche
française sur le cancer (1997-1999). Recherches Sociologiques, 35(3), 3–28.
Lazega, E., & Snijders, T. A. B. (2016). Multilevel network analysis for the social sciences:
Theory, methods and applications. In E. Lazega & T. A. B. Snijders (Eds.), Multilevel
network analysis for the social sciences (pp. 355–361). Springer International Publishing.
https://doi.org/10.1007/978-3-319-24520-1_15

167

Lee, J., & Lee, H. (2010). The computer-mediated communication network: Exploring the
linkage between the online community and social capital. New Media & Society, 12(5),
711–727. https://doi.org/10.1177/1461444809343568
Lee, S., & Monge, P. R. (2011). The coevolution of multiplex communication networks in
organizational communities. Journal of Communication, 61(4), 758–779.
https://doi.org/10.1111/j.1460-2466.2011.01566.x
Lee, S., Monge, P. R., Bar, F., & Matei, S. A. (2007). The emergence of clusters in the global
telecommunications network. Journal of Communication, 57(3), 415–434.
https://doi.org/10.1111/j.1460-2466.2007.00350.x
Leenders, R. T. (1995). Structure and influence. Statistical Models for the Dynamics of Actor
Attributes, Network Structure and Their Interdependence.
Leimeister, J. M., Huber, M., Bretschneider, U., & Krcmar, H. (2009). Leveraging
crowdsourcing: Activation-supporting components for IT-based ideas competition.
Journal of Management Information Systems, 26(1), 197–224.
Lévy, P., & Bononno, R. (1997). Collective intelligence: Mankind’s emerging world in
cyberspace. Perseus books.
Liang, H., Wang, M. M., Wang, J. J., & Xue, Y. (2018). How intrinsic motivation and extrinsic
incentives affect task effort in crowdsourcing contests: A mediated moderation model.
Computers in Human Behavior, 81, 168–176. https://doi.org/10.1016/j.chb.2017.11.040
Lindberg, A., Berente, N., Gaskin, J., & Lyytinen, K. (2016). Coordinating interdependencies in
online communities: a study of an open source software project. Information Systems
Research, 27(4), 751–772. https://doi.org/10.1287/isre.2016.0673

168

Lomi, A., Robins, G., & Tranmer, M. (2016). Introduction to multilevel social networks. Social
Networks, 44, 266–268. https://doi.org/10.1016/j.socnet.2015.10.006
Lowrey, W. (2017). The emergence and development of news fact-checking sites: Institutional
logics and population ecology. Journalism Studies, 18(3), 376–394.
https://doi.org/10.1080/1461670X.2015.1052537
Lubbers, M. J., & Snijders, T. A. (2007). A comparison of various approaches to the exponential
random graph model: A reanalysis of 102 student networks in school classes. Social
Networks, 29(4), 489-507.
Ma, L., Krishnan, R., & Montgomery, A. L. (2014). Latent homophily or social influence? An
empirical analysis of purchase within a social network. Management Science, 61(2), 454–
473. https://doi.org/10.1287/mnsc.2014.1928
Majchrzak, A., Faraj, S., Kane, G. C., & Azad, B. (2013). The contradictory influence of social
media affordances on online communal knowledge sharing. Journal of Computer-
Mediated Communication, 19(1), 38–55. https://doi.org/10.1111/jcc4.12030
Majchrzak, A., & Malhotra, A. (2016). Effect of knowledge-sharing trajectories on innovative
outcomes in temporary online crowds. Information Systems Research, 27(4), 685–703.
https://doi.org/10.1287/isre.2016.0669
Malhotra, A., & Majchrzak, A. (2014). Managing crowds in innovation challenges. California
Management Review, 56(4), 103–123. https://doi.org/10.1525/cmr.2014.56.4.103
March, J. G. (1991). Exploration and exploitation in organizational learning. Organization
science, 2(1), 71-87.
Margolin, D. B., Shen, C., Lee, S., Weber, M. S., Fulk, J., & Monge, P. R. (2015). Normative
influences on network structure in the evolution of the children’s rights ngo network,

169

1977-2004. Communication Research, 42(1), 30–59.
https://doi.org/10.1177/0093650212463731
McKelvey, B. (1982). Organizational systematics—Taxonomy, evolution, classification.
University of California Press.
Merton, R. K. (1968). The Matthew Effect in Science: The reward and communication systems
of science are considered. Science, 159(3810), 56–63.
https://doi.org/10.1126/science.159.3810.56
Meyskens, M., Carsrud, A. L., & Cardozo, R. N. (2010). The symbiosis of entities in the social
engagement network: The role of social ventures. Entrepreneurship & Regional
Development, 22(5), 425–455. https://doi.org/10.1080/08985620903168299
Moliterno, T. P., & Mahony, D. M. (2011). Network theory of organization: A multilevel
approach. Journal of Management, 37(2), 443–467.
https://doi.org/10.1177/0149206310371692
Monge, P. R., Heiss, B. M., & Margolin, D. B. (2008). Communication network evolution in
organizational communities: Network evolution. Communication Theory, 18(4), 449–
477. https://doi.org/10.1111/j.1468-2885.2008.00330.x
Monge, P. R., & Contractor, N. S. (2003). Theories of communication networks. Oxford
University Press, USA.
Monge, P. R., & Poole, M. S. (2008). The evolution of organizational communication. Journal of
Communication, 58(4), 679–692. https://doi.org/10.1111/j.1460-2466.2008.00408.x
Muchnik, L., Aral, S., & Taylor, S. J. (2013). Social influence bias: A randomized experiment.
Science, 341(6146), 647–651. https://doi.org/10.1126/science.1240466

170

Niederman, F., & Lukyanenko, R. (2018). Categorization: A source of theory and output of
research. Proceedings of the 2018 ACM SIGMIS Conference on Computers and People
Research, 143–150. https://doi.org/10.1145/3209626.3209717
Nickerson, J. V., & Sakamoto, Y. (2010). Crowdsourcing creativity: Combining ideas in
networks. Workshop on Information in Networks.
Nonnecke, B., & Preece, J. (2001). Why lurkers lurk. Americas Conference on Information
Systems 2001, 10.
Noriega-Campero, A., Almaatouq, A., Krafft, P., Alotaibi, A., Moussaid, M., & Pentland, A.
(2018). The wisdom of the network: How adaptive networks promote collective
intelligence. 50.
O’Brien, N. F., Pilny, A., Atouba, Y. C., Shumate, M., Fulk, J. L., & Monge, P. R. (2019). How
does ngo partnering change over time? A longitudinal examination of factors that
influence ngo partner selection. Nonprofit and Voluntary Sector Quarterly, 48(6), 1229–
1249. https://doi.org/10.1177/0899764019854546
Ognyanova, K., & Monge, P. R. (2013). A multitheoretical, multilevel, multidimensional
network model of the media system: Production, content, and audiences. Annals of the
International Communication Association, 37(1), 67–93.
https://doi.org/10.1080/23808985.2013.11679146
Ohly, S., Kase, R., & Škerlavaj, M. (2010). Networks for generating and for validating ideas:
The social side of creativity. Innovation, 12(1), 41–52.
https://doi.org/10.5172/impp.12.1.41
Pan, Y., Xu, Y. (Calvin), Wang, X., Zhang, C., Ling, H., & Lin, J. (2015). Integrating social
networking support for dyadic knowledge exchange: A study in a virtual community of

171

practice. Information & Management, 52(1), 61–70.
https://doi.org/10.1016/j.im.2014.10.001
Paruchuri, S., Goossen, M. C., & Phelps, C. (2019). Conceptual foundations of multilevel social
networks. In S. E. Humphrey & J. M. LeBreton (Eds.), The handbook of multilevel
theory, measurement, and analysis. (pp. 201–221). American Psychological Association.
https://doi.org/10.1037/0000115-010
Passe, J., Drake, C., & Mayger, L. (2018). Homophily, echo chambers, & selective exposure in
social networks: What should civic educators do? The Journal of Social Studies
Research, 42(3), 261–271. https://doi.org/10.1016/j.jssr.2017.08.001
Pattison, P., & Wasserman, S. (1999). Logit models and logistic regressions for social networks:
II. Multivariate relations. British Journal of Mathematical and Statistical Psychology,
52(2), 169–193.
Pénin, J., & Helmchen, T. B. (2011). Crowdsourcing of inventive activities: Definition and
limits. International Journal of Innovation and Sustainable Development, 5(2/3), 246.
https://doi.org/10.1504/IJISD.2011.043068
Pilny, A., Schecter, A., Poole, M. S., & Contractor, N. (2016). An illustration of the relational
event model to analyze group interaction processes. Group Dynamics: Theory, Research,
and Practice, 20(3), 181–195. https://doi.org/10.1037/gdn0000042
Podolny, J. M., Stuart, T. E., & Hannan, M. T. (1996). Networks, knowledge, and niches:
competition in the worldwide semiconductor industry, 1984-1991. American Journal of
Sociology, 102(3), 659–689.
Poquet, O., & Dawson, S. (2018). Network patterns of direct and indirect reciprocity in edX
MOOC forums. In R. Alhajj, H. U. Hoppe, T. Hecking, P. Bródka, & P. Kazienko (Eds.),

172

Network intelligence meets user centered social media networks (pp. 49–64). Springer
International Publishing. https://doi.org/10.1007/978-3-319-90312-5_4
Powell, W. W., White, D. R., Koput, K. W., & Owen‐Smith, J. (2005). Network dynamics and
field evolution: The growth of interorganizational collaboration in the life sciences.
American Journal of Sociology, 110(4), 1132–1205. https://doi.org/10.1086/421508
Quintane, E., Conaldi, G., Tonellato, M., & Lomi, A. (2014). Modeling relational events: A case
study on an open source software project. Organizational Research Methods, 17(1), 23–
50. https://doi.org/10.1177/1094428113517007
Renard, D., & Davis, J. G. (2019). Social interdependence on crowdsourcing platforms. Journal
of Business Research, 103, 186–194. https://doi.org/10.1016/j.jbusres.2019.06.033
Rivera, G., & Cox, A. M. (2016). A practice-based approach to understanding participation in
online communities. Journal of Computer-Mediated Communication, 21(1), 17-32.
Robins, G., Pattison, P., & Wasserman, S. (1999). Logit models and logistic regressions for
social networks: III. Valued relations. Psychometrika, 64(3), 371–394.
Rogers, R., & Ben-David, A. (2008). The Palestinian—Israeli peace process and transnational
issue networks: The complicated place of the Israeli NGO. New Media & Society, 10(3),
497–528. https://doi.org/10.1177/1461444807085321
Safadi, H., Johnson, S. L., & Faraj, S. (2018). Who contributes knowledge? Embeddedness and
marginality in online communities. Academy of Management Proceedings, 2018(1),
11588. https://doi.org/10.5465/AMBPP.2018.35
Salganik, M. J. (2006). Experimental study of inequality and unpredictability in an artificial
cultural market. Science, 311(5762), 854–856. https://doi.org/10.1126/science.1121066

173

Sánchez, D. Á., Gimilio, D. P., & Altamirano, J. I. (2015). Crowdsourcing: a new way to citizen
empowerment. In Advances in Crowdsourcing (pp. 73–86). Springer.
Sargent, M., Fulk, J., & Monge, P. R. (2020). The impacts of spanning implicit keyword
categories on product sales in online markets.
Schecter, A., & Contractor, N. (2017). Understanding and assessing collaborative processes
through relational events. In A. A. von Davier, M. Zhu, & P. C. Kyllonen (Eds.),
Innovative Assessment of Collaboration (pp. 223–231). Springer International Publishing.
https://doi.org/10.1007/978-3-319-33261-1_14
Scott, W. R. (2004). Reflections on a half-century of organizational sociology. Annual Review of
Sociology, 30, 1-21.
Shaw, A. (2012). Centralized and decentralized gatekeeping in an open online collective. Politics
& Society, 40(3), 349–388. https://doi.org/10.1177/0032329212449009
Shen, C., Monge, P. R., & Williams, D. (2014a). The evolution of social ties online: A
longitudinal study in a massively multiplayer online game. Journal of the Association for
Information Science and Technology, 65(10), 2127–2137.
Shen, C., Monge, P. R., & Williams, D. (2014b). Virtual brokerage and closure: network
structure and social capital in a massively multiplayer online game. Communication
Research, 41(4), 459–480. https://doi.org/10.1177/0093650212455197
Shen, X.L., Lee, M. K., & Cheung, C. M. (2014). Exploring online social behavior in
crowdsourcing communities: A relationship management perspective. Computers in
Human Behavior, 40, 144–151.
Shumate, M., & Contractor, N. (2013). Emergence of multidimensional social networks. The
SAGE Handbook of Organizational Communication, 449–474.

174

Shumate, M., & Lipp, J. (2008). Connective collective action online: An examination of the
hyperlink network structure of an NGO issue network. Journal of Computer-Mediated
Communication, 14(1), 178–201. https://doi.org/10.1111/j.1083-6101.2008.01436.x
Shumate, M., & Palazzolo, E. T. (2010). Exponential Random Graph (p*) Models as a method
for social network analysis in communication research. Communication Methods and
Measures, 4(4), 341–371. https://doi.org/10.1080/19312458.2010.527869
Singh, J. V., & Lumsden, C. J. (1990). Theory and research in organizational ecology. Annual
Review of Sociology, 16(1), 161–195.
https://doi.org/10.1146/annurev.so.16.080190.001113
Snijders, T. A. (2002). Markov chain Monte Carlo estimation of exponential random graph
models. Journal of Social Structure, 3(2), 1-40.
Snijders, T. A. B., van de Bunt, G. G., & Steglich, C. E. G. (2010). Introduction to stochastic
actor-based models for network dynamics. Social Networks, 32(1), 44–60.
https://doi.org/10.1016/j.socnet.2009.02.004
Steglich, C., Snijders, T. A. B., & Pearson, M. (2010). Dynamic networks and behavior:
separating selection from influence. Sociological Methodology, 40(1), 329–393.
https://doi.org/10.1111/j.1467-9531.2010.01225.x
Stephen, A. T., Zubcsek, P. P., & Goldenberg, J. (2016). Lower connectivity is better: the effects
of network structure on redundancy of ideas and customer innovativeness in
interdependent ideation tasks. Journal of Marketing Research, 53(2), 263–279.
https://doi.org/10.1509/jmr.13.0127

175

Stephens, K. K., Malone, P. C., & Bailey, C. M. (2005). Communicating with stakeholders
during a crisis: evaluating message strategies. The Journal of Business Communication
(1973), 42(4), 390–419. https://doi.org/10.1177/0021943605279057
Stohl, C. (2014). Crowds, clouds, and community: crowds, clouds, and community. Journal of
Communication, 64(1), 1–19. https://doi.org/10.1111/jcom.12075
Su, C., & Contractor, N. (2011). A multidimensional network approach to studying team
members’ information seeking from human and digital knowledge sources in consulting
firms. Journal of the American Society for Information Science and Technology, 62(7),
1257–1275. https://doi.org/10.1002/asi.21526
Sun, Y., & Majchrzak, A. Innovation Challenges. Oxford Research Encyclopedia of Business
and Management. Retrieved 15 Jun. 2021, from
https://oxfordre.com/business/view/10.1093/acrefore/9780190224851.001.0001/acrefore-
9780190224851-e-201.
Surowiecki, J. (2004). The wisdom of crowds: Why the many are smarter than the few and how
collective wisdom shapes business. Economies, Societies and Nations, 296, 5.
Swaminathan, A., & Wade, J. B. (2001). Social movement theory and the evolution of new
organizational forms.: In the entrepreneurship dynamic in industry evolution. Edited by
CB Schoonhoven and E. Romanelli. Palo Alto, CA: Stanford University Press.
Taylor, C., & Nowak, M. A. (2007). Transforming the dilemma. Evolution, 61(10), 2281–2292.
https://doi.org/10.1111/j.1558-5646.2007.00196.x
TeBlunthuis, N., Shaw, A., & Hill, B. M. (2020). The Population ecology of online collective
action. 4.

176

Tilly, C. (1978). From mobilization to revolution. In collective violence, contentious politics, and
social change (pp. 71–91). Routledge.
Tucci, C. L., Afuah, A., & Viscusi, G. (2018). Creating and capturing value through
crowdsourcing. Oxford University Press.
Walker, E. T., McCarthy, J. D., & Baumgartner, F. (2011). Replacing members with managers?
Mutualism among membership and nonmembership advocacy organizations in the united
states. American Journal of Sociology, 116(4), 1284–1337.
https://doi.org/10.1086/655753
Wang, P., Robbins, G., Philippa, P., & Koskinen, J. (2014). MPNet.
Wang, D. J., Rao, H., & Soule, S. A. (2019). Crossing categorical boundaries: a study of
diversification by social movement organizations. American Sociological Review, 84(3),
420–458. https://doi.org/10.1177/0003122419846111
Wang, D. J., & Soule, S. A. (2012). Social movement organizational collaboration: networks of
learning and the diffusion of protest tactics, 1960–1995. American Journal of Sociology,
117(6), 1674–1722. https://doi.org/10.1086/664685
Wang, D. J., & Soule, S. A. (2016). Tactical innovation in social movements: The effects of
peripheral and multi-issue protest. American Sociological Review, 81(3), 517–548.
https://doi.org/10.1177/0003122416644414
Wang, D., Piazza, A., & Soule, S. A. (2018). Boundary-spanning in social movements:
antecedents and outcomes. Annual Review of Sociology, 44(1), 167–187.
https://doi.org/10.1146/annurev-soc-073117-041258

177

Wang, L., Yang, A., & Thorson, K. (2019). Serial participants of social media climate discussion
as a community of practice: A longitudinal network analysis. Information,
Communication & Society, https://doi.org/10.1080/1369118X.2019.1668457
Wang, M., Chen, W., Huang, Y., Contractor, N. S., & Fu, Y. (2016). Modeling customer
preferences using multidimensional network analysis in engineering design. Design
Science, 2, e11. https://doi.org/10.1017/dsj.2016.11
Wang, R. (2016). Building social legoland through collaborative crowdsourcing: marginality,
functional diversity and team success.
Wang, R., Tanjasiri, S. P., Palmer, P., & Valente, T. W. (2016). Network structure, multiplexity,
and evolution as influences on community-based participatory interventions. Journal of
Community Psychology, 44(6), 781–798. https://doi.org/10.1002/jcop.21801
Wang, X., Butler, B., & Ren, Y. (2013). The impact of membership overlap on growth: an
ecological competition view of online groups. Organization Science, 24(2), 414–431.
https://doi.org/10.1287/orsc.1120.0756
Wasko, M. M., & Faraj, S. (2005). Why should i share? Examining social capital and knowledge
contribution in electronic networks of practice. MIS Quarterly, 29(1), 35–57.
https://doi.org/10.2307/25148667
Wasko, M. M., Teigland, R., & Faraj, S. (2009). The provision of online public goods:
Examining social structure in an electronic network of practice. Decision Support
Systems, 47(3), 254–265.
Wasserman, S., & Pattison, P. (1996). Logit models and logistic regressions for social networks:
I. An introduction to Markov graphs andp. Psychometrika, 61(3), 401–425.

178

Weber, M. S. (2012). Newspapers and the long-term implications of hyperlinking. Journal of
Computer-Mediated Communication, 17(2), 187–201. https://doi.org/10.1111/j.1083-
6101.2011.01563.x
Weber, M. S., Fulk, J., & Monge, P. R. (2016). The emergence and evolution of social
networking sites as an organizational form. Management Communication Quarterly,
0893318916629547. https://doi.org/10.1177/0893318916629547
Weber, M. S., Fulk, J., & Monge, P. (2016). The Emergence and Evolution of Social Networking
Sites as an Organizational Form. Management Communication Quarterly, 30(3), 305–
332. https://doi.org/10.1177/0893318916629547
Weiss, S. A., & Domingo, D. (2010). Innovation processes in online newsrooms as actor-
networks and communities of practice. New Media & Society, 12(7), 1156-1171.
Welles, B. F., Vashevko, A., Bennett, N., & Contractor, N. (2014). Dynamic models of
communication in an online friendship network. Communication Methods and Measures,
8(4), 223–243. https://doi.org/10.1080/19312458.2014.967843
Wenger, E., McDermott, R. A., & Snyder, W. (2002). Cultivating communities of practice: A
guide to managing knowledge. Harvard business press.
Wenger, E. C., & Snyder, W. M. (2000). Communities of practice: The organizational frontier.
Harvard business review, 78(1), 139-146.
Xu, R. (2018). Alternative estimation methods for identifying contagion effects in dynamic
social networks: A latent-space adjusted approach. Social Networks, 54, 101–117.
https://doi.org/10.1016/j.socnet.2018.01.002

179

Xu, S., & Zhou, A. (2020). Hashtag homophily in twitter network: Examining a controversial
cause-related marketing campaign. Computers in Human Behavior, 102, 87–96.
https://doi.org/10.1016/j.chb.2019.08.006
Xu, Y. (2018). The ecological dynamics of organizational change: density dependence in the rate
of weibo adoption by populations of news organizations. International Journal of
Communication, 12(0), 26.
Xu, Y. (2019). The evolution of multidimensional and multilevel networks in online
crowdsourcing.
Xu, Y., Fulk, J., & Monge, P. R. (2021). Ecological Influences on Network Ties Dissolution in
the Evolution of Affiliation Networks in the International Communication Association,
2009—2015. Human Communication Research.
Yan, B., Jian, L., Ren, R., Fulk, J., Sidnam-Mauch, E., & Monge, P. R. (2020). The paradox of
interaction: communication network centralization, shared task experience, and the
wisdom of crowds in online crowdsourcing communities. Communication Research,
0093650220915033. https://doi.org/10.1177/0093650220915033
Yang, A. (2020). An issue niche theory of NGOs’ interorganizational network ecology.
Communication Theory.
Zappa, P., & Robins, G. (2016). Organizational learning across multi-level networks. Social
Networks, 44, 295–306. https://doi.org/10.1016/j.socnet.2015.03.003
Zhang, S., & van der Schaar, M. (2017). From acquaintances to friends: homophily and learning
in networks. IEEE Journal on Selected Areas in Communications, 35(3), 680–690.
https://doi.org/10.1109/JSAC.2017.2672238

180

Zhao, E. Y., Fisher, G., Lounsbury, M., & Miller, D. (2018). Optimal distinctiveness:
Broadening the interface between institutional theory and strategic management: Optimal
distinctiveness. Strategic Management Journal, 38(1), 93–113.
https://doi.org/10.1002/smj.2589

181

Appendices
Appendix A. Goodness of Fit statistics for the ERGM model on the review stage (H1)
Model 1
(Hypothesized and Control
Variables Only)
Model 2
(Full Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Edges 1155.00 1072.82
Outdegree (0) 146.00 154.46
Indegree (2) 69.00 80.21
Edge covariates
Past network (ideation stage) 49.00 49.66 49.00 48.49
Nodal covariates
Node match (other opportunity
area; H1)
3.00 3.22 3.00 4.35
Control
Indegree (exp) 67.80 35.36 67.80 63.58
Indegree (tenure) 35.84 21.22 35.84 42.10
Absolute differences (exp) 534.97 520.47 534.97 563.07
Absolute differences (tenure) 1148.29 1133.18 1148.29 1040.34
Other global network structures
Indegree
0 7.00 0.11 7.00 3.47
1 88.00 0.56 88.00 16.36
2 69.00 1.54 69.00 80.21
3 44.00 3.46 44.00 43.17
4 11.00 6.95 11.00 46.51
5 15.00 11.02 15.00 38.58
6 7.00 16.16 7.00 26.15
7 4.00 20.60 4.00 15.35
8 3.00 24.05 3.00 8.19
9 0.00 25.65 0.00 3.68
10 2.00 26.42 2.00 1.46
11 2.00 26.40 2.00 0.59
12 9.00 24.77 9.00 0.18
13 6.00 21.77 6.00 0.07
14 2.00 18.64 2.00 0.01
15 0.00 15.19 0.00 0.02
16 1.00 11.21 1.00 0.00
17 2.00 9.67 2.00 0.00
18 2.00 6.85 2.00 0.00
19 2.00 4.52 2.00 0.00
20 3.00 3.53 3.00 0.00
21 0.00 2.09
22 1.00 1.35 1.00 0.00
23 0.00 0.60
24 1.00 0.41 1.00 0.00
25 0.00 0.22
26 2.00 0.15 2.00 0.00
27 0.00 0.10
29 0.00 0.01
46 1.00 0.00

182

Appendix A. (continued)
Model 1
(Hypothesized and Control Variables Only)
Model 2
(Full Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Outdegree
0 146.00 0.30 146.00 154.46
1 57.00 1.08 57.00 0.35
2 21.00 1.94 21.00 1.14
3 17.00 3.85 17.00 3.38
4 6.00 7.50 6.00 6.87
5 5.00 8.11 5.00 11.08
6 1.00 11.45 1.00 14.93
7 6.00 16.19 6.00 17.10
8 0.00 20.90 0.00 18.22
9 5.00 26.04 5.00 15.50
10 1.00 31.05 1.00 12.98
11 0.00 32.08 0.00 10.21
12 1.00 28.51 1.00 7.02
13 1.00 25.35 1.00 4.09
14 0.00 20.17 0.00 2.74
15 0.00 16.63 0.00 1.60
16 0.00 11.21 0.00 1.15
17 0.00 7.59 0.00 0.36
18 1.00 5.63 1.00 0.29
19 0.00 3.53 0.00 0.20
20 0.00 2.27 0.00 0.10
21 0.00 1.15 0.00 0.10
22 0.00 0.63 0.00 0.05
23 0.00 0.34 0.00 0.03
24 1.00 0.29 1.00 0.01
25 0.00 0.05
26 0.00 0.05 0.00 0.02
27 0.00 0.07 0.00 0.01
28 2.00 0.01 2.00 0.00
29 0.00 0.01 0.00 0.01
30 0.00 0.01
31 0.00 0.01
32 2.00 0.00 2.00 0.00
33 1.00 0.00 1.00 0.00
34 1.00 0.00 1.00 0.00
38 1.00 0.00 1.00 0.00
39 1.00 0.00 1.00 0.00
44 1.00 0.00 1.00 0.00
46 1.00 0.00 1.00 0.00
51 1.00 0.00 1.00 0.00
69 1.00 0.00 1.00 0.00
77 1.00 0.00 1.00 0.00
95 1.00 0.00 1.00 0.00
140 1.00 0.00 1.00 0.00

183

Appendix A. (continued)
Model 1
(Hypothesized and Control Variables Only)
Model 2
(Full Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Edgewise
shared partners

esp0 372.00 1457.19 372.00 956.90
esp1 235.00 901.22 235.00 108.86
esp2 76.00 367.72 76.00 6.78
esp3 55.00 161.98 55.00 0.28
esp4 120.00 77.29 120.00 0.00
esp5 114.00 44.67 114.00 0.00
esp6 75.00 36.01 75.00 0.00
esp7 32.00 29.97 32.00 0.00
esp8 30.00 20.85 30.00 0.00
esp9 22.00 11.84 22.00 0.00
esp10 6.00 6.21 6.00 0.00
esp11 10.00 2.47 10.00 0.00
esp12 2.00 0.60 2.00 0.00
esp13 1.00 0.12 1.00 0.00
esp14 2.00 0.00 2.00 0.00
esp15 1.00 0.00 1.00 0.00
esp16 1.00 0.00 1.00 0.00
esp21 1.00 0.00 1.00 0.00

184

Appendix B. Goodness of Fit statistics for the multilevel ideation network (H2a and H2b)
Model 3 (Hypothesized and Control
Variables Only; the Final Model)
Statistics Observed Mean StdDev t-ratio
Included in the model
TriangleXAX 243.00 241.94 17.69 0.06
L3XAX
Standardized
tenure activity
-136.65 -138.72 34.71 0.06
Standardized exp.
activity
-87.52 -89.50 33.61 0.06
Standardized
tenure difference
560.87 559.68 37.86 0.04
Standardized exp.
difference
347.65 346.20 31.52 0.05
Not included in the model
stddev_degreeA 16.66 2.51 0.11 127.32
skew_degreeA 9.98 0.47 0.16 59.09
clusteringA 0.07 0.02 0.004 13.40
stddev_degreeX_A 0.94 1.21 0.02 -10.9
skew_degreeX_A -1.31 1.14 0.06 -2.86
stddev_degreeX_B 79.73 68.54 0.44 25.25
skew_degreeX_B -0.80 -0.67 0.01 -10.41
clusteringX 0.00 0.23 0.025 -9.24
Mahalanobis distance = 315723
Note. There were 243 observed TriangleXAX, which was fewer than the total number of 439 reported in the result
section. The reason is that this multilevel network was not weighted, and there were cases when pairs of ideators
shared more than one comment tie.

185

Appendix C. Goodness of Fit statistics for the multilevel review network models (H2a and H2b)
Model 4 (Hypothesized and Control
Variables Only)
Model 5 (Final Model)
Statistics Observed Mean StdDev t-ratio Observed Mean StdDev t-ratio
Included in the model
TriangleXAX 184.00 184.41 12.99 -0.04 184.00 183.90 13.67 0.007
L3XAX 582.00 581.97 32.94 0.001
Standardized
tenure activity
178.59 177.58 49.29 0.02 178.59 182.78 45.39 -0.09
Standardized exp.
activity
-30.02 -26.26 42.44 -0.09 -30.02 -28.35 42.50 -0.04
Standardized
tenure difference
983.80 985.84 38.52 0.05 983.80 987.03 37.21 -0.09
Standardized exp.
difference
460.62 461.22 35.87 -0.02 460.62 459.16 36.27 0.04
Not included in the
model

stddev_degreeA 14.22 2.70 0.11 100.95 14.22 2.71 0.12 97.99
skew_degreeA 5.33 0.45 0.16 29.76 5.33 0.44 0.17 29.21
clusteringA 0.16 0.03 0.003 42.24 0.16 0.02 0.003 41.92
stddev_degreeX_A 0.92 1.14 0.02 -11.78 0.92 1.14 0.02 -10.57
skew_degreeX_A -1.35 -1.16 0.05 -4.11 -1.36 -1.16 0.05 -4.10
stddev_degreeX_B 75.55 67.15 0.38 22.13 75.55 67.35 0.41 19.85
skew_degreeX_B -0.80 -0.66 0.01 -13.78 -0.80 -0.67 0.01 -11.37
clusteringX 0.00 0.20 0.02 -10.07 0.00 0.20 0.02 -9.64
Mahalanobis distance = 187383 Mahalanobis distance = 148523

186

Appendix D. Goodness of Fit statistics for the ERGM model on the ideation stage (H3 and H4a)
Model 6
(Hypothesized and Control
Variables Only)
Model 7
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Edges 934.00 901.58
mutual 156.00 142.58
Geometrically weighted
dyadwise shared partner
(outgoing two-path)

260.00 231.81
Indegree 2 68.00 71.49
Nodal covariates
Node mismatch (new entrants
and existing members)
499.00 498.99
499.00 482.27
Node outdegree (new entrants
and existing members)
553.00 549.61
553.00 543.17
Control
Indegree (exp) 36.32 20.74 36.32 8.47
Indegree (tenure) 18.90 12.38 18.90 -2.54
Absolute differences (exp) 425.53 415.66 425.53 396.13
Absolute differences (tenure) 672.62 661.41 672.62 638.41
Indegree
0 4.00 0.66 4.00 15.28
1 75.00 3.39 75.00 35.67
2 68.00 8.61 68.00 71.49
3 53.00 18.11 53.00 47.78
4 18.00 29.71 18.00 38.17
5 12.00 39.09 12.00 26.70
6 12.00 44.28 12.00 16.99
7 7.00 39.33 7.00 10.27
8 9.00 32.78 9.00 5.81
9 9.00 22.39 9.00 2.95
10 1.00 15.07 1.00 1.74
11 3.00 9.63 3.00 0.88
12 0.00 5.68 0.00 0.43
13 0.00 3.19 0.00 0.26
14 0.00 1.47 0.00 0.19
15 1.00 0.94 1.00 0.14
16 0.00 0.36 0.00 0.05
17 1.00 0.19 1.00 0.04
18 0.00 0.06 0.00 0.04
19 0.00 0.05 0.00 0.02
22 0.00 0.01 0.00 0.02
23 0.00 0.01
28 0.00 0.01
29 1.00 0.00 1.00 0
31 0.00 0.01
36 0.00 0.01
41 0.00 0.01
44 0.00 0.01
54 1.00 0.00 1.00 0
55 0.00 0.01
57 0.00 0.01

187

Appendix D. (continued)
Model 6
(Hypothesized and Control Variables Only)
Model 7
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Outdegree
0 160.00 19.01 160.00 25.64
1 44.00 45.67 44.00 50.30
2 27.00 55.88 27.00 54.83
3 9.00 46.42 9.00 45.74
4 6.00 31.35 6.00 34.13
5 4.00 16.69 4.00 20.87
6 6.00 8.79 6.00 13.49
7 2.00 4.80 2.00 8.94
8 3.00 3.11 3.00 5.91
9 1.00 1.82 1.00 4.44
10 0.00 2.09 0.00 3.11
11 0.00 2.02 0.00 2.31
12 0.00 2.19 0.00 1.76
13 1.00 1.99 1.00 1.03
14 0.00 1.79 0.00 0.97
15 0.00 1.81 0.00 0.48
16 3.00 1.79 3.00 0.37
17 0.00 1.81 0.00 0.21
18 1.00 1.68 1.00 0.14
19 0.00 1.47 0.00 0.10
20 0.00 1.25 0.00 0.02
21 0.00 1.15 0.00 0.06
22 0.00 1.01 0.00 0.03
23 0.00 0.87 0.00 0.02
24 0.00 0.79 0.00 0.01
25 0.00 0.61 0.00 0.01
26 0.00 0.69
27 0.00 0.62
28 0.00 0.47
29 0.00 0.33
30 3.00 0.32 3.00 0.01
31 0.00 0.25 0.00 0.01
32 0.00 0.23
33 0.00 0.22
34 0.00 0.21
35 0.00 0.30
36 0.00 0.24 0.00 0.01
37 0.00 0.16
38 0.00 0.27
39 0.00 0.23
40 0.00 0.47
41 0.00 0.31
42 0.00 0.59 0.00 0.01
43 0.00 0.40
44 0.00 0.48
45 1.00 0.62 1.00 0.00
46 0.00 0.63
47
0.00 0.50

188

Appendix D. (continued)
Model 6
(Hypothesized and Control Variables Only)
Model 7
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Outdegree
48 0.00 0.78
49 0.00 0.72 0.00 0.02
50 0.00 0.68
51 0.00 0.88
52 0.00 0.64
53 1.00 0.71 1.00 0.00
54 0.00 0.65
55 0.00 0.57
56 0.00 0.46
57 0.00 0.58
58 0.00 0.38
59 0.00 0.41
60 0.00 0.27
61 0.00 0.25 0.00 0.01
62 0.00 0.13
63 0.00 0.11
64 0.00 0.10
65 0.00 0.08
66 0.00 0.06
67 0.00 0.03
68 0.00 0.04
70 0.00 0.02
71 0.00 0.01
73 1.00 0.01 1.00 0.00
78 0.00 0.01
81 0.00 0.01
96 0.00 0.01
120 1.00 0.00 1.00 0.00
124 0.00 0.01
222 1.00 0.00 1.00 0.00

189

Appendix D. (continued)
Model 6
(Hypothesized and Control Variables Only)
Model 7
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Edgewise
shared partners

esp0 217.00 1108.97 217.00 659.70
esp1 260.00 468.05 260.00 231.81
esp2 145.00 164.83 145.00 8.20
esp3 53.00 48.06 53.00 1.23
esp4 66.00 12.61 66.00 0.42
esp5 78.00 3.01 78.00 0.12
esp6 59.00 0.54 59.00 0.05
esp7 26.00 0.08 26.00 0.02
esp8 12.00 0.00 12.00 0.02
esp9 3.00 0.00 3.00 0.01
esp10 5.00 0.00 5.00 0.00
esp11 3.00 0.00 3.00 0.00
esp12 1.00 0.00 1.00 0.00
esp13 1.00 0.00 1.00 0.00
esp15 1.00 0.00 1.00 0.00
esp16 2.00 0.00 2.00 0.00
esp19 1.00 0.00 1.00 0.00
esp21 1.00 0.00 1.00 0.00

190

Appendix E. Goodness of Fit statistics for the ERGM model on the review stage (H4b and H4c)
Model 8
(Hypothesized and Control
Variables Only)
Model 9
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Edges 1155.00 1042.51
Outdegree0 146.00 157.47
Indegree (2) 69.00 82.91
Indegree (3) 44.00 46.82
Edge covariate
Past network (ideation stage) 49.00 48.72 49.00 38.39
Nodal covariates
Node mismatch (new entrants
and existing members) 476.00 483.19 476.00 450.00
Indegree (No. mutualist ties) 18525.00 17849.47 18525.00 16833.79
Control
Indegree (exp) 67.80 27.66 67.80 61.34
Indegree (tenure) 35.84 3.71 35.84 40.80
Absolute differences (exp) 534.97 508.01 534.97 467.17
Absolute differences (tenure) 1148.29 1112.43 1148.29 1002.68
Indegree
0 7.00 0.22 7.00 5.35
1 88.00 1.41 88.00 20.22
2 69.00 3.54 69.00 82.91
3 44.00 7.46 44.00 46.82
4 11.00 13.15 11.00 46.27
5 15.00 18.76 15.00 35.26
6 7.00 24.45 7.00 22.54
7 4.00 29.24 4.00 12.57
8 3.00 29.38 3.00 6.09
9 0.00 28.65 0.00 2.99
10 2.00 24.21 2.00 1.09
11 2.00 21.03 2.00 0.49
12 9.00 16.64 9.00 0.22
13 6.00 12.60 6.00 0.09
14 2.00 10.07 2.00 0.07
15 0.00 7.53 0.00 0.02
16 1.00 6.12 1.00 0.00
17 2.00 4.83 2.00 0.00
18 2.00 3.90 2.00 0.00
19 2.00 3.26 2.00 0.00
20 3.00 3.01 3.00 0.00
21 0.00 2.56
22 1.00 2.13 1.00 0.00
23 0.00 2.16
24 1.00 1.48 1.00 0.00
25 0.00 1.37 0.00 0.01
26 2.00 1.24 2.00 0.05
27 0.00 0.76 0.00 0.03
28 0.00 0.70 0.00 0.02
29 0.00 0.58 0.00 0.07

191

Appendix E. (continued)
Model 8
(Hypothesized and Control
Variables Only)
Model 9
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
30 0.00 0.50 0.00 0.03
31 0.00 0.32 0.00 0.09
32 0.00 0.23 0.00 0.05
33 0.00 0.18 0.00 0.08
34 0.00 0.07 0.00 0.08
35 0.00 0.08 0.00 0.09
36 0.00 0.12 0.00 0.05
37 0.00 0.04 0.00 0.11
38 0.00 0.01 0.00 0.02
39 0.00 0.01 0.00 0.07
40 0.00 0.09
41 0.00 0.01
42 0.00 0.01
44 0.00 0.02
46 1.00 0.00 1.00 0.01
51 0.00 0.01

192

Appendix E. (continued)
Model 8
(Hypothesized and Control
Variables Only)
Model 9
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Outdegree
0 146.00 0.51 146.00 157.47
1 57.00 1.22 57.00 0.38
2 21.00 2.94 21.00 1.49
3 17.00 4.86 17.00 3.39
4 6.00 11.55 6.00 6.92
5 5.00 15.41 5.00 10.95
6 1.00 21.35 1.00 14.44
7 6.00 26.94 6.00 17.05
8 0.00 31.65 0.00 17.39
9 5.00 33.61 5.00 15.26
10 1.00 31.25 1.00 12.77
11 0.00 26.96 0.00 9.26
12 1.00 20.13 1.00 6.54
13 1.00 13.32 1.00 3.98
14 0.00 10.05 0.00 2.71
15 0.00 6.49 0.00 1.76
16 0.00 4.24 0.00 0.81
17 0.00 2.82 0.00 0.46
18 1.00 2.13 1.00 0.33
19 0.00 1.46 0.00 0.25
20 0.00 1.34 0.00 0.12
21 0.00 1.61 0.00 0.11
22 0.00 1.53 0.00 0.06
23 0.00 1.20 0.00 0.02
24 1.00 1.61 1.00 0.04
25 0.00 1.28 0.00 0.02
26 0.00 1.31 0.00 0.01
28 0.00 1.11 2.00 0.00
30 2.00 0.91 0.00 0.01
31 0.00 0.90
32 2.00 0.60 2.00 0.00
33 1.00 0.36 1.00 0.00
34 1.00 0.50 1.00 0.00
35 0.00 0.24
36 0.00 0.20
37 0.00 0.18
38 1.00 0.12 1.00 0.00
39 1.00 0.03 1.00 0.00
41 0.00 0.01
42 0.00 0.02
44 1.00 0.03 1.00 0.00
46 1.00 0.01 1.00 0.00
51 1.00 0.01 1.00 0.00
69 1.00 0.00 1.00 0.00
77 1.00 0.00 1.00 0.00
95 1.00 0.00 1.00 0.00
140 1.00 0.00 1.00 0.00

193

Appendix E. (continued)
Model 8
(Hypothesized and Control Variables Only)
Model 9
(Final Model)
Structural features Observed Model Estimation
(Mean)
Observed Model Estimation
(Mean)
Edgewise
shared partners

esp0 372.00 1467.65 372.00 906.25
esp1 235.00 813.97 235.00 120.38
esp2 76.00 301.34 76.00 13.23
esp3 55.00 121.54 55.00 2.09
esp4 120.00 51.35 120.00 0.44
esp5 114.00 33.55 114.00 0.08
esp6 75.00 23.55 75.00 0.03
esp7 32.00 15.70 32.00 0.01
esp8 30.00 9.00 30.00 0.00
esp9 22.00 4.12 22.00 0.00
esp10 6.00 1.49 6.00 0.00
esp11 10.00 0.56 10.00 0.00
esp12 2.00 0.14 2.00 0.00
esp13 1.00 0.00 1.00 0.00
esp14 2.00 0.01 2.00 0.00
esp15 1.00 0.00 1.00 0.00
esp16 1.00 0.00 1.00 0.00
esp21 1.00 0.00 1.00 0.00

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The evolution of multidimensional and multilevel networks in online crowdsourcing

PDF

The evolution of multilevel organizational networks in an online gaming community

PDF

The coevolution of multimodal, multiplex, and multilevel organizational networks in development communities

PDF

The evolution of knowledge creation online: what is driving the dynamics of Wikipedia networks

PDF

The patterns, effects and evolution of player social networks in online gaming communities

PDF

Organizational mimicry in American social movement communities: an analysis of form communication effects on the evolution of crisis pregnancy centers, 1989-2009

PDF

Communicating organizational knowledge in a sociomaterial network: the influences of communication load, legitimacy, and credibility on health care best-practice communication

PDF

Crowdsourcing for integrative and innovative knowledge: knowledge diversity, network position, and semantic patterns of collective reflection

PDF

The evolution of scientific collaboration networks

PDF

Understanding alliance building and communication across multiple technology platforms among California's immigrant-serving NGOs

PDF

Effects of economic incentives on creative project-based networks: communication, collaboration and change in the American film industry, 1998-2010

PDF

A multitheoretical multilevel explication of crowd-enabled organizations: exploration/exploitation, social capital, signaling, and homophily as determinants of associative mechanisms in donation-...

PDF

Crisis and stasis: understanding change in online communities

PDF

Media reinvented: the transformation of news in a networked society

PDF

Building social Legoland through collaborative crowdsourcing: marginality, functional diversity, and team success

PDF

The formation and influence of online health social networks on social support, self-tracking behavior and weight loss outcomes

PDF

Food is the medium: food movements, social justice and the communication ecology approach

PDF

How social and human capital create financial capital in crowdfunding projects

PDF

Adjusting the algorithm: how experts intervene in algorithmic hiring tools

PDF

Community structure as collective identity construction and resource search

Asset Metadata

Creator Li, Yiqi (author)

Core Title Ecology and network evolution in online innovation contest crowdsourcing

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Degree Conferral Date 2021-08

Publication Date 07/19/2021

Defense Date 06/11/2021

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

Tag community of practice,crowdsourcing contests,ecology theory,evolutionary theory,multilevel,network evolution,networks,niches,OAI-PMH Harvest,online communities,populations

Format application/pdf (imt)

Language English

Advisor Monge, Peter (committee chair), Fulk, Janet (committee member), Yang, Aimei (committee member)

Creator Email yiqili@usc.edu,yli360@syr.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC15612361

Unique identifier UC15612361

Legacy Identifier etd-LiYiqi-9799

Document Type Dissertation

Format application/pdf (imt)

Rights Li, Yiqi

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu