University of Southern California Dissertations and Theses

The evolution of multidimensional and multilevel networks in online crowdsourcing

(USC Thesis Other)

The evolution of multidimensional and multilevel networks in online crowdsourcing

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i

THE EVOLUTION OF MULTIDIMENSIONAL AND MULTILEVEL NETWORKS IN
ONLINE CROWDSOURCING

by

Yu Xu

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 2019

Copyright 2019 Yu Xu
ii
Acknowledgements
The writing of this dissertation would not have been possible without guidance, support,
and help from many people. First and foremost, I would like to acknowledge my indebtedness to
my advisor and mentor, Dr. Peter Monge, who introduced me to network analysis, organizational
ecology, and socio-cultural evolution and gave me timely advice in different stages of my
academic exploration at the Annenberg School. I have learned tremendously from his enduring
commitment to mentorship and rigorous research.
I am deeply indebted to the other excellent members of my dissertation committee: Dr.
Janet Fulk and Dr. Dmitri Williams. They have provided me with a great amount of invaluable
support and guidance on this and many other projects over the past few years. In addition, I
gratefully acknowledge my qualifying exam committee member, Dr. Peer Fiss, who has inspired
me in the field of organizational theory.
My sincere gratitude also goes to my wonderful colleagues in the Annenberg Networks
Network (ANN) and the Annenberg school. I am deeply grateful to my coauthors, Yao Sun,
Larry Zhiming Xu, Ignacio Cruz, Yiqi Li, and Jingyi Sun, for their consistent support and help. I
want to acknowledge Dr. Alessandro Lomi, Dr. Su Jung Kim, Dr. Nathan Walter, Emily Sidnam,
Dr. Nirit Weiss-Blatt, Aveva Yusi Xu, Dr. Carmen Lee, Dr. Ken Sereno, Hye Min Kim, and
Steffie Kim, who gave me insightful advice on earlier versions of this work.
This dissertation is supported in part by funding from the National Science Foundation
(IIS-1514505) and the Annenberg School for Communication and Journalism at the University
of Southern California. I wish to thank Yulu Chen for her assistance in data collection.
Additionally, I would like to express my heartfelt gratitude to Dr. James Polk for proofreading
my work with great patience over the past four years.
iii
My thanks also go to the social and emotional support I received from Dr. Jiawei Sophia
Fu, Dr. Yusi Liu, and Dr. Wang Liao. I am also indebted to my friends, many of whom I have
not met in any offline settings in three online Wechat groups: “FIFAology Research Center,”
“Pokemon Go,” and “Excited”. Our conversations enabled me to have a balance between work
and life.
Finally, I am indebted to my parents for their loving encouragement throughout my life.
They always give me unconditional support and believe that I can do excellent work in the field.

iv

Table of Contents
Acknowledgements ......................................................................................................................... ii
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Abbreviations ............................................................................................................................... viii
Chapter 1: Introduction ................................................................................................................... 1
Purpose of the Study ................................................................................................................... 1
Chapter Summaries ..................................................................................................................... 4
Chapter 2: Literature Review .......................................................................................................... 6
Online Communities ................................................................................................................... 6
Crowdsourcing ............................................................................................................................ 9
Definition of Crowdsourcing ................................................................................................ 10
The Debate About the Judgement of Crowds ....................................................................... 12
Existing Empirical Research on Crowdsourcing .................................................................. 16
Crowdsourcing Communities as Online Communities ......................................................... 20
Stitchly .................................................................................................................................. 22
Multidimensional and Multilevel Networks ............................................................................. 26
The Multi-theoretical, Multi-level Model ............................................................................. 26
Multidimensional Networks .................................................................................................. 26
Multilevel Networks ............................................................................................................. 28
Multilayer Networks ............................................................................................................. 30
Conclusion ............................................................................................................................ 31
Evolutionary and Ecological Theories ...................................................................................... 33
Organizational Ecology and Communication ....................................................................... 33
An Ecology of Categories ..................................................................................................... 37
Density Dependence Theory, Fuzzy Density, and Contrast ................................................. 41
Niche Theory and Fuzzy Niche Theory ................................................................................ 44
Evolutionary Mechanisms of Variation, Selection, and Retention ....................................... 45
Network Evolution ................................................................................................................ 51
Chapter 3: Hypothesis Development ............................................................................................ 54
Building a Multidimensional and Multilevel Network in Design Crowdsourcing ................... 54
Cross-Level Ties Between Designs and Categories ................................................................. 59
Cross-Level Ties Between Designers and Designs .................................................................. 67
Cross-Level Ties Between Designers and Categories .............................................................. 77
Within-Level Ties Among Designers ....................................................................................... 83
Chapter 4: Method ........................................................................................................................ 89
Research Setting ........................................................................................................................ 89
Sampling and Data Collection .................................................................................................. 89
Sample 1 ................................................................................................................................ 89
Sample 2 ................................................................................................................................ 90
Network Construction ............................................................................................................... 91
Cross-Level Ties Between Designs and Categories ............................................................. 91
v
Cross-Level Ties Between Designers and Designs .............................................................. 91
Cross-Level Ties Between Designers and Categories .......................................................... 92
Within-Level Ties Among Designers ................................................................................... 93
Measures ................................................................................................................................... 94
Dependent Variables ............................................................................................................. 94
Independent Variables .......................................................................................................... 96
Control Variables .................................................................................................................. 99
Analytical Procedures ............................................................................................................. 102
Chapter 5: Results ....................................................................................................................... 105
Cross-Level Ties Between Designs and Categories ............................................................... 105
Cross-Level Ties Between Designers and Designs ................................................................ 109
Cross-Level Ties Between Designers and Categories ............................................................ 115
Within-Level Ties Among designers ...................................................................................... 119
Chapter 6: Discussion and Conclusion ....................................................................................... 122
Review of Findings ................................................................................................................. 122
Cross-Level Ties Between Designs and Categories ........................................................... 123
Cross-Level Ties Between Designers and Designs ............................................................ 126
Cross-Level Ties Between Designers and Categories ........................................................ 128
Within-Level Ties Among Designers ................................................................................. 130
Contributions .......................................................................................................................... 132
Limitations .............................................................................................................................. 135
Directions for Future Research ............................................................................................... 138
Practical Implications .............................................................................................................. 142
Conclusion .............................................................................................................................. 143
References ................................................................................................................................... 144

vi
List of Tables
Table 1. Number of Submitted and Printed Designs on Stitchly's Site (2014-2018) 23
Table 2. Summary of Hypotheses and Research Question 87
Table 3. A Category's Fuzzy Density, Population Density, and Contrast 97
Table 4. Descriptive Statistics and Spearman Correlation Coefficients Among Study
Variables for Hypotheses 1 and 2 106
Table 5. Fixed-Effects Negative Binominal Regression Models Predicting the Rate of
Entry into a Category from November 2013 to July 2018 (Hypotheses 1 and 2) 106
Table 6. Descrptive Statistics and Spearman Correlation Coefficients Among Study
Variables for Research Question 1 and Hypotheses 3 and 4 108
Table 7. Linear Regression Models Predicting the Fitness of a Design (Research
Question 1 and Hypotheses 3 and 4) 109
Table 8. Exponential Random Graph Models Predicting Tie Formation in the Design
Co-Creation Network (Hypotheses 5 and 7a) 112
Table 9. Descrptive Statistics and Spearman Correlation Coefficients Among Study
Variables for Hypotheses 6, 7b, and 8 113
Table 10. Discrete-Time Event History Analysis Predicting the Decay of a
Commensalist Tie (Hypotheses 6, 7b, and 8) 114
Table 11. Descrptive Statistics and Spearman Correlation Coefficients Among Study
Variables for Hypotheses 9a, 10a, and 11 116
Table 12. Discrete-Time Event History Analysis Predicting the Entry into a Product
Category (Hypotheses 9a, 10a, and 11) 117
Table 13. Descrptive Statistics and Spearman Correlation Coefficients Among Study
Variables for Hypotheses 9b and 10b 118
Table 14. Discrete-Time Event History Analysis Predicting the Exit from a Product
Category (Hypotheses 9b and 10b) 119
Table 15. Descriptive Statistics and Spearman Correlation Coefficients Among Study
Variables for Hypotheses 12, 13, and 14 120
Table 16. Multilevel Mixed-Effects Logistic Regression Models Predicting
Unfollowing Behavior (Hypotheses 12, 13, and 14) 121
Table 17. Summary of Hypothesis-Testing Results 122

vii
List of Figures
Figure 1. Hypothetical Cross-Level Ties Between Designs and Categories on the
Stitchly Digital Platform 57
Figure 2. Hypothetical Cross-Level Ties Between Designers and Designs on the
Stitchly Digital Platform 58
Figure 3. Hypothetical Cross-Level Ties Between Designers and Categories on the
Stitchly digital platform 58
Figure 4. Hypothetical Within-Level Ties Among Designers on the Stitchly Digital
Platform 59
Figure 5. A Visualization of Commensalist Ties Among 1,861 Successful Designers on
the Stitchly Digital Platform 110

viii
Abbreviations
AIC Akaike Information Criterion
BIC Bayesian Information Criterion
BS Broadcasting Search
DHIT Distributed Human Intelligence Tasking
EHA Event History Analysis
ERGM Exponential Random Graph Models
GoM Grades of Membership
GWESP Geometrically Weighted Edgewise Shared Partners
ICTs Information and Communication Technologies
KDM Knowledge Discovery and Management
MCMCMLE Markov Chain Monte Carlo Maximum Likelihood Estimation
MTML Multi-Theoretical Multi-Level
PVCP Peer-Vetted Creative Production
REM Relational Event Modeling
SIENA Simulation Investigation for Empirical Network Analysis
V-S-R Variation-Selection-Retention
WoC Wisdom of Crowds

ix
Abstract
Crowdsourcing has been identified as a new organizational form that mobilizes the
efforts of a distributed crowd for soliciting creative ideas and solutions. This dissertation treats
crowdsourcing as a new case of online communities and establishes a multidimensional and
multilevel network framework that recognizes both human and non-human entities in online
crowdsourcing as hierarchically structured in a dynamic and complex system. Using digital trace
data scraped from Stitchly’s website, the current research tests hypotheses regarding the
evolution of a three-level network composed of designers, designs, and product categories. The
empirical findings highlight the importance of both nodal attributes and ecological dynamic
forces in shaping network structures and outcomes.
The first part of the analysis examines what drives the evolution of cross-level ties
between designs and categories. The empirical evidence shows that increases in the fuzzy density
of a category lead to increases in the rate of entry into the category by designs. In addition,
categorical niche width is a positive predictor of the fitness of a design. Category contrast also
negatively moderates this relationship such that the positive effect of categorical niche width is
stronger when a design has a higher level of category contrast.
The second part of the analysis investigates the formation and dissolution mechanisms of
cross-level ties between designers and designs. The final results confirm that designer-level
attributes, such as design experience and expertise, significantly influence the chances of tie
creation and decay. Tie duration also predicts the likelihood that an existing tie disappears.
Specifically, the longer a cross-level tie has been maintained, the less likely the tie will decay.
The third part of analysis focuses on the longitudinal changes of cross-level ties between
designers and categories. Contrary to expectation, designers have a higher propensity of entering
x
or exiting a low-contrast category than a high-contrast category. In addition, inexperienced
designers have a much higher tendency to enter a low-contrast product category than do
experienced designers.
The fourth part of the analysis concentrates on unfollowing behavior among designers on
the Stitchly digital platform. The findings demonstrate that designers are more likely to unfollow
others who have lower levels of design expertise. Moreover, designers who are members of high-
closure groups are not necessarily less likely to unfollow others than designers who belong to
low-closure groups.

Keywords: networks, crowdsourcing, organizational ecology, evolutionary theory, organizational
communication, categories, multilevel

1
Chapter 1: Introduction
Purpose of the Study
Drawing on evolutionary and ecological theories, this dissertation establishes a
multidimensional and multilevel network framework to study the crowdsourcing phenomenon.
Specifically, the objective of this research is to examine the formation, maintenance,
transformation, and dissolution mechanisms of a three-level network composed of both human
(designers) and non-human entities (designs and categories). Since the ecological and
evolutionary perspective is concerned with dynamic change, and our social systems are
multilevel in nature, what drives the evolution of both within-level and cross-level network ties
is of considerable interest to many researchers. The research setting of this study is the digital
platform of Stitchly (a pseudonym). This crowdsourcing site for graphic design is an ideal test
bed because it includes a complete ecosystem in which dynamic networks are constructed from
multiple dimensions and are shaped by a variety of exogenous and endogenous factors. The
features of Stitchly’s site are quite representative of other digital platforms because it also
involves the components of self-categorization (e.g., eBay), value-rating (e.g., Yelp), and social
networking (e.g., Facebook). Thus, an investigation of the evolution of multidimensional and
multilevel networks on Stitchly’s website helps to shed light on the role of communication in
driving changes in several different types of online communities.
Several trends influence the development of this dissertation. First, social phenomena are
becoming increasingly complex and often span multiples levels of analysis (Klein & Kozlowski,
2000; Zappa & Lomi, 2015). Thus, taking interdependence into account is essential to
understanding communication processes. Second, our society has evolved into a social structure
organized around networks (Castells, 2000; Friedland, Howe, & Rojas, 2006). As the new central
2
form of social integration, networks play a key role in guiding action of social entities. It is
increasingly crucial for researchers to adopt a network perspective to understand complex
systems. Third, the coming age of computational social science opens a remarkable opportunity
to track digital traces of social entities over time and enables researchers to get theoretical
insights that would be otherwise unavailable (Contractor, 2018; Lazer et al., 2013; Shah,
Cappella, & Neuman, 2015).
The overarching question that guides the development of this dissertation is: How do
different sets of entities tied together by multiple networks coevolve with each other? Answering
this question has important implications for social scientists, strategic communicators, and
society at large. Using digital trace data scraped from Stitchly’s site, the current study partially
addresses the question by highlighting the importance of nodal attributes and ecological dynamic
forces in shaping various network structures and outcomes.
The first part of the analysis investigates the factors that drive dynamic changes in cross-
level ties between designs and categories. Drawing on the revised density dependence theory and
fuzzy niche theory (Hannan, Pólos, & Carroll, 2007; Hsu, Hannan, Kovács, 2009), this research
hypothesizes that increases in a category’s fuzzy density and contrast result in increases in the
rate of entry into the category by designs. It is also expected that the ecological variables of
categorical niche width and category contrast work together to constrain the fitness of a design in
crowdsourcing competitions.
The second part of the analysis focuses on the evolution of cross-level ties between
designers and designs. This research integrates insights from the liability of newness theory
(Stinchcombe, 1965), network or relational inertia theory (Kim, Oh, & Swaminathan, 2006;
Maurer & Ebers, 2006), and expertise theory (Treem, 2012) into the evolutionary mechanisms of
3
variation, selection, and retention (Campbell, 1965). This integrated theoretical framework is
built to explain how design experience, design expertise, and tie duration influence the
probability of tie formation and decay over time.
The third part of the analysis pays attention to the evolutionary patterns of cross-level ties
between designers and categories by testing several hypotheses around the antecedents of a
designer’s decision to enter or exit a product category on Stitchly’s website. Category entry and
exit are considered as a form of explorative or exploitative decision-making activities (March,
1991) at the designer level. Guided by the recent line of research on ecologies of categories
(Pontikes & Hannan, 2014), performance feedback theory (Greve, 1998), and cognitive
entrenchment theory (Dane, 2010), this research predicts that the decision-making process is
determined by category contrast, design experience, and design expertise.
The fourth part of the analysis examines what drives unfollowing behavior among
designers on the Stitchly digital platform. Based on economic and normative explanations for
network change (Cheon, Choi, Kim, & Kwak, 2015, Kim et al., 2006), this research hypothesizes
that designers are more likely to unfollow others who have lower levels of design expertise. In
addition, designers who belong to high-closure groups are less likely to unfollow others than
designers who are members of low-closure groups. Moreover, network closure negatively
moderates the relationship between design expertise and unfollowing behavior.
By bringing these results to light, this research makes several important contributions to
the literature on communication, networks, information systems, and management. First, it
advances the ecological approach in communication research by drawing theoretical insights
from ecologies of categories (Hannan, 2010; Pontikes & Hannan, 2015). In doing so, this study
contributes to reformulating density dependence theory and niche theory in communication
4
research. Second, prior research on crowdsourcing largely assumes that online crowds are fluid
and independent (e.g., Brabham, 2008; Surowiecki, 2005). This work casts doubt on the
assumption by showing that crowd members are embedded in networks through various direct
and indirect forms of communication. Third, it contributes to the literature on organizational
communication (Oh & Monge, 2013; Sargent, Clark, Monge, & Fulk, 2018; Stohl, 2014; Treem
& Leonardi, 2016, 2017) by revealing that categorization plays an important role in organizing
crowd behavior and that design expertise is a key driver of tie formation, maintenance, and
decay. Fourth, it leverages the evolutionary perspective in explaining network change. Although
the evolutionary approach has been applied to examine the evolution of one-mode networks
(e.g., Kleinbaum, 2018; Margolin et al., 2015; Shen, Monge, & Williams, 2014), little research
has been conducted to track how multidimensional and multilevel networks change over time.
This is worthy of investigation due to the multilevel nature of social structure. By analyzing how
the evolution of multilevel and multidimensional networks proceeds through tie formation and
decay, this study offers a more complete and accurate view of network change. Fifth, it
contributes to further developing the multi-theoretical, multi-level (MTML) network model
(Monge & Contractor, 2003) and conceptualizing multidimensional and multilevel networks
(Contractor, Monge, & Leonardi, 2011; Lomi, Robins, & Tranmer, 2016) by considering the
nested network structures composed of both human and non-human entities.
Chapter Summaries
This dissertation is organized as follows. Chapter 2 reviews the key concepts and
theories, including online communities, crowdsourcing, multidimensional and multilevel
networks, and ecological and evolutionary approaches. Guided by the relevant literature, Chapter
3 proposes a series of hypotheses regarding the emergence, maintenance, transformation, and
5
dissolution of a three-level network composed of designers, designs, and categories. By
recognizing human and non-human entities in design crowdsourcing as structured in a system of
multidimensional and multilevel networks (Xu et al., 2018), this research takes an important step
toward examining how network nodes and ties at different levels are related to each other.
Chapters 4 and 5 introduce the research design and present the empirical results, respectively.
Specifically, the current study utilizes computational methods, including web scraping and text
mining, to collect and preprocess digital trace data retrieved from Stitchly’s website. Network
analysis, fixed-effects models, multilevel modeling, and event history analysis are employed to
explain structural interdependence and complexity. Finally, Chapter 6 discusses the empirical
findings, theoretical contributions, and directions for future research.

6
Chapter 2: Literature Review
Online Communities
As an evolving concept, community has received widespread attention from scholars over
the past few decades, possibly due to the emergence and development of various new
information and communication technologies (ICTs) (Johnson, Safadi, & Faraj, 2015; Malinen,
2015). The traditional notion of communities emphasizes the importance of physical location,
social cohesion, shared norms and values in shaping the sense of togetherness (Carpenter, Nah,
& Chung, 2015). Constrained by the boundaries of “space of places” (Castells, 2000), traditional
communities are largely built upon small, homogenous, close-knit, and geographically bounded
groups.
The rise of online communities is largely facilitated by the widespread use of new ICTs
that shorten the social distance among people around the world (Fisher, 2019; Rheingold, 2000).
Many scholars have provided definitions of online communities. For instance, Preece and
Maloney-Krichmar (2005) argued that online communities were “named by the activity and
people they serve or the technology that supports them” and connoted “the intense feelings of
camaraderie, empathy and support” among people in cyberspace. Kraut and Resnick (2012)
referred to online communities as collective spaces in which “people come together with others
to converse, exchange information or other resources, learn, play, or just be with each other” (p.
1). Faraj and Johnson (2011) emphasized that online communities “bring together individuals
with mutual interests using electronic mediation to overcome the same-place, same-time
limitation inherent in face-to-face settings” (p. 1464).
Overall, researchers tend to conceptualize an online community as a virtual space that
bonds people with common interests, goals, and experience from all over the world. Here, a
7
virtual space does not refer to a specific physical location, but is considered as an online
environment or platform where people have real-time interactions and discussions with each
other through computer-mediated communication (Castells, 2000; Reynold, 2000). The Internet-
based technologies have also fundamentally changed the way communities are organized (Faraj,
von Krogn, Monteiro, & Lakhani, 2016). The formation and maintenance of traditional
geographically-based communities are primarily driven by well-delineated family,
neighborhood, organizational, and national boundaries or other existing normative institutional
arrangements. By contrast, the organizing processes of online communities are less constrained
by physical location and exclusive commitment and thus reflect higher levels of individual
agency and autonomy, as people “come together in community and disperse more often,
contributing as much or as little as they would like to multiple communities and carrying norms
from one community to another” (Massa, 2017, p. 962). Due to a lack of formal structures and
commitment, entry and exit rates in online communities are relatively high (Massa, 2017; Yang,
Li, & Huang, 2017) and the level of participation of community members is largely driven by
intrinsic motivation (Faraj et al., 2011; Fisher, 2019).
As online communities continue to gain popularity, many communication, sociology, and
information systems scholars have been theorizing how new ICTs bring fundamental change to
the structure of community. For instance, Castells (2000) proposed that the “network society”
was a new paradigm of social structure in the information age. He argued that new technologies
contributed to the emergence of the network society by creating “space of flows,” where social
interaction was not constrained by the boundaries of geographical locations. Similarly, Wellman
and colleagues’ (Wellman, Quan-Haase, Witte, & Hampton, 2011; Rainie & Wellman, 2012)
descriptions of “networked individualism” recognized that physical proximity played a
8
decreasing role in shaping social interactions. Networked individuals moved in loosely bounded,
digitally mediated social environments and joined different Internet-based communities based on
their personal interests, thus reducing the social cohesion of traditional groups. Recently,
Hampton’s (2016) theorization about “persistent-pervasive community” combined the mobility
afforded by new ICTs with the constraints and opportunities of the preindustrial community in
explaining the structure of community in contemporary society. He argued that we had entered
an era of “metamodernity” in which life was organized in multiple social and physical contexts
and social ties were loosely knit but could persist over a life course.
Although there are many differences between online and traditional, geographically
based communities, these two types of communities have common characteristics. For instance,
online communities are similar to offline communities in terms of shared goals, network
building, and a sense of belonging and togetherness (Malinen, 2015; Shukla & Drennan, 2018).
While some online communities exist on the Internet only, many of them co-evolve with their
offline counterparts and are mutually intertwined in a complex manner (Preece & Maloney-
Krichmar, 2005; Rufas & Hine, 2018; Sessions, 2010). When considering the origin of online
communities and their consequences for interaction in the offline context, researchers have
conflicting views of which comes first (Shen & Cage, 2015). On the one hand, some studies have
reported an offline to online trend (Ellison, Steinfield, & Lampe, 2007; Hampton & Wellman,
2003), suggesting that online communities emerged from pre-existing physical communities.
Research has shown that online communities are especially important for older adults with
limited mobility to maintain existing social circles (Quan-Haase, Mo, & Wellman, 2017).
Sharing economy companies have also initiated digital platforms that are tightly tied to offline
geographic, relational, and valued-based communities (Vaskelainen & Piscicelli, 2018). On the
9
other hand, the online to offline directionality has also received support. It is very prevalent that
online interaction leads to face-to-face activities and further collective action (Walther & Parks,
2002). For instance, Bulger, Bright, and Cobo (2015) revealed that members in virtual learning
communities were motivated to have offline meetups. Shen and Cage (2015) argued that the
emergence of websites like Meetup.com provided a technological infrastructure for individuals
to arrange offline gatherings in the same geographic area.
Online community is not a monolithic concept, as digitally mediated communities differ
from each other in many dimensions, including participation architectures (Lessig, 1999;
Majchrzak & Malhotra, 2014), norms (Lessig, 1999; Shaw & Hill, 2014), social functions
(Norris, 2002), temporariness (Majchrzak & Malhotra, 2016), entrance and exit costs or barriers
(Bregman & Haythornthwaite, 2003; Williams, 2007), and knowledge-sharing trajectories
(Majchrzak & Malhotra, 2016). Researchers have not reached a consensus on what are defining
features of online communities, as “a wide range of community types varying in terms of
structure, purpose and user base have been compared under the heading online community”
(Maline, 2015, p. 229).
Online communities co-evolve with the development of new ICTs and social change
(Bennett & Segerberg, 2013; Malinen, 2015). Given this evolving nature, the current research
treats crowdsourcing as a new organizational form mediated by Internet-based technologies and
as a new case of online communities. The next section reviews the literature on crowdsourcing,
compares different crowdsourcing models, and introduces the community features of the
research setting.
Crowdsourcing
10
Definition of Crowdsourcing
Since its widespread use beginning in the mid-2000s, crowdsourcing has been recognized
as a new organizational form using the knowledge, expertise, and judgments of crowds
(Surowiecki, 2005). As a portmanteau of crowd and outsourcing, crowdsourcing describes a
process in which organizations obtain ideas or services from a large and unknown population
(generally labeled as the crowd; Howe, 2006, 2008). In contrast to the traditional outsourcing
model where focal organizations negotiate with and choose service providers or vendors,
crowdsourcing refers to the practice where organizations solicit ideas from defined crowds. To
implement the idea of crowdsourcing, organizations typically build intermediary online
platforms in which crowd members self-select to perform the designated tasks (Zogaj,
Bretschneider, & Leimeister, 2014). A well-cited definition of crowdsourcing was provided by
Estellés-Arolas and González-Ladrón-de-Guevara (2012),
Crowdsourcing is a type of participative online activity in which an individual, an
institution, a non-profit organization, or company proposes to a group of individuals of
varying knowledge, heterogeneity, and number, via a flexible open call, the voluntary
undertaking of a task. (pp. 197)

Although the term “crowdsourcing” was not coined until 2006 (Howe, 2006), the idea of
crowdsourcing can be traced back to as early as the early eighteenth century when the British
government outsourced the task of precisely measuring a ship’s position at sea to the general
public via an open call (Afuah & Tucci, 2012). A carpenter, John Harrison, solved the so-called
“Longitude Problem” by inventing the marine chronometers and was thus awarded more than
£20,000 (equivalent to around £3 million in 2019) by the government.
Currently, crowdsourcing mainly refers to Internet-based activities in which sponsoring
organizations rely on the efforts of online crowds to find and gather novel ideas for new product
development or problem-solving (Brabham, 2017; Majchrzak & Malhotra, 2014). Compared
11
with participatory social media (e.g., YouTube) and commons-based peer production (e.g.,
Wikipedia), crowdsourcing platforms do not offer a purely bottom-up system. Rather, the
crowdsourcing process is a blend of top-down, hierarchical production and bottom-up, open
production, in which the locus of control is between organizations and online crowds (Brabham,
Ribisl, Kirchner, & Bernhardt, 2014). The process begins when organizations launch an open
call for solutions to a problem. Crowd members come up with creative ideas and provide peer
evaluations of the solutions proposed by others. In the final stage, crowdsourcing organizations
select from the winning ideas and even sell them back to the crowd for profit (Brabham, 2008,
2010; Howe, 2008). The utilization of crowdsourcing entails benefits for both sponsoring
organizations and crowds because (a) organizations can build competitive advantage and achieve
market success by utilizing expertise, knowledge, resources, and contributions from diverse and
heterogeneous collections of individuals (Afuah & Tucci, 2012; Di Pietro, Prencipe, &
Majchrzak, 2018) and by engaging with existing or potential customers (Fisher, 2019), and (b)
online crowds obtain “the satisfaction of a given type of need, be it economic, social recognition,
self-esteem, or the development of individual skills” (Estellés-Arolas & González-Ladrón-de-
Guevara, 2012, p. 197). By allowing companies and online crowds to meet in the middle for co-
creation, the crowdsourcing model also generates considerable value for customers because they
can actively participate in the design of new products that satisfy their own needs (Füller, 2005).
As for governmental and non-profit organizations, crowdsourcing is an effective strategy for
them to address issues of public interest and to engage and mobilize their target communities
simultaneously (Zhao & Zhu, 2014).
12
The Debate About the Judgement of Crowds
The embrace of crowdsourcing has largely been driven by the belief in the wisdom of
crowds (WoC, Surowiecki, 2005). Despite a long tradition depicting crowd behavior as irrational
(e.g., Le Bon, 1952), scholars have acknowledged “the cognitive and productive capabilities of
crowds” enabled by large-scale digitally mediated communication (Stohl, 2014, p. 5). For
instance, Surowiecki (2005) claimed that a crowd produced good prediction and judgment when
it satisfied the four conditions: (a) diversity of opinion, (b) independence, (c) decentralization,
and (d) aggregation. Diversity and independence are the conditions under which systematic
errors in estimation at the individual level are minimal (Simmons et al., 2011). Decentralization
emphasizes the importance of local knowledge on which people in central locations may not be
able to draw. The assumption of aggregation highlights that “mathematical or statistical
aggregates (e.g., measures of central tendency) of the judgments of a group of individuals will be
more accurate than those of the average individual by exploiting the benefit of error
cancellation” (Budescu & Chen, 2015, p. 267).
Although researchers have documented the remarkable ability of crowds to make
accurate predictions and judgements, critics of the wisdom-of-the-crowd phenomenon have
pointed to a systematic bias in the generative process of crowd evaluations. For example,
Simmons and colleagues (2011) demonstrated that crowd members were strongly biased in a
sports gambling context. Driven by intuitive but misleading confidence, the judgmental biases
among gamblers intensified over time and further undermined the wisdom of crowds.
Based on analyses of publicly available ratings data retrieved from Amazon.com and Yelp.com,
Le Mens, Kovács, Avrahami, and Kareev (2018) revealed an asymmetry in which items with
higher average scores received more additional judgements from crowd members in later time
13
periods than did the ones with lower average scores. Further, items with small numbers of judges
suffered penalties in terms of quality rankings.
Other scholars expressed doubt regarding the assumptions of independence and diversity
that make crowds wise. For instance, Lorenz and colleagues’ (2011) experimental study allowed
participants to reconsider their initial responses after receiving the estimates of the others. The
final results confirmed that the wisdom of crowd effect did not hold true in the presence of even
mild social influence. The byproduct of social influence was the loss of diversity due to the
formation of consensus over time. They further argued that obtaining truly independent
judgements were not practically feasible because individuals were connected to each other
through ubiquitous social networks.
These mixed findings suggest that it is important to investigate the boundary conditions
of the judgement of crowds. Simmons and colleagues (2011) speculated that “predicting whether
a crowd will be wise or unwise demands an understanding of the psychological processes
induced by the judgment task” (p. 13). Clearly, more work needs to be done to clarify the
conditions under which crowds provide accurate estimates.
Typologies of Crowdsourcing
Researchers have relied on different standards to define typologies of crowdsourcing.
One widely-used typology is the distinction between tournament- and collaboration-based
crowdsourcing (Afuah & Tuccci, 2012; Azar, 2018; Blohm, Leimeister, & Krcmar, 2013). In
tournament crowdsourcing, crowd members submit independent solutions to challenges and
compete with each other for the winning ideas. By contrast, collaboration-based crowdsourcing
requires participants to work as groups to create complex artifacts. These two forms of
14
crowdsourcing are not mutually exclusive because crowdsourcing platforms can entail both
competitive and collaborative components (Azar, 2018; Blohm et al., 2013).
Zhao and Zhu (2014) developed a two-dimensional model to classify existing
crowdsourcing applications. The first dimension is context, as the crowdsourcing model can be
utilized for either business or non-profit purposes. The other dimension, function, refers to “the
part of the product and/or service lifecycle that is being crowdsourced” (Zhao & Zhu, 2014, p.
424). Based on an open coding of 126 existing crowdsourcing cases, Zhao and Zhu (2014, p.
425) found that the most common crowdsourcing functions were (a) design and development, (b)
idea and consultant, and (c) test and evaluation.
Malhotra, Majchrzak, Kesebi, and Looram (2017) emphasized the difference between
internal and external crowdsourcing. In internal crowdsourcing, organizations solicit ideas from
their own employees, facilitating intra-organizational knowledge creation, sharing, and diffusion
(Stephens, Chen, & Butler, 2016). External crowdsourcing involves requesting novel ideas from
stakeholders (e.g., suppliers and consumers) outside the focal organizations. The advantages of
using external crowds for new product development and problem-solving include “diversity of
ideas,” “direct engagement with customers,” “implicit customer feedback on solutions,” and
“increased marketability of solutions” (Malhotra et al., 2017, p. 75).
Brabham (2017) identified four approaches to crowdsourcing, including knowledge
discovery and management (KDM), distributed human intelligence tasking (DHIT), broadcast
search (BS), and peer-vetted creative production (PVCP). KDM crowdsourcing requires an
organization to call on “an online community to find and gather information in a prescribed
format to build a common set” (Brabham, 2017, p. 594). DHIT crowdsourcing is ideal for
solving “problems where the information is already in hand, and the efforts of an online
15
community are needed to process it” (Brabham, 2017, p. 594). The BS approach is particularly
useful when a sponsoring organization needs to address “problems with empirical verifiable
solutions like scientific problems” (Stephens et al., 2016, p. 212). Finally, the PVCP type is
suitable for dealing with “problems where solutions are derived from taste, aesthetical appeal, or
market support” (Stephens et al., 2016, p. 212). Dell’s IdeaStorm should be considered as an
example of the PVCP approach because the company held ongoing open challenges that allowed
members to submit creative ideas and to judge the novelty of others’ ideas (Bayus, 2013).
Clearly, KDM and DHIT are ideal for information management problems, and BS and PVCP
crowdsourcing are primarily used to solve ideation problems.
Blohm, Zogaj, Bretschneider, and Leimeister (2018) argued that organizations could
engage in crowdsourcing activities in four different approaches, including microtasking,
information pooling, broadcast search platforms, and open collaboration platforms. These four
archetypes of crowdsourcing differed in two dimensions: (a) diversity of contributions
(homogenous or heterogeneous contributions) and (b) aggregation of contribution (selective and
integrative contributions). As for the first dimension, contributions are homogenous when target
challenges are well specified and highly standardized. A well-known example of homogeneous
crowdsourcing platforms is Amazon’s Mechanical Turk in which task requirements are typically
straightforward and highly repetitive. Heterogeneous contributions arise from “open and
unstructured tasks for which numerous alternative solutions are contributed, for example, single
contributions in GE’s Ecomagination Challenge are likely to be highly differentiated from each
other” (Blohm et al., 2018, p., 124). As for the dimension of aggregation, selective contributions
are different from integrative contributions in the aspect that the latter requires integration of
16
diverse resources, knowledge, expertise, and assessments. Therefore, single contributions have
value only when they are aggregated to the collective level.
In terms of incentives, there is a distinction between outcome-based and contribution-
based crowdsourcing (Majchrzak & Malhotra, 2014). In the outcome-based crowdsourcing
model, sponsoring organizations reward those who create the winning ideas. For example, Dell’s
IdeaStorm program sent gifts to crowd members who proposed creative ideas that were
eventually accepted and implemented by the company (Bayus, 2013). In the other form of
crowdsourcing, top contributors who submit the largest number of creative ideas or useful
comments receive social and economic recognition. Contribution-based crowdsourcing is
especially important for a sponsoring organization when the latter is “particularly concerned
about fostering enough crowd activity on the crowdsourcing platform” (Majchrzak & Malhotra,
2014, p. 260). These two forms of crowdsourcing are not necessarily exclusive, as crowd
members who contribute most to the platform can generate high-quality ideas through the
learning process (Riedl & Seidel, 2018).
Existing Empirical Research on Crowdsourcing
As an emerging social phenomenon, crowdsourcing has received considerable scholarly
attention. Typical units of analysis in prior empirical research on crowdsourcing included
submitted ideas, crowd members, and sponsoring organizations. On the one hand, great scholarly
attention has been paid to the antecedents of creative participation, success, and judgement of
proposed ideas in crowdsourcing competitions. The empirical findings demonstrated that
crowdsourcing outcomes were determined by multilevel factors. For example, Guth and
Brabham (2017) developed the Peer Vetted Ideas model and empirically showed that a design’s
visual style, visual content, textual narrative, and number of received comments were significant
17
predictors of the likelihood that the design would receive high scores from audiences. Using both
quantitative and qualitative data from a creative crowdsourcing platform, Chua, Roth, and
Lemoine (2015) found that an idea’s success rate was contingent on the level of cultural
tightness, defined as the degree to which “a society is characterized by strong social norms and
low tolerance for deviant behavior” (p. 190), in both the innovator’s and the audience’s
countries. Based on a randomized field experiment, Liu, Yang, Adamic, and Chen (2014)
revealed that an increase in reward size would lead to an increase in both the quantity and quality
of submitted ideas, highlighting the importance of website design. By using a controlled
experiment approach, Stephen, Zubcsek, and Goldenberg (2016) addressed the endogeneity
concern and confirmed that a highly clustered personal network significantly reduced the
innovativeness of the person’s submitted idea due to a high level of idea redundancy. Other
researchers demonstrated the key role the learning process played in achieving individual success
in crowdsourcing competitions. Empirical evidence consistently showed that contributors
improved their performance evaluations by learning the preferences of both crowd members and
sponsoring organizations directly through participation or indirectly through observation (Huang,
Singh, & Srinivasan, 2014; Riedl & Seidel, 2018; Schemmann, Herrmann, Chappin, &
Heimeriks, 2016).
On the other hand, many scholars have established theoretical models to explain the
decision to crowdsource a task (e.g., Afuah & Tucci, 2012; Allen, Chandrasekaran, & Basuroy,
2018) and to filter out suggestions (Blohm et al., 2013; Piezunka & Dahlander, 2015) at the
organizational level. For example, Afuah and Tucci’s (2012) paper on organizational search
theory proposed that the probability that a focal agent would use crowdsourcing as a mechanism
for problem-solving was high under the following five conditions:
18
(1) the problem is easy to delineate and broadcast to the crowd, (2) the knowledge
required to solve the problem falls outside the focal agent’s knowledge neighborhood
(requires distant search), (3) the crowd is large, with some members of the crowd
motivated and knowledgeable enough to self-select and solve the problem, (4) the final
solution is easy to evaluate and integrate into the focal agent’s value chain, and (5)
information technologies are low cost and pervasive in the environment that includes the
focal agent and the crowd. (pp. 356)

Other empirical research showed that, despite the positive impact of a crowdsourcing decision on
subsequent market performance (Allen et al., 2018; Nishikawa, Schreier, Fuchs, & Ogawa,
2017), organizations tended to pay attention to only a limited number of suggestions that they
were familiar with (Piezunka & Dahlander, 2015), suggesting a lack of absorptive capability that
enabled organizations to transform crowd contributions into business value (Blohm et al., 2013;
Malhotra & Majchrzak, 2014).
There are some limitations of empirical research on crowdsourcing. The first issue is the
disconnection from existing typologies of crowdsourcing. While prior literature has pointed out
different organizational forms of crowdsourcing (Blohm et al., 2018; Brabham, 2017; Malhotra
et al., 2017; Zhao & Zhu, 2014), empirical researchers tend to lump them together again, making
none of the distinctions described in the previous section. In view of this, more work needs to be
conducted to reveal how platform-specific characteristics shape crowdsourcing processes and
dynamics.
Another observation is that most of the existing literature on crowdsourcing assumes that
online crowds act independently (for notable exceptions, see Le Mens et al., 2018; Stephen et al.,
2016; Stephens et al., 2016). Particularly, this assumption is prevalent in the research on the
factors that drive a contributor’s creativity engagement, success, and judgment in crowdsourcing
competitions (e.g., Blohm, Riedl, Füller, & Leimeister, 2016; Chua et al., 2015; Huang et al.,
2014; Liu et al., 2014), largely because the WoC argument builds upon the condition in which
19
crowd members make independent decisions (Surowiecki, 2005). However, in doing so,
researchers neglect the possibility that crowd members influence each other’s actions. Yet, true
independence is not easy to achieve in most crowdsourcing contexts. For example, some
crowdsourcing sites allow voters to see others’ votes before casting their own (e.g., Bighash, Oh,
Monge & Fulk, 2018). The assumption of complete independence is also contradictory to the fact
that crowd members are embedded in social networks through various socializing processes
afforded by crowdsourcing platforms, such as “following” others, commenting on others’ ideas,
and participating in interactive forums. Empirical research has revealed the antecedents of
crowdsourcing networks and their consequences for creativity performance. Using observational
data from an internal crowdsourcing competition where the sponsoring organization solicited
creative ideas from its employees, Stephens and colleagues (2016) found that popularity and
transitivity were significant characteristics in the bipartite upvote network between employees
and ideas. Homophily also played an important role, such that employees from the same
geographic area were more likely to upvote the same idea(s). Stephen and colleagues (2016)
constructed a communication network based on how crowd members were exposed to others’
ideas. The experimental results confirmed that an increase in network clustering resulted in a
decrease in the innovativeness of a crowd member’s submitted idea due to information
redundancy in the personal network. Despite these insightful findings, it is clear that more work
needs to be done to unpack the mechanism through which central communication processes such
as organizing, socializing, and mediating shape crowd behavior in online crowdsourcing (Stohl,
2014).
20
Crowdsourcing Communities as Online Communities
Digitally mediated communities co-evolve with new ICTs and social change (Bennett &
Segerberg, 2013; Malinen, 2015). Given this evolving nature, the current research conceives of
crowdsourcing communities as a new case of online communities. A crowdsourcing platform
provides a common space that enables individuals who do not know each other to work
individually or collaboratively on a collective problem or challenge. Crowd members with
similar interests develop shared goals and norms, establish social connections, and form a sense
of belonging and togetherness over time (Brabham, 2010; Bauer, Franke, & Tuertscher, 2016;
Stephen et al., 2016). In this regard, these features of online crowdsourcing are the same as those
of traditional, physical communities.
Thanks to the emergence and development of various new ICTs, the rise of online
communities transforms the traditional notion of communities into technologically-supported,
loosely-connected networks based on common interests. Crowdsourcing platforms perform
similar functions by allowing geographically dispersed individuals to transcend geographic and
institutional constraints to engage in various forms of computer-mediated communication for
idea generation, knowledge sharing, resource combination, and collaboration (Faraj, Jarvenpaa,
& Majchrzak, 2011; Faraj & Johnson, 2014).
In addition, crowdsourcing communities are similar to online communities in terms of
temporariness or transience (Bighash et al., 2018; Majchrzak & Malhortra, 2016) because a
crowd exists for a short period of time and dissolves once a crowdsourced task is completed
(Majchrzak & Malhortra, 2016). While some crowd members will leave immediately after
making contributions and may never go back to the community again, others will work with new
entrants to form another temporary crowd to find solutions to the next problem. Even in a
21
repeatable and ongoing crowdsourcing challenge, crowd composition is changing all the time
(Riedl & Seidel, 2018). Given the fluid nature of crowds, member motivation to persist on
crowdsourcing platforms varies from person to person.
One feature that distinguishes crowdsourcing communities from existing forms of online
communities is the role that sponsoring organizations play in mediating collective action among
crowd members. Specifically, a sponsoring organization, or a crowdsourcer who solicits ideas
from external crowds, establishes a digital platform where geographically dispersed individuals
can gather together to perform certain tasks. The organization guides crowd’s collective action
by identifying unsolved problems, building collective identity, offering economic and social
incentives, and designing platform infrastructure. In this sense, crowdsourcing activities are not
purely driven by self-organization processes as in previous online communities, but are largely
constrained by top-down organizational control (Brabham et al., 2014). When the sponsor is a
for-profit organization, collective action in a crowdsourcing platform produces private goods
because the organization generally owns all property rights of outputs contributed by crowd
members (Majchrzak & Malhotra, 2014). In other words, community members propose creative
ideas to get a chance of receiving social and economic benefits at the expense of giving up
intellectual property rights. In contrast, collective action in prior forms of online communities
often results in the creation of public goods for the whole community (Bighash et al., 2018).
While interactions in digitally mediated communities often result from existing
geographically bounded communities (Quan-Haase et al., 2017; Vaskelainen & Piscicelli, 2018)
or lead to offline meetings (Bulger et al., 2015; Shen & Cage, 2015), it remains unclear whether
crowdsourcing communities are also intertwined with offline counterparts. Clearly, more work
22
needs to be done to address the question about the offline origins or extensions of crowdsourcing
communities.
Stitchly
Founded in 2001, Stitchly (a pseudonym) is a clothing company that crowdsources the
creation of artistic designs through an ongoing competition on its digital platform. The major
challenge for Stitchly is to find creative designs that can be printed on t-shirts, phone cases,
accessories, and home goods that the company sells to the public.
The business model of Stitchly is described as follows. Crowd members serve as
contributors of designs, and others cast votes or give comments indicating their quality
judgment. Each submitted design is scored on a five-point scale (1 to 5) and is available for
evaluation for seven days. The voting process is relatively independent. Voters cannot view the
others’ ratings on the design until the evaluation period has closed. Neither can they be exposed
to the arithmetic mean of all ratings when making evaluations, thus preventing information
cascades. In the final stage, the company selects some of the top designs as products and sells
them back to the online crowd for profit. This selection mechanism remains unknown because
the company does not make its criteria publicly available to site users.
The competition for wining designs among crowd members is fierce on Stitchly’s site.
As of December 31, 2018, there were 446,937 submitted designs in the company’s history but
only 9,051 of these were eventually printed for sale. In other words, the overall success rate of a
design was as low as 2%.

23
Table 1
Number of Submitted and Printed Designs on Stitchly's Site (2014-2018)
Time
Number of
Submitted
Designs
(Annual)
Number
of Printed
Designs
(Annual)
Number of
Submitted
Designs
(Daily)
Number
of Printed
Designs
(Daily)
Success
Rate
(Annual)
January 1, 2014 –
December 31, 2014
34,960 549 95.78 1.50 1.57%
January 1, 2015 –
December 31, 2015
29,917 1,337 81.74 3.65 4.47%
January 1, 2016 –
December 31, 2016
28,625 745 78.42 2.04 2.60%
January 1, 2017 –
December 31, 2017
26,091 336 71.48 0.92 1.29%
January 1, 2018 –
December 31, 2018
23,120 462 63.34 1.27 2.00%

Table 1 describes the daily and annual volumes of submitted and printed designs on
Stitchly’s site. The data in the second and the third columns were collected through the Wayback
Machine (https://archive.org/web/) sponsored by the Internet Archive, a nonprofit organization
that offers free access to billions of archived webpages. Scholars have already utilized web data
extracted from the Internet Archive to analyze communication phenomena (e.g., Weber, 2018;
Weber, Ognyanova, & Kosterich, 2017; Weber & Monge, 2017).
The archival records revealed that the Stitchly company had made the most recent data
about the number of submitted and printed designs publicly available since October 2013, which
enabled tracking the changes in these numbers between 2014 and 2018. The overall trend was
that the number of designs has declined over the past five years. The success rate (annual)
displayed in the final column, defined as the ratio of the number of printed ideas (shown in the
third column) to the number of submitted designs (column two), fluctuated over time, ranging
between 1.29% (year 2017) and 4.47% (year 2015).
24
The subsequent analysis places the Stitchly platform inside a multidimensional spectrum
that characterizes different typologies of crowdsourcing platforms and online communities. First,
Stitchly adopts tournament-based rather than collaboration-based crowdsourcing because it
launches an ongoing design challenge that requires designers to compete for economic rewards.
While the platform allows crowd members to co-create a design, most of them work
independently on their own ideas and do not often engage in collaboration.
Second, Stitchly adopts external rather than internal crowdsourcing. It is because ideas
are solicited from the crowd beyond the organizational boundary. In other words, the company
does not request ideas from its employees for new product development or problem solving
although they are free to participate if they wish.
Third, Stitchly adopts outcome-based rather than contribution-based crowdsourcing.
According to the rules of Stitchly, rewards are given to those who submit the winning ideas.
Designers who frequently contribute their ideas do not receive any economic rewards from the
company if none of their designs are selected as a printed product.
Fourth, participation architectures are defined as “sociotechnical systems design elements
that encourage and integrate contributions made by participants to an open online forum focused
on developing innovative solutions” (Majchrzak & Malhotra, 2014, p. 258). In the case of
Stitchly, participants are required to submit ideas in the form of both visual designs and textual
descriptions (i.e., tags and categories). There is no guideline on the content of comments, which
allows crowd members to express their opinions in a flexible way. Similarly, as the architecture
does not offer any explicit criteria in judging the quality of a submitted design, crowd members
may have conflicting views on what the best ideas are. As a result, top-rated designs may not be
the most ideal choices for the Stitchly company.
25
Empirical evidence has shown that crowd judgment is an important consideration for the
company in deciding whether to transform a submission into a product, but it may not be a
sufficient condition for product selection (Mukherjee, Xiao, Wang, & Contractor, 2018; Xu, Sun,
& Cruz, 2018).
Fifth, temporary communities are characterized as “a group of strangers coming together
for a short period of time for a particular purpose, disbanding once the time period or purpose is
achieved, and with minimal normative effects created by a shadow of the future” (Majchrzak &
Malhortra, 2016, p. 686). The Stitchly digital platform fits this fluid nature because the majority
of participants join and leave the crowd frequently and do not spend much time in the design
challenge. Prior research once reported that nearly 40 percent of all designers made only one
submission on the site (Riedl & Seidel, 2018). While the architecture of Wikipedia facilitates the
emergence of hierarchical structures where formal leaders use their authority to coordinate peer
production (Shaw & Hills, 2014), crowd members on Stitchly’s site primarily come up with
ideas independently and do not often co-create designs with others. A descriptive analysis of the
digital trace data (see the methods section for details) retrieved from Stitchly’s site on August 21,
2018 showed that design co-creation occurred rarely. There had been 439,348 designs submitted
to Stitchly since 2001, but only 2,917 of them (.66%) had more than one author.
Sixth, a knowledge-sharing trajectory refers to “the types of knowledge shared (i.e.,
“what”) and the sequence (i.e., in what order) in which different types of knowledge are shared”
(Majchrzak & Malhortra, 2016, p. 686). Knowledge sharing is integral to the crowdsourcing
process on Stitchly’s platform and is primarily achieved through commenting behavior. The
knowledge-sharing process begins after a design is submitted to the system. Then, online crowds
can comment to refine or support others’ completed designs (Xu et al., 2018). As most crowd
26
members work independently on the design challenge, they do not generally share knowledge
with others through co-creation activities.
Multidimensional and Multilevel Networks
The Multi-theoretical, Multi-level Model
Communication scholars have been long interested in investigating “the formation,
growth, maintenance, and eventual demise of communication and other network linkages”
(Monge et al., 2008, p. 449). The MTML network model (Monge & Contractor, 2003) proposes
that the formation of a network tie is determined by both endogenous and exogenous variables.
Endogenous variables refer to “structural tendencies based on configurations of the focal relation
itself” (Contractor, Wasserman, & Faust., 2006, p. 686), which capture the self-organizing
process of a focal network. Exogenous variables focus on how properties outside the network
(e.g., nodal attributes) shape the probability of tie presence in the same network (Contractor et
al., 2006). This network model is multi-theoretical because it relates various social science
theories (e.g., self-interest theories, homophily theories, and cognitive theories, etc.) with
structural configurations in a network. It highlights the importance of a multi-level perspective
because network structures span multiple units of analysis (e.g., individual, dyadic, triadic,
group, and global; Monge & Contractor, 2003). Exploring single levels of analysis will sacrifice
the richness of network data (Zappa & Lomi, 2015).
Multidimensional Networks
Recent advancements in information and communication technologies (ICTs) and their
widespread applications in the process of organizing have encouraged communication
researchers to theorize about “multidimensional networks” (Contractor, 2009; Su & Contractor,
2011; Ognyanova & Monge, 2013). Contractor and colleagues (2011) developed a typology for
27
networks, including unimodal uniplex networks, unimodal multiplex networks, multimodal
uniplex networks, and multidimensional networks (or multimodal multiplex networks).
Specifically, multidimensional networks are composed of “different types of nodes and relations
that are embedded in the same network” (Shumate & Contractor, 2013, p. 450).
Multidimensional networks are multimodal in the sense that they contain more than one type of
node, such as humans and non-human entities (e.g., technologies, databases, documents, and
other material artifacts, etc.). They are also multiplex because there are multiple sets of relations
involved (e.g., trust, friendship, and financial relationship, etc.). Fully multidimensional
networks include “multiple sets of nodes and multiple sets of relations with relations both within
sets of nodes and among sets of nodes” (Contractor et al., 2011, p. 14). For example, Brennecke
and Rank (2016) investigated the interplay between project membership and advice-seeking
behaviors within an organization from a multidimensional perspective. There were two types of
nodes: knowledge workers and project teams. They established a full multidimensional network
by examining ties among workers (a one-mode network) and interactions between workers and
teams (a two-mode network). By contrast, partially multidimensional networks have relations
among the same set(s) of nodes only, or they have multiple relations between different sets of
nodes only.
Organizational communication scholars originally theorized about multidimensional
networks to capture how the social and the material realms are interdependent in organizational
systems. This conceptualization of multidimensional networks generated important implications
for the network field as well. As Contractor and his colleagues (2011) explained:
While much has been learned from the large corpus of published research on
unidimensional networks, there is little doubt that unidimensionality is a significant
oversimplification of the rich complexity that exists in most social networks. As a general
analytic system, network analysis can be applied to both an amazingly diverse set of
28
objects and a similarly diverse set of relations. However, in most instances, any single
study typically looks at networks comprised of only one type of node and, at most, a
handful of relations among these similar objects. (pp. 4-5)

Multilevel Networks
The conceptualization of multidimensional networks parallels a recent stream of studies
using the term “multilevel networks” to describe the hierarchical nature of social structures (e.g.,
Lomi et al., 2016; Wang, Robins, Pattison, & Lazega, 2013, 2016). Multilevel networks refer to
“distinct types of nodes defined at different multiple levels (e.g., individuals and groups) with
ties possible between all nodes, both within and across levels” (Lomi et al., 2016, p. 266).
According to this definition, there are two preconditions for being considered as a multilevel
network. First, the network consists of at least two different sets of nodes. Second, the network
includes both within-level and cross-level ties.
In a more general form, a multilevel network is equivalent to a k-level (k ≥ 2) network
composed of k different types of nodes, with each type representing a level. Taking a two-level
network as an example, Wang and colleagues (2013) explicated that:
For a two-level network with u nodes at the macro-level, and v nodes at the micro-level,
we label the macro-level network as network A, the micro-level network as network B,
and the meso-level bipartite network as network X. We refer the overall network as a (u,
v) two-level network, labelled as M. (pp. 97)

Lazega and colleagues (2008) were among the first to empirically examine the formation
mechanism of a two-level network. They constructed the network based on two distinct sets of
nodes: researchers and laboratories. The data captured relations at both the researcher (advice-
seeking ties) and the laboratory levels (collaboration ties). Besides the two types of within-level
ties, they established a complete two-level network by considering cross-level ties between
researchers and laboratories (affiliation ties).
In general, the unit of analysis in social science research can be categorized into three
29
types: micro-level, meso-level, and macro-level (Blalock, 1979). However, using these labels to
define the three types of ties in the two-level network may be misleading. First, given that a two-
level network consists of pupils and classes, it is evident that pupils are micro-level nodes and
classes are macro-level nodes. But when moving to a network composed of humans and
technologies (or other material artifacts), it is somewhat arbitrary to determine which set of
nodes is the lower or upper level. Second, conceptualizing the network ties between the micro
and the macro levels of nodes as something at the meso level is also inconsistent with our
general understanding of hierarchical data structures. For conceptual clarity, the relations that
span different levels of analysis (Network X) are better understood as the cross-level ties
between the two levels of nodes. Accordingly, the other two types of relations (Network A and
Network B) in the two-level network should be labeled as within-level ties.
Network scholars have demonstrated the utility of multilevel network analysis in
enhancing our understanding of multilevel networks because this methodology considers
multiple node types and relations simultaneously and models the within-level and cross-level
interdependence together (Wang et al., 2013). Multilevel network models are different from the
traditional multilevel non-network analysis (Bryk & Raudenbush, 1992) because they (a) treat
ties rather than actors as the focal element of analysis and (b) assess the factors that drive the
presence of network interdependence across levels (Zappa & Lomi, 2015). The embrace of
multilevel network analysis is largely driven by the advances in exponential random graph
models (ERGMs) for multilevel networks (Wang et al., 2013) and the development of the MPNet
software for analyzing such networks (Wang, Robins, Pattison, & Koskinen, 2014). The current
version of MPNet has the capability to simulate and estimate a two-level network composed of
two within-level one-mode networks and a cross-level bipartite network. In early 2016, the
30
journal Social Networks released a special issue, directing attention to the empirical application
of multilevel ERGMs to analyze social networks (Brennecke & Rank, 2016; Hollway &
Koskinen, 2016; Lomi et al., 2016; Wang, Robins, Pattison, & Lazega, 2016; Zappa & Robins,
2016). To date, network scholars have applied the techniques to empirically examine the
formation mechanisms of multilevel networks in many organizational contexts. For instance,
Meredith, Van den Noortgate, Struyve, Gielen, and Kyndt (2017) found that shared membership
on subject departments significantly predicted the likelihood that teachers sought information
from each other across 12 secondary schools. Gondal (2018) confirmed that an increase in shared
specializations resulted in an increase in the probability of the exchange of doctoral candidates
between sociology departments. Although multilevel ERGMs enable researchers to empirically
test how within-level and cross-level network structures are mutually shaped, the network
models still have limitations, including “issues of model degeneracy, lack of computational
efficiency, and a possibility of inadequate representation of the modeled social processes”
(Wang, Robins, & Matous, 2016, p. 141). Clearly, statisticians and network scholars need to
further develop and refine multilevel ERGMs.
Multilayer Networks
Another related term is multilayer networks. Multilayer networks are defined as
“interconnected networks with nodes of the same nature in each layer, but interacting with nodes
of different nature in a different layer,” or “a multiplex structure with nodes of the same nature
interacting via a different network in each layer” (Diakonova, Nicosia, Latora, & San Miguel,
2016, p. 1). In a broad sense, a multilayer network is composed of “a node u in layer α that can
be connected to any node v in any layer β” (Aleta & Moreno, 2018, p. 48). Some network
scholars advocate the multilayer network approach because it provides a more accurate
31
approximation of the complexity of social structures in real contexts than what is offered by
other alternatives (De Domenico et al., 2015; Kivelä et al., 2014).
Unimodal multiplex networks and multilevel networks are special cases of multilayer
networks. On the one hand, unimodal multiplex networks fit the framework of multilayer
networks, as each layer can consist of the same set of nodes. For example, Twitter users can
initiate and maintain relations in the follower-followee (layer 1), the mention (layer 2), and the
retweet (layer 3) networks. A combination of the three types of networks constitute a unimodal
multiplex network because there are intra-layer (within-level) ties only. In other words, a node in
one layer cannot build an inter-layer tie with others in another layer. On the other hand,
multilevel networks are subsets of multilayer networks because the set of ties in multilevel
networks can be partitioned into two parts: intra-layer (within-level) ties among nodes of the
same nature in the same layers (levels) and inter-layer (cross-level) ties that connect nodes of
different nature in different layers (levels). However, the scope of multilayer networks is
broader, as inter-layer links can exist among the same set of nodes in different layers (Aleta &
Moreno, 2018; Fank & Bassett, 2017; Kivelä et al., 2014).
Conclusion
Based on the above observations, this work draws three conclusions about the MTML
model, multidimensional networks, multilevel networks, and multilayer networks. First, one
commonality of the four network frameworks is that they all highlight the importance of
multilevel theory and analysis. This echoes Moilterno and Mahony’s (2011) argument that
despite the multilevel nature of social systems, prior network scholarship has predominantly
operated at the single level of analysis.
Second, the term “multilevel” is appropriated differently by the MTML model and
32
multilevel networks. One the one hand, the MTML model uses the term “multilevel” to
emphasize that network analysis needs to consider both network structures spanning different
levels (e.g., individual, dyadic, triadic, group, and global levels) and nodal attributes. Therefore,
“multilevel” here refers to an analytical strategy on which researchers can rely. This
conceptualization of multilevel is not consistent with Blalock’s (1979) micro, meso, and macro
approaches to empirical research or Klein and Kozlowski’s (2000) classical view of multilevel
theory because the MTML model acknowledges that a multilevel analysis can be conducted in a
simple within-level or single-level network (Contractor et al., 2006; Welles & Contractor, 2015).
On the other hand, the notion of “multilevel networks” employs “multilevel” to highlight the
presence of both within-level and cross-level ties. Thus, the term indicates the nested nature of
social networks and highlights the need for going beyond a single- or within-level network
analysis (Moilterno & Mahony, 2011).
Third, multidmensional networks and multilevel networks are not interchangeable, but
fully multidimensional networks are the same as multilevel networks. In other words, being a
multidimensional network is a necessary but not sufficient condition for being a multilevel
network. For example, suppose we have collected network data about how people contribute
information to and retrieve information from several kinds of new ICTs. This network is
multidimensional because there are two types of nodes (people and new ICTs) and two types of
relations (contributing information to and retrieving information from). However, this is not a
multilevel network because there is no within-level relation (relations among people or relations
among new ICTs).
Fourth, multilevel networks and fully multidimensional networks are special cases of
multilayer networks. In a typical multilayer network, a node belongs to one or more layers and
33
can be connected to another node in any layers. Thus, multilayer networks include more types of
social structures than do the other network terms.
This dissertation chooses (fully) multidimensional networks and multilevel networks to
explain structural complexity in online crowdsourcing. While these two concepts were proposed
by scholars from different disciplinary backgrounds (e.g., communication, sociology, and
statistics), they overlap each other and are more concise than the term “multilayer networks” in
describing structural embeddedness in this research setting because cross-level ties exist between
different sets of nodes only. To engage in dialogue with different camps in network research and
to avoid conceptual ambiguity, this dissertation uniformly uses the term “multidimensional and
multilevel networks” in the following parts to provide a more accurate understanding of network
dynamics in online crowdsourcing.
Evolutionary and Ecological Theories
Organizational Ecology and Communication
The past 40 years have witnessed the development of theories of organizational ecology
(Aldrich & Reuf, 2006; Bogaert, Boone, Negro, & van Witteloostuijn, 2016; Hannan et al., 2007;
Lander & Heugens, 2017). Inspired by Darwin’s (1859) theory of biological evolution and
Hawley’s (1950) ideas regarding human ecology, organizational researchers introduced the
ecological perspective to explain the diversity of organizational forms, structures, and behaviors
over time (Hannan & Freeman, 1977, 1984). Specifically, theories of organizational ecology
emphasize the roles of organizational characteristics and environmental conditions in
determining the founding, disbanding, and transformation rates of organizations (Baum &
Amburgey, 2002; Hannan & Freeman, 1977, 1984, 1989). The basic premise of this theoretical
approach is that organizations collectively evolve as open systems. Consequently, organizational
34
change should be understood as something resulting from a focal entity’s interactions with other
constituents (e.g., consumers, competitors, suppliers, governmental agencies) in the
organizational field and with the resource environment.
Population is an important term in organizational ecology research. Here, this term is not
used to describe the total set of similar entities about which researchers use statistical samples to
make inferences. Rather, populations refer to “all those organizations that compete for resources
in the same environmental niche” (Scott, 2004, p. 8). Prior research relied mainly on industrial
categories to define populations (Delacroix & Carroll, 1983) because categorization provides
cognitive foundations for social entities to make sense of the industrial world (Vergne, 2012). As
some organizations (e.g., social movement organizations) do not fit the conventional domain of
industry, Hsu and Hannan (2005) developed an identity-based approach to defining populations.
Identities are “a set of codes held by audiences specifying the feature that an organization is
expected to possess” (Hsu & Hannan, 2005, p. 487). These audiences are both internal and
external to a focal organization, including employees, investors, suppliers, consumers, critics,
industrial analysts, among others. According to this identity-based approach, a population is
composed of organizations that adhere to a coherent collective identity.
While organizational ecology mainly uses the term population to refer to a set of
organizations with similar forms, this population thinking can be applied to explain dynamic
changes of other social and material entities. Up to now, researchers have defined populations in
terms of routines, categories, individuals, groups, communities, and others (Baum & Amburgey,
2002; Baum & Shipilov, 2006; Hannan et al., 2007). As a result, populations are also multilevel
in nature.
Rogers (1994) once argued that the roots of the communication discipline could be traced
35
back to Charles Darwin and his ecological thinking. Interestingly, it was not until quite recently
that communication researchers began to empirically test, develop, and refine theories of
organizational ecology. There are three related lines of research in the existing literature.
The first stream of studies has paid much attention to the birth (Bryant & Monge, 2008;
Lowrey, 2017; Weber, Monge, & Fulk, 2016), transformation (Dimmick, 2003; Bryant &
Monge, 2008; Weber & Monge, 2017; Xu, 2018), and death (Kim, Konieczna, Yoon, &
Friedland, 2016; Larson & Linder, 2018) of organizational populations in the contemporary
media landscape. For example, Weber and colleagues’ (2016) analysis of historical data showed
that the social networking site (SNS) form emerged through two ecological processes: (a)
replication of features across different sites, producing greater density of particular SNS forms,
and (b) recognition from adjacent industries such as advertisers. Lowrey (2017) combined
population ecology with the institutional logic approach to explain the emergence of news fact-
checking sites. He showed that the founding rate of these websites was determined by both the
existing number of organizations within the population and the external institutional
environment. Larson and Linder (2018) explained the relative success of some citizen journalism
sites during a period of substantial decline in such sites. They found that sites were less likely to
disband if they had been in existence longer but also if they adopted conventions of traditional
journalism such as a for-profit or community-based business model.
The second line of research has examined the ecological forces that shape the fitness of
online or offline groups, populations, and communities (Lai, 2014; Lowrey & Kim, 2016;
Hilbert, Oh, & Monge, 2016; TeBlunthuis, Shaw, & Hill, 2017). For example, Lai’s (2014)
longitudinal analysis showed that the survival rate of voluntary Meet-Up groups in digital space
was influenced by the ecological factor of density, indicated by the number of groups in
36
existence that were targeting the same resource space (i.e., competitors). Hilbert and colleagues
(2017) revealed that environmental selection acted not just upon entities per se, but also on ties
between entities. Their analysis showed that ecological factors favored certain online
communities because of their network positions and partitions as well as traits of community
members. TeBlunthuis and colleagues (2017) drew on organizational ecology theory to examine
whether collective actions in the forms of e-petitions competed with each other for signatures
because they tapped similar resource niches. Their analysis of more than 400, 000 Change.org
petitions confirmed that competition for resources influenced the number of signatures received.
The third stream of empirical studies has applied the ecological perspective to explain tie
formation and decay of interorganizational or digitally mediated networks (Lee & Monge, 2011;
Margolin et al., 2015; Shumate, Fulk, & Monge, 2005; Shen et al., 2014). For example, Lee and
Monge (2011) focused on the evolution of multiplex communication networks among
organizations. They demonstrated that generalist organizations were more likely to initiate
multiplex ties than were specialist organizations. Margolin and colleagues (2015) examined the
formation and dissolution of network ties among organizations in the children’s right
community. They found that institutional codification of shared principles in the form of the
Convention on the Rights of the Child yielded a significant impact on interorganizational tie
survival, such that the risk of tie dissolution was higher for younger links after the ratification of
normative rules in the organizational community. Shumate and colleagues (2005) drew on
ecological principles to predict the formation and dissolution of patterns of interorganizational
collaborations within the community of HIV-AIDS international nongovernmental organizations.
They found that patterns of collaboration between organizations were predicted by past
collaborations, geographic proximity, and common connections with intergovernmental
37
organizations.
While these communication studies have shed light on the ecological dynamics of
organizational change and network evolution, they neglected the theoretical insights from the
most recent of ecological research such as ecologies of categories (e.g., Pontikes & Hannan,
2014; Pontikes & Kim, 2017). A category-based explanation for evolutionary changes of social
entities better captures ecological dynamics because this view acknowledges that entities have
differential grades of membership in socially constructed populations. In other words, an entity
does not necessarily have a full identity in a population. Given the importance of research on
ecologies of categories, the next three sections review some key concepts and propositions in this
area and discuss how a category-based ecological explanation refines density dependence theory
and niche theory.
An Ecology of Categories
Over the past decade, organizational ecologists have paid considerable attention to the
consequences of “categories” in classification systems and the process of “categorization” (e.g.,
Hannan et al., 2007; Hannan, 2010; Hsu, 2006; Negro, Koçak, & Hsu, 2010) for decision making
and quality judgement. The widespread interest in categories and categorization was stimulated
largely by Zuckerman’s (1999) earlier work on the categorical imperative. As Zuckerman (2017)
explained,
This imperative derives from the tendency for evaluators to place less value on
“offerings” that do not fit in the categories they use to organize valuation. Insofar as this
tendency to devalue the ill-fitting is strong, “candidates” feel a strong pressure to
conform – that is, to shape themselves or their offerings so that they fit in existing
categories, and insofar as this categorical imperative is heeded, the aggregate result will
be stasis. (pp. 32)

As a useful concept for understanding organization-environment relationships, categories
are defined as “an audience’s collective agreement that members of a set belong to it based on
38
the extent to which they share similarities” (Durand & Paolella, 2013, p. 1104). Thus,
organizational forms can be better understood as socially constructed categories in
institutionalized classification systems. Organizations affiliated with the same category have
shared identity. Market participants, including consumers, investors, suppliers, and critics, rely
on the existing evaluative schemas about the meanings of a category and the larger classification
system to make sense of the outside world (Kovács, Hannan, & Hsu, 2014; Lounsbury & Rao,
2004; Navis & Glynn, 2010). For instance, restaurants can be assigned to one or more cuisine
categories such as Italian, Mediterranean, Middle Eastern, American, Mexican, Chinese, Indian,
Japanese, Korean, Vietnamese, and so on. Consumers make evaluations about a restaurant’s
authenticity by perceiving the degree to which the food it serves fits the general expectation of
the restaurant’s self-claimed category or categories (Kovács, Carroll, & Lehman, 2014).
According to Navis and Glynn (2010), categories have two basic elements. The first is
constituent members. Although category members share collective identity (Negro et al., 2010),
not all members have the same level of identification with the category because they want to
balance their similarity with the other category members and their own uniqueness (Hannan et
al., 2007; Vergne & Wry, 2014). Consequently, category members often have partial
membership because their offerings are not focused enough and can be affiliated with other
categories (Hannan et al., 2007; Hannan, 2010). The second element is a label that reflects “the
commonalities that link together the members of the category” (Navis & Glynn, 2010, p. 440).
Languages are central to categories because they cognitively construct classification systems
(Lounsbury & Rao, 2004). As a label “conveys broad agreement about the pairing of a concept
with a label, which gives the label a social meaning” (Pontikes & Hannan, 2014, p. 313),
39
classification systems typically contain categories that are largely socially constructed (Lerner &
Lomi, 2018; Pontikes & Hannan, 2014).
Categorization involves “lumping similar things into distinct clusters, rendering them
cognizable, and creating shared understanding” (Lounsbury & Rao, 2004, p. 969). This process
is ubiquitous because it occurs across a wide range of social and market contexts (Vergne &
Wry, 2014). For example, movies are categorized into genres (Hsu, 2006). Corporations are
affiliated with industrial fields (Pontikes, 2012). Articles on Wikipedia are assigned different
categories to reflect their positions in the hierarchical knowledge classification system (Lerner &
Lomi, 2018). The process of categorization is important because it shapes the identities of market
producers and what other audiences expect them to do (Negro et al., 2010).
The main proposition of the ecological theories of categories is that producers use
categories primarily to communicate with audiences (or consumers) and to position themselves
against competitors in the larger classification system (Hannan et al., 2007; Pontikes & Kim,
2017). Socially constructed categories produce an exogenous constraint on producers and guide
their action. Drawing insights from fuzzy set theory, ecological researchers acknowledge that
categorical boundaries are fuzzy sets. In other words, social entities can have differential grades
of membership (GoM) in categories (Hannan et al., 2007; Hannan, 2010; Pontikes & Hannan,
2015). GoM refers to an entity’s “partial membership in multiple categories rather than full or
non-membership in a focal category” (Kennedy & Fiss, 2013, p. 1148). A category’s boundary is
fuzzy when members in a category have memberships in others (or category memberships are
partial). A producer’s GoM reflects the extent to which it fits the schemas set by audience
members (Kvocas & Hannan, 2010). A full grade of membership indicates that a focal entity fits
the category neatly or has a high degree of typicality as a member of the focal category (Kovács
40
& Hannan, 2010, 2015; Kuilman & Wezel, 2013). GoM of an entity can be determined through
audience or critic assignment (Kovács & Hannan, 2010; Light & Odden, 2017), self-claimed
labels (Kovács & Johnson, 2014; Kuilman & Wezel, 2013; Pontikes, 2012; Pontikes & Barnett,
2015; Pontikes & Hannan, 2014), and similarity (Hsu, 2006), among others (Hannan, 2010;
Pontikes & Barnett, 2015). GoM can be used to evaluate the categorical niche width of an entity
(Kovács & Hannan, 2010; Negro, Hannan, & Rao, 2010). It is also a baseline for constructing
category-level variables, including contrast (Kovács & Hannan, 2015), contrast share (Kovács &
Hannan, 2010), fuzziness (Pontikes, 2012), and leniency (Pontikes, 2012; Pontikes & Hannan,
2014; Pontikes & Barnett, 2015). Contrast refers to “the degree to which a set stands out from
the background, the clarity of its boundary” (Kovács & Hannan, 2015, p. 264). A category will
have a high level of contrast when most members are affiliated only with the category. Category
share is “the ratio of the maximum contrast of the assigned categories to their sum.” (Kovács &
Hannan, 2010, p. 186). This ratio will be close to zero when a market offering belongs to
multiple low-contrast categories and these categories have equal contrast score. Fuzziness and
leniency are both used to measure a category’s ambiguity (Pontikes, 2012). Fuzziness reflects
“the extent to which members of a label identify elsewhere” (Pontikes, 2012, p. 94). A category
will have no fuzziness when all affiliated members are assigned only to the category.
Mathematically, fuzziness is computed by subtracting the contrast score of the category from
one. Similarly, leniency indicates how categories are close to many others (Pontikes & Hannan,
2014). It is calculated by multiplying fuzziness by “the (natural log of the) number of distinct
other labels members identified with” (Pontikes, 2012, p. 94). Thus, high fuzziness does not
necessarily lead to high leniency.
41
The research on ecologies of categories generates important implications for theory
development in organizational ecology. The shift to categories and categorization processes has
resulted in revision of existing theories, including density dependence theory and niche theory.
The next two sections introduce the original formulations of these theoretical frameworks and
how organizational ecologists reformulate them by conceptualizing socially constructed
categories in the classification system as fuzzy sets.
Density Dependence Theory, Fuzzy Density, and Contrast
Density dependence is one of the most important ecological processes of organizations
identified by the literature on organizational ecology (Baum & Amburgey, 2002). First
developed by Hannan and Freeman (1989), the classical density dependence model is generally
used to explain the impact of population density on important events in the organizational life
cycle like birth, transformation, and death. The concept of density dependence describes “the
relationships between population growth processes and the size of population itself” (Aldrich &
Reuf, 2006, p. 214). Population density is the total number of organizations in a population
(Aldrich & Reuf, 2006).
The main proposition of the density dependence model is that population density is a
proxy measure for constitutive legitimation and diffuse competition within a population and
therefore predicts the population’s subsequent growth patterns (Carroll & Hannan, 2000; Hannan
& Freeman, 1989). Constitutive legitimation describes the extent to which “stakeholders
perceive and support a certain organizational form as a natural, taken-for-granted way to perform
a certain kind of action” (Bogaert et al., 2016, p. 1348), reflecting the structural context of a
market niche (Überbacher, 2014). This term is often used interchangeably with cognitive
legitimacy or taken-for-grantedness by ecologists (Lander & Heugens, 2017). Diffuse
42
competition stems from crowding in niche space (Lander & Heugens, 2017). As organizations do
not usually engage in direct competition with all the remaining members in an organizational
population, this term provides a fine-grained structure of competition at the population level
(Baum & Singh, 1994).
Density dependence theory predicts that legitimation grows with population density at a
decreasing rate, while competition increases at an increasing rate (Hannan, Carroll, Dundon, &
Torres, 1995; Haveman, 1993). More specifically, an increase in density from an initial low level
(with zero as the lowest level) mainly reflects the growing external legitimacy of a population,
thus encouraging more entries into the population. Population density is a crucial indicator of
legitimacy because “growing numbers of a thing make it seem both less anomalous and more
like a new population of organizations” (Lo & Kennedy, 2015, p. 123). Later, competition
processes overwhelm legitimation processes and predominantly shape a population’s future
growth when population density reaches the environmental carrying capacity. Carrying capacity
reflects “the maximum density that an environment can support” (Aldrich & Reuf, 2006, p. 212).
As the competition among existing organizations for limited resources becomes intensified in
this stage, there are fewer new entries due to a lack of environmental opportunities (Carroll &
Hannan, 2000; Hannan & Freeman, 1989). Prior empirical research has consistently reported an
inverted U-shaped relationship between population density and the rate of entries across
organizational fields (Aldrich & Reuf, 2006; Baum & Amburgey, 2002; Lander & Heugens,
2017; Stretesky, Huss, & Lynch, 2012). In addition, the curvilinear relationship between
population density and organizational entries is moderated by time-specific, place-specific, and
field-specific variables. For instance, a meta-analysis conducted by Lander and Heugens (2017)
43
demonstrated that this non-monolithic relationship was contingent on population age, historical
time, socio-political legitimacy, and the emergence of prototypical categories.
The revised density dependence theory integrates insights from cognitive psychology and
categorization theory (Rosch & Lloyd, 1978) into the ecological framework and emphasizes the
importance of fuzzy density and contrast in driving organizational entry and exit rates (Hannan et
al., 2007). Both concepts take into account the cognitive ambiguity of category-based or identity-
based populations (Bogaert et al., 2016). For example, audiences may not perceive every
newspaper organization equally well as a member of the population of newspapers because a lot
of the papers have become completely Internet-based. As a refined measure of population
density (defined as the total number of organizations in a population) in ecological analysis,
fuzzy density is the sum of the GoM of all organizations in a population (Hannan et al., 2007).
Contrast refers to “the degree to which a set stands out from the background, the clarity of its
boundary” (Kovács & Hannan, 2015, p. 264). High contrast means low fuzziness of a
population, indicating that members of a population do not identify elsewhere (Hannan, 2010).
Operationally, contrast is computed as the average GoM of member organizations (Hannan et al.,
2007).
Prior research has shown that both fuzzy density and contrast well capture the
legitimation process of population evolution by taking into account unequal contributions to
legitimacy by organizations with differential grades of membership. Integrating these two
concepts into the existing theoretical framework enables researchers to reformulate the
legitimation part of the density dependence theory (Bogaert, Boone, & Carroll, 2010; Bogaert et
al., 2016). Empirical results have shown that fuzzy density and contrast increase organizational
entry rates (Kuilman & Li, 2009) and decrease organizational exit rates (Bogaert et al., 2010;
44
Kuilman & Wezel, 2013). Increasing levels of fuzzy density or contrast contribute to the external
legitimacy of a population thus encouraging entry and discouraging exit from the population.
Niche Theory and Fuzzy Niche Theory
In evolutionary and ecological research, niche theory provides an explanatory framework
for competition and coexistence within and between populations (Hannan & Freeman, 1977).
Initially borrowed from biology, the concept of “niche” plays an important role in the ecological
context (Hannan & Freeman, 1977). In animal ecology, niche refers to “a location in
multidimensional space defined by the resources in the environment” (McPherson, 1983, p. 520).
In organizational scholarship, niche is defined as “the n-dimensional resource space within
which a population can exist” (Carroll, 1985, p. 1266) or “a distinct combination of resources
and other constraints that are sufficient to support a population” (Aldrich & Reuf, 2006, p. 183).
The degree of competition between two populations of social entities is usually a function of the
extent to which they overlap in various community niches (Aldrich & Reuf, 2006; Baum &
Singh, 1994; Rao, 2007). Niche overlap reflects the degree to which two or more populations
share the same set of ecological resources (Dobrev, Kim, & Carroll, 2003). When competitors
from other populations are present, a population’s realized niche is usually smaller than its
fundamental niche. A fundamental niche represents the entire set of environmental conditions in
which a population can survive (Aldrich & Reuf, 2006; Hannan & Freeman, 1977).
Niche theory provides ideal-type distinctions between generalists and specialists that
reflect the niche width of social entities (Aldrich & Reuf, 2006; Carroll, 1985; Freeman &
Hannan, 1983; Hannan & Freeman, 1977). Niche width refers to the number of niches entities
inhabit out of the total number of niches that exist. On the one hand, generalists occupy broad
niches, meaning that they spread their activities over a wide and heterogeneous range of niches.
45
On the other hand, specialists have a narrow range of niches, possess fewer extra resources, and
build their fitness by concentrating on the narrowly defined resource space. By using this niche
strategy, specialists may be able to exploit more of the available niche resources without
engaging in direct competition with generalists.
Research on ecologies of categories has recognized that social entities’ category
memberships can reflect the entities’ niches (Hsu et al., 2009; Hannan et al., 2008; Pontikes &
Hannan, 2014). Drawing insights from fuzzy set theory, ecological researchers developed fuzzy
niche theory that defined niches in a space of fuzzy rather than crispy categories (Hsu et al.,
2009; Kovács & Hannan, 2010; Negro et al., 2010). In crispy sets, social entities are either
included or excluded from categories. By contrast, category memberships can take values on a
[0, 1] interval in fuzzy sets. Treating niches as fuzzy sets offers a nuanced way to capture the
position of social entities in environmental spaces because it acknowledges the possibility that
the entities have partial memberships in category schemas (Hannan et al., 2007; Carroll, Feng,
Le Mens, & McKendrick, 2010). As Hsu and her colleagues (2009) explained:
A producer’s niche is defined by a grade-of-membership function that tells the degree to
which each position in the space belongs in the niche. This fuzzy representation allows
explicit treatment of variations in the degree to which social positions belong to a niche,
ensuring that specialists have niches with high grades of membership in one or a few
positions, while generalists’ niches have lower (but positive) grades of membership in
several positions. (…) Targeting a diverse array of categories can be regarded as a kind of
categorical generalism. A generalist distributes its degrees of membership across
categories fairly evenly. In contrast, a specialist has a highly unequal distribution of
memberships across categories. This treatment captures a core insight of niche theory:
generalists target a greater diversity of resource positions (in this case, categories). (pp.
154)

Evolutionary Mechanisms of Variation, Selection, and Retention
The three Darwinian principles of variation, selection, and retention (V-S-R) (Darwin,
1859) provide a generalized theory of dynamic change (Campbell, 1965). Variation is the source
46
of diversity in populations. The occurrence of variations can be blind, random, haphazard,
planned, intentional, or heterogeneous (Campbell, 1965; Simonton, 1999). Selection involves
accepting some of the new variations and rejecting the others based on their relative fitness in an
environmental niche. In other words, selection mainly occurs through competition among
available options (Campbell, 1965; Monge et al., 2008). Retention is the duplication, persistence,
reproduction, or institutionalization of variations that were selected in the past. While variation
and selection forces are dominant when change is favored, the stage of retention is driven by
inertia or resistance to change (Campbell, 1965; Monge & Poole, 2008).
By examining the evolution of gas stations in Edmonton, Canada over a 30-year period,
Usher and Evans (1996) demonstrated the utility of the Darwinian V-S-R principles in
explaining longitudinal changes of organizational populations. At the beginning of the study
period, the emergence of several new competencies (e.g., car washes, convenience stores, and
gas bars) introduced variation into the population. The selection process favored some of the new
variations that outperformed old competencies, leading to failures of many old-style gas stations.
The stage of retention occurred when these selected variations were further reproduced and
institutionalized by new entrants to the organizational population. Over time, the population of
gas stations “reconfigured itself from a relatively homogeneous state dominated by a single form
to one characterized by several successful alternative forms” (Usher & Evans, 1996, p. 1461).
Although Darwin himself recognized the potential utility of the three core principles to
cover social evolution (Darwin, 1859), it was not until the 1960s that social scientists borrowed
the ideas from evolutionary biology and applied the three-stage model to explain evolutionary
processes in the social-cultural realm (Alrdich & Pfeffer, 1976; Aldrich & Reuf, 2006; Campbell,
1965; Nelson & Winter, 1982). Up to now, the cycle of V-S-R has been conceptualized as a
47
framework to understand evolutionary changes in routines (Feldman & Pentland, 2003),
vocabularies (Ocasio & Joseph, 2005), organizational forms (Aldrich & Reuf, 2006; Hannan &
Freeman, 1984), among others. These phenomena operate at multiple levels of analysis, ranging
from within organizations to organizational populations or communities (Aldrich & Reuf, 2006).
Researchers agree that a simple linear causal sequence of variation, selection, and retention
events does not occur in social evolution (Lippmann & Aldrich, 2014). Social evolutionary
processes are better understood as “a cumulative progression of numerous interrelated acts of
variation, selection, and retention over an extended period of time” (Van de Ven & Garud, 1994,
p. 427).
Some scholars have proposed that replication should be viewed as “the fourth possible
component of the evolutionary process, located between selection and retention” (Lomi & Zollo,
2008, p. 1069, see Zollo and Winter’s (2002) study as an example). Y should be considered as a
replication of X only if: “(i) X and Y are similar (in some relevant respects), and (ii) X was
causally involved in the production of Y in a way responsible for the similarity of Y to X”
(Godfrey-Smith, 2000, p. 414). In biological evolution, the fundamental logic of replication is to
copy genetic information from parental organisms to their offspring through the medium of DNA
(Metcalfe, 2008). However, the process of replication is not always without errors. The failure to
precisely or faithfully copy the hereditary information is termed as mutation (Monge & Oh,
2016; Sperber, 2000). A decrease in mutation rates favors retention over variation because it
limits the diversity of the next generation in biological reproduction (Margolin & Monge, 2013).
The self-replication mechanism also takes place in non-biological entities. For instance,
organizational theorists have employed the idea of replication to explain the diffusion of
organizational forms, routines, habits, ideas, knowledge, rules, among others (Aldrich et al.,
48
2008; Lomi & Zollo, 2008; Nicholson & Narayanan, 2008; Zollo & Winter, 2002). Here, the
notion of replication captures “the adaptation of the change proposal to the local context and the
progressive discovery of the power and the boundaries of the novel insight” (Lomi & Zollo,
2008, p. 1070). In other words, replication does not simply mean a direct copy of organizational
practices. It also serves to offer information needed to start another cycle of V-S-R. At the intra-
organizational level, the replication process is especially important for large, transnational,
multidimensional organizations because selected practices need to be implemented internally in
spatially diverse contexts (Zollo & Winter, 2002).
The key differences between biological evolution and social evolution have drawn much
attention from scholars. First, the evolutionary replicators in the two processes are distinct. While
the self-replication mechanism in biological evolution involves genes and DNA, social evolution
is embodied in treating the meme as the replicator (Ball, 1984; Dawkins, 1976; Roy, 2017). By
definition, the meme is “information copied from person-to-person by imitation” (Blackmore,
2001, p. 225). Dawkins first coined the term and conceived memes as socio-cultural equivalents
to genes. As Dawkins (1976) explained:
Just as genes propagate themselves in the gene pool by leaping from body to body via
sperms or eggs, so memes propagate themselves in the meme pool by leaping from brain
to brain via a process which, in the broad sense, can be called imitation. (pp. 227)

The characteristics of memes have significant consequences for evolution in the social
realm. As memes can be transmitted between any two individuals and can be blended or
recombined with each other easily (Dawkins, 1982), social evolution is expected to have faster
reproduction rates and lower copying-fidelity than genetic-based evolution does (Heylighen &
Chielens, 2009). Thus, social evolution suffers from a greater selection bias and a higher
mutation rate (Sperber, 2000; Creanza, Kolodny, & Feldman, 2017).
49
Second, the two evolutionary processes are fundamentally divergent in the aspect that the
evolution of social entities involves human agency (Vanberg, 2013). While biological evolution
is largely driven by natural selection, evolutionary processes in the social realm are often guided
by human entities to meet the requirements of environmental conditions (Lovas & Ghoshal,
2000). In that sense, social evolution is not purely Darwinian but Lamarckian (Heylighen &
Chielens, 2009; Rose, 1998; Usher & Evans, 1996). In biology, Lamarckian evolution means that
offspring can “inherit changes made to the phenotype during the lifetime of the parent organism”
(Rose, 1998, p. 1364). A Lamarckian-based view of organizational evolution highlights the
importance of human agency and adaptive learning. Population-level changes occur because
“existing members adapt to environmental pressures by replacing less favored competencies with
more favored competencies” (Usher & Evans, 1996, p 1436). By contrast, a Darwinian
perspective states that an organizational population evolves because “new members holding
more favored competencies compete into failure members holding less favored competencies”
(Usher & Evans, 1996, p. 1436).
The criticism of social evolution rests on the belief that the Lamarckian view of
biological evolution receives no scientific evidence. However, biologists have acknowledged the
existence of Lamarckian processes that influence the variation of phenotype generationally. For
instance, Skinner (2015) proposed a unified theory of evolution by considering both Lamarckian
and Darwinian aspects of evolutionary mechanisms and further argued that “environmental
epigenetics and genetic mutations both promote phenotypic variation on which natural selection
acts” (p. 1297).
Although social scientists have applied the V-S-R model to explain social evolution for
many years, the idea of generalizing the Darwinian principles is still controversial due to huge
50
differences between biological and social entities (Rose, 1998; Sperber, 2000). In defense of
generalized Darwinism in social sciences, some researchers have argued that the three Darwinian
principles should be better understood as meta-theory where other context-specific explanations
can be placed (Lippmann & Aldrich, 2014; Vanberg, 2013). In responses to many
misconceptions and misunderstandings from critics, Aldrich and colleagues (2008) have
emphasized that generalized Darwinism does not mean the mechanisms of biological evolution
and social evolution are exactly the same and that social scientists need to incorporate other
theoretical explanations into the core Darwinian principles. As they further explained:
To repeat: acknowledging the role of Darwinian principles in social evolution does not
imply that the detailed mechanisms of selection, variation and inheritance are analogous
or similar to biological evolution. Consequently, the application of general Darwinian
principles cannot do all the explanatory work for the social scientist. Darwinism alone is
not enough. (pp. 585)

Following this line of thinking, this dissertation applies the three Darwinian principles to
examine the evolution of multidimensional and multilevel networks in online crowdsourcing. To
better capture the evolutionary processes, this work integrates other ecological theories (e.g.,
density dependence theory, niche theory, the liability of newness theory, and structural inertia
theory) in organizational research into the baseline mechanisms of V-S-R. Organizational
ecology fits the evolutionary framework well because the emergence and development of the
former is largely inspired by Darwin’s theory of biological evolution. As organizational ecology
also draws insights from network theory, cognitive psychology, categorization theory, and fuzzy
set theory (Hannan et al., 2007; Kim et al., 2006; Pontikes & Hannan, 2014), the ecological
framework can potentially supplement the three Darwinian principles by offering multiple
additional theoretical explanations.
51
Network Evolution
Like biological entities, social networks have their own life cycles and evolve over time
(Kossinets & Watts, 2006). Researchers have applied the evolutionary mechanism of V-S-R to
explain dynamic change in communication networks (Doerfel & Taylor, 2017; Hilbert et al.,
2016; Margolin & Monge, 2013; Monge et al., 2008; Shen et al., 2014; Shumate, 2012; Shumate,
Monge, & Fulk, 2005). Monge and colleagues (2008) stress that a V-S-R framework is
particularly helpful to understand the creation, growth, transformation, decline, and decay of
network links over time. In network terms, variation means experimenting with different network
ties for planned or unplanned reasons. As a focal entity’s social network has a limited carrying
capacity (Dunbar, Arnaboldi, Conti, & Passarella, 2015), not all network links of the entity will
be retained. Therefore, the stage of selection involves choosing some ties (variations) and
eliminating the others through a process of trial and error. In other words, selector systems select
both for and against variations (Monge & Poole, 2008), but this does not necessarily mean all
chosen ties have good fitness. Retention occurs when social entities reaffirm past selections and
make selected ties as routinized practices. The stage of retention is also labeled as “selective
retention” (Aldrich & Pfeffer, 1976; Campbell, 1965) in the sense that only positively selected
ties (variations) will be persistently maintained. The sequence of the three stages can be
described as a pipeline through which individuals need to pass in managing network ties. The
pipeline metaphor does not imply a simple causal relationship among the three stages. Rather, it
acknowledges the co-production of variation, selection, and retention events (Van de Ven &
Garud, 1994).
Network evolution generally proceeds through two fundamental events: the creation of
new ties and the decay of old ones (Kleinbaum, 2018; Snijders, 2001, 2005; Yue, 2012). In other
52
words, examining network ties from an evolutionary perspective needs to explain what drives tie
formation and decay. Empirically, researchers have mainly employed advanced network
analytics and event history analysis to capture the processes of network evolution (Miller et al.,
2011; Monge et al., 2011). For instance, exponential random graphs are methodological tools
that examine the formation mechanism of network ties (Fu & Shumate, 2016; Huang & Sun,
2014; Pilny & Atouba, 2018; Wang et al., 2013). ERGMs allow researchers to determine
whether a specific structural configuration occurs more or less frequently than what is predicted
by randomness when controlling for network interdependencies or nodal attributes. Event history
analysis models tie decay along with other time-constant or time-varying covariates (Burt, 2000;
Margolin et al., 2015; Shen et al., 2014; Yue, 2012). Analyzing the event of tie decay is critical
to understand how networks evolve over time because this event is “the most clear-cut example
of selection processes at work” (Miller et al., 2011, p. 32), capturing both “selective variation”
and “selective retention” (Feldman & Pentland, 2003). Assuming network evolution results from
“many (usually non-observed) small changes occurring between the consecutively observed
networks (Snijders, 2005, p. 215), Simulation Investigation for Empirical Network Analysis
(SIENA) is a statistical program for analyzing dynamic network data with repeated measures and
the coevolution of networks and behavior (Shumate, 2012; Weber, 2012).
The utility of the V-S-R framework for network evolution has yet to be realized. Up to
now, little research has employed it to investigate how multidimensional and multilevel networks
evolve over time. Rather, researchers are limited in focusing only on the evolution of one-mode
networks (e.g., Bryant & Monge, 2008; Doerfel & Taylor, 2017; Shen et al., 2014; Shumate,
2012; Shumate et al., 2005). As multidimensional and multilevel networks include both cross-
53
level and within-level ties, examining the evolutionary mechanisms of such networks can better
capture the complexity of social systems.

54
Chapter 3: Hypothesis Development
Building a Multidimensional and Multilevel Network in Design Crowdsourcing
Networks in crowdsourcing platforms are multidimensional and multilevel in nature
(Stephens et al., 2016). In general, many crowdsourcing participants not only have direct
communication with each other, but also interact with artifacts they have produced. Recognizing
both human and non-human entities in online crowdsourcing as structured in a system of
multidimensional and multilevel networks enables us to consider “how the levels within a system
of nested networks relate to each other” (Moilterno & Mahony, 2011, p. 447).
This dissertation focuses on design crowdsourcing (Allen et al., 2018) that involves
PVCP (Brabham, 2017; Guth & Brabham, 2017; Stephens et al., 2016). In the process of external
design crowdsourcing, sponsoring organizations do not solicit ideas from internal employees but
solicit “external inputs in the form of functional design solutions for new product development
from the ‘crowd’” (Allen et al., 2018, p. 106). Specifically, crowdsourcing organizations rely on
the wisdom (or accuracy) of online crowds (in this case, designers) to generate attractive ideas
(in the present case, designs) and to select the top ones for sale. PVCP is suitable for addressing
“problems where solutions are derived from taste, aesthetic appeal, or market support” (Stephens
et al., 2016, p. 212). Categorization is an important element in PVCP because it denotes
similarity and difference among submitted ideas (Murphy, 2002) and thus act as a market signal
(Hannan, 2010; Oh & Monge, 2013; Pontikes & Hannan, 2014). For example, crowdsourcing
platforms provide an interface where some participants present creative ideas and others judge
them. Crowd members rely on categories to claim or evaluate the identity of ideas in socially
constructed classification systems. Incorporating the categorization process into the existing
analytical framework of crowdsourcing enables researchers and practitioners to better understand
55
how online crowds organize and understand information resources (Negro, Hannan, & Fassiotto,
2015; Oh & Monge, 2013).
By using online observational data from the Stitchly digital platform (see the section of
Online Communities in Chapter 2 for details), this dissertation examines the evolution of a three-
level network of designers, designs, and classification categories in online crowdsourcing.
Classification categories and designs are material entities and designers are social entities. These
three types of nodes are hierarchically nested because designers create designs and each design is
labeled with different categories.
The philosophy of biology (Hull, 1988) provides a framework to establish the structural
hierarchy in design crowdsourcing by emphasizing the differential roles that replicators and
interactors play in evolutionary processes. A replicator is “an entity that passes on its structure
largely intact in successive replications” (Hull, 1988, p. 408). An interactor is described as “an
entity that directly interacts as a cohesive whole with its environment in such a way that this
interaction causes replication to be differential” (Hull, 1988, p. 408). Differential replication
means that the reproduction rates of interactors are not always the same. In classical biological
evolution, replicators are genes (DNA), whereas interactors are organisms (Nayay, 2002). A
possible instance of differential replication is that taller organisms produce more offspring of
their own kind than shorter ones under certain circumstances (Brandon, 1996). In social
evolution, memes are conceived as replicators (Ball, 1984; Heylighen & Chielens, 2009; Roy,
2017), and human artifacts are usually recognized as interactors (Hodgson & Knudsen, 2004).
Just as genes rely primarily on organisms for biological reproduction, so do memes need
connections with human artifacts for replications (Hodgson & Kundsen, 2004). Differential
replication is likely to occur because human artifacts do not always have the same reproduction
56
rate.
Drawing on Hull’s (1988) conceptualizations of replicators and interactors, Baum and
Singh (1994) made a distinction between genealogical and ecological entities. “Genealogical
entities pass on their information largely intact in successive replications. Ecological entities, the
structural and behavioral expressions of the genealogical entities, interact with the environment
and this interaction causes replication to be differential” (Baum & Singh, 1994, p. 4). In design
crowdsourcing, categories and designs should be regarded as genealogical entities because these
material artifacts are replicator products (Ball, 1984). Memes recorded in categories and designs
are parallel with genes because they encode information and can be transmitted to later carriers
with or without modifications (Adamic, Lento, Adar & Ng, 2014). Accordingly, designers
should be treated as ecological entities in the sense that their direct interactions with the
environment and with each other lead to differential replications of memes. Replication is
enacted by these designers and takes place through imitation and social learning (Blackmore,
2001; Flinn, 1997; Minor & Raghavan, 1999). The likelihood of a meme’s replication can be
biased by specific modes of imitation. For example, Minor and Raghavan (1999) argued that
replication occurred through frequency-based, trait-based, and outcome-based imitation. Hilbert
and colleagues (2016) revealed that social evolution favored certain population types because of
traits as well as network positions. Creanza and colleagues (2017) found that social-cultural
transmission was largely shaped by the frequency of a trait in a population (e.g., conformity bias
and novelty bias) and the popularity of the people who have the trait (e.g., prestige bias and
success bias).
Based on the above discussion, this study builds a three-level network based on the
structural hierarchy in design crowdsourcing. This multidimensional and multilevel network is
57
composed of (a) cross-level ties between designs and categories (see Figure 1), (b) cross-level
ties between designers and designs (see Figure 2), (c) cross-level ties between designers and
categories (see Figure 3), (d) within-level ties among designers (see Figure 4), (e) within-level
ties among designs, and (f) within-level ties among categories. As the last two forms of networks
do not exist in the research setting, they are excluded from the analysis. This feature of the data
structure does not enable empirical tests of (a) how the evolution of ties among designs or
categories is shaped by various endogenous and exogenous mechanisms and (b) how within-
level ties among designs determine the fitness of designs. Building on the relevant literature, the
next few sections propose a series of hypotheses regarding the evolution of the multidimensional
and multilevel network on the Stitchly digital platform.
Figure 1. Hypothetical Cross-Level Ties Between Designs and Categories on the Stitchly Digital
Platform.

58
Figure 2. Hypothetical Cross-Level Ties Between Designers and Designs on the Stitchly Digital
Platform.

Figure 3. Hypothetical Cross-Level Ties Between Designers and Categories on the Stitchly
Digital Platform.

59
Figure 4. Hypothetical Within-Level Ties Among Designers on the Stitchly Digital Platform.

Cross-Level Ties Between Designs and Categories
The first part of the dissertation focuses on cross-level ties between designs and
categories. The first two hypotheses connect density dependence theory to the evolutionary
perspective. Density dependence theory proposes that population density is a proxy measure for
legitimation within a population and can thus predict the growth of a population in later periods
(Carroll & Hannan, 2000; Hannan & Freeman, 1989). The theory also assumes that “legitimation
is an ongoing process of replication of features” (Weber et al., 2016, p. 305). One important
dependent variable in the density-dependent model is the rate of entry into a population, which is
operationalized as the number of social entities in the population at time t. The main predictor of
the entry rate is population density, which refers to the existing number of entities at the previous
time point, t - 1. From an evolutionary perspective, what the density dependence model predicts
is the reproductive fitness (Metcalfe, 2008) or the retention rate of the population (Margolin &
Monge, 2013). The coefficient of the population density at time t in regression models can be
60
interpreted as the expansion or contraction of the population over the observed time periods
(Heylighen & Chielens, 2009; Hilbert et al., 2016; Metcalfe, 2008).
Density dependence theory has received widespread criticism for the assumption that
each member in a population contributes equally to the legitimation or competition process
(Bogaert et al., 2016). As a modification to the assumption of homogeneity, the recent version of
the theory proposes that the legitimation effect of population density is stronger when audiences
perceive less ambiguity about the population (Bogaert et al., 2016). Hannan and his colleagues
(2007) used fuzzy density and contrast to redefine population density. These two concepts
capture the cognitive ambiguity of a population by recognizing that organizations, or other social
entities, with high grades of membership are more typical as members of the claimed population
(Hannan et al., 2007). A population is fuzzy when its members also have multiple memberships
in other populations. Thus, fuzzy density represents the existing number of entities in a
population while considering the GoM of these entities. The contrast of a population is high
when the average GoM of all its members are high, meaning that the members are not usually
assigned to multiple populations (Kovács & Hannan, 2010). While both fuzzy density and
contrast capture fuzziness, the major difference between the two concepts is that fuzzy density
weights fuzziness by population density (Bogaert et al., 2010). Mathematically, population
density times contrast equals fuzzy density. Contrast ranges from 0 to 1. An increase in contrast
results in an increase in the degree to which a population overlaps others.
Prior research has shown that fuzzy density better captures the process of legitimation
than the traditional measure of population density does because it acknowledges that
memberships in populations can be graded and take values on a [0, 1] interval (Bogaert et al.,
2010). In addition, an increase in contrast leads to an increase in the identification of a class of
61
entities with similar characteristics by audience members. Thus, the concept of contrast shapes
cognitive legitimacy independently from density (Kuilman & Wezel, 2013). Empirical studies
have shown that both fuzzy density and contrast were positively associated with the rate of entry
into a population when other factors were held constant (Kuilman & Li, 2009). As populations
can be defined in terms of categories (Delacroix & Carroll, 1983; Hannan et al., 2007; Pontikes
& Hannan, 2014), this research extends the revised density dependence theory (Hannan et al.,
2007) to explain population evolution in the context of design crowdsourcing. It is hypothesized
that both fuzzy density and contrast are positively associated with the rate of entry into a product
category by designs. These two hypotheses test the impact of the cross-level ties between designs
and categories because the conceptualizations of both fuzzy density and contrast are built on
entities’ (in the present case, designs’) affiliations with categories. Compared with the traditional
measure, population density, in density dependence theory, fuzzy density and contrast consider
unequal contributions to categories by entities with differential grades of membership, thus better
capturing legitimation processes in category evolution. Thus, it is posited that:
Hypothesis 1 (H
1
). Increases in the fuzzy density of a category lead to increases in the
rate of entry into the category by designs.
Hypothesis 2 (H
2
). Increases in the contrast of a category lead to increases in the rate of
entry into the category by designs.
The third hypothesis focuses on the relationship between a design’s niche width and its
fitness. Niche width refers to the actual resource space that a social entity inhabits (Freeman &
Hannan, 1983). This representation classifies social entities into generalists (broad niche) or
specialists (narrow niche). Fitness is the relative adaptiveness or performance of an entity in a
resource environment (Darwin, 1859; Hannan & Freeman, 1977). In this case, fitness reflects the
62
degree to which a design appeals to external audiences including online crowds and the
sponsoring company.
There has been a long research tradition examining the relationship between an entity’s
niche width and its fitness since Hannan and Freeman (1977) first introduced the organizational
ecology approach in the 1970s. Conventional strategic management literature emphasizes the
importance of generalism (or diversification) in enhancing the survival rates of organizations
(Dobrev, Kim, & Hannan, 2011). A generalist strategy entails benefits because it helps
organizations to achieve economies of scale and scope. In addition, organizations can reduce the
risk of failure by spreading available resources over different niches (Dobrev et al., 2001). This
argument was further revised by two theoretical approaches in organizational ecology, niche
theory and resource partitioning theory.
On the one hand, Hannan and Freeman (1977; Freeman & Hannan, 1983) developed
niche theory and specified the environmental conditions under which generalists were favored
over specialists. Their findings showed that generalists outperformed specialists only when (a)
environmental variation was unstable and (b) the pattern of variation was coarse-grained. Grain
is defined as “the length of typical periodicities” (Freeman & Hannan, 1983, p. 1119). A
variation is coarse-grained (or fine-grained) when the duration of an environmental fluctuation is
typically longer (or shorter) than the lifetime of organizations (Baum & Amburgey, 2002;
Hannan & Freeman, 1977; Freeman & Hannan, 1983). At the same time, stable and fine-grained
environments favored specialists because generalists were burdened by their extra capacity and
slack resources (Hannan & Freeman, 1977).
On the other hand, resource partitioning theory acknowledges that while generalists may
benefit from possessing extra capacity, they are exposed to intense competition as well (Carroll,
63
Dobrev, & Swaminathan, 2002; Dobrev et al., 2011). When the environment is characterized by
economies of scale and the resource distribution is unimodal, specialists have higher survival
rates than do generalists because the former can exploit peripheral resources freed by the
competition among generalists in the concentrated market (Carroll, 1985). Inspired by the
“categorical imperative” thesis (Zuckerman, 1999) and by fuzzy niche theory (Hsu et al., 2009;
Kovács & Hannan, 2010; Negro et al., 2010), the most recent line of research in organizational
ecology has produced mixed findings about the relationship between niche width and fitness. As
niches are usually defined in a space of categories (Hsu et al., 2009), a producer/product’s niche
width can be reflected by category membership. In this sense, a generalist (wide niche) targets
many diverse categories, while a specialist (narrow niche) allocates resources over a limited
number of categories.
One group of researchers has confirmed that niche narrowness elicits positive reactions
from audiences (Hsu, 2006; Hsu et al., 2009; Kovács & Hannan, 2015; Kovács & Johnson, 2014;
Lerner & Lomi, 2018; Light & Odden, 2017; Negro et al., 2010; Negro & Leung, 2013; Olzak,
2016; Pontikes, 2012). This line of research assumes that a social entity has a finite amount of
resources to invest. Therefore, an increase in the number of target niches (categories) will lead to
a decrease in investment in each (Hsu, 2006; Hsu et al., 2009). This trade-off is generally termed
“the principle of allocation” and “the jack-of-all-trades phenomenon” (Hsu et al., 2009).
Spanning multiple niches results in market disadvantages because this strategy (a) confuses
audiences and poorly fits with their categorical expectations and (b) impedes an entity’s
specialized skill acquisition and expertise development (Hsu et al., 2009; Lo & Kennedy, 2015;
Negro et al., 2010).
Other research has demonstrated the positive reactions among audience members to
64
market offerings that belong to multiple categories (e.g., Lo & Kennedy, 2015; Pontikes, 2012;
Paolella & Durand, 2016; Vergne, 2012). For example, Vergne (2012) found that category
spanning helped a firm to dilute its prior association with a stigmatized label and thus reduced
disapproval. Pontikes (2012) showed that category generalists in the software industry were
more appealing to consumers than were category specialists. Niche broadness elicits positive
evaluations when institutional logics encourage novelty and non-conformity (Lo & Kennedy,
2015) and when audiences have non-monolithic preferences (Goldberg, Hannan, & Kovács,
2016; Pontikes, 2012; Smith, 2011) or complex goals (Paolella & Durand, 2016; Durand,
Granqvist, & Tyllström, 2017; Durrand & Paolella, 2013).
As niche broadness can be viewed positively or negatively (Durrand & Paolella, 2013;
Durand et al., 2017; Keuschnigg & Whimmer, 2017; Wry, Lounsbury, & Jennings, 2014), this
study proposes a research question about the relationship between the niche width of an entity (in
this case, design) and its fitness. Since this work relies on the assigned categories of a design to
determine its niche width, this research question reflects the cross-level relations between
designs and categories. On the one hand, audiences on crowdsourcing platforms may dismiss
designs that span multiple categories because these designs imperfectly fit the schema of each
label. As a result, broadening niche width comes at the expense of fitness in the competition for
best designs. On the other hand, not all market participants respond to the categorical imperative
in a similar way (Pontikes, 2012). A positive association between niche width and fitness is also
applicable to the crowdsourcing context. As audiences (online crowds) expect novel solutions to
sophisticated crowdsourced tasks (Majchrzak & Malhotra, 2014), occupying multiple categorical
niches allows recombining and integrating existing features into new ones. A submitted idea (in
our case, design) that targets more categories is likely to redefine or even challenge the
65
prevailing logic in crowdsourcing platforms, thus better meeting the complex demands of
audiences. As a result, the idea is likely to receive exceptional praise and attention from
audiences.
Research Question 1 (RQ
1
). What is the relationship between the categorical niche
width of a design and its fitness?
Ecological researchers have stressed the importance of category contrast in the process of
categorization (Hannan et al., 2007). The notion of contrast reflects the extent to which a
category stands out from the background (Hannan et al., 2007; Hannan, 2010; Negro et al.,
2015). In other words, high (or low) levels of contrast mean low (or high) levels of fuzziness
(Pontikes & Hannan, 2014) and cognitive ambiguity (Kovács & Hannan, 2010). Operationally,
category contrast is defined as the average GoM of all affiliated members in a focal category
(Hannan et al., 2007). The higher the average GoM of category members, the higher the degree
of category contrast will be, meaning that the affiliated members are not usually assigned to
other categories (Kovács & Hannan, 2010). As Kuilman and Wezel (2013) stated:
A category with a low degree of contrast has more members that only partly fit in, i.e.
they have a lower GoM. The boundaries that demarcate such a hypothetical category
from other organizational categories tend to be blurred. Conversely, a category with a
high degree of contrast contains a high proportion of organizations with a high grade of
membership, i.e., organizations that can be regarded as highly typical and have a high
degree of resemblance to the rest of the population. (pp. 58)

Previous studies have consistently demonstrated that category contrast produces a
positive effect on the valuation of a category member (Boone, Declerck, Rao, & Van den Buys,
2012; Kovács & Hannan, 2010, 2015; Pontikes, 2012). A decrease in category contrast indicates
that fewer affiliated members in the category “conform fully to the relevant schemata” (Hannan,
2010, p. 175). These category members generally suffer in terms of external evaluations. Due to
the growing difficulty of interpretation caused by reduced category contrast, observers can be
66
quite confused about the given set of meanings assigned to the category members (Hannan,
2010; Kovács & Hannan, 2010).
This study posits that category contrast plays an important role in design crowdsourcing.
In the present case, crowd members rely on socially constructed product categories to claim or
evaluate the identity of submitted designs. In a high-contrast product category, the affiliated
designs are primarily assigned only to that product category rather than to multiple categories.
As a decrease in category contrast leads to an increase in fuzziness (Pontikes & Hannan, 2014)
and cognitive ambiguity (Kovács & Hannan, 2010), a design’s membership in a low-contrast
product category poses a threat to the design’s overall appeal to audiences. Thus, it is expected
that:
Hypothesis 3 (H
3
). Increases in the category contrast of a design lead to increases in the
fitness of the design.
Prior research has also shown that the relationship between the categorical niche width of
a social entity and its fitness is contingent on the contrast of an entity’s affiliated categories. For
example, Kovács and Hannan (2010) reported that the negative impact of straddling is more
severe when the spanned categories have high contrast. Negro, Hannan, and Rao (2010) revealed
that the reduced contrast of spanned categories lowered the advantages of specialism and
increased the penalties of generalism. Kovács and Hannan (2010) argued that the contrast of
categories moderated the association between niche width and fitness because:
(1) high-contrast categories possess stronger codes, and spanning them results in stronger
code clashes, (2) when categories have low contrast, ambiguity is high to begin with, and
spanning ambiguous categories does not add much additional ambiguity, and (3) high-
contrast categories are more likely than others to entail a given set of capabilities. (pp.
180)

67
To put it simply, reduced contrast of categories leads to decreased salience of category
spanning, which makes it harder for audiences to identify niche broadness. Under such
circumstances, category spanning does not bring extra rewards or penalties. But when spanned
categories are high in contrast, audiences will have increased awareness of the offering and will
be confused about its meaning. Consequently, the rewards or penalties associated with category
spanning should be enhanced. As the relationship between the categorical niche width of a
design and its fitness remains underspecified, this dissertation proposes the following hypothesis:
Hypothesis 4 (H
4
). The relationship between categorical niche width and fitness is
moderated by category contrast, such that the intensity of the association is strengthened when
the level of category contrast is high.
Cross-Level Ties Between Designers and Designs
The second part of the dissertation focuses on the evolution of cross-level ties between
designers and designs. Specifically, this work investigates how commensalist ties among
designers change over time. The idea of commensalist ties builds upon the concept of
commensalism in ecological research (Monge & Poole, 2008; Weber, 2012). By definition,
commensalism is “co-action of like forms” (Hawley, 1950, p. 209), meaning that “units make
similar demands on the environment” (Aldrich & Reuf, 2006, p. 243). At the dyadic level, a
commensalist tie consists of two entities from the same population (Audia, Freeman, &
Reynolds, 2006; Weber, 2012). In most cases, commensalism leads to partial or full mutualism
(Aldrich & Reuf, 2006). In the form of partial mutualism, the commensalist tie is asymmetrical
because only one side benefits from the presence of the other. In the form of full mutualism, the
two sides are mutually beneficial. One type of commensalist ties on Stitchly’s digital platform is
the co-creation relations among designers, which can be constructed based on designers’
68
affiliations with designs (or cross-level ties between designers and designs). Design co-creation
can be viewed as an important strategy for designers to exchange knowledge and to buffer
against uncertainty.
While previous studies focus on the competitive nature of crowdsourcing platforms
(Boudreau, Lacetera, & Lakhani, 2013; Majchrzak & Malhotra, 2014; Zhao & Zhu, 2014), little
is known about how online crowds participate in competition and cooperation simultaneously in
tournament-based crowdsourcing. Hutter and colleagues (2011) argued that crowdsourcing
challenges allowed online crowds “both to competitively disclose their creative ideas to
corporations and also to interact and collaborate with like-minded peers” (p. 3). Their
conceptualization of ‘communitition’ acknowledges that crowd members can “engage in
competitive as well as co-operative behavior with the same individuals at the same time” (Hutter
et al., 2011, p. 13). By investigating the formation and dissolution mechanisms of commensalist
relations among designers in the context of Stitchly, this dissertation considers the co-existence
of cooperation and competition in crowdsourcing platforms. Drawing upon ecological and
evolutionary theories and the literature on expertise, the next few paragraphs will develop several
hypotheses about antecedents of tie formation and decay.
In organizational ecology, Stinchcombe (1965) coined the phrase “liability of newness”
to describe age dependence in the process of organizational death. He observed that new
organizations generally have greater failure rates than old ones do because the former lack
experience, knowledge, resources, managerial expertise, external legitimacy, and ties with
stakeholders (Stinchcombe, 1965). All these factors constrain the ability of new entrants to
become reliable and accountable in market environments (Hannan & Freeman, 1977, 1984). In
other words, emerging organizations were susceptible to the liability of newness and had a high
69
risk of failure. Although the original idea of “liability of newness” was put forward five decades
ago, it is still one of the most enduring topics in the literature on organizational transformation
and evolution (Larson & Linder, 2018; Li, Bruton, & Filatotchev, 2016), entrepreneurship
(Überbacher, 2014; Yang & Aldrich, 2017; Zhang & White, 2016), strategy (McDowell, Harris,
& Geho, 2016), and public policy (DeVaughn & Leary, 2018).
As the liability of newness is conceived as a general rule that spans multiple levels of
analysis in organizational settings and has been adapted to explain network evolution (Burt,
2000; Koka, Madhavam, & Prescott, 2006), this dissertation extends this idea to explain why
new entrants choose to initiate many commensalist ties to buffer against uncertainty and why
network ties formed by new entrants in design crowdsourcing are susceptible to failure. In design
crowdsourcing, the liability of newness refers to the possibility that inexperienced designers have
disproportionately low success rates. This disadvantage results from the challenge that new
entrants face in the learning process (DeVaughn & Leary, 2018). Specifically, new entrants do
not have enough time to absorb knowledge, develop routines, acquire expertise, and adapt to the
external environment. Even if prior experience in crowdsourcing-related activities may buffer
against some uncertainties in the new context, inexperienced designers still need to make some
effort to familiar themselves with the specific rules or routines of crowdsourcing platforms
through experiential learning.
Prior research has identified structural embeddedness as an effective strategy to mitigate
the liability of newness (Hager, Galaskiewicz, & Larson, 2004; Kim et al., 2006; Stitchcombe,
1965; Yang & Aldrich, 2017). In the context of design crowdsourcing, inexperienced designers
can take advantage of commensalist ties to accumulate experience and make knowledge and
skills transferrable to them. From an evolutionary perspective, new entrants focus more attention
70
on the stage of variation and thus experiment with a wide range of network ties. By embracing
variation in a trial-and-error fashion, inexperienced designers become open-minded, explorative,
and innovative (March, 1991). These benefits help them to overcome the liability of newness.
Altogether, inexperienced designers tend to establish many commensalist ties after initially
joining the crowdsourcing community. This behavioral tendency should apply mostly to
designers who joined Stitchly’s site at later time points during the observation period. For early
entrants to the platform, there were fewer available collaborators to choose from and good
examples to learn from. In other words, it was unlikely that experienced designers sought to
establish collaborative ties to minimize the liability of newness when they were inexperienced.
Then, it is predicted that:
Hypothesis 5 (H
5
). Inexperienced designers establish more commensalist ties than do
experienced designers.
While inexperienced designers have more incentives to establish more such cooperative
ties than experienced designers do to maximize the benefits, the likelihood of tie decay will be
higher for the former due to the initial learning curve disadvantage. Although commensalist ties
are particularly beneficial, inexperienced designers still have the difficulty of preserving and
maintaining such ties because it takes time to develop expertise, shape trust, build legitimacy,
and accumulate relationship-specific assets (Kim et al., 2006). These constraints lead to the
instability of established commensalist ties.
The carrying capacity of network ties also contributes to tie dissolution. Carrying
capacity indicates the limit in the number of network links that a social entity can maintain at any
particular time (Dunbar et al., 2015; Kleinbaum, 2018; Monge et al., 2008). Networks cannot
grow without any bound, so designers must select some of the existing ties, especially those that
71
best fit their needs, and eliminate the others through a recursive process of trial and error (Lovas
& Ghoshal, 2000; Feldman & Pentland, 2003). As inexperienced designers have no prior
experience in the crowdsourcing platform and are uncertain about the net benefits brought by
collaboration, experimenting with different network ties for planned or unplanned reasons
enables them to partially overcome the liability of newness. However, as carrying capacity poses
cognitive and economic constraints on these new organizational members, they also need to
consider removing some of the less useful first-trail ties.
Based on the above discussion, inexperienced designers should be characterized as
having high rates of both tie creation and tie decay. Thus, it is posited that commensalist ties
formed by not yet established designers are less likely to persist into the future than those created
by more established designers. This leads to the following hypothesis:
Hypothesis 6 (H
6
). Commensalist ties involving inexperienced designers are more likely
to decay than ties with no inexperience designers involved.
Expertise refers to mastery of knowledge in a specific domain (Ackerman, Pipek, &
Wulf, 2003). It has been regarded as a valuable asset owned by social entities (Dane, 2010).
Treem (2012) discussed three approaches to conceptualizing expertise. Psychologists generally
adopt a cognitive view of expertise and suggest that experts have specific competencies that are
developed through learning processes over time. Expertise is chiefly identified and recognized in
knowledge domains that involve stable, repeatable problem-solving situations and objective
judgement standards. The sociological approach highlights the importance of professionalism in
determining who is an expert. Specifically, established and institutionalized professional groups
“set the boundaries for what counts as expertise and allow others to distinguish among those who
legitimately lay claims to expert status” (Treem, 2012, p. 25). A communicative view of
72
expertise treats expertise as “an attribution that emerges through social interaction and is
communicated to others through the process of organizing” (Treem, 2012, p. 25). This
communication approach directs attention to expertise emergence and perception in knowledge-
intensive environments that involve invisible and ambiguous tasks.
This research focuses solely on expertise in a specific domain, design expertise. In the
present case, design expertise is treated as a competency developed and possessed by idea
contributors in the context of design crowdsourcing. Expert designers have superior ability to
abstract underlying principles from the examples that they have encountered. The complexity of
this process makes knowledge transfer a challenging task in design education (Dorst & Reymen,
2004). As a result, “learning by doing” is a gold standard in the field of design. Learning from
books is less effective than immersing oneself in the domain through continuous practice (Dorst
& Reymen, 2004; Collins & Evans, 2007). At first glance, this definition of design expertise is
consistent with the cognitive perspective on expertise, but design expertise has its own
uniqueness. For instance, prior research has shown that expertise in design industries is
significantly different from that of other fields, chiefly because the standards in judging good
designs are less articulated and uniform but tend to vary in different contexts (Lawson & Dorst,
2013). In other words, there are no objective and consensus standards of judgment that can be
used to evaluate design outputs. In a recently published article, Aldrich (2019) made a similar
observation:
There are some well-established and standardized procedures for acquiring the skills
needed to potentially do well as an angler or artist, but for any given performance,
whether it will be a “success” or “failure” is highly uncertain. They control their
performance but not how it is perceived by audiences or reviewed by critics. (pp. 4)

The idea of crowdsourcing is built upon the input of external expertise (Majchrzak &
Malhotra, 2014). Sponsoring organizations can generate competitive advantage from utilizing
73
domain expertise of crowd members (Di Pietro et al., 2018; Fisher, 2019). Expertise in
crowdsourcing processes can be understood as something that is dispersed, dynamic, and
communicative (Treem, 2012; Treem & Leonardi, 2016, 2017; van den Hooff & Kotlarsky,
2016). First, expertise is dispersed because online crowds come from diverse geographical
locations worldwide. Second, expertise is dynamic in the sense that what is required in an
ongoing crowdsourcing competition may evolve over time. Given the fluid or temporary nature
of crowdsourcing communities (Majchrzak & Malhortra, 2016; Riedl & Seidel, 2018), what is
valued by crowd members may be subject to rapid change. Third, communication is necessary
for the creation of expertise in crowdsourcing processes. As sponsoring companies do not
usually make judgment criteria publicly available to crowd members (Mukherjee et al., 2018;
Piezunka & Dahlander, 2015), it is ambiguous for online crowds to determine what counts as a
novel or creative idea. Consequently, idea submitters need to employ various communicative
acts such as the effective use of tags to shape the way in which idea evaluators perceive
expertise.
In the present case, design expertise refers to the extent to which a crowd member is
successful in submitting attractive designs on Stitchly’s site. This form of expertise is dispersed
in that those who have the required competencies, resources, and skills are not geographically
bounded. Design expertise is also dynamic because aesthetic taste among crowd members may
change rapidly, threatening the stability of expertise over time. Considering the ambiguity of
what constitutes a novel design resulting from a lack of clearly defined judgement standards,
idea submitters need to rely on various communicative acts to shape the crowd’s perception of
expertise. In this sense, design expertise is communicative.
74
This study emphasizes the importance of design expertise in shaping a designer’s
decision to initiate and break network ties with others. People do not establish their networks
indiscriminately. For example, prior research has shown that high-performance organizations
receive significantly more links from firms with varying levels of performance (Kim et al., 2006;
Podolny, 2001). One potential benefit of this linking strategy is that third-party observers may
infer that the tie initiator has good connections with high-performance organizations in the
market. The evaluations of the focal firm by observers become more positive due to the implicit
quality transfer (Stuart, Hoang, & Hybels, 1999). Establishing network ties with high-
performance organizations produces additional benefits for low-performance organizations
because it allows the latter to build competitive advantage through the inflow of resources,
knowledge, and expertise (Stuart et al., 1999). Thus, it is reasonable to expect that designers with
varying degrees of expertise prefer to collaborate with experts in crowdsourcing competitions.
However, design experts select collaborative ties carefully since they have more options in the
choice set (Podolny, 2001). As initiating ties with non-experts entails less opportunity to enhance
their own capability and competence and delivers a negative signal of the quality to potential
audiences, design experts would like to adhere to a principle of exclusivity (Podolny, 1994) to
choose only those with similar levels of design expertise as their collaborators. In other words, it
is less likely for designers with lower levels of expertise to succeed in establishing connections
with design experts. Even though a commensalist tie exists between two designers who have a
large expertise gap, this network link should have a high probability of dissolution. Given that an
individual’s decision-making is not purely driven by economic considerations and can be
influenced by normative factors, it is entirely possible that a co-creation tie exists between an
expert and a non-expert. However, as commensalist ties often allow two entities from the same
75
population to benefit from each other (Weber, 2012), unbalanced expertise between the two sides
may produce relational tension over time, which impairs the relationship itself and eventually
leads to tie decay. Altogether, it is hypothesized that:
Hypothesis 7a (H
7a
). Designers are more likely to collaborate with those whose design
expertise is similar to their own.
Hypothesis 7b (H
7b
). The greater the gap in design expertise between designers, the
more likely that their commensalist ties will decay.
As a general concept, inertia has received considerable scholarly attention in
organizational theory (Briscoe & Tsai, 2011). Hannan and Freeman (Freeman & Hannan, 1989;
Hannan & Freeman, 1977, 1984) were the first to claim that structures of organizations were
subject to strong inertia pressures. They used the term “structural inertia” to refer to “a
correspondence between the individual capabilities of a class of organizations and their
environments” (Hannan & Freeman, 1984, p. 151). The level of structural inertia is higher when
the speed of reorganization is lower than the rate of change in environments (Hannan &
Freeman, 1984). In general, older and larger organizations face stronger inertial forces (Hannan
& Freeman, 1984; Kim et al., 2006). From an evolutionary perspective, inertia is a by-product of
rather than a precondition for natural selection (Hannan & Freeman, 1984). As an indicator of
“good management”, structural inertia shows the establishment of well-tuned organizational
arrangements (Kim et al., 2006).
Drawing upon the structural inertia theory in organizational ecology, Kim, Oh, and
Swaminathan (2006) defined “network inertia” as “a persistent organizational resistance to
changing interorganizational network ties or difficulties that an organization faces when it
attempts to dissolve old relationships and form new network ties” (Kim et al., 2006, p. 704).
76
Inertial forces are produced by factors that span different levels of analysis, including
organizational characteristics, interorganizational dyadic ties, interorganizational network
positions, and the external environment (Kim et al., 2006). Although the theorization on
“network inertia” focuses on the evolution of interorganizational networks, Kim and colleagues
expect that the inertia-based perspective is also applicable to other forms of network ties. This
proposition has received considerable empirical support. For example, network scholars have
revealed the path-dependent nature of individual relationships (e.g., Briscoe & Tsai, 2011; Okoli
& Oh, 2007; Shen et al., 2014).
The conceptualization of “network inertia” parallels another line of research using the
term “relational inertia” or “relational locked-in” (Cheon et al., 2015; Gargiulo & Benassi, 2000;
Maurer & Ebers, 2006) to describe the resistance to network change. Unlike the network inertia
perspective, this stream of studies pays more attention to the conditions under which inertial
forces can be overcome (Maurer & Ebers, 2006; Srivastava, 2015) and the negative
consequences of a general tendency toward inertia for performance (Demirkian et al., 2013;
Okoli & Oh, 2013).
This dissertation extends the literature on “network inertia,” “relational inertia,” and
“relational locked-in” to examine whether network ties tend to persist over time in design
crowdsourcing. It is expected that an increase in the duration of a commensalist tie will lead to a
decrease in the likelihood of tie decay. In general, increased duration of a network tie indicates
the relative success of the tie in the past (Marsden & Campbell, 1984). Even though maintenance
costs may overwhelm potential benefits from a persistent tie, forming a new relationship to
replace the existing tie involves search, opportunity, and coordination costs (Cheon et al., 2015;
Gargiulo & Benassi, 2000). Network inertia then becomes a natural strategy to preserve a variety
77
of benefits from existing structural arrangements and to avoid uncertainty and risks (Briscoe &
Tsai, 2011; Walker, Kogut, & Shan, 1997). Apart from rational choice and cost-benefit analysis,
an increase in tie duration suggests an increase in attachment and commitment in the relationship
(Kim et al., 2006), making crowdsourcing designers subject to strong inertia. Altogether, it is
posited that designers are reluctant to change the existing co-creation ties with others if such ties
have already been maintained for a long time. This proposition can also be viewed as an
adaptation of the “liability of newness” hypothesis to the context of network evolution.
Hypothesis 8 (H
8
). The longer commensalist ties have been maintained between two
designers, the less likely the ties will decay.
Cross-Level Ties Between Designers and Categories
The third part of the dissertation focuses on cross-level ties between designers and
categories to shed light on market evolution in design crowdsourcing. Organizational ecologists
have long been interested in examining the ecological processes that drive organizational entry
and exit from emerging or existing populations (Aldrich & Reuf, 2006; Hannan & Freeman,
1977; Haveman, 1993). For instance, the main proposition of density dependence theory is that
population density is a proxy measure for legitimation and competition dynamics within a
population (Hannan & Freeman, 1989). The theory predicts an inverted U-shaped relationship
between population density and the organizational founding rate. In addition, the rate of
organizational exit should first decline and then rise with an increase in population density
(Hannan & Freeman, 1989).
The most recent line of research on ecologies of categories follows this theoretical
tradition and investigates why some categories draw entry and drive exit from social entities
(Carnabuci, Operti, & Kovács, 2015; Hannan et al., 2007; Hannan, 2010; Pontikes & Hannan,
78
2014; Pontikes & Barnett, 2015). Prior research has found that category contrast plays an
important role in the two fundamental events (Pontikes & Barnett, 2015). As described in the
previous sections, the contrast of a category is high when its members are assigned only to the
category rather than to multiple other categories (Kovács, & Hannan, 2010). Thus, boundaries of
low-contrast categories are fuzzy, meaning that the affiliated category members lack coherent
collective identities. Empirical evidence has consistently revealed that decreases in the average
contrast of entities’ assigned categories lead to decreases in the entities’ appeal to external
evaluators (Boone et al., 2012; Hannan, 2010; Hsu et al., 2009; Kovács & Hannan, 2010, 2015).
This negative impact can be explained by increased fuzziness (Pontikes & Hannan, 2014) and
cognitive ambiguity (Kovács & Hannan, 2010) perceived by audiences. Recently, Pontikes and
Barnett (2015) found that entering low-contrast categories generated benefits as well because this
strategy allowed social entities “to change over time and also to cultivate an identity that fits into
multiple frames” (Pontikes & Barnett, 2015, p. 1420). In the context of design crowdsourcing,
low-contrast product categories may attract entries from designers due to their flexible nature.
Specifically, low-contrast categories give designers room for conveying their unique identities
and values in crowdsourcing competitions and can be viewed as opportunities to reshape the
prevailing logics and tastes in crowdsourcing communities. Thus, it is expected that:
Hypothesis 9a (H
9a
). Designers have a higher propensity of entering a low-contrast
product category than a high-contrast product category.
Although designers are likely to use low-contrast product categories to label their
submissions, not every one of them will achieve success in creativity performance because such
categories “do not evoke commonly agreed-upon expectations and so provide weak signals”
(Pontikes & Barnett, 2015, p. 1420). Altogether, the flexibility of low-contrast categories is a
79
two-edged sword. Despite the benefit of keeping options open, designers face the difficulty of
interpreting the meaning and attracting the right types of audiences. This implies the following
hypothesis:
Hypothesis 9b (H
9b
). Designers have a higher propensity of exiting a low-contrast
product category than a high-contrast product category.
Decision-making processes often involve “trade-offs between exploiting known
opportunities and exploring for better opportunities elsewhere” (Hills et al., 2015, p. 46). This
tension is generally labeled as exploitation-versus-exploration (March, 1991), which is akin to
“the problem of deciding whether the present should be hedged for the future” (Lavie, Stettner,
& Tushman, 2010, p. 116). As a multilevel concept, exploitation-versus-exploration operates at
the individual, group, organization, and community levels (Durcikova, Fadel, Butler, & Galletta,
2011; Lavie et al., 2011). By tracking the evolution of category entry and exit by designers, this
work treats exploitation and exploration as decision-making processes at the individual level.
This conceptualization is consistent with the existing literature. On the one hand, exploitation is
defined as “the refinement and extension of existing competences, technologies, and paradigms”
(March, 1991, p. 85). On the other hand, exploration refers to experimenting with new
alternatives or making radical changes (March, 1991). Individuals may suffer from the costs of
exploration. As the search for novel alternatives is partly random and departs from an existing
direction, expected returns are uncertain or even negative.
In this case, entering an unexplored product category or exiting an explored product
category reflects a high level of exploration in crowdsourcing competitions, whereas staying in
explored categories indicates a tendency toward exploitation. It is expected that an increase in
design expertise will lead to a decrease in the likelihood that the focal designer engages in
80
explorative activities or makes categorical changes. There are three streams of studies that can
support this prediction.
First, drawing insights from learning theory, the performance feedback model (Greve,
1998) assumes that past success and failure guides behavioral evolution. The theoretical model
proposes that unsatisfactory results contribute to a shift from exploiting existing knowledge to
exploring new possibilities. Meanwhile, positive feedback activates a strong bias toward
exploitation and reduces the likelihood of risky changes (Lavie et al., 2010; Greve, 2008). This
tendency for exploitation is stronger when performance feedback is much higher than aspiration
levels (Greve, 1998, 2007). These theoretical propositions have received empirical support in
various research settings (Bayus, 2013; Audia & Goncalo, 2007). For example, in a study of the
Dell IdeaStorm community, Bayus (2013) found that successful ideators tended to submit ideas
that are similar to past ideas in subsequent work. This finding suggests that in crowdsourcing
contexts successful designers may engage in a type of learning that results in replications of past
ideas, thus hindering novelty and innovativeness in problem-solving. Similarly, Audia and
Goncalo (2007) found that among patent inventors, those who achieved success were more likely
to continue to produce more patents; however, subsequent patents were less divergent in nature.
Second, cognitive entrenchment theory (Dane, 2010) provides a more direct
explanation for why expertise determines exploratory and exploitative activities in
generating creative ideas. Cognitive entrenchment refers to “a high level of stability in
one’s domain schemas” (Dane, 2010, p. 579) and arises largely from domain expertise.
As expertise is acquired through continual practice, “the content and relations comprising
an expert’s domain schemas are likely to be activated and applied innumerable times”
(Dane, 2010, p. 583). Consequently, the acquisition of expertise results in cognitive
81
entrenchment. This theory may help to clarify the mechanism through which expertise
puts individuals into a success trap that results in a bias against exploration (Levinthal &
March, 1993; Rhee & Kim, 2015). The inflexibility-related limitations of expertise are
applicable to crowdsourcing processes in that contributors also perform tasks in dynamic
and competitive environments, attain expertise through continuous performance feedback
from the environment, and make strategic decisions in real time. Cognitive entrenchment
is expected to be more pronounced in design crowdsourcing because “a blind adherence
to a set of ideas or concepts” (Jansson & Smith, 1991, p. 3) has been widely observed in
design processes (Sio, Kotovsky, & Cagan, 2015).
Third, research on ecologies of categories views entry and exit from product categories as
a form of evolutionary change (Carnabuci et al., 2015). Design experts may suffer from stronger
inertial forces because they face higher levels of pressure to signal consistent reliability and
accountability (Hannan & Freeman, 1984; Kim et al., 2006). Exploiting the existing affiliation
ties with product categories not only enables design experts to avoid uncertainty and switching
costs, but also contributes to the maintenance of their advantageous status on the Stitchly site. As
a result, they are subject to strong inertia forces and thus behave conservatively in category entry
and exit. Specifically, design experts are expected to continue to stay in product categories in
which they have been successful and are thus less likely to enter unexplored categories.
Altogether, it is hypothesized that the overall tendency for design experts to enter a new category
is low and that an existing tie between a design expert and a product category is persistent over
time.
Hypothesis 10a (H
10a
). Designers with high levels of expertise have a lower propensity
of entering a product category than do those with low levels of expertise.
82
Hypothesis 10b (H
10b
). Designers with high levels of expertise have a lower propensity
of exiting a product category than do those with low levels of expertise.
Although designers tend to enter low-contrast categories (stated in Hypothesis 9b), high-
contrast categories draw significant attention from inexperienced designers for two reasons
(Carnabuci et al., 2015): First, without any preexisting identity, new entrants are structurally
flexible to adapt themselves to the idiosyncratic features of the high-contrast categories (Sosa,
2013). Second, entering high-contrast categories helps new organizations to mitigate the liability
of newness. As high-contrast categories are less ambiguous, it takes less time for the new
entrants to develop skills and competitiveness (Carnabuci et al., 2015). Accordingly, this study
proposes that the underlying theoretical mechanism through which high-contrast categories are
attractive to new entrants is also applicable to design crowdsourcing. The above two reasons
suggest that entering product categories with clear boundaries may help inexperienced designers
to alleviate their weakness.
Designers with high levels of expertise are expected to pay considerable attention to low-
contrast categories. Design experts are successful on Stitchly’s site because their uniqueness and
creativity have been well recognized by other crowd members. Thus, these designers should not
fit neatly into the existing categories. The ambiguity of low-contrast categories is extremely
suitable for those successful designers because it allows them to keep design activities flexible
and open so that their competitive advantage can be further sustained (Pontikes & Barnett, 2015;
Pontikes, 2012). Altogether, it is hypothesized that:
Hypothesis 11a (H
11a
). Inexperienced designers have a lower propensity of entering a
low-contrast product category than do experienced designers.
83
Hypothesis 11b (H
11b
). Designers with high levels of expertise have a higher propensity
of entering a low-contrast product category than do those with low levels of expertise.
Within-Level Ties Among Designers
The fourth part of the dissertation focuses on the longitudinal change of follower-
followee ties among designers. For a focal designer, those who follow him or her are identified
as followers; those whom the designer follows are his or her followees. While there are many
types of communication networks in online communities (e.g., mention, retweet, reply, and
comment networks on Twitter, Huang & Sun, 2014; Lai, She, & Tao, 2017; Peng, Liu, Wu, &
Liu, 2015), this dissertation focuses on the follower-followee network because it is the only
communication network articulated by designers through technical features in the research
setting. Communication networks are defined as “relations among various types of nodes that
illustrate the ways in which messages are transmitted or interpreted” (Shumate et al., 2013, p.
97). The formation of follower-followee networks requires information transmission or exchange
among nodes. For instance, when people follow others on Twitter, the followees will receive a
message that they have a new follower. Then, these followees can decide whether to follow back.
Follower-followee ties also build a communication infrastructure that enables further
communicative acts among Twitter users. For instance, followers can comment on followees’
newly posted tweets. If followers believe that tweet content has important implications for the
larger community, they will probably engage in retweet behavior. Given the prevalence of
follower-followee relations in online space, investigating what drives some ties to be more stable
or transient than others should be of deep interest to many scholars. In the context of Stitchly’s
site, designers who receive new follower-followee links will be notified that others have
followed them. After this communication infrastructure is established, followers will get email
84
notifications every time a specific followee submits a new design for crowd evaluation or
produces a printed design for sale. Followers can further engage in other communicative acts,
including commenting on the submitted design and exchanging ideas with the followee.
Researchers have already paid considerable attention to the phenomenon of tie formation
and have identified various endogenous and exogenous factors that drive the creation of
follower-followee networks (Huang & Sun, 2014; Peng et al., 2014). However, what is less well
studied is the other important side of network evolution: the decay of existing follower-followee
ties. By examining how unfollowing behavior is shaped by designer-level attributes in design
crowdsourcing, this study joins the recent efforts in analyzing the decay of follower-followee ties
in online communities (e.g., Kwak, Moon, & Lee, 2012; Liang & Fu, 2016; Xu, Huang, Kwak,
& Contractor, 2013). Unfollowing behavior occurs when a crowdsourcing participant followed
another one at time t but stopped following at time t + 1 (Xu et al., 2013).
From an economic perspective, social entities do not maintain their network ties
indiscriminately. They make a cost benefit analysis when deciding to retain or drop a tie (Cheon
et al., 2015; Dermirkan et al., 2013). For example, a focal person may consider dropping a tie
with another one if maintenance costs are much larger than potential benefits (Kim et al., 2006).
In the context of Stitchly, persistence of networks should be contingent on the design expertise of
others in a designer’s personal network. Maintaining follower-followee relations with designers
with lower levels of expertise generates less utility. Following others’ profiles allows designers
to receive updates and notifications so that they can access followees’ newly submitted designs
in a timely fashion. As observing good examples contributes to a designer’s creativity
performance in crowdsourcing competitions (Riedl & Seidel, 2018), it is unlikely for the
85
designer to get clues about the path to success by continuing subscribing to non-experts’ profiles.
Thus, the following hypothesis was formulated:
Hypothesis 12 (H
12
). Designers are more likely to unfollow others who have lower levels
of design expertise than those who have higher levels of design expertise.
Network closure describes the degree to which individuals are embedded in structures
where their contacts are already connected with each other (Burt, 2005). For example, there are
two hypothetical networks X and Y composed of persons A, B, and C. In network X, person A
serves as an intermediary between persons B and C who are not directly connected to each other;
that is, there is a “structural hole” (Burt, 1992) between B and C that A fills by channeling
information and resources between them. By contrast, Network Y is fully connected, meaning
that there is a network tie between any pair of nodes. In this hypothetical scenario, network X
has a lower level of closure than network Y does.
Social scientists propose that network closure contributes to the accumulation of bonding
or internal social capital (Alder & Kwon, 2002) by facilitating norm emergence, imposing
external effects, offering collective sanctions, and promoting social trust at the intra-group level
(Coleman, 1988). Network scholars have extended this idea to investigate how network closure
brings social benefits for individual participants in online communities. For example, Shen,
Monge, and Williams (2014) found that network closure was positively associated with mutual
trust among players in Massively Multiplayer Online games (MMOs). Meng, Chung, and Cox
(2016) confirmed that the degree of closure positively predicted the amount of emotional and
esteem support in an online health community. Stephen, Zubcsek, and Goldenberg (2016)
focused specifically on the impact of network closure in the context of crowdsourcing. Their
research showed that high clustering produced a negative effect on the innovativeness of a
86
contributor’s idea in the sense that “idea inspirations are more likely to be similar or redundant
when their sources (i.e., other customers to which one is connected) are clustered” (Stephen et
al., 2016, p. 263).
While previous studies have provided important insights into various behavioral
consequences of network closure, it remains unclear how the degree of closure constrains the
decision to drop an existing tie. Prior research on network or relational inertia suggests that
closure may impede an individual’s ability to overcome the resistance to network change
(Briscoe & Tsai, 2011; Kim et al., 2006; Maurer & Ebers, 2006). A high degree of closure
generates a variety of benefits, but it also poses limitations on individuals by reducing network
flexibility and autonomy. By contrast, when individuals are embedded in low-closure structures
where there are many structural holes (Burt, 1992), they feel less obligated to commit themselves
to existing relationships. In addition, shared expectations, norms, and values that are developed
in high-closure networks make individuals more aware of the negative consequences of tie decay
in a cohesive group (e.g., collective sanctions). In the context of Stitchly, follower-followee
relations constitute a primary source of information exchanges among individual designers.
Considering the normative constraints described above, designers who have high-closure
personal networks are less likely to engage in unfollowing behavior. This leads to the following
hypothesis:
Hypothesis 13 (H
13
). Designers who are members of high-closure groups are less likely
to unfollow others than designers who belong to low-closure groups.
The tendency for a focal designer to unfollow others who have low levels of expertise
should be lower when the designer is embedded in a high-closure personal network. In this case,
network closure exerts normative influences on tie decay choices, such that designers need to
87
consider exit costs of relationships in a cohesive structure. As a decrease in network closure
leads to an increase in flexibility and autonomy, designers have fewer normative constraints in
deleting the existing ties that poorly fit the environment (Cheon et al., 2015; Kim et al., 2006).
Consequently, designers who possess lower levels of network closure are driven more by cost-
benefit considerations and are more likely to break follower-followee relations with non-experts.
Thus, it is hypothesized:
Hypothesis 14 (H
14
). The negative relationship between design expertise and
unfollowing behavior is moderated by network closure, such that the negative association is
stronger when the focal designer belongs to a low-closure group.
The hypotheses and research question are summarized in Table 2.
Table 2
Summary of Hypotheses and Research Question
Cross-level Ties Between Designs and Categories
H
1
Increases in the fuzzy density of a category lead to increases in the rate of entry into
the category by designs.
H
2
Increases in the contrast of a category lead to increases in the rate of entry into the
category by designs.
RQ
1
What is the relationship between the categorical niche width of a design and its
fitness?
H
3
Increases in the category contrast of a design lead to increases in the fitness of the
design.
H
4
The relationship between categorical niche width and fitness is moderated by
category contrast, such that the intensity of the association is strengthened when the
level of category contrast is high.
Cross-level Ties Between Designers and Designs
H
5
Inexperienced designers establish more commensalist ties than do experienced
designers.
H
6
Commensalist ties involving inexperienced designers are more likely to decay than
ties with no inexperience designers involved.
H
7a
Designers are more likely to collaborate with those whose design expertise is
similar to their own.
H
7b
The greater the gap in design expertise between designers, the more likely that their
commensalist ties will decay.
H
8
The longer commensalist ties have been maintained between two designers, the less
likely the ties will decay.

88
Cross-level Ties Between Designers and Categories
H
9a
Designers have a higher propensity of entering a low-contrast product category than
a high-contrast product category.
H
9b
Designers have a higher propensity of exiting a low-contrast product category than
a high-contrast product category.
H
10a
Designers with high levels of expertise have a lower propensity of entering a
product category than do those with low levels of expertise.
H
10b
Designers with high levels of expertise have a lower propensity of exiting a product
category than do those with low levels of expertise.
H
11a
Inexperienced designers have a lower propensity of entering a low-contrast product
category than do experienced designers.
H
11b
Designers with high levels of expertise have a higher propensity of entering a low-
contrast product category than do those with low levels of expertise.
Within-level Ties Among designers
H
12
Designers are more likely to unfollow others who have lower levels of design
expertise.
H
13
Designers who are members of high-closure groups are less likely to unfollow
others than designers who belong to low-closure groups.
H
14
The negative relationship between design expertise and unfollowing behavior is
moderated by network closure, such that the negative association is stronger when
the focal designer belongs to a low-closure group.

89
Chapter 4: Method
Research Setting
The research setting of this study is Stitchly, a clothing company that solicits ideas from
crowd members beyond the organizational boundary for new product development. The major
task of Stitchly is to collect attractive graphic designs that can be printed on clothes and other
goods for profit. In response to this challenging task, the company adopts the model of design
crowdsourcing and outsources the ideation process to online crowds through an ongoing
tournament-based competition. This dissertation utilizes online observational data collected from
Stitchly’s website for hypothesis testing.
Sampling and Data Collection
Two different samples are selected for data analysis. Sample 1 was obtained through the
use of the public search engine on the Stitchly digital platform to retrieve all creative designs
submitted to the company. Since the number of designers is huge and only a limited number of
them have printed designs, sample 2 focuses specifically on a small group of successful
designers and their submitted designs.
Sample 1
Designers and designs. The search engine on Stitchly’s site was used to extract all
archived designs in its history and to identify their authors (designers). Python scripts were run
on August 21, 2018 to scrape all 439,348 designs created by 118,766 designers. The first design
on Stitchly’s site was submitted on July 21, 2001.
Categories. Product categories were constructed by the Stitchly company to classify
printed designs to satisfy the needs of consumers. Examination of the Wayback Machine
provided by the Internet Archive (https://archive.org/web/) showed that 32 categories
90
(“Abstract,” “Animals,” “Bikes,” “Comics,” “Creativity,” “Cute,” “Fantasy,” “Film,” “Food,”
“Funny,” “Gaming,” “Halloween,” “Historical,” “Ironic,” “Literacy,” “Luv,” “Monsters,”
“Music,” “Nature,” “Nerdy,” “Patterns,” “Political,” “Pop-culture,” “Religion,” “Robots,”
“Slogans,” “Space,” “Sports,” “Typography,” “Violent,” “WTF,” and “Zombies” ) emerged in
November 2012 and have been stable and unchanged since then.
Product categories impose an exogenous structure on designers. The company requires
designers to assign at most 10 tags to categorize their designs before moving to the voting stage.
At least one tag must be an existing product category. A design without any product category
tags cannot be submitted for competition. To extract the assigned categories of submitted
designs, a Python script was first run to scrape all self-claimed tags in all designs. A dictionary-
based approach in automated text mining was then employed to count how often each of the 32
product categories occurred in each design. A matrix was established to reflect the degree to
which a design was affiliated with the existing 32 categories. Each row represents a single
design, and each column indicates one category. For instance, suppose a design is composed of
four unique tags and one of them belongs to the category “Gaming”. Then, the cell between the
design and the category is automatically coded as “1”. The remaining 31 cells for this design are
recorded as “0” because the design is assigned only to one category. This automated analysis was
performed by the quanteda package in R (Welbers, Van Atteveldt, & Benoit, 2017).
Sample 2
Designers. The second sample focuses on successful designers, those who have
generated at least one printed design after crowd judgment on Stitchly’s site. At the end of
March 2017, Python scripts were run to scrape the author information of each printed product to
retrieve the list of successful designers. Stitchly assigns a unique ID to each product, thus
91
enabling researchers to track who developed the idea. There is also a symbol of “alumni”
displayed beside the username if a focal person has produced at least one design that was
selected by the company after the voting process. Following this procedure, 1,861 successful
designers were identified.
Designs. Python programming techniques were employed to track the submission history
of successful designers. All the designs they had submitted since the emergence of the website
were also recorded.
Categories. Following the same procedure in sample 1, this research employed
automated text mining to assign the designs in sample 2 to the existing 32 categories. The
analysis was also conducted by the quanteda package in R (Welbers et al., 2017).
Network Construction
Cross-Level Ties Between Designs and Categories
As designs are labeled with different categories, an affiliation matrix is created to reflect
the cross-level ties between designs and categories in sample 1. A binary value was assigned to
each cell. A design-category pair is coded as “1” if a focal design is affiliated with a specific
category. This matrix is used for computing measures for key explanatory variables.
Cross-Level Ties Between Designers and Designs
Cross-level ties between designers and designs refer to commensalist ties among 1,861
successful designers in sample 2. This research views design co-creation as an important form of
commensalist ties on Stitchly’s site. The Stitchly company allows for a design to have at most
two authors. But co-creation does not occur very often. Among 439,348 submitted designs over
the 18 years, only 2,917 of them (0.66%) had more than one author. There was no co-creation
design prior to 2012. This research focuses on co-created designs that were produced by pairs of
92
successful designers, which accounted for 45.35% (or 1,323 out of 2,917) of all cases. The year
of submission ranged from 2012 to 2017.
The presence or absence of a cross-level network tie is reflected in a two-mode matrix
consisting of both designers and designs. The network was constructed on an annual basis and
was then transformed into several one-mode matrices, where the number in each cell indicates
the number of co-created designs between any pair of designers in a specific year between 2012
and 2017. As design co-creation occurs rarely on Stitchly’s site, constructing variables on a
monthly or weekly basis would have resulted in too many zeros in the network matrices, making
it empirically difficult to run data analysis.
Cross-Level Ties Between Designers and Categories
Cross-level ties between 1,861 successful designers and 32 categories (designer-category
pairs) in sample 2 were created on an annual basis. The one-year interval is artificially designed,
given that no prior research has identified a reasonable observation period for network evolution
on crowdsourcing platforms. A designer-category pair is coded as “1” if a focal designer
assigned a specific category to his design(s) in a given year. As it was not until October 2013
that designers were required to choose at least one of these existing categories to label their
designs before submission, this research established five designer-category affiliation matrices
running from 2013 through 2017. Using one month or one week as the time interval is not
desirable. The necessary condition for investigating the evolution of cross-level ties was the
presence of nodes during the observation period. However, in the present case, no single
designer produced at least one design every week or month, giving rise to the issue of interrupted
observation.
93
To examine the antecedents of tie formation and decay during the observation period, it
was important to ensure that designers were available for category entry or exit throughout the
five consecutive years. In other words, they should submit at least one design in any year.
Finally, 155 out of 1,861 successful designers in sample 2 met this criterion. From 2013 to 2017,
the chosen designers created 14,645 designs that contained self-claimed categories.
Within-Level Ties Among Designers
Within-level ties among designers refer to follow-followee relations among 1,861
successful designers in sample 2. Social media have become an essential component of Stitchly’s
site. The company has launched a public social media platform that allows designers to establish
a directed communication network by clicking on the Follow button displayed on another’s
profile. In fact, this is the only possible communication network articulated by crowd members
through technical features on the Stitchly digital platform. The creation of follower-followee ties
on Stitchly’s site involves information transmission. When designers follow others, the followees
will receive a message that they have a new follower. Following behavior enables the subscriber
to receive a notification each time the followee submits a new design for crowd evaluation or
produces a printed design for sale. More important, the follower-followee network constitutes a
communication infrastructure that enables further collective action among designers, including
commenting on others’ submitted ideas. Like Twitter, a follower-followee tie is not necessarily
reciprocal, as the link receiver can choose not to follow the link sender.
The first wave of the follower-followee network among 1,861 successful designers was
collected in March 2017 and the second wave of network data was gathered in October 2018.
Since this study is interested in unfollowing behavior, inactive designers between March 2017
and October 2018 were excluded as senders of follower-followee ties. Inactive designers were
94
those who did not engage in any activity (initiate any thread, submit any design, or score any
design) on the Stitchly digital platform at any time point between Wave 1 and Wave 2. Tie
receivers include all 1,861 successful designers in sample 2. Following these procedures, the
current research identified 494 active designers who established 42,899 and 44,836 ties in Wave
1 and Wave 2, respectively. Seventeen active designers were isolates in Wave 1, meaning that
they were not connected to anyone in the follower-followee network. Since this research is
interested in explaining the dissolution of within-level ties among social entities, the empirical
analysis focuses on the remaining 477 designers who were at the risk of tie decay in Wave 2.
Measures
Dependent Variables
The rate of entry into a category (H
1
and H
2
). Following prior research in organizational
ecology and ecology of categories (Aldrich & Reuf, 2006; Kuilman & Li, 2009), the rate of entry
into a category is operationalized as the total number of designs that belong to a specific category
in a given month between November 2013 to July 2018. The one month interval is artificially
designed, considering that no prior work has identified an appropriate time interval when
analyzing density dependence process in crowdsourcing platforms. Constructing this dependent
variable on a yearly basis is also viable but reduces the number of observations over time in
panel data analysis from 57 to four, which sacrifices the richness of the dataset. In addition,
unlike other types of cross-level ties on Stitchly’s site, using one month as the time interval does
not result in too many zeros in the network matrices and does not generate the issue of
interrupted observation (see the section of Network Construction for details).
The fitness of a design (RQ
1
, H
3
, and H
4
). This variable is measured as the average score
of a design. Mathematically, it equals the mean score that a design receives from online crowds
95
on Stitchly’s site.
The presence of a commensalist tie (tie formation) (H
5
). This is a binary variable where
“1” indicates the presence of a co-creation relation between two designers. It is coded as “0”
when a tie between the two designers is absent.
The decay of a commensalist tie (tie decay) (H
6
, H
7
, and H
8
). This is a binary variable in
which “1” represents that an existing tie between two designers decayed in a specific year.
Alternatively, the variable is coded as “0” if the tie is persistent in the present year. Treating the
decay of a tie as “1” is necessary because the dependent variable in event history analysis (EHA)
is the occurrence of an event (Burt, 2000; Miller et al., 2011).
Entry into a product category (H
9a
, H
10a
, and H
11
). This is a binary variable where “1”
indicates that a focal designer enters the category in a given year and “0” indicates that he or she
does not. Entry into a specific product category can occur multiple times. The variable is coded
as “1” when designers enter a category for the first time or when designers re-enter the category
after exit in the prior year(s).
Exit from a product category (H
9b
and H
10b
). This is a binary variable that equals “1”
when a focal designer exits a product category in the year and “0” when he or she does not. The
variable is coded as “1” only if the designer-category pair exists in the prior year. Designers also
can leave a product category more than once. As it is impossible for a designer to break an
affiliation tie with a category in the first year during the observation period, this outcome
variable is observed between 2014 and 2017 only.
Unfollowing behavior (H
12
, H
13
, and H
14
). This variable is measured by observing
whether designers’ existing follower-followee ties in Wave 1 (March 2017) were removed in
Wave 2 (October 2018). Two hundred and eleven follower-followee ties initiated by 48
96
successful designers decayed during the two waves. Unlike other digital trace data retrieved from
the Stitchly site, there is no time stamp for follower-followee relations. In other words, the exact
time when a tie is initiated or dropped remains unknown. This explains why this research
observes the change in unfollowing behavior in discrete time only.
Independent Variables
The fuzzy density of a category (H
1
). This variable is measured as the sum of the GoM of
category members (Hannan et al., 2007). A design’s GoM in a category equals 0.00 if the design
is not affiliated (tagged) with the category. Otherwise, the score equals one divided by the
number of assigned product categories in the design (Kovács & Hannan, 2010). Suppose that
there are four distinct categories (Categories A, B, C, and D) and two designs (Designs x and y)
at time t. Design x is affiliated with and tagged by Categories A, B, and C. Design y is affiliated
with and tagged by Categories B and D. Design x is assigned a GoM of one-third (1.00/3.00) in
each of the three categories. Similarly, Design y contributes 0.50 (1.00/2.00) to Category B and
Category D, respectively.
Then, the fuzzy density of Categories A and C equals 0.33 (1.00/3.00) because both
categories have only one member and the member (Design x) belongs to three categories in all.
As Category B has two affiliated designs (Designs x and y), its fuzzy density equals 0.83
(1.00/3.00+1.00/2.00). The fuzzy density score of Category D is 0.50 (1.00/2.00) because the
only category member (Design y) is assigned to two categories.
The contrast of a category (H
2
, H
9
, and H
11
). Contrast is the average GoM of category
members (Hannan et al., 2007). On the Stitchly digital platform, a category has a high level of
contrast when it is rarely used together with others to describe a design. Mathematically, contrast
equals fuzzy density divided by population density (Bogaert et al., 2010; Kuilman & Li, 2009).
97
The way to compute fuzzy density scores was described in the last paragraph. The population
density of a category is operationalized as the total number of affiliated members in the category.
The population density of Category B equals 2.00 because two designs use the category. The
population density score of Categories A, C, and D equals 1.00 since only one design is labeled
with each of the three categories. After dividing fuzzy density by population density, the last
column of Table 3 reports the contrast scores of all categories.
Table 3
A Category’s Fuzzy Density, Population Density, and Contrast

Design x’s
GoM
Design y’s
GoM
Fuzzy density
Population
Density
Contrast
Category A 0.33 0.00 0.33 1.00 0.33
Category B 0.33 0.50 0.83 2.00 0.42
Category C 0.33 0.00 0.33 1.00 0.33
Category D 0.00 0.50 0.50 1.00 0.50

For H
1
and H
2
, this work follows prior ecological research (Aldrich & Reuf, 2006) and
lags the fuzzy density and the contrast of a category by one month in the predicting model. The
rationale is that when social entities decide to enter a population at time t (in the present case, a
given month), they are constrained by ecological variables at the previous time point, t - 1. As
the dependent variable is constructed on an annual instead of monthly basis for H
9
and H
11
,
category contrast is lagged by one year in model estimation. All the above variables are
computed based on 136,287 designs submitted between October 2013 and July 2018 in sample 1.
The categorical niche width of a design (RQ
1
). Ecological research generally treats the
number of distinct categories under which an entity is classified as an indicator of niche width
(Hsu, 2006; Kovács & Hannan, 2015). Thus, the present research measures the categorical niche
width of a design as the number of assigned product categories in the design.
The category contrast of a design (H
3
and H
4
). This variable is measured as the average
contrast of the assigned categories in a design. The contrast score of a category is constructed on
98
a monthly basis and is directly retrieved from H
2
.
Experience (H
5
, H
6
, and H
11a
). Since the outcome variables in H
5
and H
11a
are constructed
on a yearly basis, the experience of a focal designer is measured as years since registration on the
Stitchly digital platform. Inexperienced designers are characterized as being registered into
Stitchly’s site quite recently. For H
6
, this variable is the lower value of years since registration
between two successful designers in a dyad. If both designers have the same length of years
since registration, then the number is assigned to the independent variable. Although it is entirely
possible that two designers who started to use Stitchly’s digital platform in the same year
registered in different months, this information is not publicly available on the website.
Design expertise (H
10
, H
11b
, and H
12
). This variable is operationalized as the ratio of the
number of printed designs to the total number of submitted designs. As competition on Stitchly’s
site is extremely fierce and only a small percentage of submitted designs can be printed by the
company, design expertise can be reflected by designers’ competence in creating designs that are
selected for sale to the public, that is, printed designs. The measure of design expertise also
considers the total number of submitted designs because of the assumption that two designers
who generate the same number of printed designs should not have an equal level of design
expertise if one submits more designs than the other. Gap in design expertise (H
7a
) is measured
at the dyadic level. It equals the absolute difference in the ratio of the number of printed designs
to the total number of submitted designs between the two nodes in each dyad.
The duration of a commensalist tie (tie duration) (H
8
). The panel data about
commensalist ties span from 2012 to 2017. The duration of a tie is computed as the total number
of consecutive years for which a design co-creation relation has been maintained between any
pair of successful designers. For example, if a tie existed from 2012 to 2014, the tie duration was
99
three years.
Network closure (H
13
and H
14
). This study uses the network constraint score to measure
network closure. Network constraint describes the extent to which a person invests relations in
those who are also connected by his or her network “neighbors.” This measure is calculated by
formula (1) (Burt, 1992, p. 55):
(1)
In this formula, p
ij
is the proportion of a focal person i’s investment in relationship with
others in their network, labeled j. p
iq(qj)
is the tie strength from person i (q) to person q (j). The
higher the constraint score is, the higher the degree of closure will be. Network closure of each
designer in the sample was computed through UCINET 6 (Borgatti, Everett, & Freeman, 2002).
Control Variables
For hypotheses 1 and 2, the two control variables are (a) the total number of submitted
designs and (b) the percentage of designs submitted to the general challenge. Like the dependent
and independent variables, they are also constructed on a monthly basis. The total number of
submitted designs reflects competition intensity on Stitchly’s site in a given time period. As for
the second control variable, the Stitchly company invites crowd members to submit designs to
either the general challenge or themed challenges. While the former is open-ended, the latter are
announced by the company at some time and each challenge lasts for only several weeks.
Controlling the percentage of designs submitted to the general challenge is necessary because a
theme challenge may draw disproportionately more or fewer entries into a specific category.
For research question 1 and hypotheses 3 and 4, the current research includes (a) type of
challenge, (b) number of tags, and (c) time of submission as three design-level control variables.
Type of challenge is coded as “1” if a focal design is submitted to the general challenge. A
100
submission to the theme challenge is coded as “0.” This variable is used to control the alternative
explanation that the fitness of a design is determined by the type of challenge to which the design
is submitted. Number of tags is a count variable ranging from 1 to 10 because the company
requires designers to assign at most 10 tags before final submission. Controlling this variable is
important because the relationship between categorical niche width and fitness may be
confounded by the total number of tags to which a design belongs. Time of submission is
measured at the month level. As there are 57 months during the observation period, time of
submission is transformed into a series of dummy variables in model estimation. The first month,
November 2013, is set as the reference group. Including time dummies enables this research to
rule out the alternative explanation that submitted designs have significantly better or worse
fitness during a specific period of time.
For hypotheses 5 and 7a, this study controls the effects of individual attributes at the
nodal level (number of printed designs and the absolute value of design expertise) and
homophily variables at the dyadic level (gap in the number of printed designs and gap in
experience) on tie formation. Controlling the two designer-level attributes helps to rule out the
alternative explanation that increases in (a) the number of printed designs or (b) the degree of
design expertise at the nodal level lead to increases (or decreases) in the likelihood of tie
formation. The two homophily variables are computed as the absolute differences in the number
of printed designs and experience on Stitchly’s site between any pair of nodes. These variables
are used to rule out the possibility that designers are more (or less) likely to collaborate with
those whose (a) number of printed designs or (b) prior experience is similar to their own.
For hypotheses 6, 7b, and 8, this research controls for (a) year dummies, (b) the network
size of both the first and the second authors of a design, (c) gap in the number of printed designs,
101
and (d) gap in experience. Year dummies are included to allow the probability of tie decay to
vary with time. The first year, 2013, is set as the reference group. Network size is measured as
the number of unique co-creation ties of a focal designer. As the limit of carrying capacity does
not allow individuals to maintain an unlimited number of ties (Dunbar et al., 2015), it is entirely
possible that those with larger network size are more inclined to drop existing ties. The same
approach in Hypotheses 4 and 6a is employed to construct the remaining two control variables.
These two variables consider the homophily mechanisms at the designer level in driving tie
decay.
For hypotheses 9 through 11, the control variables include (a) year dummies, (b) the
number of submitted designs of a focal designer, (c) experience, and (d) tie duration. The
inclusion of year dummies allows the event of category exit to vary with time. The first year,
2014, is set as the reference group. Controlling the total number of submitted designs is
necessary because it rules out the alterative explanation that a designer will be more (or less)
likely to enter or exit a product category if he or she has more submissions in the past.
Experience is also a time-varying variable, which is measured as the year of submission minus
the year of registration. For hypotheses 9b and 10b, this research also controls tie duration
because network inertia theory (Kim et al., 2006) posits that social entities are subject to strong
inertial forces in tie choice and that an increase in the duration of an existing tie leads to a
decrease in the probability of tie decay. This variable is computed as the total number of
consecutive years for which a designer-category pair has persisted.
For hypotheses 12 through 14, this study controls for both sender-specific and receiver-
specific characteristics extracted from the first wave. These control variables include the number
of printed designs and experience at both the sender and the receiver levels. Failure to control the
102
number of printed designs may inflate the effect of design expertise. Experience has been chosen
as a control variable in prior research on tie decay (e.g., Liang & Fu, 2016).
Analytical Procedures
As this study includes panel data with repeated measures, an OLS-based regression
analysis may lead to biased estimates due to unobserved heterogeneity (Allison, 2009). In view
of this, conditional fixed-effects negative binomial regression models were employed to test H
1

and H
2
. Fixed-effects models have been widely employed to examine ecological dynamics
(Alrdich & Reuf, 2006) and network evolution (Demirkan et al., 2013). By focusing merely on
intra-category variation over time, the fixed-effects models capture time-invariant unobserved
heterogeneity. As the dependent variable, the rate of entry into a category, is an over-dispersed
count variable (see Table 3 and the next chapter for details), it is more suitable to run the analysis
using negative binomial regression instead of Poisson regression.
Linear regression models were run to test or answer RQ
1
, H
3
, and H
4
. To account for
heteroscedasticity and autocorrelation in cluster-correlated data, this research estimated robust
standard errors. These numbers were clustered at the author level through the use of the “cluster”
command in Stata 14.0 (Williams, 2000).
Exponential random graph models using Markov chain Monte Carlo maximum likelihood
estimation (MCMCMLE) were employed to test H
5
and H
7a
. ERGM is an inferential technique
that enables researchers to test hypotheses regarding the effects of network properties (or
structural parameters) on the probability of tie formation in a network. The underlying logic of
MCMCMLE is “to simulate a distribution of random graphs from a set of starting parameter
values, …, and to redefine the estimated parameter values by comparing this distribution with the
observed graph” (Zappa & Lomi, 2015, p. 557). When a structural parameter occurs more (or
103
less) frequently in the observed network than in random networks, researchers can conclude that
this parameter represents a significant underlying mechanism of tie creation. Data analysis was
performed by the statnet package in R (Handcock et al., 2003).
Discrete-time event history analysis was employed to test H
6
, H
7b
, H
8
and H
9
through
H
11
. The dependent variable in each hypothesis is the occurrence of tie decay, which can be
described as the likelihood that a network tie will be susceptible to decay at any particular time.
Communication network scholars have largely relied on event history analysis to examine the
factors that drive tie decay over time (Burt, 2000; Margolin et al., 2015; Miller et al., 2011).
Event history analysis (also known as survival analysis) is one of the most commonly used
analytical technique in ecological and evolutionary research (Alrdich & Reuf, 2006; Miller et al.,
2011). This technique is well suited for modeling dynamic and longitudinal change because it (a)
incorporates time-varying covariates and (b) corrects for the problem of right-censoring, which
occurs when an event is not experienced by the completion of an observation period (Allison,
1984). For hypotheses 5, 6b, and 7, the unit of analysis is the commensalist tie between two
designers. Suppose that Design X is created by Designer A and B, and Design Y is created by
Designer B and A. This research treats the A-B and the B-A pairs as two unique commensalist
ties in event history analysis based on the assumption that the first author devotes more time and
energy than does the second one in the design process. The authorship order is determined by
designers themselves. While it is possible that the two authors of a design have equal
contributions, none of the existing co-created designs includes such acknowledgments. For
hypotheses 9 through 11, the unit of analysis is the designer-category pair. Robust standard
errors are obtained for coefficients that are clustered by unique commensalist ties or designer-
category pairs in the estimation process (Pontikes & Barnett, 2015; Shen et al., 2014).
104
Multilevel logistic regression models were used to test H
12
, H
13
, and H
14
, given that this
analytical technique has been successfully employed to examine ego-centric network data (Liang
& Fu, 2016; Leonardi & Meyer, 2015; Song, 2018; Stefanone, Kwon, & Lackaff, 2012). The unit
of analysis in these hypotheses is the tie between any pair of successful designers. As the
complex interdependency of the network data violates the assumption of independent
observations in ordinary linear regression, multilevel models (known as hierarchical linear
models or mixed models; Raudenbush & Bryk, 2002) were employed to address the problem of
non-independence. Prior research has shown the utility of multilevel modeling as an analytical
technique to control for generic forms of dependence (Liang & Fu, 2016; Zappa & Lomi, 2015)
and to leverage multilevel theorization in network research (Moilterno & Mahony, 2011). In this
case, a designer’s follower-followee relations with others are nested under the same designer. As
the dependent variable was binary, multilevel logistic regression models were selected. Data
analysis was performed with Stata 14.0.
105
Chapter 5: Results
Cross-Level Ties Between Designs and Categories
Table 4 reports descriptive statistics and Spearman correlation coefficients among study
variables for Hypotheses 1 and 2. The data are organized in the form of 57 separate months from
November 2013 to July 2018. Although 32 product categories emerged in November 2012 and
have been stable and unchanged since then, it was not until October 2013 that designers were
required to choose at least one of these existing categories to label their designs before
submitting to the general or themed challenges. As the two explanatory variables are lagged by
one month, November 2013 was selected as the starting point in panel data analysis. The
distribution of the outcome variable is clearly over-dispersed, leading to the use of the negative
binominal regression model.
Table 5 presents the results of negative binomial regression models that consider both
category and time fixed effects. Model 1 estimates the baseline effects of the two control
variables. H
1
H
2
Model 2 adds the two ecological variables of fuzzy density and category
contrast. This model explains an additional 12.09% of the variance in the dependent variable for
a total of 31.10%.
Hypothesis 1 stated that increases in the fuzzy density of a category led to increases in the
rate of entry into the category by designs. As reported in H
1
H
2
Model 2, the final results
revealed that fuzzy density was significantly and positively associated with the rate of entry (β =
0.23, SE = 0.02, p < .001). Thus, H
1
was supported.
Hypothesis 2 predicted that increases in the contrast of a category would lead to increase
the rate of entry into the category by designs. Contrary to the prediction, category contrast was
106
not a significant predictor of the outcome variable (β = -0.03, SE = 0.17, p = .87). Thus, H
2
was
rejected.
Table 4
Descriptive Statistics and Spearman Correlation Coefficients Among Study Variables for
Hypotheses 1 and 2 (n = 32, T = 57, N = 1,824)
M SD Min Max
1 2 3 4 5
1. Rate of entry into
a category
180.71 296.90 0.00 1,659.00 1.00
2. Total number of
submitted designs
2,369.49 589.26 1,046.00 3,573.00 0.10*** 1.00
3. Percent of designs
submitted to the
general challenge
0.44 0.10 0.12 0.67 -0.06** -0.43*** 1.00
4. Fuzzy density of a
category
73.80 118.41 0.00 685.18 0.94*** 0.06** -0.02 1.00
5. Contrast of a
category
0.40 0.10 0.14 1.00 0.30*** -0.02 -0.03 0.39*** 1.00
Note. a. p* < .05, p** < .01, p*** < .001.

Table 5
Fixed-Effects Negative Binominal Regression Models Predicting the Rate of Entry into a
Category from November 2013 to July 2018 (Hypotheses 1 and 2)
Model 1 H
1
H
2
Model 2
β SE β SE
Intercept -1.86 3.33 -2.68 3.16
Total number of submitted
designs/1000
0.20 0.52 0.14 0.50
Percentage of designs submitted to
the general challenge
7.54 10.96 9.82 10.41
Fuzzy density of a category/100
(H
1
)
0.23*** 0.02
Contrast of a category (H
2
) -0.03 0.17
Category fixed effects Included Included Included Included
Time fixed effects Included Included Included Included
Number of categories 32 32
Number of time periods 57 57
Number of observations 1,824 1,824
Log likelihood -7815.48 -7730.18
Wald χ
2
362.26*** 690.48***
Note. a. p* < .05, p** < .01, p*** < .001.
b. Time fixed effects are controlled by using 56 dummy variables to represent different months.
The first month, November 2013, is set as the reference group. Due to page limit, the coefficients
of time fixed effects are not reported in the table.

107
Table 6 shows the descriptive statistics and Spearman correlation coefficients among
study variables for research question 1 and hypotheses 3 and 4. The final dataset includes
130,806 submitted designs that received ratings from online crowds. The unit of the analysis is at
the design level. The observation period also runs from November 2013 through July 2018. Due
to the large sample size, it is not surprising that a coefficient as small as -0.01 reaches a 0.001
level of statistical significance.
Table 7 reports the results of the linear regression models predicting the fitness of a
design for research question 1 and hypotheses 3 and 4. Considering that multiple designs can be
single-authored or co-authored by the same person or group of people, standard errors are
adjusted for 32,031 unique combinations of the authorship. Model 1 estimates the baseline
effects of the type of change, the number of tags, and the time of submission. These control
variables explain 9.91% of the variance in the fitness of a design. RQ
1
Model 2 and H
3
Model 3
include the categorical niche width and the category contrast of a design, respectively. H
4
Model
4 adds the interaction term. The final model explains 11.40% of the variance in the outcome
variable.
Research question 1 asked what the relationship would be between the categorical niche
width of a design and its fitness. The final results (see RQ
1
Model 2) showed that niche width
was a positive and significant predictor of the dependent variable (B = 0.03, robust SE = 0.003, p
< .001). In other words, niche broadness was viewed positively in the context of design
crowdsourcing.
Hypothesis 3 posited that increases in the category contrast of a design would result in
increases in the fitness of the design. As shown in H
3
Model 3, category contrast was a
108
significant and negative predictor of a design’s fitness (B = -1.14, robust SE = 0.06, p < .001).
This result rejected H
3
.
Hypothesis 4 stated that the relationship between categorical niche width and fitness was
moderated by category contrast, such that the intensity of the association was strengthened when
the level of category contrast is high. Contrary to the prediction, H
4
Model 4 revealed that this
variable negatively moderated the positive association between categorical niche width and
fitness (B = -0.26, robust SE = 0.06, p < .001), such that the positive relation was stronger when a
design had a lower level of category contrast. Thus, H
4
did not receive empirical support.
Table 6
Descriptive Statistics and Spearman Correlation Coefficients Among Study Variables for
Research Question 1 and Hypotheses 3 and 4 (N = 130,806)
M SD Min Max 1 2 3 4 5
1. Fitness of a design 2.56 0.47 1.13 4.61 1.00
2. Type of challenge 0.42 N.A. 0.00 1.00 0.07* 1.00
3. Number of tags 5.75 2.90 1.00 10.00 0.19* -0.01* 1.00
4. Categorical niche width of a
design
2.44 1.04 1.00 7.00 0.12* 0.04* 0.29* 1.00
5. Category contrast of a design 0.42 0.05 0.20 1.00 -0.11* -0.06* -0.04* -0.39* 1.00
Note. a. p* < .001.
b. Fifty-six dummy variables for time of submission are not reported in the table.

109
Table 7
Linear Regression Models Predicting the Fitness of a Design (Research Question 1 and
Hypotheses 3 and 4)
Model 1 RQ 1 Model 2 H 3 Model 3 H 4 Model 4

B Robust SE B
Robust
SE B Robust SE B Robust SE
Intercept 2.53* 0.01 2.48* 0.01 3.14* 0.03 2.89* 0.06
Type of challenge (1 =
the general challenge)
0.07* 0.007 0.07* 0.007 0.06* 0.007 0.06* 0.007
Number of tags 0.03* 0.002 0.03* 0.002 0.03* 0.002 0.03* 0.002
Time of submission
(dummy variables)
Included Included Included Included
Categorical niche
width of a design
(RQ 1)

0.03* 0.003 0.11* 0.03
Category contrast of a
design (H 3)

-1.14* 0.06 -0.56* 0.13
Categorical niche
width of a design ×
Category contrast of a
design (H 4)

-0.26* 0.06
Number of observation 130,806 130,806 130,806 130,806
F 72.40 73.88 79.92 78.09
R
2
(%) 9.91 10.26 11.31 11.40
Note. a. p* < .001.
b. Due to page limit, the coefficients of 56 dummy variables for time of submission are not
reported in the table.

Cross-Level Ties Between Designers and Designs
Figure 5 presents a visualization of all non-directed commensalist ties among 1,861
successful designers on Stitchly’s website. In this network graph, nodes represent designers, and
lines indicate the presence of commensalist ties between pairs of designers. The authorship order
is not considered. Design co-creation occurred among designers between 2012 and 2017 only.
There were 516 unique ties initiated by 271 nodes (14.6% of all designers) during the six years.
In other words, the average number of commensalist ties for this group of designers is around
1.90. These nodes are positioned in the central area of the graph. The remaining 1,590 nodes are
network isolates, revealing that most of the existing successful designers did not work on the
same designs with others. The node with the largest size has the highest degree centrality and
once collaborated with 82 successful designers in the network. As exponential random graph
modeling requires the input of binary network data, the strength of network ties was
dichotomized before hypothesis testing.
110
Figure 5. A Visualization of Commensalist Ties Among 1,861 Successful Designers on the
Stitchly Digital Platform. The graph was made with Gephi 2.0. The size of a node is proportional
to degree centrality, or the total number of ties the designer has. The thickness of an edge is
proportional to tie strength, or the total number of co-created designs between two designers.
Degree centrality ranges from 0 to 82. Tie strength ranges from 0 to 55. Average degree = 0.553.
Network density = 0.0004. Average clustering coefficient = 0.168. Average path length = 3.504.

111
Table 8 presents the final results of ERGMs predicting the formation of commensalist
ties among 1,861 designers in the sample. Like a logistic regression, an ERGM reports the log
odds of a structural parameter to tie creation. A positive and significant coefficient indicates that
a specific parameter in the observed network occurs more frequently than what is predicted by
randomness. Model 1 estimates the baseline effects of purely structural parameters (including
edges and shared partners), nodal attributes, and homophily. H
5
H
7a
Model 2 adds two
parameters of theoretical interest: experience and gap in design expertise.
Hypothesis 5 posited that inexperienced designers established more commensalist ties
than did experienced designers. As expected, this hypothesis was supported because experience
was a negative and significant predictor of the probability of tie creation (log odds = -0.05, SE =
0.01, p < .001). Specifically, a one-year increase in experience on Stitchly’s site would lead to a
decrease in the odds of tie formation by 5.3% (exp(-0.05)= 0.947).
Hypothesis 7a stated that designers were more likely to collaborate with those whose
design expertise was similar to their own. Contrary to expectation, the final results revealed a
heterophily effect. Specifically, a unit increase in the gap in design expertise at the dyadic level
would result in an increase in the probability of tie formation (log odds = 5.56, SE = 0.95, p
< .001). Thus, H
7a
was rejected.

112
Table 8
Exponential Random Graph Models Predicting Tie Formation in the Design Co-Creation
Network (Hypotheses 5 and 7a)
Model 1 H
5
H
7a
Model 2
Log Odds SE Log Odds SE
Purely structural effects
Edges -7.63*** 0.10 -6.15*** 0.24
Shared partners (GWESP) 3.18*** 0.07 2.86*** 0.12
Nodal attributes
Number of printed designs 0.01* 0.005 0.04*** 0.01
Design expertise -2.61*** 0.31 -7.25*** 0.85
Experience (H
5
) -0.05*** 0.01
Homophily
Gap in the number of printed
designs
0.01* 0.005 -0.01 0.01
Gap in experience -0.14*** 0.01 -0.21*** 0.03
Gap in design expertise (H
7a
) 5.56*** 0.95
Residual deviance -3,900 -3,831
AIC 7,812 7,678
BIC

7,866 7,777
Note. a. p* < .05, p** < .01, p*** < .001.
b. GWESP = Geometrically Weighted Edgewise Shared Partners.
c. AIC = Akaike Information Criterion.
d. BIC = Bayesian Information Criterion.

Table 9 reports descriptive statistics and Spearman correlations among study variables for
Hypotheses 6, 7b, and 8. The empirical results of the discrete-time event history analysis are
presented in Table 10. Standard errors are adjusted for 595 unique ties. Model 1 estimates the
baseline effects of the control variables. H
6
H
7b
H
8
Model 2 examines the effects of three
independent variables on the probability of tie decay. A positive and significant coefficient
means that an increase in a study variable results in an increase in the likelihood that an existing
commensalist tie disappears.
Hypothesis 6 stated that commensalist ties involving inexperienced designers were more
likely to decay than ties with no inexperienced designers involved. As shown in H
6
H
7b
H
8
Model
2 in Table 10, the impact of experience on the outcome variable was not statistically significant
(log odds = 0.03, robust SE = 0.06, p = .69). Thus, H
6
did not receive empirical support.
113
Hypothesis 7b predicted that the greater the gap in design expertise between designers,
the more likely that their commensalist ties would decay. This hypothesis was not supported
either, as gap in design expertise produced a negative and significant effect on the dependent
variable (log odds = -1.17, robust SE = 0.41, p = .005). In other words, a one unit increase in gap
in design expertise would result in a decrease in the odds of tie decay by 69.0% (exp(-1.17) =
0.310).
Hypothesis 8 proposed that the longer commensalist ties had been maintained between
two designers, the less likely the ties would decay. As expected, tie duration was a significant
and negative predictor of tie decay (log odds = -0.58, robust SE = 0.13, p < .001), lending
support to H
8
. The hazard function showed that a one unit increase in tie duration would lead to a
decrease in the odds of tie decay by 44.0% (exp(-0.58) = 0.560).
Table 9
Descriptive Statistics and Spearman Correlation Coefficients Among Study Variables for
Hypotheses 6, 7b, and 8 (N = 779)
M SD Min Max
1. Tie decay 0.78 N.A. 0.00 1.00
2. Year 2014 0.31 N.A. 0.00 1.00
3. Year 2015 0.23 N.A. 0.00 1.00
4. Year 2016 0.08 N.A. 0.00 1.00
5. Year 2017 0.11 N.A. 0.00 1.00
6. Network size (the
first author)
14.27 17.35 1.00 76.00
7. Network size (the
second author)
14.27 18.96 1.00 76.00
8. Gap in the number of
printed designs
13.99 11.89 0.00 73.00
9. Gap in experience 2.25 1.68 0.00 8.00
10. Experience 4.87 2.23 1.00 13.00
11. Gap in design
expertise
0.13 0.20 0.00 0.99
12. Tie duration 1.27 0.62 1.00 5.00

114
1 2 3 4 5 6 7 8 9 10 11 12
1 1
2 0.36*** 1
3 0.29*** -0.37*** 1
4 0.15*** -0.19*** -0.15*** 1
5 0.19*** -0.24*** -0.19*** -0.10** 1
6 -0.12** 0.13*** -0.15*** -0.10* -0.06 1
7 -0.11** 0.12*** -0.11** -0.07 -0.09* -0.22*** 1
8 -0.10** -0.06 -0.04 0.05 -0.04 0.19*** -0.02 1
9 -0.02 -0.07 -0.00 0.02 0.09** 0.03 -0.07 0.13*** 1
10 -0.03 -0.34** 0.05 0.18*** 0.31*** -0.01 -0.06 0.07 -0.31*** 1
11 -0.06 -0.03 -0.05 0.04 0.02 0.01 0.01 0.27*** -0.03 0.09* 1
12 -0.21*** -0.26** 0.03 0.14*** 0.01 -0.01 0.01 0.11** 0.04 0.16*** 0.08* 1
Note. a. p* < .05, p** < .01, p*** < .001.

Table 10
Discrete-Time Event History Analysis Predicting the Decay of a Commensalist Tie (Hypotheses
6, 7b, and 8)
Model 1 H
6
H
7b
H
8
Model 2
Log Odds Robust SE Log Odds Robust SE
Intercept 1.28*** 0.35 1.92*** 0.45
Year 2014 1.07*** 0.30 1.10*** 0.31
Year 2015 0.97** 0.31 1.16*** 0.35
Year 2016 0.07 0.36 0.45 0.44
Year 2017 0.52 0.36 0.83 0.45
Network size (the first
author)
-0.02*** 0.005 -0.02*** 0.005
Network size (the second
author)
-0.02** 0.006 -0.02*** 0.005
Gap in the number of printed
designs
-0.01 0.008 -0.01 0.008
Gap in experience 0.01 0.06 0.02 0.06
Experience (H
6
) 0.03 0.06
Gap in design expertise (H
7b
) -1.17** 0.41
Tie duration (H
8
) -0.58*** 0.13
Number of unique ties 595 595
Number of observations 779 779
Log pseudolikelihood -391.12 -377.46
Wald χ
2
41.59*** 90.33***
Pseudo R
2
(%) 5.47 8.77
Note. a. p* < .05, p** < .01, p*** < .001.
b. Standard errors are adjusted for 595 clusters (unique ties).

115
Cross-Level Ties Between Designers and Categories
Table 11 reports descriptive statistics and Spearman correlations among study variables
for Hypotheses 9a, 10a, and 11. As category contrast exists between 2014 and 2017 only, any
observation of the dependent variable in 2013 is not included. Table 12 presents the results of
discrete-time event history analysis predicting a designer’s entry into a product category.
Standard errors are adjusted for 4,402 unique designer-category pairs. Model 1 estimates the
baseline effects of all control variables. H
9a
H
10a
Model 2 tests the impacts of category contrast
and design expertise on the outcome variable. H
11a
H
11b
Model 3 further adds the two interaction
terms. A positive and significant coefficient means that an increase in a study variable results in
an increase in the probability that a designer enters a specific category.
Hypothesis 9a stated that designers had a higher propensity of entering a low-contrast
product category than a high-contrast product category. Contrary to the prediction, the final
results revealed that a high-contrast category drew significantly more entries from designers (log
odds = 1.99, robust SE = 0.32, p < .001). Thus, H
9a
was unsupported.
Hypothesis 10a predicted that designers with high levels of expertise had a lower
propensity of entering a product category than did those with low levels of expertise. H
9a
H
10a

Model 2 revealed that design expertise was not a significant predictor of the probability of
category entry (log odds = -0.09, robust SE = 0.43, p = .83). This result rejected H
10a
.
Hypothesis 11a posited that inexperienced designers had a lower propensity of entering a
low-contrast product category than did experienced designers. H
11a
H
11b
Model 3 confirmed that
the interaction term composed of experience and category contrast produced a positive and
significant effect on the event of category entry. In other words, designers with little experience
116
on Stitchly’s site were actually more likely to enter a low-contrast category (log odds = 0.46,
robust SE = 0.12, p < .001). Thus, H
11a
did not receive empirical support.
Hypothesis 11b proposed that designers with high levels of expertise had a higher
propensity of entering a low-contrast product category than did those with low levels of
expertise. The final results showed that the interaction term composed of design expertise and
category contrast was not a significant predictor of category entry (log odds = -3.91, robust SE =
3.91, p = .32). H
11b
was thus not supported.
Table 11
Descriptive Statistics and Spearman Correlation Coefficients Among Study Variables for
Hypotheses 9a, 10a, and 11 (N = 14,092)

M SD Min Max 1 2 3 4 5 6 7 8
1. Entry into a
product category
0.14 N.A. 0.00 1.00 1.00
2. Year 2015
0.23 N.A. 0.00 1.00 -0.04** 1.00
3. Year 2016
0.24 N.A. 0.00 1.00 -0.04** -0.31** 1.00
4. Year 2017
0.24 N.A. 0.00 1.00 -0.07** -0.31** -0.32** 1.00
5. Number of
submitted designs
19.40 27.36 1.00 280.00 0.17** 0.08** -0.04** -0.24** 1.00
6. Experience
5.90 2.53 1.00 13.00 -0.02* -0.09** 0.12** 0.33** -0.04** 1.00
7. Category contrast
0.41 0.08 0.23 0.74 0.14** -0.23** -0.13** -0.08** 0.03** -0.12** 1.00
8. Design expertise 0.07 0.07 0.004 0.38 -0.02 0.00 0.01 0.01 -0.13** 0.18** 0.01 1.00
Note. a. p* < .01, p** < .001.

117

Table 12
Discrete-Time Event History Analysis Predicting the Entry into a Product Category (Hypotheses
9a, 10a, and 11)
Model 1 H
9a
H
10a
Model 2 H
11a
H
11b
Model 3
Log Odds Robust SE Log Odds Robust SE Log Odds Robust SE
Intercept -1.62** 0.07 -2.56** 0.18 -1.61** 0.31
Year 2015 -0.80** 0.07 -0.62** 0.08 -0.63** 0.08
Year 2016 -0.80** 0.06 -0.65** 0.07 -0.64** 0.07
Year 2017 -0.99** 0.07 -0.84** 0.08 -0.83** 0.08
Number of submitted
designs
0.01** 0.001 0.01** 0.001 0.01** 0.001
Experience 0.03 0.01 0.03 0.01 -0.17* 0.05
Category contrast (H
9a
)
1.99** 0.32 -0.20 0.67
Design expertise (H
10a
)
-0.09 0.43 1.59 1.71
Experience × Category
contrast (H
11a
)
0.46** 0.12
Design expertise ×
Category contrast (H
11b
)
-3.91 3.91
Number of unique
designer-category pairs
4,402 4,402 4,402
Number of observations
14,092 14,092 14,092
Log pseudolikelihood
-5,489.34 -5,471.15 -5463.96
Wald χ
2
420.74** 504.51** 505.24**
Pseudo R
2
(%) 3.51 3.83 3.95
Note. a. p* < .01, p** < .001.
b. Standard errors are adjusted for 4,402 clusters (unique designer-category pairs).

Table 13 presents descriptive statistics and Spearman correlations among study variables
for Hypotheses 9b and 10b. As category contrast is observed between 2014 and 2017, this
research does not take into account any observation of the dependent variable, exit from a
product category, in 2013. The results of discrete-time event history analysis predicting a
designer’s exit from a product category are reported in Table 14. Standard errors are adjusted for
2,328 unique designer-category pairs. Model 1 shows the baseline effects of all control variables.
H
9b
H
10b
Model 2 adds category contrast and design expertise. This model explains 13.49% of
the variance in the outcome variable. A positive and significant coefficient means that an
increase in an explanatory variable leads to an increase in the probability that a designer chooses
to exit a specific category.
118
Hypothesis 9b predicted that designers had a higher propensity of exiting a low-contrast
product category than a high-contrast product category. Contrary to this prediction, the final
results showed that category contrast was a positive and significant driver of the exit from the
category (log odds = 2.02, robust SE = 0.54, p < .001). In other words, designers were more
likely to break an existing affiliation tie with a high-contrast category. Altogether, the findings
rejected H
9b
.
Hypothesis 10b posited that designers with high levels of expertise had a lower
propensity of exiting a product category than did those with low levels of expertise. As indicated
in H
9b
H
10b
Model 2, design expertise was not a significant predictor of the outcome variable (log
odds = -0.004, robust SE = 0.47, p = .993). Thus, H
10b
was not supported.
Table 13
Descriptive Statistics and Spearman Correlation Coefficients Among Study Variables for
Hypotheses 9b and 10b (N = 5,602)
M SD Min Max 1 2 3 4 5 6 7 8 9
1. Exit from a
product category
0.30 N.A. 0.00 1.00 1.00
2. Year 2015 0.30 N.A. 0.00 1.00 0.00 1.00
3. Year 2016 0.27 N.A. 0.00 1.00 0.03* -0.40*** 1.00
4. Year 2017 0.26 N.A. 0.00 1.00 0.07*** -0.38*** -0.36*** 1.00
5. Number of
submitted designs
28.88 38.84 1.00 280.00 -0.30*** 0.02 0.00 -0.13*** 1.00
6. Experience 6.26 2.52 1.00 13.00 0.00 -0.16*** 0.09*** 0.33*** 0.02 1.00
7. Tie duration 1.91 1.02 1.00 4.00 -0.19*** -0.23*** 0.23*** 0.40*** -0.04** 0.25*** 1.00
8. Category
contrast
0.42 0.06 0.23 0.74 0.03* -0.11*** -0.07*** 0.05*** 0.03* -0.02 -0.04*** 1.00
9. Design
expertise
0.07 0.07 0.004 0.38 0.02 -0.00 -0.01 -0.01 -0.15*** 0.18*** -0.00 -0.02 1.00
Note. a. p* < .05, p** < .01, p*** < .001.

119
Table 14
Discrete-Time Event History Analysis Predicting the Exit from a Product Category (Hypotheses
9b and 10b)
Model 1 H
9b
H
10b
Model 2
Log Odds Robust SE Log Odds Robust SE
Intercept -0.01 0.12 -0.90*** 0.26
Year 2015 1.07*** 0.10 1.13*** 0.10
Year 2016 1.82*** 0.12 1.87*** 0.12
Year 2017 2.36*** 0.14 2.40*** 0.14
Number of submitted designs -0.02*** 0.002 -0.02*** 0.002
Experience -0.87*** 0.04 -0.87*** 0.04
Tie duration -0.04*** 0.01 -0.04** 0.01
Category contrast (H
9b
) 2.02*** 0.54
Design expertise (H
10b
) -0.004 0.47
Number of unique designer-category
pairs
2,328 2,328
Number of observations 5,602 5,602
Log pseudolikelihood -2,961.24 -2,954.20
Wald χ
2
615.60*** 630.19***
Pseudo R
2
(%) 13.28 13.49
Note. a. p* < .05, p** < .01, p*** < .001.
b. Standard errors are adjusted for 2,328 clusters (unique designer-category pairs).

Within-Level Ties Among designers
Table 15 presents descriptive statistics and Spearman correlations among study variables
for Hypotheses 12 through 14. The final results of the multilevel mixed-effects logistic
regression models are shown in Table 16. Model 1 estimates the baseline effects of control
variables. H
12
H
13
Model 2 examines the impacts of design expertise and network closure on
unfollowing behavior. H
14
Model 3 further tests the moderating effect of network closure on
unfollowing behavior. Compared with Model 1, the final model explains additional 0.65% of the
variance in the dependent variable for a total of 5.45%. A positive and significant coefficient
means that an increase in a control or independent variable leads to an increase in the probability
of tie decay.
Hypothesis 12 stated that designers were more likely to unfollow others who had lower
levels of design expertise. The final results in H
12
H
13
Model 2 showed that the target’s design
120
expertise negatively predicted the likelihood that an existing follower-followee tie disappeared
(log odds = -2.24, SE = 0.89, p = .01), lending empirical support to H
12
. In other words, a one
unit decrease in the target’s design expertise added 89.4% greater odds of tie decay (exp(-2.24) =
0.106).
Hypothesis 13 proposed that designers who were members of high-closure groups were
less likely to unfollow others than designers who belonged to low-closure groups. As indicated
in H
12
H
13
Model 2, the findings confirmed that tie initiators’ network closure did not
significantly predict unfollowing behavior (log odds = 3.86, SE = 3.94, p = .33). Thus, H
13
was
not supported.
Hypothesis 14 posited that the negative relationship between design expertise and
unfollowing behavior is moderated by network closure, such that the negative association was
stronger when the focal designer belonged to a low-closure group. H
14
Model 3 revealed that the
moderating effect of network closure was not statistically significant (log odds = -10.02, SE =
20.21, p = .62). This result rejected H
14
.
Table 15
Descriptive Statistics and Spearman Correlation Coefficients Among Study Variables for
Hypotheses 12, 13, and 14 (N = 44,499)

M SD Min Max
1 2 3 4 5 6 7 8
1. Unfollowing
behavior
0.005 N.A. 0.00 1.00 1.00
2. Number of printed
designs (tie sender)
8.19 10.86 1.00 80.00 0.00 1.00
3. Number of printed
designs (tie receiver)
11.10 15.41 1.00 84.00 0.00 -0.03* 1.00
4. Experience (tie
sender)
8.72 2.83 2.00 18.00 -0.04* 0.37* -0.01 1.00
5. Experience (tie
receiver)
8.98 2.88 2.00 18.00 0.02* 0.08* 0.35* 0.16* 1.00
6. Design expertise (tie
sender)
0.09 0.13 0.002 1.00 -0.03* 0.49* 0.03* 0.25* 0.09* 1.00
7. Design expertise (tie
receiver)
0.12 0.14 0.002 1.00 -0.01 0.05* 0.49* 0.09* 0.26* 0.07* 1.00
8. Network closure (tie
sender)
0.03 0.03 0.00 0.95 -0.03* -0.57* 0.15* -0.16* 0.04* 0.03* 0.05* 1.00
Note. a. p* < .001.

121
Table 16
Multilevel Mixed-Effects Logistic Regression Models Predicting Unfollowing Behavior
(Hypotheses 12, 13 and 14)
Model 1 H
12
H
13
Model 2 H
14
Model 3
Log Odds SE Log Odds SE Log Odds SE
Intercept -117.74 212.41 -38.02 214.41 -33.90 214.92
Number of printed
designs (tie sender)
0.05* 0.02 0.05** 0.02 0.05** 0.02
Number of printed
designs (tie receiver)
-0.001 0.005 0.005 0.006 0.005 0.006
Experience (tie
sender)
-0.29** 0.10 -0.26* 0.10 -0.26* 0.10
Experience (tie
receiver)
0.24*** 0.03 0.25*** 0.03 0.25*** 0.03
Design expertise (tie
sender)
-1.02 2.45 -1.04 2.46
Design expertise (tie
receiver) (H
12
)
-2.24** 0.89 -1.96 1.02*
Network closure (tie
sender) (H
13
)
3.86 3.94 4.94 4.37
Design expertise (tie
receiver) × Network
closure (tie sender)
(H
14
)
-10.02 20.21
Number of designers 477 477 477
Number of ties 44,499 44,499 44,499
Log Likelihood -724.49 -719.95 -719.80
Wald χ
2
69.44*** 73.35*** 73.73***
Note. a. p* < .05, p** < .01, p*** < .001.

122
Chapter 6: Discussion and Conclusion
Review of Findings
Based on ecological and evolutionary theories, this dissertation builds a multidimensional
and multilevel network framework to study the crowdsourcing phenomenon. The aim of the
research is to examine how the evolution of multidimensional and multilevel networks in design
crowdsourcing is shaped by the dynamics of both human and non-human entities. Since the
ecological and evolutionary perspective is concerned with longitudinal change, and our social
systems are complex and multilevel in nature, what drives the formation, maintenance,
transformation, or decay of both cross-level and within-level network ties should be of
considerable interest to many researchers. Table 17 summarizes all the hypothesis-testing results
of the study. Overall, the empirical findings reveal that both nodal attributes and ecological
dynamic forces significantly constrain network structures and outcomes. The major findings of
this research are presented separately in the next few sections.
Table 17
Summary of Hypothesis-Testing Results
Hypotheses Results
Cross-level Ties Between Designs and Categories
H
1

Increases in the fuzzy density of a category lead to increases
in the rate of entry into the category by designs. Supported
H
2

Increases in the contrast of a category lead to increases in the
rate of entry into the category by designs. Unsupported
RQ
1

What is the relationship between the categorical niche width
of a design and its fitness?
Significantly
Positive
H
3

Increases in the category contrast of a design lead to
increases in the fitness of the design. Unsupported
H
4

The relationship between categorical niche width and fitness
is moderated by category contrast, such that the intensity of
the association is strengthened when the level of category
contrast is high. Unsupported
Cross-level Ties Between Designers and Designs
H
5

Inexperienced designers establish more commensalist ties
than do experienced designers. Supported
123
H
6

Commensalist ties involving inexperienced designers are
more likely to decay than ties with no inexperience designers
involved. Unsupported
H
7a

Designers are more likely to collaborate with those whose
design expertise is similar to their own. Unsupported
H
7b

The greater the gap in design expertise between designers,
the more likely that their commensalist ties will decay. Unsupported
H
8

The longer commensalist ties have been maintained between
two designers, the less likely the ties will decay. Supported
Cross-level Ties Between Designers and Categories
H
9a

Designers have a higher propensity of entering a low-contrast
product category than a high-contrast product category. Unsupported
H
9b

Designers have a higher propensity of exiting a low-contrast
product category than a high-contrast product category. Unsupported
H
10a

Designers with high levels of expertise have a lower
propensity of entering a product category than do those with
low levels of expertise. Unsupported
H
10b

Designers with high levels of expertise have a lower
propensity of exiting a product category than do those with
low levels of expertise. Unsupported
H
11a

Inexperienced designers have a lower propensity of entering
a low-contrast product category than do experienced
designers. Unsupported
H
11b

Designers with high levels of expertise have a higher
propensity of entering a low-contrast product category than
do those with low levels of expertise. Unsupported
Within-level Ties Among designers
H
12

Designers are more likely to unfollow others who have lower
levels of design expertise. Supported
H
13

Designers who are members of high-closure groups are less
likely to unfollow others than designers who belong to low-
closure groups. Unsupported
H
14

The negative relationship between design expertise and
unfollowing behavior is moderated by network closure, such
that the negative association is stronger when the focal
designer belongs to a low-closure group. Unsupported

Cross-Level Ties Between Designs and Categories
The first part of the analysis considers cross-level ties between designs and categories. By
connecting density dependence theory to the evolutionary perspective, this study shows that
when category and time fixed effects and other factors are held constant increases in the fuzzy
density of a category lead to increases in the rate of entry into the category by designs. This
124
result suggests that fuzzy density captures the legitimation process of category evolution in
design crowdsourcing quite well. The traditional measure of population density assumes that two
entities with different levels of category identity contributes equally to the legitimacy of the
product category. As a refined measure of population density identified by ecology research,
fuzzy density considers unequal contributions to institutionalized categories by entities with
differential grades of membership (Bogaert et al., 2016; Hannan et al., 2007; Kuilman & Li,
2009). Another interesting finding is that the contrast of a category is not a significant driver of
the rate of entry into the category by designs, which is not in line with the findings to date that
have been generated in non-creative task situations (Kuilman & Li, 2009). Perhaps crowd
members are not sensitive to category contrast when positioning their designs in the
classification system. Clearly, more work is needed to explain why this variable is not a good
indicator of legitimation processes in the context of crowdsourcing.
The findings on categorical niche width reveal that adopting a generalist rather than a
specialist strategy at the design level contributes to receiving a higher average score. Prior
ecological research on categories and categorization outside the crowdsourcing context has
provided mixed results about the relationship between categorical niche width and fitness (e.g.,
Hsu, 2006; Kovács & Hannan, 2015; Pontikes, 2012; Vergne, 2012). This research supports the
view that occupying multiple niches confers benefits especially when institutional logics
encourage novelty and non-conformity (Lo & Kennedy, 2015). Enlarging niche width produces a
significantly positive impact on fitness on Stitchly’s site, possibly because both the
crowdsourcing organization and crowd members embrace the logic of non-conformity and
expect novel or even revolutionary solutions to the crowdsourced challenge (Majchrzak &
Malhotra, 2014). By bringing this result to light, this research extends the debate about the effect
125
of categorical niche width on fitness to the context of online crowdsourcing and suggests that the
relationship between the two constructs may be contingent on the unique expectation of
audiences.
While prior research outside the crowdsourcing context has consistently shown
audiences’ positive reactions to high-contrast categories (e.g., Boone et al., 2012; Kovács &
Hannan, 2010, 2015; Pontikes, 2012), the current study confirms that increases in the category
contrast of a design lead to decreases in the fitness of the design in crowdsourcing competitions.
In the present case, designs labeled as “Cute” (a low-contrast) generally receive better
evaluations from crowd members than the ones labeled as “Sports” (a high-contrast category).
Although a member affiliated with one or more low-contrast categories is expected to have a
greater difficulty of interpretation due to cognitive ambiguity (Hannan, 2010; Kovács & Hannan,
2010), online crowds on Stitchly’s site may conceive this unclear identity in the classification
system as a form of innovation and thus strongly support the non-conformity behavior. Another
possible explanation derives from the ceiling effect. Suppose a crowd member likes designs that
belong to the category of “Zombies” and keeps looking at all the new designs on Stitchy’s site.
Over time, the person may find that the various sub-themes of the “Zombies” category are all
there already and that new submissions are just repetitions of the same tropes. After getting
bored, he or she may start to look for innovative and novel designs at the fuzzy borders.
Further, a design’s average category contrast is found to negatively moderate the positive
association between categorical niche width and fitness, such that the positive relationship is
stronger when a design had a lower level of category contrast. One possible explanation is that
reduced contrast of categories results in increased salience of niche broadness, thus making it
easier for online crowds to identify the non-conformity nature of the design. Broadening
126
categorical niche width involves spanning two or more categories. The presence of multiple low-
contrast categories in a single design strengthens evaluators’ perceived fuzziness and
innovativeness of the design. Under such circumstances, increased niche width brings extra
rewards in terms of external evaluations.
Cross-Level Ties Between Designers and Designs
Previous studies have emphasized the competitive nature of crowdsourcing platforms
(Boudreau et al., 2013; Majchrzak & Malhotra, 2014; Zhao & Zhu, 2014), but online crowds can
engage in collaborative activities at the same time (Hutter et al., 2011). In other words, while
prior research has made a distinction between tournament- and collaboration-based forms of
crowdsourcing (Afuah & Tuccci, 2012; Azar, 2018; Blohm, Leimeister, & Krcmar, 2013), these
two types of organizing are not mutually exclusive. In the present case, considering the low
success rate (see Table 1 for details), there is no doubt that competition on Stitchly’s site is very
fierce. However, this competitive environment does not necessarily mean designers do not form
collaborative relationships. This research identifies design co-creation as an important form of
commensalist ties (Aldrich & Reuf, 2006) on Stitchly’s site. This cross-level network can be
established based on designers’ affiliations with designs. By examining the factors that drive the
evolution of commensalist ties, the current study considers cooperative activities among crowd
members in tournament-based crowdsourcing, thus offering a more complete picture of crowd
behavior in online crowdsourcing.
Guided by the liability of newness theory (Stitchcombe, 1965), this research confirms
that new entrants in crowdsourcing competitions establish more commensalist ties than do
experienced community members. It is likely that this connection strategy helps the former to
buffer against environmental uncertainty and failure risks. Multiple cooperative ties may
127
compensate these inexperienced designers for a lack of knowledge and expertise. From a
network evolution perspective, inexperienced designers tend to engage more in the stage of
variation by experimenting with a wide range of ties with others. With regard to the impact of
experience on Stitchly’s site on the probability of tie decay, the findings do not support the
prediction that commensalist ties involving inexperienced designers are more likely to dissolve.
This result should be interpreted with caution because the sample includes only successful
designers, or those who have produced at least one printed design. It remains unknown whether
experience on Stitchly’s site significantly constrains the occurrence of tie decay among non-
experts. One possible explanation for the non-significant effect in this research is that network
size for inexperienced designers is not large enough to reach carrying capacity. Among all
successful designers who collaborated with others during the observation period, the average
number of unique commensalist ties was less than two. Therefore, these inexperienced designers
do not have much difficulty in preserving and maintaining such collaborative ties.
Drawing upon the conceptualization of network or relational inertia (Kim et al., 2006;
Maurer & Ebers, 2006), the current research provides empirical evidence for the tendency for
existing network ties to persist over time. Specifically, an increase in tie duration results in a
decrease in the probability of tie decay. Designers are probably subject to strong inertial forces
because of attachment and commitment embedded in a long-term relationship (Kim et al., 2006)
and the high costs associated with tie replacement (Briscoe & Tsai, 2011; Cheon et al., 2015).
Overall, this empirical finding can be conceived of as an adaptation of the “liability of newness”
thesis to the context of network evolution.
This research highlights the importance of design expertise in shaping a designer’s
decision to initiate or drop commensalist ties with others. On the one hand, designers tend to
128
establish ties with those with different levels of expertise. On the other hand, an increase in the
gap in design expertise at the dyadic level leads to a decrease in the likelihood of tie decay. The
above findings show that experts do not adhere to a principle of exclusivity (Podolny, 1994) to
target those whose design expertise is similar to their own as collaborators only. Similarly, a
commensalist tie involving two designers with large gaps in their levels of expertise is stable
over time, implying that tie decay may not be purely driven by economic considerations. It is
worthwhile to mention that the above analyses focus only on successful designers, thus making
the less “expert” designers very elite. Given this range restriction on expertise, future research
should include non-experts in crowdsourcing competitions in the sample and address this
generalizability issue.
Cross-Level Ties Between Designers and Categories
This dissertation also examines the evolution of cross-level ties between designers and
categories by tracking how designers enter or exit product categories on Stitchly’s site over time.
On the one hand, although prior research has suggested that the flexibility of low-contrast
categories allows social entities to keep options open and to develop a unique form of identity,
the empirical findings reveal that designers have a lower propensity of entering a low-contrast
category than a high-contrast category. This unexpected result can be explained by the
unambiguous features of high-contrast categories. In general, high-contrast categories entail
“clear capability requirements and stringent audience expectations” (Carnabuci et al., 2015, p.
1736). Thus, it does not seem to be quite challenging to come up with ideas that can be
positioned in categories with clear and crispy boundaries.
On the other hand, despite the conventional notion that social entities are more likely to
break affiliation ties with low-contrast categories due to the difficulty of meaning interpretation
129
and the risk of attracting the right type of audiences (Kovács & Hannan, 2010; Pontikes &
Barnett, 2015), this research supports the opposite direction: Designers have a higher tendency to
retain an existing tie with a low-contrast category than a high-contrast category after initial entry.
This opposite effect is not surprising when it is considered in the context of another finding that
increases in the average category contrast of a design significantly lead to decreases in its fitness
on Stitchly’s site. Over time, designers may gradually learn that designs affiliated with high-
contrast categories are not attractive to external evaluators and thus decide not to use these
categories to label their future submissions after initial trials.
Moreover, the moderation analysis confirms that inexperienced designers have a higher
propensity of entering a low-contrast category than do experienced designers. This result is not
in line with the traditional view that high-contrast categories are generally more attractive to new
entrants (Carnabuci et al., 2015; Sosa, 2013). One possible explanation is that while
inexperienced designers in crowdsourcing competitions can partially overcome the liability of
newness by choosing categories with clear or crispy boundaries, most of them have not enough
time to notice this potential benefit in their early stage. Alternatively, they are desperate to
achieve early success and intuitively believe that entering low-contrast categories helps to
accomplish the goal, since this strategy is often associated with positive external evaluations.
Altogether, the above findings demonstrate that designers are more likely to enter or exit a high-
contrast category. The positive effect of contrast on category entry is strengthened when the
focal designer is a relatively old community member.
Category entry and exit are conceived in this study as a form of exploration and
exploitation at the designer level. Entering an unexplored or exiting an explored product category
indicates a tendency toward exploration, while maintaining an existing relationship with a
130
category reflects a high level of exploitation. Contrary to the prediction guided the performance
feedback model (Greve, 1998), cognitive entrenchment theory (Dane, 2010), and ecological
theory (Carnabuci et al., 2015), design expertise did not produce a main effect on a designer’s
decision to enter or exit a product category. In addition, design expertise did not work together
with experience on Stitchly’s site to constrain category entry. These findings suggest that
expertise does not necessarily guide behavioral evolution across all contexts and thus direct
attention to the possible boundary conditions under which expertise plays an important role in
shaping strategic behavior. Considering that the sample is composed of all successful designers,
future research is needed to test whether the above findings also apply to non-experts.
Within-Level Ties Among Designers
This study contributes to recent research efforts to analyze unfollowing behavior in
online communities (e.g., Kwak et al., 2012; Liang & Fu, 2016; Xu et al., 2013) by examining
how designer-level attributes shape the decay of follower-followee ties among designers in
online crowdsourcing. It is noteworthy that removing someone from the following lists on the
Stitchly digital platform is not prevalent. Only 0.5 percent of follower-followee ties decayed
during the two waves, demonstrating the enduring nature of the communication network. This
percentage is a bit lower than that on Twitter. For instance, Xu and colleagues (2013) reported
that 858,702 out of 34 million (2.5%) following relations disappeared during the observation
periods. Similarly, Liang & Fu (2016) showed that 47,962 out of 1,658,069 ties were removed by
Twitter users. One possible explanation for network persistency in the current research is that
social networking is not a defining feature of the Stitchly site. Following behavior allows a
follower to receive an email notification each time a followee submits something new for crowd
judgment or produces a printed design for sale. More important, the creation of the follower-
131
followee network enables further communicative acts, thus making information exchange and
knowledge transfer among Stitchly designers possible. Despite these potential benefits, following
behavior is not quite common in the present case. Liang & Fu (2016) once reported that the
average Twitter user followed about 159 accounts. By contrast, the average number of followees
of 1,861 successful designers in Wave 1 was only 77.89. More than half of these follower-
followee ties (39.87/77.89, or 51.19%) were connected to other selected designers in the sample.
It is speculated that crowdsourcing participants do not care much about the strategic use of the
following lists and thus follow and unfollow others less frequently than typical social media
users do.
Consistent with an economic view of network evolution (Cheon et al., 2015; Dermirkan
et al., 2013), the empirical findings demonstrate that a decrease in a target’s design expertise
results in an increase in the likelihood that an existing tie disappears. Maintaining follower-
followee relations with design experts creates values for subscribers because these followees
tend to continue to submit ideas of good quality. As observing good examples through vicarious
learning can enhance creativity performance in crowdsourcing competitions (Riedl & Seidel,
2018), it becomes a reasonable choice for subscribers to retain the relationships with design
experts.
Contrary to the argument that closure impedes an individual’s ability to overcome
network or relational inertia change (Briscoe & Tsai, 2011; Kim et al., 2006; Maurer & Ebers,
2006), the final results confirmed that network closure did not pose normative constraints on a
designer’s decision to unfollow others. In addition, network closure failed to moderate the
negative association between design expertise and the tendency to unfollow others. One possible
reason for these findings is that designers do not care much about the negative consequences of
132
tie decay in a cohesive group, given the temporariness of online crowds (Majchrzak &
Malhortra, 2016).
Contributions
By examining the evolution of multidimensional and multilevel networks on the Stitchly
digital platform, this dissertation makes several contributions to the existing literature. First, the
research advances the refinement and development of ecological theories in communication
research. Although prior research has demonstrated the utility of ecological theories in
explaining longitudinal change of entities (Miller et al., 2011; Monge et al., 2008; Shumate,
2012; Weber & Monge, 2017), their application in the communication discipline has yet to be
fully realized. The current study takes an important step toward that direction by drawing
theoretical insights from the most recent line of ecological research on categories (Pontikes &
Hannan, 2014; Pontikes & Kim, 2017; Zuckerman, 2017) to reformulate density dependence
theory and niche theory in the communication literature. In addition, while communication
scholars have employed the ecological approach to examine the populations of routines (Monge
& Poole, 2008), groups (Lai, 2014), organizations (Larson & Linder, 2018), and communities
(Bryant & Monge, 2008), this research extends the literature by focusing on the populations of
designs and categories, or a set of designs with similar categorical affiliations. The empirical
findings reveal that an increase in a category’s fuzzy density results in an increase in the rate of
entry into the category by designs. In addition, strategic categorization in ecological niches,
reflected by the variables of categorical niche width and category contrast, constrains the fitness
of a design in crowdsourcing competitions. Specifically, categorical niche width is positively
associated with fitness. Category contrast is a negative predictor of a design’s fitness and
negatively moderates the positive relationship between niche width and fitness.
133
Second, this study advances the crowdsourcing literature by considering ecological
variables such as fuzzy density, category contrast, and niche width. While prior research has
revealed communicative (e.g., Guth & Brabham, 2017), cultural (e.g., Chua et al., 2015),
relational (e.g., Stephens et al., 2016), motivational (e.g., Azar, 2018), cognitive (e.g., Riedl &
Seidel, 2018), and material (e.g., Liu et al., 2014) antecedents of crowdsourcing behaviors and
outcomes, little attention has been paid to the impact of ecological factors. Building on the recent
development of an ecological theory of categories (Pontikes & Hannan, 2014), the current study
addresses this knowledge gap by demonstrating the theory’s applicability in the context of
crowdsourcing. The results suggest that design crowdsourcing operates on some of the same
ecological principles as other types of markets where category factors play an important role. In
addition, some empirical findings are not in line with traditional views of category contrast held
by existing ecology research outside the crowdsourcing context (e.g., Boone et al., 2012;
Carnabuci et al., 2015; Kovács & Hannan, 2010, 2015; Pontikes, 2012; Pontikes & Barnett,
2015), highlighting the necessity of theory refinement.
Third, this study highlights the importance of structural interdependence in explaining
crowd behavior. Most crowdsourcing research has assumed that online crowds act independently
because independent decision-making is a necessary condition for the wisdom of crowds
(Surowiecki, 2005). However, true independence is hard to achieve in crowdsourcing processes
(Bighash et al., 2018; Stephen et al., 2016; Stephens et al., 2016). In the present case, crowd
members are embedded in social networks through various socializing processes afforded by the
digital platform, such as co-creating designs with others and “following” others. By revealing
how the evolution of multidimensional and multilevel networks in design crowdsourcing is
134
shaped by designer-level, design-level, and category-level attributes and dynamics, this research
offers additional evidence of the complex interdependence in online crowdsourcing.
Fourth, the current study contributes to the literature on organizational communication in
two ways. On the one hand, this study emphasizes the importance of categorization processes
(Oh & Monge, 2013) in guiding crowd behavior. Crowdsourcing platforms provide an interface
where some crowd members submit creative ideas and others evaluate them. The dissertation
develops the argument that socially constructed classifications (e.g., product categories and tags)
constitute a communication system in which crowds engage in both sense-giving and sense-
making. By revealing the key role of categorization in organizing and coordinating collective
activities, this work deepens our understanding of crowds as “a significant feature of our
communities and the emerging communication landscape” (Stohl, 2014, p. 1). On the other hand,
this study advances the literature on expertise (Collins & Evans, 2007; Treem, 2012; Treem &
Leonardi, 2016, 2017) by demonstrating that design expertise is a driver of tie formation,
maintenance, and decay in the context of crowdsourcing. Specifically, the empirical findings
confirm that design expertise can well explain the evolution of design co-creation and follower-
followee ties.
Fifth, this study leverages the evolutionary approach in explaining network change by
integrating ecological theories into the baseline mechanisms of variation-selection-variation.
Although the evolutionary thinking has been applied to investigate the evolution of one-mode (or
unimodal) networks composed of the same sets of nodes (e.g., Kossinets & Watts, 2006;
Kleinbaum, 2018; Lee & Monge, 2011; Margolin et al., 2015; Shen et al., 2014; Shumate, 2012;
Weber, 2012; Yue, 2012), little attention has been paid to the evolutionary patterns of
multidimensional and multilevel networks. Considering the hierarchical nature of social structure
135
in the contemporary network society (Castells, 2000; Zappa & Lomi, 2015), it is increasingly
crucial for researchers to capture longitudinal changes in such complex networks. By analyzing
how the evolution of the multidimensional and multilevel network on Stitchly’s site proceeds
through tie formation and decay, this research addresses an important knowledge gap and thus
provides a more complete view of network change.
Sixth, this study contributes to further developing the multi-theoretical, multi-level
(MTML) network model (Monge & Contractor, 2003) and conceptualizing multidimensional and
multilevel networks (Contractor et al., 2011; Lomi et al.,2016). The traditional MTML model
employs different sets of theories to reveal that the creation of a unimodal or uniplex network
can be explained by endogenous and exogenous factors that operate at different levels of
analysis. The current study joins scholarly efforts in extending the MTML framework to explain
the coevolution of multiple human and non-human entities tied together by multiplex networks
(e.g., Ognyanova & Monge, 2013; Xu et al., 2018), thus expanding the model’s applicability for
explaining the evolutionary change of multidimensional and multilevel networks.
Limitations
Despite the contributions to the literature, this dissertation has several limitations. The
first concern is about generalizability. As Stitchly’s site is only one specific case of design
crowdsourcing that involves peer-vetted creative production, it remains unknown how far the
findings will generalize to other digitally mediated communities. Nonetheless, this issue may not
be serious, given that Stitchly’s site has the same elements as other digital platforms that
incorporate the elements of self-categorization (e.g., eBay), value ratings (e.g., Yelp and
TripAdvisor), and social networking (e.g., Facebook and Twitter). While the features of the
Stitchly digital platform are fairly common and representative, it is still worthwhile to investigate
136
how the observed relationships are contingent on other context-specific factors, including the
degree of collaboration (Afuah & Tucci, 2012), the purpose of sponsoring organizations (Zhu &
Zhu, 2014), reward systems (Liu et al., 2014), participation architectures (Majchrzak &
Malhotra, 2014), the temporariness of crowds (Majchrzak & Malhortra, 2016), among others. To
lend further evidence to the external validity of this study, a systematic comparison across
different crowdsourcing applications is necessary. Investigating this issue is particularly
important to advance our current understanding of crowdsourcing, given that most prior work
adopted a case study approach only (e.g., Bayus, 2013; Chua et al., 2015; Riedl & Seidel, 2018;
Stephens et al., 2016).
Second, this study fails to analyze a full three-level network because within-level ties
among designs and within-level ties among categories do not exist on Stitchly’s site. This
limitation prevents empirical testing of the formation, maintenance, transformation, and decay of
within-level ties among non-human entities (e.g., designs and categories) and the co-evolution of
within-level network structures and the fitness of non-human entities. Further, this research does
not examine the evolution of complex network interdependence across levels. While the recent
development of advanced network analytics (e.g., multilevel exponential random graph
modeling) has enabled researchers to estimate and simulate within-level and cross-level relations
simultaneously, this study does not test any hypotheses regarding how within- or cross-level
structural configurations shape network change at another level.
Third, although the use of digital trace data helps researchers to obtain theoretical
insights that would be otherwise unavailable and to avoid self-report response bias (Lazer et al.,
2013; Shah et al., 2015), it is not without limitations. For instance, the current study fails to
include motivational (Liu et al., 2014), cognitive (Blohm et al., 2016) or normative (Bauer et al.,
137
2016) factors in modeling the evolution of multidimensional and multilevel networks in online
crowdsourcing. As it is almost impossible to collect these designer-level variables by relying
merely on web scraping, future studies should consider combining survey data with digital trace
data to confirm that the observed relationships are not spurious.
Fourth, this research tests some hypotheses based on a sample consisting entirely of
successful designers who have produced at least one printed design. This selection bias suggests
that it is still unknown whether some of the observed relationships apply to the group of non-
experts on Stitchly’s site. However, this may not be a serious concern, given that low-potential
idea providers are generally not persistent contributors in online crowdsourcing (Huang et al.,
2014). In other words, even if non-experts, or those whose submitted designs have not been
selected by the company for sale, are included in the sample, it is empirically difficult, if not
impossible, to track their behavioral evolution.
There are some measurement limitations as well. First, due to the unavailability of
timestamps for follower-followee relations, the measure of unfollowing behavior captures
network change in discrete time only. In other words, the dataset does identify whether an
existing follower-followee tie was retained or decayed across the two waves, but the exact time
when a network tie disappeared is unknown. Information about the sequences of network change
is also missing. Consequently, this data structure does not allow the use of continuous-time
network models.
Second, this research overlooks other forms of the fitness of a design. In addition to the
average score of a design, the sales number is a good indicator of the design’s appeal to online
crowds on the Stitchly platform. However, only printed designs have product sales and they are
not publicly available. Future research needs to access the data from internal Stitchly sources.
138
Another desirable outcome variable would be audiences’ willingness to pay, which can be
measured in a survey or experiment (O’Connor, Carroll, & Kovács, 2017).
Third, the current research focuses on institutionalized product categories only and
overlooks other personalized ones when creating ecological variables. The use of personalized
categories is prevalent among designs. Table 6 shows that 57.6% (1-2.44/5.75) of all self-
claimed tags in 130,806 submitted designs do not include product categories constructed by the
Stitchly company. An examination of personalized categories is necessary in the future because
it contributes to deepening our understanding of crowd-enabled connective action based on the
logics of self-organization and deinstitutionalization (Bennett & Segerberg, 2013).
Fourth, this study dichotomizes the strength of commensalist ties into binary numbers
and thus sacrifices the richness of valued network data. Pilny and Atouba (2018) have
empirically shown that dichotomizing valued network data “can have significant effects on
which local processes or structures emerge as significant in shaping the network” (p. 259-260).
Future research can overcome this limitation by utilizing ERGMs developed to analyze valued
rather than binary network data (Krivitsky, 2012).
Directions for Future Research
There are several important future directions for research that can be based on the present
endeavor. First, as mentioned in the section on limitations, future studies need to replicate the
findings in the contexts of other crowdsourcing platforms or online communities. In addition, it
is important to jointly analyze digital trace data and survey data and to refine the measurement of
key variables, thus providing a more nuanced understanding of network and behavioral dynamics
in online crowdsourcing. Qualitative research using ethnography or interview data is also
139
desirable so that designers’ lived experiences and their cognitive and psychological processes
associated with decisions to form, maintain, and break network ties can be captured.
Second, this dissertation may inform research on marketing communication by
demonstrating the potential of ecological theories to re-conceptualize authenticity. Authenticity
fits the framework of ecologies of categories because marketing researchers have defined
authenticity in terms of categorical memberships. For instance, Davies (2001) argued that
“something is an authentic X if it is an instance or member of the class of Xs” (p. 203).
Similarly, Carrol and Wheaton (2009) emphasized that authenticity means “something is true to
its (alleged) type (or genre or category) classification” (p. 257). In other words, the existing
conceptualizations have already viewed authenticity as the extent to which a producer/product
adheres to an existing category. Another important reason for examining authenticity from an
ecological perspective is that, like the process of categorization, authenticity is largely socially
constructed and is evaluated in subjective terms (Carroll, 2015). While consumers seek
authenticity in a wide range of market contexts (Carroll, 2015; O’Connor, Carroll, & Kovács,
2017), their evaluations of authenticity have contextual and historical contingencies, creating a
flexibility for producers to position themselves in a favorable manner through marketing
communication. As authenticity has quality and price implications, future research can
investigate how consumers react to producers’ communication of authenticity based on
categorical identities or affiliations.
Third, future research should employ relational event modeling (REM) to examine the
evolution of multidimensional and multilevel networks (Kitts, Palloti, Lomi, Quintane, &
Mascia, 2017; Pilny, Schecter, Poole, & Contractor, 2016; Vu, Pattison, & Robins, 2015). As a
new statistical methodology in social network analysis, REM models network change by
140
considering the continuous sequence of social interactions. As traditional ERGMs and SIENA
treat network change as events in discrete time, time-stamped network data need to “be
aggregated into one cross-sectional dataset, or at best into panels of data at specific time points,
thereby removing detailed information about timing and sequence” (Patison, Quintane, Swain,
Robins, & Pattison, 2015, p. 842). REM is different from these two analytical approaches
because it does not require the aggregation of continuous records of network data. Another
unexplored area in network evolution is the empirical examination of “dormant ties” (Levin,
Walter, & Murnighan, 2011). Dormant ties refer to “a relationship between two individuals who
have not communicated with each other for a long time” (Levin et al., 2011, p. 923). The
conceptualization of dormant ties recognizes that decayed ties can be reconnected again in the
future. In the present case, designers who have exited a product category may return to the
category again. It is also possible for designers to reconnect to others whom they have
unfollowed. Investigating dormant ties provides a more complete view of network evolution
because social networks are “formed, dissolved, and reformed on a continuous basis” (Kim et al.,
2006, p. 704). While the idea of network evolution is largely inspired by Darwin’s (1859) theory
of biological evolution, social networks are different from biological entities in that the former
can be reconnected after decay. By contrast, biological entities have no opportunity to be alive
again after death.
Fourth, future research should continue to leverage the dialogue between ecological
theories and network research (Cattani, Ferriani, Negro, & Perretti, 2008). Ecological theories
can also inform network research by demonstrating ecological foundations of network change.
Although this study takes a further step toward that direction by examining ecological
antecedents of multidimensional and multilevel networks in online crowdsourcing, this issue still
141
warrants future investigation. Further, ecological theories are implicitly relational because they
consider the interaction between social entities and the external others and environments.
Network research can inform the ecological perspective by providing the evidence that network
connections within and across social entities at different levels can lead to population and
institutional change. The incorporation of a network perspective contributes to reformulating
existing ecological theories. For instance, the original density dependency theory assumed that
each entity in a population contributes equally to the legitimation and competition process in the
population (Bogaert et al., 2016). Network research can refine the theory by acknowledging that
an entity’s decision to enter into or exit from a population is more influenced by direct ties than
indirect ties. A network perspective provides news insights into resource partitioning theory as
well. By analyzing how resources flow are exchanged through social networks, researchers can
gain a better understanding of cooperative and competition dynamics between generalists and
specialists (Kitts et al., 2017).
Fifth, while this study shows that category contrast moderates the relationship between
categorical niche width and fitness, future research needs to consider other boundary conditions
of ecological influences. For instance, the impact of niche width may also be contingent on the
extent to which the spanned categories are distant or similar (Pontikes & Kim, 2017; Wry &
Castor, 2017). Prior research has followed a co-occurrence approach to measure the pairwise
distance or similarity between the categories to which an entity was assigned (Goldberg et al.,
2016; Kovács & Hannan, 2015). As measuring similarity and distance has a long history in
social network analysis as well (Borgatti et al., 2002), it is promising for future communication
scholars to integrate the existing measures from both the ecological and the network traditions
142
when exploring how categorical niche width exerts non-monolithic effects on the fitness of
social entities.
Practical Implications
There are practical implications as well. For instance, this research provides guidance for
idea submitters who desire to enhance their creativity performance in crowdsourcing
communities that involve self-categorization and value ratings. The findings about the ecological
antecedents of the fitness of a creative idea demonstrate that both niche broadness and reduced
category contrast contribute to an idea’s appeal to the crowd. Practically, idea submitters should
be aware of the importance of strategic categorization in constraining social evaluations because
crowd members treat self-categorization as a form of identity claim in the socially constructed
classification system and rely on this identity information for quality and novelty judgements.
The effective use of product categories to label submitted ideas is high valued in crowdsourcing
competitions and may enhance evaluators’ perceived expertise of idea submitters.
The finding about the positive impact of the fuzzy density of category on the rate of entry
into the category generates implications for the team of user experience in Stitchly. For instance,
real-time data about (a) the existing number of designs labeled by a specific product category
(population density) and (b) the sum of the GoM of all affiliated designs in the category (fuzzy
density) can be displayed on the webpage. These numbers can inform online crowds about the
legitimation and competitive dynamics in ecological niches.
The findings of this research can also guide the strategic management of crowdsourcing
organizations. In the present case, it is not necessary for the Stitchly company to clearly specify
the meanings of a category so that the existing fuzzy boundaries suitable for the emergence of
novel or even revolutionary ideas can be maintained. In addition, the company can offer
143
additional incentives to encourage category spanning and entry into a low-contrast category.
Considering the positive impacts of niche broadness and reduced category contrast on crowd
evaluations, this strategy may lead to an increased likelihood that the company receives more
highly rated ideas from designers. Consequently, the product team will probably have many
more good choices when deciding what to print.
Conclusion
Over the past decade, crowdsourcing has been identified as a new organizational form
and unique business model that mobilizes the effort of a distributed crowd for soliciting creative
ideas and solutions. Given growing interest in self-organizing, digitally mediated crowds and
communities (e.g., Stohl, 2014), this dissertation offers a novel approach to understanding
complex interdependence among large, anonymous, and geographically distributed
crowdsourcing participants and the current communication landscape. Drawing on the literature
on community, crowdsourcing, networks, organizational ecology, and socio-cultural evolution,
this research establishes a multidimensional and multilevel network framework that recognizes
both human and non-human entities in crowdsourcing processes as hierarchically structured in a
dynamic and complex system. This has led to the development of an ecological theory that
explains network and behavioral change of different sets of social entities tied together by
multiplex relations over time. Using digital trace data scraped from Stitchly’s site, the current
research tests hypotheses regarding the evolution of a three-level network composed of
designers, designs, and categories. The empirical findings highlight the importance of both nodal
attributes and ecological dynamic forces in shaping network structures and outcomes.

144
References
Ackerman, M., Pipek, V., & Wulf, V. (2003). Sharing expertise: Beyond knowledge
management. Cambridge, MA: MIT Press.
Adamic, K, A, Lento, T. M., Adar, E., & Ng, P. C. (2014). Information evolution in social
networks. In WSDM’16: Proceedings of the Ninth ACM International Conference on
Web Search and Data Mining (pp. 473–482). New York, NY: ACM.
Adler, P. S., & Kwon, S. (2002). Social capital: Prospects for a new concept. The Academy of
Management Review, 27(1), 17–40.
Afuah, A., & Tucci, C. L. (2012). Crowdsourcing as a solution to distance search. Academy of
Management Review, 37(3), 355–375.
Alder, P. S., & Kwon, S-W. (2002). Social capital: Prospects for a new concept. The Academy of
Management Review, 27(1), 17–40.
Aldrich, H. E. (2019). Guidepost: What sustains a belief in success among the unsuccessful?
Academy of Management Discoveries. Advance online publication. Retrieved from
https://journals.aom.org/doi/abs/10.5465/amd.2019.0139?journalCode=amd
Aldrich, H. E., Hodgson, G. M., Hull, D. L., Knudson, T., Mokyr, J., & Vanberg, V. J. (2008). In
defense of generalized Darwinism. Journal of Evolutionary Economics, 18(5), 577–596.
Aldrich, H. E., & Pfeffer, J. (1976). Environments of organizations. Annual Review of Sociology,
2, 79–105.
Aldrich, H. E., & Ruef, M. (2006). Organizations evolving (2
nd
ed.). Thousand Oaks, CA: Sage.
Allen, B. J., Chandrasekaran, D., & Basuroy, S. (2018). Design crowdsourcing: The impact on
new product performance of sourcing design solutions from the “crowd”. Journal of
Marketing, 82(2), 106–123.
145
Allison, P. D. (1984). Event history analysis: Regression for longitudinal event data. Beverly
Hills, CA: Sage.
Allison, P. D. (2009). Fixed effects regression models. Los Angeles, CA: Sage.
Aleta, A., & Moreno, Y. (2018). Multilayer networks in a nutshell. Annual Review of Condensed
Matter Physics, 10, 45–62.
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.
Audia, P. G., & Goncalo, J. A. (2007). Past success and creativity over time: A study of
inventors in the hard disk drive industry. Management Science, 53(1), 1–43.
Azar, O. A. (2018). Motivations and solution appropriateness in crowdsourcing challenges for
innovation. Research Policy. Advance online publication. Retrieved from
https://www.sciencedirect.com/science/article/abs/pii/S0048733318302889
Ball, J. A. (1984). Menes as replicators. Ethology and Sociobiology, 5(3), 145–161.
Bauer, J., Franke, N., & Tuertscher, P. (2016). Intellectual property norms in online
communities: How user-organized intellectual property regulation supports innovation.
Information Systems Research, 27(4), 724–750.
Baum, J. A. C., & Amburgey, T. L. (2002). Organizational ecology. In J. A. C. Baum (Ed.), The
Blackwell companion to organizations (pp. 304–326). Malden, MA: Blackwell.
Baum, J. A. C., & Shipilov, A. V. (2006). Ecological approaches to organizations. In S. R.
Clegg, C. Hardy, W. Nord, & T. Lawrence (Eds.), The SAGE handbook for organization
studies (pp. 55–110). London: Sage Publications.
146
Baum, J. A. C., & Singh, J. V. (1994). Organizational niches and the dynamics of organizational
mortality. American Journal of Sociology, 100(2), 346–380.
Baum, J. A. C., & Singh, J. V. (1994). Organizational hierarchies and evolutionary processes:
Some reflections on a theory of organizational evolution. In J. A. C. Baum & J. V . Singh
(Eds.), Evolutionary dynamics of organizations (pp. 3–20). New York, NY: Oxford.
University Press.
Bayus, B. (2013). Crowdsourcing new product ideas over time: An analysis of the Dell
IdeaStorm community. Management Science, 59(1), 226–244.
Bennett, W. L., & Segerberg, A. (2013). The logic of connective action: Digital media and the
personalization of contentious politics. New York, NY: Cambridge University Press.
Bickhard, M. H., & Campbell, D. T. (2003). Variations in variation and selection: The ubiquity
of the variation-and-selective-retention ratchet in emergent organizational complexity.
Foundations of Science, 8(3), 215–282.
Bighash, L., Oh, P., Fulk, J., & Monge, P. (2018). The value of questions in organizing:
Reconceptualizing contributions to online public information goods. Communication
Theory. 28(1), 1–21.
Blackmore, S. (2001). Evolution and memes: The human as a selective imitation device.
Cybernetics and Systems: An International Journal, 32(1–2), 225–255.
Blalock, H. M. (1979). Social statistics. New York, NY: McGraw-Hill.
Blau, P. M., & Schwartz, J. E. (1984). Crosscutting social circles: Testing a macrostructural
theory of intergroup relations. Orlando, FL: Academic Press.
Blohm, I., Leimeister, J. M., & Krcmar, H. (2013). Crowdsourcing: How to benefit from (too)
many great ideas. MIS Quarterly Executive, 12(4), 199–211.
147
Blohm, I., Riedl, C., Füller, J., & Leimeister, J. M. (2016). Rate or trade? Identifying winning
ideas in open idea sourcing. Information Systems Research, 27(1), 27–48.
Blohm, I., Zogaj, S., Bretschneider, U., & Leimeister, J. M. (2018). How to manage
crowdsourcing platforms effectively. California Management Review, 60(2), 122–149.
Bogaert, S., Boone, C., & Carroll, G. R. (2010). Organizational form emergence and
competing professional schemata of Dutch accounting, 1884–1939. Research in the
Sociology of Organizations, 31, 115–150.
Bogaert, S., Boone, C., Negro, G., & van Witteloostuijn, A. (2016). Organizational form
emergence: A meta–analysis of the ecological theory of legitimation. Journal of
Management, 42(5), 1344–1373.
Boone, C., Declerck, C. H., Rao, H., & Van den Buys, K. (2012). Out of tune: The rise and fall
of modernistic music in Brussels, 1919–1939. Poetics, 40(1), 44–66.
Boudreau, K. J., Lacetera, N., Lakhani, K. R. (2013). Incentives and problem uncertainty in
innovation contests: An empirical analysis. Management Science, 57(5), 843–863.
Borgatti, S. P., Everett, M. G., & Freeman, L. C. (2002). Ucinet for Windows: Software for
social network analysis. Harvard, MA: Analytic Technologies.
Brabham, D. C. (2008). Crowdsourcing as a model for problem solving: An introduction and
cases. Convergence: The International Journal of Research into New Media
Technologies, 14(1), 75–90.
Brabham, D. C. (2010). Moving the crowd at Threadless: Motivations for participation in a
crowdsourcing application. Information, Communication & Society, 13(8), 1122–1145.
148
Brabham, D. C. (2017). Crowdsourcing. In C. R. Scott & L. Lewis (Eds.), The international
encyclopedia of organizational communication (pp. 593–597). Hoboken, NJ: Wiley &
Sons.
Brabham, D. C., Ribisl, K. M., Kirchner, T. R., & Bernhardt, J. M. (2014). Crowdsourcing
applications for public health. American Journal of Preventive Medicine, 46(2), 179–
187.
Brandon, R. N. (1996). Concepts and methods in evolutionary biology. New York, NY:
Cambridge University Press.
Bregman, A., & Haythornthwaite, C. (2003). Radicals of presentation: visibility, relation, and co-
presence in persistent conversation. New Media & Society, 5(1), 117–140.
Brennecke, J., & Rank, O. N. (2016). The interplay between formal project memberships and
informal advice seeking in knowledge-intensive firms: A multilevel network approach.
Social Networks, 44, 307–318.
Briscoe, F., & Tsai, W. (2011). Overcoming relational inertia: How organizational members
respond to acquisition events in a law form. Administrative Science Quarterly, 56(3),
408–440.
Bryant, J. A., & Monge, P. R. (2008). The evolution of the children’s television community,
1953–2003. International Journal of Communication, 2, 160–192.
Budescu, D. V., & Chen, E. (2015). Identifying expertise to extract the wisdom of crowds.
Management Science, 61(2), 267–280.
Bulger, M., Bright, J., & Cobo, C. (2015). The real component of virtual learning: motivations
for face-to-face MOOC meetings in developing and industrialised countries. Information,
Communication & Society, 18(10), 1200–1216.
149
Burt, R. S. (1992). Structural holes. Cambridge, MA: Harvard University Press.
Burt, R. S. (2000). Decay functions. Social Networks, 22, 1-28.
Burt, R. S. (2005). Brokerage and closure: An introduction to social capital. New York, NY:
Oxford University Press.
Campbell, D.T. (1965). Variation and selective retention in socio–cultural evolution. In H. R.
Barringer, G. I. Blanksten, & R. W. Mack (Eds.), Social change in developing areas: A
reinterpretation of evolutionary theory (pp. 19–48). Cambridge, MA: Schenkman.
Carnabuci, G., Operti, E., & Kovács, B. (2015). The categorical imperative and structural
reproduction: Dynamics of technological entry in the semiconductor industry.
Organization Science, 26(6), 1734–1751.
Carpenter, S., Nah, S., & Chung, D. (2015). A study of US online community journalists and
their organizational characteristics and story generation routines. Journalism, 16(4), 505–
520.
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. (2015). Authenticity: Attribution, value, and meaning. In R. A. Scott & M. C.
Buchmann (Eds.), Emerging trends in the social and behavioral sciences: An
interdisciplinary, searchable, and linkable resource. New York, NY: John Wiley & Sons.
Carroll, G. R., Dobrev, S. D., & Swaminathan, A. (2002). Organizational processes of resource
partitioning. Research in Organizational Behavior, 24, 1–40.
Carroll, G. R., Feng, M., Le Mens, G., & McKendrick, D. G. (2010). Organizational evolution
with fuzzy technological formats: Tape drive producers in the world market, 1951–1998.
Research in the Sociology of Organizations, 35, 203–233.
150
Carroll, G. R., & Hannan, M. T. (2000). The demography of corporations and industries.
Princeton, NJ: Princeton University Press.
Carroll, G. R., & Wheaton, D. R. (2009). The organizational construction of authenticity: An
examination of contemporary food and dining in the U.S. Research in Organizational
Behavior, 29, 255–282.
Castells, M. (2000). The rise of the network society. Oxford; Malden, MA: Blackwell Publishers.
Cattani, G., Ferriani, S., Negro, G., Perretti, F. (2008). The structure of consensus: Network ties,
legitimation, and exit rates of U.S. feature film producer organizations. Administrative
Science Quarterly, 53(1), 145–182.
Cheon, Y. J., Choi, S. K., & Kim, J., & Kwak, K. T. (2015). Antecedents of relational inertia and
information sharing in SNS usage: The moderating role of structural autonomy.
Technological Forecasting & Social Change, 95, 32–47.
Chua, R. Y. J., Roth, Y., & Lemoine, J. F. (2015). The impact of culture on creativity: How
cultural tightness and cultural distance affect global innovation crowdsourcing work.
Administrative Science Quarterly, 60(2), 189–227.
Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of
Sociology, 94, S95–S121.
Collins, H., & Evans, R. (2007). Rethinking expertise. Chicago, IL: University of Chicago Press.
Contractor, N. (2009). The emergence of multidimensional networks. Journal of Computer-
Mediated Communication, 14(3), 743–747.
Contractor, N. (2018). How can computational social science motivate the development of
theories, data, and methods to advance our understanding of communication and
organizational dynamics? In B. Foucault Welles & S. Gonzalez-Bailon (Eds.), The
151
Oxford handbook of networked communication. Oxford, UK: Oxford University Press.
Retrieved from
http://www.oxfordhandbooks.com/view/10.1093/oxfordhb/9780190460518.001.0001/oxf
ordhb-9780190460518-e-7
Contractor, N. S., Monge, P. R., & Leonardi, P. M. (2011). Multidimensional networks and the
dynamics of sociomateriality: Bring technology inside the network. International Journal
of Communication, 5, 682–720.
Contractor, N. S., Wasserman, S., & Faust, K. (2006). Testing multi-theoretical multilevel
hypotheses about organizational networks: An analytic framework and empirical
example. Academy of Management Review, 31(3), 681–703.
Creanza, N., Kolodny, O., & Feldman, M. W. (2017). Cultural evolutionary theory: How culture
evolves and why it matters. Proceedings of the National Academy of Sciences of the
United States of America, 114(30), 7782–7789.
Cross, N. (2004). Expertise in design: an overview. Design Studies, 25(5), 427–441.
Dawkins, R. (1976). The selfish gene. Oxford: Oxford University Press.
Dawkins, R. (1982). The extended phenotype. Oxford: Oxford University Press.
Dane, E. (2010). Reconsidering the trade-off between expertise and flexibility: A cognitive
entrenchment perspective. Academy of Management Review, 35(4), 579–603.
Darwin, C. R. (1859). On the origin of species by means of natural selection, or the preservation
of favoured races in the struggle for life. London: John Murray.
Davies, S. (2001). Musical works and performances: A philosophical exploration. Oxford, UK:
Clarendon Press.
152
De Domenico, M., Solé-Ribalta, A., Omodei, E., Gómez, S., & Arenas, A. (2015). Ranking in
interconnected multilayer networks reveals versatile nodes. Nature Communications, 6,
6868.
Delacroix, J. & Carroll, G. R. (1983). Organizational foundings: An ecological study of the
newspaper industries of Argentina and Ireland. Administrative Science Quarterly, 28(2),
631–661.
Demirkan, I., Deeds, D. L., & Demirkan, S. (2013). Exploring the role of network
characteristics, knowledge quality, and inertia on the evolution of scientific networks.
Journal of Management, 39(6), 1462–1489.
DeVaughn, M. L., & Leary, M. M. (2018). Learn by doing or learn by failing? The paradoxical
effect of public policy in averting the liability of newness. Group & Organization
Management, 43(6), 871–905.
Di Pietro, F., Prencipe, A., & Majchrzak, A. (2018). Crowd equity investors: An underutilized
asset for open innovation in startups. California Management Review, 60(2), 43–70.
Diakonova, M., Nicosia, V., Latora, V., & San Miguel, M. (2016). Irreducibility of multilayer
network dynamics: the case of the voter model. New Journal of Physics, 18.
Dimmick, J. W. (2003). Media competition and coexistence: The theory of the niche. Mahwah,
NJ: Erlbaum.
Dobrev, S. D., Kim T-Y, Hannan, M. T. (2001). Dynamics of niche width and resource
partitioning. American Journal of Sociology, 106(5), 1299–1337.
Dobrev, S. D., Kim, T-Y., & Carroll, G. R. (2003). Shifting gears, shifting niches:
Organizational inertia and change in the evolution of the U.S. automobile industry, 1885–
1981. Organizational Science, 14(3), 264–282.
153
Doerfel, M. L., & Taylor, M. (2017). The story of collective action: The emergence of
ideological leaders, collective action network leaders, and cross-sector network partners
in civil society. Journal of Communication, 67(6), 920–943.
Dorst, K., & Reymen, I. (2004). Levels of expertise in design education. In P. Lloyd, N.
Roozenburg, C. McMahon, & L. Brodhurst. (Eds.), Proceedings of the 2nd International
Engineering and Product Design Education Conference, 2-3 September 2004, Delft (pp.
159–166). Delft, Netherlands: TU Delft.
Durand, R., Granqvist, N., & Tyllström, A. (2017). From categories to categorization: A social
perspective on market categorization. Research in the Sociology of Organizations, 51, 3–
30.
Durand, R., & Paolella, L. (2013). Categories in strategy, entrepreneurship, and organizational
theory. Journal of Management Studies, 50(6), 1100–1123.
Durcikova, A., Fadel, K. J., Butler, B. S., & Galletta, D. F. (2011). Knowledge exploration and
exploitation: The impacts of psychological climate and knowledge management system
access. Information Systems Research, 22(4), 855–866.
Dunbar, R. I. M., Arnaboldi, V., Conti, M., & Passarella, A. (2015). The structure of online
social networks mirrors those in the offline world. Social Networks, 43, 39–47.
Ellison, N. B., Steinfield, C., & Lampe, C. (2007). The benefits of Facebook "Friends:" Social
capital and college students' use of online social network sites. Journal of Computer-
Mediated Communication, 12(4), 1143–1168.
Estellés-Arolas, E., & González-Ladrón-de-Guevara, F. (2012). Towards an integrated
crowdsourcing definition. Journal of Information Science, 38(2), 189–200.
154
Fank, E. B., & Bassett, D. S. (2017). Brain and social networks: Fundamental building blocks of
human experience. Trends in Cognitive Science, 21(9), 674–690.
Faraj, S., Jarvenpaa, S. L., & Majchrzak, A. (2011). Knowledge collaboration in online
communities. Organizational Science, 22(5), 1224–1239.
Faraj, S., & Johnson, S. L. (2011). Network exchange patterns in online communities.
Organizational Science, 22(6), 1464–1480.
Faraj, S., von Krogn, G., Monteiro, E., & Lakhani, K. R. (2016). Online community as space for
knowledge flows. Information Systems Research, 27(4), 668–684.
Feldman, M. S., & Pentland, B. T. (2003). Reconceptualizing organizational routines as a source
of flexibility and change. Administrative Science Quarterly, 48(1), 94–118.
Fisher, G. (2019). Online communities and firm advantages. Academy of Management Review,
44(2), 279–298.
Flinn, M. V. (1997). Culture and the evolution of social learning. Evolution and Human
Behavior, 18(1), 23–67.
Freeman, J., & Hannan, M. T. (1983). Niche width and the dynamics of organizational
populations. American Journal of Sociology, 88(6), 1116–1145.
Freeman, J., & Hannan, M. T. (1989). Setting the record straight on organizational ecology:
Rebuttal to Young. American Journal of Sociology, 95(2), 425–439.
Freeman, J., Carroll, G. R., & Hannan, M. T. (1983). The liability of newness: Age dependence
in organizational death rates. American Sociological Review, 48(5), 692–710.
Friedland, L. A., Howe, T., & Rojas, H. (2006). The networked public sphere. Javnost - The
Public, 13(4), 5–26.
155
Fu, J. S., & Shumate, M. (2016). Hyperlinks as institutionalized connective public goods for
collective action online. Journal of Computer-Mediated Communication, 21(4), 298–311.
Füller, J. (2005). Refining virtual co-creation from a consumer perspective. California
Management Review, 52(2), 98–122.
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.
Godfrey-Smith, P. (2000). The replicator in retrospect. Biology and Philosophy, 15(3), 403–
423.
Goldberg, A., Hannan, M. T., & Kovács, B., (2016). What does it mean to span cultural
boundaries? Variety and atypicality in cultural consumption. American Sociological
Review, 81(2), 215–241.
Gondal, N. (2018). Duality of departmental specializations and PhD exchange: A Weberian
analysis of status in interaction using multilevel exponential random graph models
(mERGM). Social Networks, 55, 202–212.
Greve. H. R. (1998). Performance, aspirations, and risky organizational change. Administrative
Science Quarterly, 43(1), 58–86.
Greve, H. R. (2003). Organizational learning from performance feedback: A behavioral
perspective on innovation and change. New York, NY: Cambridge University Press.
Greve, H. R. (2007). Exploration and exploitation in product innovation. Industrial and
Corporate Change, 16(5), 945–975.
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.
156
Hager, M. A., Galaskiewicz, J., & Larson, J. A. (2004). Structural embeddedness and the liability
of newness among nonprofit organizations. Public Management Review, 6(2), 159–188.
Handcock, M. S., Hunter, D. R., Butts, C. T., Goodreau, S. M., & Morris, M. (2003).
statnet: Software tools for the Satistical Modeling of Network Data. Retrieved
from http://statnetproject.org
Hannan, M. T. (2010). Partiality of memberships in categories and audiences. Annual Review of
Sociology, 36, 159–181.
Hannan, M. T., Carroll, G. R., Dundon, E. A., & Torres, J. C. (1995). Organizational evolution in
a multinational context: Entries of automobile manufacturers in Belgium, Britain, France,
Germany and Italy. American Sociological Review, 60(4), 509–528
Hannan, M. T., & Freeman, J. (1977). The population ecology of organizations. American
Journal of Sociology, 82(5), 929–964.
Hannan, M. T., & Freeman, J. (1984). Structural inertia and organizational change. American
Sociological Review, 49, 149–164.
Hannan, M. T., & Freeman, J. (1989). Organizational ecology. Cambridge, MA: Harvard
University Press.
Hannan, M. T., Pólos, L., & Carroll, G. R. (2007). The logics of organizational theory:
Audiences, codes, and ecologies. Princeton, NJ: Princeton University Press.
Hampton, K. N. (2016). Persistent and pervasive community: New communication technologies
and the future of community. American Behavioral Scientist, 60(1), 101–124.
Hampton, K., & Wellman, B. (2003). Neighboring in Netville: How the Internet supports
community and social capital in a wired suburb. City & Community, 2(4), 277–311.
157
Haveman, H. A. (1993). Follow the leader: Mimetic isomorphism and entry into new markets.
Administrative Science Quarterly, 38(4), 593–627.
Hawley, A. H. (1950). Human ecology: A theory of community structure. New York,
NY: Ronald Press.
Heylighen, F., & Chielens, K. (2009). Evolution of culture, memetics. In R. Meyers (ed.),
Encyclopedia of complexity and systems science (pp. 3205–3220). New York, NY:
Springer.
Hilbert, M., Oh, P., & Monge, P. (2016). Evolution of what? A network approach for the
detection of evolutionary forces. Social Networks, 47, 38–46.
Hills, T. T., Todd, P. M., Lazer, D., Redish, A. D., Couzin, I. D., & Cognitive Search Research
Group. (2015). Exploration versus exploitation in space, mind, and society. Trends in
Cognitive Sciences, 19(1), 46–54.
Hodgson, G. M., & Knudsen, R. (2004). The firm as an interactor: Firms as vehicles for habits
and routines. Journal of Evolutionary Economics, 14(3), 281–307.
Howe, J. (2006). The rise of crowdsourcing. Wired Magazine, 14(6), 1–4.
Howe, J. (2008). Crowdsourcing: Why the power of the crowd is driving the future of business.
New York, NY: Crown Publisher.
Hsu, G. (2006). Jacks of all trades and masters of none: Audiences' reactions to spanning genres
in feature film production. Administrative Science Quarterly, 51(3), 420–450.
Hsu, G., & Hannan, M. T. (2005). Identities, genres, and organizational forms. Organization
Science, 16(5), 474-490.
158
Hsu, G., Hannan, M. T., & Koçak, Ö. (2009). Multiple category memberships in markets: An
integrative theory and two empirical tests. American Sociological Review, 74(1),150–
169.
Hsu, G., Negro, G., & Perretti, F. (2012). Hybrids in Hollywood: a study of the production and
performance of genre-spanning films. Industrial and Corporate Change, 21(6), 1427–
1450.
Huang, Y., Singh, P. V., & Srinivasan, K. (2014). Crowdsourcing new product ideas under
consumer learning. Management Science, 60(9), 2138–2159.
Huang, R., & Sun, X. (2014). Weibo network, information diffusion and implications for
collective action in China. Information, Communication & Society, 17(1), 86–104.
Hull, D. L. (1988). Science as a process: An evolutionary account of the social and conceptual
development of science. Chicago, IL: The University of Chicago Press.
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.
Jansson, D. G., & Smith, S. M. (1991). Design fixation. Design Studies, 12(1), 3–11.
Johnson, S, L., Safadi, H., & Faraj, S. (2015). The emergence of online community leadership.
Information Systems Research, 26(1), 165–187.
Katila, R., & Ahuja, G. (2002). Something old, something new: A longitudinal study of search
behavior and new product introduction. The Academy of Management Journal, 45(6),
1183–1194.
159
Kennedy, M. T., & Fiss, P. C. (2013). An ontological turn in categories research: From standard
of legitimacy to evidence of actuality. Journal of Management Studies, 50(6), 1138–
1154.
Keuschnigg, M., & Whimmer, T. (2017). Is category spanning truly disadvantageous? New
evidence from primary and secondary movie markets. Social Forces, 96(1), 449–479.
Kim, N., Konieczna, M., Yoon, H. Y., & Friedland, L. A. (2016). Sustainability factors of
emergent civic news websites: A qualitative comparative analysis approach. Journalism
& Mass Communication Quarterly, 93(4), 750–769.
Kim, T. Y., Oh, H., & Swaminathan, A. (2006). Framing interorganizational network change: A
network inertia perspective. Academy of Management Review, 31(3), 704–720.
Kitts, J. A., Palloti, F., Lomi, A., Quintane, E., & Mascia, D. (2017). Investigating the temporal
dynamics of interorganizational exchange: Patient transfers among Italian hospitals.
American Journal of Sociology, 123(3), 850–910.
Kivelä, M., Arenas, A., Barthelemy, M., Gleeson, J. P., Moreno, Y., & Porter, M. A. (2014)
Multilayer networks. Journal of Complex Network, 2(3), 203–271.
Klein, K. J., & Kozlowski, S. W. J. (Eds.). (2000). Multilevel theory, research, and methods in
organizations. San Francisco, CA: Jossey-Bass.
Kleinbaum, A. M. (2018). Reorganization and tie decay choices. Management Science, 64(5),
2219–2237.
Koçak, Ö., Hannan, M. T., & Hsu, G. (2014). Emergence of market orders: Audience interaction
and vanguard influence. Organization Studies, 35(5), 765–790.

160
Koka, B. R., Madhavan, R., & Prescott, J. E. (2006). The evolution of interfirm networks:
Environmental effects on patterns of network change. Academy of Management Review,
31(3), 721–737.
Kossinets, G., & Watts, D. J. (2006). Empirical analysis of an evolving social network. Science,
311(5757), 88–90.
Kovács, B., Carroll, G. R., & Lehman, D. W. (2014). Authenticity and consumer value ratings:
Empirical tests from the restaurant domain. Organization Science, 25(2), 458–478.
Kovács, B., & Hannan, M. T. (2010). The consequences of category spanning depend on
contrast. Research in the Sociology of Organizations, 35, 175–201.
Kovács, B., & Hannan, M. T. (2015). Conceptual spaces and the consequences of category
spanning. Sociological Science, 2, 252–286.
Kovács, B., & Johnson, R. (2014). Contrasting alternative explanations for the consequences of
category spanning: A study of restaurant reviews and menus in San Francisco. Strategic
Organization, 12(1). 7–37.
Kraut, R. E., & Resnick, P. (2012). Building successful online communities: Evidence-based
social design. Cambridge, MA: MIT Press.
Krivitsky, P. N. (2012). Exponential-Family random graph models for valued networks.
Electronic Journal of Statistics, 6, 1100–1128.
Kuilman, J. G., & Li, J. (2009). Grades of membership and legitimacy spillovers: Foreign banks
in Shanghai, 1847–1935. Academy of Management Journal, 52(2), 229–245
Kuilman, J. G., & Wezel, F. C. (2013). Taking off: Category contrast and organizational
mortality in the UK airline industry, 1919–64. Strategic Organization, 11(1), 56–77.
161
Kwak, H., Moon, S. B., & Lee, W. (2012). More of a receiver than a giver: Why do people
unfollow in Twitter? In Proceedings of the Sixth International AAAI Conference on
Weblogs and Social Media (pp. 499–502).
Lai, C-H., She, B., & Tao, C-C. (2017). Connecting the dots: A longitudinal observation of relief
organizations' representational networks on social media. Computers in Human Behavior,
74, 224–234.
Lander, M. W., & Heugens, P. P. (2017). Better together: Using meta-analysis to explore
complementarities between ecological and institutional theories of organization.
Organizational Studies, 38(11), 1573–1601.
Larson, R. P., & Linder, A. M. (2018). Professionalization through attrition? An event history
analysis of mortalities in citizen journalism. Information, Communication & Society,
21(5), 745–760.
Lavie, D., Stettner, U., & Tushman, M. L. (2010). Exploration and exploitation within and across
organizations. Academy of Management Annals, 4(1), 109–155.
Lawson, B., & Dorst, K. (2013). Design expertise. New York, NY: Routledge.
Lazega, E., Jourda, M-T., Mounier, L., & Stofer, R. (2008). Catching up with big fish in the big
pond? Multilevel network analysis through linked design. Social Networks, 30(2), 157–
176.
Lazer, D., Pentland, A., Adamic, L., Aral, S., Barabasi, A.L., Brewer, D., . . . Van Alstyne, M.
(2009). Life in the network: The coming age of computational social science. Science,
323, 721–723.
Le Bon, G. (1952). The crowd: A study of the popular mind. London: Ernest Benn.
162
Le Mens, G., Kovács, B., Avrahami, J., & Kareev, Y. (2018). How endogenous crowd formation
undermines the wisdom of the crowd in online ratings. Psychological Science, 29(9),
1475–1490.
Lee, S., & Monge, P. (2011). The coevolution of multiplex communication networks in
organizational communities. Journal of Communication, 61(4), 758–779.
Leonardi, P. M., & Meyer, S. R. (2015). Social media as social lubricant: How ambient
awareness eases knowledge transfer. American Behavioral Scientist, 59(1), 10–34.
Lerner, J., & Lomi, A. (2018). Knowledge categorization affects popularity and quality of
Wikipedia articles. PLoS ONE, 13(1), e0190674.
Levin, D. Z., Walter, J., & Murnighan, J. K. (2011). Dormant ties: The value of reconnecting.
Organization Science, 22(4), 923–939.
Levinthal, D. A., & March, J. G. (1993). The myopia of learning. Strategic Management
Journal, 14(S2), 95–112.
Lessig, L. (1999). Code and other laws of cyberspace. New York, NY: Basic Books.
Li, W., Bruton, G. D., & Filatotchev, I. (2016). Mitigating the dual liability of newness and
foreignness in capital markets: The role of returnee independent directors. Journal of
World Business, 51(5), 781–799.
Liang, H., & Fu, K. W. (2016). Information overload, similarity, and redundancy: Unsubscribing
information sources on Twitter. Journal of Computer-Mediated Communication, 22(1),
1–17.
Light, R., & Odden, C. (2017). Managing the boundaries of taste: Culture, valuation, and
computational social science. Social Forces, 96(2), 877–908.
163
Lippmann, S., & Aldrich, H. E. (2014). History and evolutionary theory. In M. Bucheli & D.
Wadhwani (Eds.), Organizations in time: History, theory, methods (pp. 124–146). New
York, Oxford University Press.
Liu, T. X., Yang, J., Adamic, L. D., & Chen, Y. (2014). Crowdsourcing with all-pay auctions: A
field experiment on Taskcn. Management Science, 60(8), 2020–2037.
Lo, J. Y. C., & Kennedy, M. T. (2015). Approval in nanotechnology patents: Micro and macro
factors that affect reactions to category blending. Organization Science, 26(1), 119–139.
Lomi, A., Robins, G., & Tranmer, M. (2016). Introduction to multilevel social networks. Social
Networks, 44, 255–268.
Lomi, A., & Zollo, M. (2008). Organizational evolution. In S. R. Clegg & J. R. Bailey (Eds.),
International encyclopedia of organizational studies (pp. 1069–1070). Thousand Oaks,
CA: Sage Publications.
Lorenz, J., Rauhut, H., Schweitzer, F., & Helbing, D. (2011). How social influence can
undermine the wisdom of crowd effect. Proceedings of the National Academy of Sciences
of the United States of America, 108(22), 9020–9025.
Lounsbury, M., & Rao, H. (2004). Sources of durability and change in market classifications: A
study of the reconstitution of product categories in the American mutual fund history,
1944–1985. Social Forces, 82(3), 969–999.
Lovas, B., & Ghoshal, S. (2000). Strategy as guided evolution. Strategic Management Journal,
21(9), 875–896.
Lowrey, W. (2017). The emergence and development of news fact-checking sites. Journalism
Studies, 18(3), 376–394.
164
Lowrey, W., & Kim, E. (2016). Hyperlocal news coverage: A population ecology perspective.
Mass Communication and Society, 19(6), 694–714.
Majchrzak, A., & Malhotra, A. (2016). Effect of knowledge-sharing trajectories on innovative
outcomes in temporary online crowds. Information Systems Research, 27(4), 685–703.
Malhotra, A., & Majchrzak, A. (2014). Managing crowds in innovative challenges. California
Management Review, 56(4), 103–123.
Malhotra, A., Majchrzak, A., Kesebi, L., & Looram, S. (2017). Developing innovative solutions
through internal crowdsourcing. MIT Sloan Management Review, 58(4), 73–79.
Malinen, S. (2015). Understanding user participation in online communities: A systematic
literature review of empirical studies. Computers in Human Behavior, 46, 228-238.
March, J. (1991). Exploration and exploitation in organizational learning. Organizational
Science, 2(1), 71–87.
Margolin, D., Lee, S., Shen, C., Weber, M., Monge, P. & Fulk, J. (2015). Normative influences
on network structure in the evolution of the Children’s Rights NGO network, 1977–2004.
Communication Research, 42(1), 30–59.
Margolin, D. B., & Monge, P. (2013). Conceptual retention in epistemic communities:
Examining the role of social structure. In P. Moy (Ed.), Communication and community
(pp. 1–22). New York, NY: Hampton Press.
Marsden, P. V., & Campbell, K. E. (1984). Measuring tie strength. Social Forces, 63(2), 482–
501.
Massa, F. G. (2017). Guardians of the Internet: Building and sustaining the anonymous online
community. Organization Studies, 38(7), 959–988.
165
Maurer, I., & Ebers, M. (2006). Dynamics of social capital and their performance implications:
Lessons from biotechnology start-ups. Administrative Science Quarterly, 51(2), 262–292.
McDowell, W. C., Harris, M. L., & Geho, P. R. (2016). Longevity in small business: The effect
of maturity on strategic focus and business performance. Journal of Business Research,
69(5), 1904–1908.
McPherson, M. (1983). An ecology of affiliation. American Sociological Review, 48(4), 519–
532.
Meng, J., Chung, M., & Cox, J. (2016). Linking network structure to support messages: Effects
of brokerage and closure on received social support. Journal of Communication, 66(6),
982–1006.
Metcalfe, J. S. (2008). Accounting for economic evolution: Fitness and the population method.
Journal of Bioeconomics, 10(1), 23–49.
Meredith, C., Van den Noortgate, W., Struyve, C., Gielen, S., & Kyndt, E. (2017). Information
seeking in secondary schools: A multilevel network approach. Social Networks, 50, 35–
45.
Miller, V. D., Poole, M. S., Seibold, D. R., Myers, K. J., Park, H. S., Monge, P., . . . Shumate, M.
(2011). Advancing research in organizational communication through quantitative
methodology, Management Communication Quarterly, 25(1), 4–58.
Minor, A. S., & Raghavan, S. V. (1999). Interorganizational imitation: A hidden engine of
selection. In J. A. C. Baum & B. McKelvey (Eds.), Variations in organizations: In honor
of Donald T. Campbell (pp. 35–62). Thousand Oaks, CA: Sage Publications.
Moliterno, T. P., & Mahony, D. M. (2011). Network theory of organizations: A Multilevel
approach. Journal of Management, 37(2), 443–467.
166
Monge, P. R., & Contractor, N. S. (2003). Theories of communication networks. New York, NY:
Oxford University Press.
Monge, P., Heiss, B. M., & Margolin, D. B. (2008). Communication network evolution in
organizational communities. Communication Theory, 18(4), 449–477.
Monge, P., Lee, S., Fulk, J., Weber, M., Shen, C., Schultz, C., . . . Frank, L. B. (2011). Research
methods for studying evolutionary and ecological processes in organizational
communication. Management Communication Quarterly, 25(2), 211–251.
Monge, P., & Oh, P. (2016). Evolutionary theory and models. In K. B. Jensen & R. Craig (Eds.),
International encyclopedia of communication theory and philosophy (pp. 661–670).
Malden, MA: John Wiley & Sons.
Monge, P., & Poole, S. (2008). The evolution of organizational communication. Journal of
Communication, 58(4), 679–692.
Mukherjee, A., Xiao, P., Wang, L., & Contractor, N. (2018). Does the opinion of the crowd
predict commercial success? Evidence from Threadless. Academy of Management
Proceedings.
Murphy, G. L. (2002). The big book of concepts. Cambridge: MIT Press.
Nanay, B. (2002). The return of the replicator: What is philosophically significant in a general
account of replications and selection. Biology and Philosophy, 17(1), 109–121.
Navis, C., & Glynn, M. A. (2010). How new market categories emerge: Temporal dynamics of
legitimacy, identity, and entrepreneurship in satellite radio, 1990–2005. Administrative
Science Quarterly, 55(3), 439–471.
Negro, G., Hannan, M. T., & Fassiotto, M. (2015). Category signaling and reputation.
Organization Science, 26(2), 584–600.
167
Negro, G., Hannan, M. T., & Rao, H. (2010). Categorical contrast and audience appeal: Niche
width and critical success in winemaking. Industrial and Corporate Change, 19(5),
1397–1425.
Negro, G., Koçak, Ö., & Hsu, G. (2010). Categories in markets: Origins and evolution. Research
in the sociology of organizations, 35, 3–35.
Negro, G., & Leung, M. D. (2013). “Actual” and perceptual effects of category spanning.
Organization Science, 24(3), 684–696.
Nelson, R. R., & Winter, S. G. (1982). An evolutionary theory of economic change. Cambridge,
MA: Harvard University Press.
Nicholson, N., & Narayanan, J. (2008). Evolutionary theory. In S. R. Clegg & J. R. Bailey
(Eds.), International encyclopedia of organizational studies (pp. 480–484). Thousand
Oaks, CA: Sage Publications.
Nishikawa, H., Schreier, M., Fuchs, C., & Ogawa, S. (2017). The value of marketing
crowdsourced new products as such: Evidence from two randomized field
experiments. Journal of Marketing Research, 54(4), 525–539.
Norris, P. (2002). The bridging and bonding role of online communities (Editorial). Harvard
International Journal of Press/Politics, 7(3), 3–13.
Ocasio, W., & Joseph, J. (2005). Cultural adaptation and institutional change: The evolution of
vocabularies of corporate governance, 1972-2003. Poetics, 33(3–4), 163–178.
O’Connor K., Carroll, G. R., & Kovács, B. (2017). Disambiguating authenticity: Interpretations
of value and appeal. PLoS ONE, 12(6), e0179187.
168
Ognyanova, K., & Monge, P. (2013). A multitheoretical, multilevel, multidimensional network
model of the media system: Production, content, and audiences. Annal of the
International Communication Association, 37, 67–93.
Oh, P., & Monge, P. (2013). The functionality of social tagging as a communication system.
International Journal of Communication, 7, 653–680.
Okoli, C., & Oh, W. (2007). Investigating recognition-based performance in an open content
community: A social perspective. Information & Management, 44(3), 240–252.
Paolella, L., & Durand, R. (2016). Category spanning, evaluation, and performance: Revised
theory and test on the corporate law market. Academy of Management Journal, 59(1),
330–351.
Patison, K. P., Quintane, E., Swain, D. L., Robins, G., & Pattison, P. (2015). Time is of the
essence: an application of a relational event model for animal social networks.
Behavioral Ecology and Sociobiology, 69(5), 841–855.
Peng, T.-Q., Liu, M., Wu, Y., & Liu, S. (2016). Follower-followee network, communication
networks, and vote agreement of the U.S. members of Congress. Communication
Research, 43(7), 996–1024.
Piezunka, H., & Dahlander, L. (2015). Distant search, narrow attention: How crowding alters
organizations’ filtering of suggestions in crowdsourcing. Academy of Management
Journal, 58(3), 856–880.
Pilny, A., & Atouba, Y. (2018). Modeling valued organizational communication networks using
exponential random graph models. Management Communication Quarterly, 32(2), 250–
264.
169
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.
Podolny, J. M. (1994). Market uncertainty and the social character of economic exchange.
Administrative Science Quarterly, 39(3), 458–483.
Podolny, J. M. (2001). Networks as the pipes and prisms of the market. American Journal of
Sociology, 107(1), 33–60.
Pontikes, E. G. (2012). Two sides of the same coin: How ambiguous classification affects
multiple audiences’ evaluations. Administrative Science Quarterly, 57(1), 81–118.
Pontikes, E. G., & Barnett, W. P. (2015). The persistence of lenient market categories.
Organization Science, 26(5), 1415c1431.
Pontikes, E. G., & Hannan, M. T. (2014). An ecology of social categories. Sociological Science,
1, 311–343.
Pontikes, E. G., & Kim, R. (2017). Strategic categorization. Research in the Sociology of
Organizations, 51, 71–111.
Preece, J., & Maloney-Krichmar, D. (2005). Online communities: Design, theory, and practice.
Journal of Computer-Mediated Communication, 10(4).
Quan-Haase, A., Mo, G. Y., & Wellman, B. (2017). Connected seniors: How older adults in East
York exchange social support online and offline. Information, Communication & Society,
20(7), 967–983.
Rainie, L., & Wellman, B. (2012). Networked: The new social operating system. Cambridge,
MA: The MIT Press.
170
Rao, H. (2002). Interorganizational ecology. In J. A. C. Baum (Ed.), The Blackwell companion to
organizations (pp. 541–556). Malden, MA: Blackwell.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data
analysis methods (2nd ed.). Thousand Oaks, CA: Sage.
Rhee, M., & Kim, T. (2015). Great vessels take a long time to mature: Early success traps and
competences in exploitation and exploration. Organization Science, 26(1),180–197.
Rheingold, H. (2000). The virtual community: Homesteading on the electronic frontier.
Cambridge, MA: MIT Press.
Riedl, C., & Seidel, V. P. (2018). Learning from mixed signals in online innovation
communities. Organization Science, 29(6), 1010–1032.
Rogers, E. (1994). A history of communication study: A biographical research. New York, NY:
The Free Press.
Rosch, E., & Lloyd, B. B. (1978). Cognition and categorization. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Rose, N. (1998). Controversies in meme theory. Journal of Memetics - Evolutionary Models of
Information Transmission, 2. http://www.cpm.mmu.ac.uk/jom-
emit/1998/vol2/rose_n.htm.
Roy, D. (2017). Myths about memes. Journal of Bioeconomics, 19(3), 281–305.
Rufas, A., & Hine, C. (2018). Everyday connections between online and offline: Imagining
others and constructing community through local online initiatives. New Media &
Society, 20(10), 3879–3897.
Sargent, M., Clark, J. A., Monge, P., & Fulk, J. (2018). The impacts of spanning implicit
categories in online markets: Mapping stylistic categories from keywords. Paper
171
presented at the 68
th
annual conference of the International Communication Association,
Prague, Czech Republic.
Schemmann, B., Herrmann, A. M., Chappin, M. M. H., & Heimeriks, G. J. (2016).
Crowdsourcing ideas: Involving ordinary users in the ideation phase of new product
development. Research Policy, 45(6), 1145-1154.
Scott, W. R. (2004). Reflections on a half-century of organizational sociology. Annual Review of
Sociology, 30, 1–21.
Sessions, L. F. (2010). How offline gatherings affect online communities: When virtual
community members ‘meetup’. Information, Communication & Society, 13(3), 375–395.
Shaw, A., & Hill, B. M. (2014). Laboratories of oligarchy? How the iron law extends to peer
production. Journal of Communication, 64(2), 215–238.
Shah, D. V., Cappella, J. N., & Neuman, W. R. (2015). Big Data, Digital Media, and
Computational Social Science: Possibilities and Perils. The ANNALS of the American
Academy of Political and Social Science, 659(1), 6–13.
Shen, C., & Cage, C. (2013). Exodus to the real world? Assessing the impact of offline meetups
on community participation and social capital. New Media & Society, 17(3), 394–414.
Shen, C., Monge, P. & Williams, D. (2014). The evolution of social ties online: A longitudinal
study of a massively multiplayer online game. Journal of the American Society for
Information Science and Technology (JASIST), 65(10), 2127–2137.
Shukla, P., & Drennan, J. (2017). Interactive effects of individual- and group-level variables on
virtual purchase behavior in online communities. Information & Management, 55(5),
598–607.
172
Shumate, M. (2012). The evolution of the HIV/AIDS NGO hyperlink network. Journal of
Computer-Mediated Communication, 17(2), 120–134.
Shumate, M., Fulk, J., & Monge, P. (2005). Predictors of the international HIV-AIDS INGO
network over time. Human Communication Research, 31(4), 482–510.
Shumate, M., & Contractor, N. (2013). The emergence of multidimensional social networks. In
L. L. Putnam & D. K. Mumby (Eds.), The SAGE handbook of organizational
communication (pp. 449–474). Thousand Oaks, CA: Sage.
Shumate, M., Pilny, A., Atouba, Y. C., Kim, J., Peña-y-Lillo, M., Cooper, K. R., . . . Yang, S.
(2013). A taxonomy of communication networks. Annals of the International
Communication Association, 37(1), 95–123.
Simmons, S. P., Nelson, L. D., Galak, J., & Frederick, S. (2011). Intuitive biases in choice versus
estimation: Implications for the wisdom of crowds. Journal of Consumer Research,
38(1), 1–15.
Simonton, D. K. (1999). Creativity as blind variation and selective retention: Is the creative
process Darwinian. Psychological Inquiry, 10(4), 309–328.
Sio, U. N., Kotovsky, K., & Cagan, J. (2015). Fixation or inspiration? A meta-analytic review of
the role of examples on design processes. Design Studies, 39, 70–99.
Skinner, M. J. (2015). Environmental epigenetics and a unified theory of the molecular aspects
of evolution: A Neo-Lamarckian concept that facilitates neo-Darwinian evolution.
Genetic Biology and Evolution, 7(5), 1296–1302.
Smith, E. B. (2011). Identities as lenses: How organizational identity affects audiences'
evaluation of organizational performance. Administrative Science Quarterly, 56(1), 61–
94.
173
Snijders, T. A. B. (2001). The statistical evaluations of social network dynamics. Sociological
Methodology, 31(1), 361–395.
Snijders, T. A. B. (2005). Models for longitudinal network data. In P. Carrington, J. Scott, & S.
Wasserman (Eds.), Models and methods in social network analysis (pp. 215–247). New
York, NY: Cambridge University Press.
Song, H. (2018). A primer on multilevel mediation models for egocentric social network data.
Communication Methods and Measures, 12(1), 1–24.
Sosa, M. L. (2013). Corporate structure, indirect bankruptcy costs, and the advantage of de novo
firms: The case of gene therapy research. Organizational Science, 25(3), 850–867.
Sperber, D. (2000). An objection to the memetic approach to culture. In R. Aunger (Ed.),
Darwinizing culture: The status of mimetics as a science (pp. 163–173). Oxford: Oxford
University Press.
Srivastava, S. B. (2015). Interorganizational network dynamics in times of ambiguity.
Organizational Science, 26(5), 1365–1380.
Stefanone, M. A., Kwon, K. H., & Lackaff, D. (2012). Exploring the relationship between
perceptions of social capital and enacted support online. Journal of Computer-Mediated
Communication, 17(4), 451–466.
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.
Stephens, B., Chen, W., & Butler, J. S. (2016). Bubbling up the good ideas: A two-mode
network analysis of an intra-organizational idea challenge. Journal of Computer-
Mediated Communication, 21(3), 210–229.
174
Stinchcombe, A. L. (1965). Social structure and organizations. In J. G. March (Ed.), Handbook
of organizations (pp. 142–193). Chicago, IL: Rand McNally.
Stohl, C. (2014). Crowds, clouds, and community. Journal of Communication, 64(1), 1–19.
Stretesky, P. B., Huss, S., & Lynch, M. J. (2012). Density dependence and environmental justice
organizations, 1970–2008. The Social Science Journal, 49(3), 343–351.
Stuart, T. E., Hoang, H., & Hybels, R. C. (1999). Interorganizational endorsements and the
performance of entrepreneurial ventures. Administrative Science Quarterly, 44(2), 315–
349.
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.
Surowiecki, J. (2005). The wisdom of crowds. New York, NY: Anchor Books.
TeBlunthuis, N., Shaw, A., & Hill, B. W. (2017). Density dependence without resource
partitioning: Population ecology on Change. org. In Companion of the 2017 ACM
Conference on Computer Supported Cooperative Work and Social Computing (pp. 323–
326). New York, NY: ACM Press.
Treem, J. W. (2012). Communicating expertise: Knowledge performances in professional-
service firms. Communication Monographs, 79(1), 23–47.
Treem, J. W. & Leonardi, P. M. (2016). Expertise, communication, and organizing. New York,
NY: Oxford University Press.
175
Treem, J. W. & Leonardi, P. M. (2017). Recognizing expertise: Factors promoting congruity
between individuals’ perceptions of their own expertise and the perceptions of their
coworkers. Communication Research, 44(2), 198–224.
Überbacher, F. (2014). Legitimation of new ventures: A review and research programme.
Journal of Management Studies, 51(4), 667–698.
Usher, J. M., & Evans, M. G. (1996). Life and death along gasoline alley: Darwinian and
Lamarckian processes in a differentiating population. The Academy of Management
Journal, 39(5), 1428–1466.
Vanberg, V. J. (2013). Darwinian paradigm, cultural evolution and human purposes: On F.A.
Hayek’s evolutionary view of the market. Journal of Evolutionary Economics, 24(1), 35–
57.
Van de Ven, A.H., & Garud, R. (1994). The coevolution of technical and institutional events in
the development of an innovation. In J. A. C. Baum & J. V. Singh (Eds.), Evolutionary
dynamics of organizations. New York, NY: Oxford University Press.
van den Hooff, B., & Kotlarsky, J. (2016). Managing dispersed and dynamic expertise in fluid
organizational forms. In J. W. Treem & P. M. Leonardi (Eds.), Expertise,
communication, and organizing (pp. 232–250). New York, NY: Oxford University Press.
Vaskelainen, T., & Piscicelli, L. (2018). Online and offline communities in the sharing economy.
Sustainability, 10(8), 2927.
Vergne, J. P. (2012). Stigmatized categories and public disapproval of organizations: A mixed-
methods study of the global arms industry, 1996–2007. Academy of Management
Journal, 55(5), 1027–1052.
176
Vergne, J. P., & Wry, T. (2014). Categorizing categorization research: Review, integration, and
future directions. Journal of Management Studies, 51(1), 56–94.
Vu, D., Pattison, P., & Robins, G. (2015). Relational event models for social learning in
MOOCs. Social Networks, 43, 121–135.
Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of
an industry network. Organization Science, 8(2), 109–125.
Walther, J. B., & Parks, M. R. (2002). Cues filtered out, cues filtered in: Computer-mediated
communication and relationships. In M. L. Knapp & J. A. Daly (Eds.), The handbook of
interpersonal communication (3rd ed., pp. 529–563). Tousand Oaks, CA: Sage.
Wang, P., Robins, G., Pattison, P., & Koskinen, J. (2014). Program for the simulation and
estimation of (p*) exponential random graph models for multilevel networks. Retrieved
from
https://static1.squarespace.com/static/57a1436215d5dbbcd2031828/t/581ac475e58c6
2432be83acf/1478149242644/MPNetManual.pdf.
Wang, P., Robins, G. L., Pattison, P., & Lazega, E. (2013). Exponential random graph models
for multilevel networks. Social Networks, 35(1), 96–115.
Wang, P., Robins, G., Pattison, P., & Lazega, E. (2016). Social selection models for multilevel
networks. Social Networks, 44, 346–362.
Wang, P., Robins, G., Matous, P. (2016). Multilevel network analysis using ERGM and its
extension. In E. Lazega & T.A.B. Snijders (Eds.), Multilevel network analysis for the
social sciences: Theory, methods and applications (pp. 125–143). Cham, Germany:
Springer.
177
Weber, M. S. (2012). Newspapers and the long-term implications of hyperlinking. Journal of
Computer-Mediated Communication, 17(2), 187–201.
Weber, M. S. (2018). Methods and approaches to using web archives in computational
communication research. Communication Methods and Measures, 12(2–3), 200–215.
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.
Weber, M. S., & Monge, P. R. (2017). Industries in turmoil: Driving transformation during
periods of disruption. Communication Research, 44(2), 147–176.
Weber, M. S., Ognyanova, K., & Kosterich, A. (2017). Imitation in the quest to survive: Lessons
from news media on the early web. International Journal of Communication, 11, 5068–
5092.
Welbers, K., Van Atteveldt, W., & Benoit, K. (2017). Text analysis in R. Communication
Methods and Measures, 11(4), 245–265.
Welles, B. F., & Contractor, N. (2015). Individual motivations and network effects: A multilevel
analysis of the structure of online social relationships. The ANNALS of the American
Academy of Political and Social Science, 659(1), 180–190.
Wellman, B., Quan-Haase, A., Witte, J., & Hampton, K. (2001). Does the Internet increase,
decrease, or supplement social capital? Social networks, participation, and community
commitment. American Behavioral Scientist, 45(3), 436–455.
Williams, D. (2007). The impact of time online: Social capital and cyberbalkanization.
CyberPsychology & Behavior, 10(3), 398–406.
Williams, R. L. (2000). A note on robust variance estimation for cluster-correlated data.
Biometrics, 56, 645–646.
178
Wry, T., & Castor, A. R. (2017). Opportunity, status, and similarity: Exploring the varied
antecedents and outcomes of category spanning innovation. Research in the
Sociology of Organizations, 51, 355–389.
Wry, T., Lounsbury, M., & Jennings, P. D. (2014). Hybrid vigor: Securing venture capital by
spanning categories in nanotechnology. Academy of Management Journal, 57(5), 1309–
1333.
Xu, B., Huang, Y., Kwak, H., & Contractor, N. (2013). Structures of broken ties: Exploring
unfollow behavior on twitter. In Proceedings of the 2013 ACM Conference on Computer
Supported Cooperated Work (pp. 871–876).
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, 1612–1637.
Xu, Y., Sun, Y., & Cruz, I. (May, 2018). Finding the path to success: An examination of
expertise, exploration and exploitation, and creativity performance in a crowdsourcing
platform. Paper presented at the 68
th
annual conference of the International
Communication Association, Prague, Czech Republic.
Yang, T., & Aldrich, H. E. (2017). “The liability of newness” revisited: Theoretical restatement
and empirical testing in emergent organizations. Social Science Research, 63, 36–53.
Yang, X., Li, G., & Huang, S. S. (2017). Perceived online community support, member relations,
and commitment: Differences between posters and lurkers. Information & Management,
54(2), 154–165.
179
Yue, L. Q. (2012). Asymmetric effects of fashions on the formation and dissolution of networks:
Board interlocks with internet companies, 1996–2006. Organizational Science, 23(4),
1114–1134.
Zappa, P., & Lomi, A. (2015). The analysis of multilevel networks in organizations: Models and
empirical test. Organizational Research Methods, 18(3), 542–569.
Zappa, P., & Robins, G. (2016). Organizational learning across multi-level networks. Social
Networks, 44, 295–306.
Zhang, W., & White, S. (2016). Overcoming the liability of newness: Entrepreneurial action and
the emergence of China’s private solar photovoltaic firms. Research Policy, 45(3), 604–
617.
Zhao, Y., & Zhu, Q. (2014). Evaluation on crowdsourcing research: Current status and future
direction. Information Systems Frontiers, 16(3), 417–434.
Zogaj, S., Bretschneider, U., & Leimeister, J. M. (2014). Managing crowdsourced software
testing: A case study based insight on the challenges of a crowdsourcing intermediary.
Journal of Business Economics, 84(3), 375–405.
Zollo, M., & Winter, S. G. (2002). Deliberate learning and the evolution of dynamic capabilities.
Organization Science, 13(3), 339–351.
Zuckerman, E. W. (1999). The categorical imperative: Securities analysts and the illegitimacy
discount. American Journal of Sociology, 104(5), 1398–1438.
Zuckerman, E. W. (2017). The categorical imperative revisited: Implications of categorization as
a theoretical tool. Research in the Sociology of Organizations, 51, 31–68.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Ecology and network evolution in online innovation contest crowdsourcing

PDF

The evolution of multilevel organizational networks in an online gaming community

PDF

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

PDF

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

PDF

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

PDF

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

PDF

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

PDF

The evolution of scientific collaboration networks

PDF

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

PDF

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

PDF

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

PDF

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

PDF

Crisis and stasis: understanding change in online 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

How social and human capital create financial capital in crowdfunding projects

PDF

Bounded technological rationality: the intersection between artificial intelligence, cognition, and environment and its effects on decision-making

PDF

The Silicon Valley startup ecosystem in the 21st century: entrepreneurial resilience and networked innovation

PDF

Building personal wellness communities: meaningful play in the everyday life of a network society

PDF

Staying ahead of the digital tsunami: strategy, innovation and change in public media organizations

PDF

Civic games with 'local fit': embedding with real‐world neighborhoods and place‐based networks

Asset Metadata

Creator Xu, Yu (author)

Core Title The evolution of multidimensional and multilevel networks in online crowdsourcing

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Publication Date 07/26/2019

Defense Date 05/06/2019

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

Tag categories,crowdsourcing,evolutionary theories,multilevel,networks,OAI-PMH Harvest,organizational communication,organizational ecology

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Monge, Peter (committee chair), Fulk, Janet (committee member), Williams, Dmitri (committee member)

Creator Email xuyu@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-194184

Unique identifier UC11663243

Identifier etd-XuYu-7635.pdf (filename),usctheses-c89-194184 (legacy record id)

Legacy Identifier etd-XuYu-7635.pdf

Dmrecord 194184

Document Type Dissertation

Format application/pdf (imt)

Rights Xu, Yu

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 a...

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