University of Southern California Dissertations and Theses

The evolution of multilevel organizational networks in an online gaming community

(USC Thesis Other)

The evolution of multilevel organizational networks in an online gaming community

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

THE EVOLUTION OF MULTILEVEL ORGANIZATIONAL NETWORKS IN AN
ONLINE GAMING COMMUNITY

by

Jingyi Sun

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

August 2021

Copyright 2021 Jingyi Sun
ii
Acknowledgements
I would like to express my sincerest gratitude to my advisor, Dr. Dmitri Williams, who
has given me invaluable guidance, support, mentoring and encouragement along the way. I am
especially grateful for Dr. Williams’ advice on making research meaningful for broader
audiences, which I believe will greatly benefit my future academic career.
This dissertation would also be impossible without the two excellent members of my
committee, Dr. Alessandro Lomi and Dr. Peter Monge. I am deeply indebted to Dr. Lomi, who
introduced me to social networks and organizational studies. Over the years, I have been
following Dr. Lomi’s intellectual footprints and have learned tremendously from his scholarship.
My deepest gratitude goes to Dr. Monge, who has given me enormous guidance since my
first day at the Annenberg school. Dr. Monge introduced me to statistics, empirical research, and
rigorous scholarship. It is his advice that has helped me find the research area that I truly love.
I am especially fortunate to have Dr. Janet Fulk and Dr. Aimei Yang as my quals
committee members and coauthors. It is during the many rounds of revision with Dr. Fulk on our
project that I have really learned the essence of academic writing. I have collaborated with Dr.
Yang on multiple projects, and hopefully we will write many more papers together in the future.
My thanks also go to my fellow doctoral friends at ANN and TGG, including Dr. Yu Xu,
Dr. Larry Xu, Dr. Yao Sun, Dr. Ruqin Ren, Dr. Emily Sidnam, Dr. Yiqi Li, Hye Min Kim,
Sukyoung Choi, and Mingxuan Liu. I also want to thank Eugene Kislyi and Jeremy Ballenger
and Wargaming for providing data access and continuous support.
Finally, I want to thank my dearest parents for their unconditional love. Thanks for
always being there for me, having faith in me, and giving me comfort and encouragement. I
wouldn’t be where I am today without you.
iii
Table of Contents
Acknowledgements ..................................................................................................................... ii

List of Tables ...............................................................................................................................v

List of Figures ............................................................................................................................ vi

Abstract ..................................................................................................................................... vii

Chapter 1: Introduction ................................................................................................................1
Purpose of the Study ............................................................................................................1
Summary of Studies .............................................................................................................6

Chapter 2: Literature Overview .................................................................................................11
Ecology and Evolutionary Theories...................................................................................11
Basic Concepts of Ecology Theories .........................................................................11
The Ecology of Categories .........................................................................................14
The Social Approach of Categorization .....................................................................18
The Evolution of Multilevel Organizational Networks .....................................................22
The Evolution of Networks ........................................................................................22
Organizational Networks from a Multilevel Perspective ...........................................24
Social Boundaries in Online Communities ........................................................................30
Ecology in Online Communities ................................................................................30
Common Bond and Common Identity .......................................................................35

Chapter 3: Research Context .....................................................................................................39
The Social Architecture of Gaming Communities .............................................................39
The Research Context: World of Tanks .....................................................................43
One Day as a Clan Member .......................................................................................45
An Organizing Context with Generalizability ...........................................................49
The Clan Community as an Ecological System .................................................................51
Clans Categories—A Semantic Analysis of Clan Profiles ................................................56

Chapter 4: Hypotheses Development.........................................................................................64
Study 1: The Survival of Clans ..........................................................................................64
Organizational mortality ............................................................................................65
The Classical and Revised Theory of Niche Width ...................................................67
The Theory of Density Dependence and Category Contrast .....................................74
Network Brokerage from a Collective View .............................................................81
Study 2: The Formation of Individual-Group Multilevel Networks ..................................86
The Theory of Niche Overlap ....................................................................................90
Affiliation-based Closure - Boundary Constraint ......................................................92
Affiliation-based Closure - Boundary Spanners ........................................................95
Cross-level Alignment ...............................................................................................98
Cross-level Assortativity ..........................................................................................100
Study 3: Diversification or Focus? An Agency Perspective ............................................102
iv
Personality and Networks ........................................................................................104
Multirole Network Personality ................................................................................106
Agentic Categorization ............................................................................................107

Chapter 5: Data and Method ....................................................................................................110
Data for Study 1 ...............................................................................................................110
Data for Study 2 ...............................................................................................................112
Data for Study 3 ...............................................................................................................117

Chapter 6: Results ....................................................................................................................128
Results for Study 1 ...........................................................................................................128
Results for Study 2 ...........................................................................................................132
Results for Study 3 ...........................................................................................................137

Chapter 7: Discussion ..............................................................................................................141
Summary of Findings .......................................................................................................141
Theoretical Contribution ..................................................................................................148
Practical Implication ........................................................................................................153
Limitations and Future Research .....................................................................................157

References ................................................................................................................................164

v
List of Tables
Table 1. Autoregressive moving average (ARMA) diagnostics of clan founding rate
(2016-2020). 55
Table 2. Negative binomial times series model predicting the founding rate of clans. 55
Table 3. Ten topics and associated keywords and most probable clan profile. 60
Table 4. Descriptive statistics for Cox’s proportional hazards modeling predicting the clan
mortality. 112
Table 5. Descriptive network statistics of five samples for MERGM. 115
Table 6. Visual illustration of MERGM parameters (Adapted from MPnet manual). 115
Table 7. Mixed effects linear regression models of roles predicting win rate and network
patterns (compared to quartermaster). 119
Table 8. Correlation statistics of mixed effects model predicting player’s win rate. 120
Table 9. Mixed effects linear regression model predicting players’ win rate. 121
Table 10. Comparison of roles held by 29.6% removed players and 70.4% kept players. 122
Table 11. Mixed effects regression models of network personality predicting individual
role performance. 124
Table 12. Results for Cox’s proportional hazards modeling predicting clan mortality. 131
Table 13. Estimates of MERGMs for the multilevel networks of players and clans across
four random samples. 133
Table 14. Goodness of fit for MERGMs for the multilevel networks of players and clans
across four random samples. 134
Table 15. Estimates of MERGMs for the multilevel networks of small clans. 136
Table 16. Model selection based on goodness of fit (GoF) statistics. 138
Table 17. Relational Event Model (REM) of multirole network personality to predict the
event of players joining clans. 139

vi
List of Figures
Figure 1. Hypothetical multilevel networks. 6
Figure 2. Founding rate of clans from 2016-07 to 2020-12. 53
Figure 3. Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF)
of clan founding time series. 54
Figure 4. Optimal topic number detection of structural topic modeling. 58
Figure 5. Visualization of the correlation matrix of ten topics using the stm build-in
function. 59
Figure 6. Kaplan-Meier survival estimate of clan mortality. 128
Figure 7. Model checking results based on Schoenfeld residuals. 139

vii
Abstract
This dissertation adopts a multilevel network perspective that integrates social categories and
social networks to examine the dynamics in an online gaming community, which are
characterized by both interpersonal and individual-group interaction. The first study examines
how the mortality of virtual groups is driven by groups’ niche occupation. From a multilevel
perspective, such an effect is contingent upon group boundaries and within-group networks. The
second study examines the micro-mechanisms of multilevel networks at three levels, including
an inter-group niche overlap network, an individual-clan affiliation network, and a collaborative
network among individuals. This study confirms the higher likelihood of within-boundary ties
for large groups. This study also shows that inter-group niche overlap facilitates boundary
spanning, but too much overlap erodes group boundaries and reduces a group’s capacity to
attract popular individuals. The third study shows that brokerage prone individuals across roles
are more likely to span group boundaries. That is, given that categorization and network building
both differentiate between diversification and concentration of resource allocation, an individual
follows a consistent pattern in building interpersonal networks and joining groups. These studies
contribute to a holistic perspective of organizing. Virtual group boundaries are not predetermined
but collectively constructed, and dynamically coevolve with interpersonal networks. Sharp
boundaries of virtual groups can benefit group fitness and structure the formation of
interpersonal networks.

Keywords: network evolution, category, multilevel networks, organizational networks,
individual-group interaction, massively multiplayer online game (MMOG).
1

Chapter 1: Introduction
Purpose of the Study
The recent years have witnessed a growing trend of computer-supported collaboration in
various online communities, including collective contribution to open-source software projects
(Lerner & Tirole, 2001) peer production of knowledge by leveraging wisdom of crowds (Arazy
et al., 2006), team-based battles in virtual worlds (Shen et al., 2020). The common theme of
these virtual communities is the completion of some tasks by an unsupervised crowd mediated
by digital technology. The role of organizing dynamics has attracted substantial research
attention. These virtual communities follow the rules set by program codes (Lessig, 2009), and
follow the social rules similar to offline human organizing (Vesa et al., 2017). How could an
unsupervised crowd in the absence of centralized coordination and authority, together complete
some tasks, e.g., production tasks and team-based competitive tasks for higher rankings? How do
individuals voluntarily form virtual groups with socially codified hierarchy to organize the
collaborative process more efficiently and what makes such groups successful?
The ecology framework is useful for examining complex organizing systems at the
collective level. As an essential part of the ecology framework, the theory of niche focuses on
how entities respond to their environment by occupying resource positions, or niche categories,
for survival and growth (Hannan & Freeman, 1984). Networks, as channels for acquiring and
exchanging resources, play an essential role in the ecology framework (Monge et al., 2008). The
interconnectedness between social categories and network dynamics manifests the long-held
debate and integration on the conceptualization of social space as patterns of categorical
affiliation (Breiger, 1974) as well as social networks (White et al., 1976).
2

Two theoretical trends are essential for examining virtual group organizing. First, the
recent development of niche theory takes into account fuzzy boundaries of niche categories and
proposes that membership in categories can be partial and graded (Durand et al., 2017; Hannan
et al., 2007). This extension indicates a tradeoff between breadth and depth in niche membership
and success. That is, some entities occupy a diverse range of niche categories, but this resource
allocation pattern reduces their capability to perform well in any of them in some cases (Hsu,
2006). With this theoretical revision, the theory of niche has the flexibility to examine social
categories in online communities, which usually have permeable and dynamic boundaries and
are subject to constant collective construction and changing interpretations.
Second, the social perspective posits category boundaries not as given or static but
dynamically negotiated by social actors, such that actors and social categories are embedded
within a larger system and such interdependencies are likely to influence categorization
behaviors and consequences (Durand et al., 2017; Glynn & Navis, 2013; Zuckerman, 2017). In
other words, actors’ memberships to categories are influenced by inter-actor relations and inter-
category relations. For instance, in Wry and Castor's (2017) study of research scientists and their
patent categories, scientists follow their peers of higher status as role models to span patent
categories, and they are more likely to span to similar patent categories. In Boda's (2019) study
on how friends perceive each other’s ethnic groups, ethnic categorization and friendship seeking
coevolve together.
This dissertation proposes a multilevel network framework of organizing that seeks to
understand such complex interdependencies. The multilevel network approach considers how
individual and collective agency are enabled and constrained by category boundaries and
relational hierarchy (Lazega, 2013, 2016). Within such nested multilevel networks, there are
3

cross-level bipartite multimodal networks, which involve two different types of nodes and
relations only connecting different node types (Contractor, 2009), and there are within-level one-
mode networks, which involve only one type of nodes (Monge & Contractor, 2003).
This dissertation makes two propositions regarding such nested multilevel networks.
First, categories are embedded in a hierarchical system, so there could be more than one form of
category memberships (Hannan et al., 2019). For example, attorneys are employed by laws
firms, and law firms operate in different legal specialties areas. Second, category memberships
can be reframed as a bipartite affiliation network, where a partial membership to a higher-level
category means a weighted tie in network representation. This is essentially a structural
translation of duality linking one type of social structure to another type at multiple levels
(Breiger, 1974; Mützel & Breiger, 2020). Such reframing allows for considering how category
memberships influence or are influenced by embeddedness at different levels. Following the law
firm example, law firms with multiple legal specialties are evaluated more positively because
they are viewed as capable of handling complex problems (Paolella & Durand, 2016). How well
law firms can perform in these specialty areas depends on their attorneys, who change jobs and
build interpersonal networks.
The theoretical foundation of this multilevel perspective can be traced back to hierarchic
complex systems composed of interrelated subsystems (Simon, 2019). A common type of social
hierarchy is a formal organization that consists of multiple levels of clustered relations
(Paruchuri et al., 2019). This perspective also accords with Padgett and Powell's (2012) concept
of emergence, which posits that “emergence of organizations is grounded in the transformation
of networks” (p.7). Whether actors create relations or relations create actors is a matter of time
scale, because in the long term new organizations emerge from multiple intertwined networks
4

that are contexts for each other (Padgett & Powell, 2012). Taken together, this multilevel
network perspective provides a conceptual anchor to consider various types of interdependencies
and the coevolution of different systems.
Applied to virtual group organizing, groups are not homogenous entities but consist of
heterogenous individuals. Similar-minded individuals form group categories, which are special
social categories defined by a complex series of factors including gender, ethnicity, skills,
interests, values (T. T. Postmes & Spears, 2000). These individuals build interpersonal networks
within and outside their groups, and they can also quit and join other groups. These groups are
further categorized into broader types that are collectively constructed by communities through
sensemaking and sensegiving. Represented in multilevel networks (Figure 1), there are cross-
level ties between organizations and niches, and cross-level ties between individuals and
organizations. There are within-level ties among individuals, among organizations, and among
niche categories. Cross-level ties can also be projected to within-level ties. The theory of niche is
applicable for understanding the relation between individuals and organizations, and the relation
between organizations and organizational categories because it generally examines entities and
the resources they consume (Hannan et al., 2007). The theory of niche basically reflects the
duality principle that applies to a variety of clustered relations (Breiger, 1974; Mützel & Breiger,
2020). Guided by this multilevel network structure, this dissertation seeks to understand how
social boundaries are reinforced, negotiated, and eroded through the dynamic evolution of
networks.
The empirical research setting is World of Tanks (WoT), a massively multiplayer online
game (MMOG) as a form of online community (Malinen, 2015). Video games have become an
increasingly prevalent form of digital communication. The latest study shows that 64% of
5

American adults play video games and 75% U.S. households have at least one player (ESA,
2020). In 2020, the game industry generated $90.3 billion to U.S. economy. Thus, online gaming
is a huge digital phenomenon worthy of studying in its own right. Additionally, large-scale game
communities are natural laboratories for informing the multilevel perspective of organizing,
particularly because the social architecture usually incentivizes competition and collaboration for
experience, skills and ranking.
Therefore, players are motivated to access resources through building interpersonal
networks and joining in-game organizations (Ducheneaut, 2009). In the current context, players
have to team up with others in battles, and better teamwork usually means a better chance of
victory. Players can also join clans, which are in-game organizations with formal leadership
structures, in order to access resources such as skill development and social coordination
(Williams et al., 2006). As argued by many game scholars, the collective action between clan
recruiters and clan seekers gives rise to two broad categories of clans: competitive clans and
social clans (Snodgrass et al., 2017; Vesa et al., 2017; Williams et al., 2006). Clans are
positioned on a spectrum ranging from very social to very competitive. Such categorization
organizes players’ expectations and motivations for joining clans (Williams et al., 2006). In other
words, individuals’ niche affiliation is not directly manifested but through their interactions with
interest groups. Clans and clan categories constitute a hierarchical categorization system,
because clans are more fine-grained categories to organize players’ interests, mindsets and skill
levels given the broad competitive and social categorization. Moreover, groups as a special type
of social categories are also defined by other factors including values, nationality, geographical
areas, gender, ethnicity (T. T. Postmes & Spears, 2000). Therefore, for individuals, clans make
their niches; for clans, they are categorized into competitive and social clans.
6

Figure 1. Hypothetical multilevel networks.

In the language of multilevel networks, the three-level networks of players, clans and
clan categories are the empirical representation of the networks of individuals, organizations and
niche categories (Figure 1). Apart from the two types of cross-level ties, there are also within-
level interpersonal collaborative ties. In the current case, there are no additional within-level ties,
because the social architecture in the current context does not support formally declaring inter-
clan alliances, and there are only two broad clan categories.
Summary of Studies
The empirical analysis involves three quantitative studies. The first study examines the
survival of clans by assuming their independence while taking into account inter-clan
interdependency brought by membership overlap. Adopting a multilevel approach, this study
proposes that the mortality of clans is a function of both niche occupation at the higher level, and
clan boundaries and within-clan co-play networks at the lower level. This study follows the
guidance of the revised theory of niche width (Hannan, 2010; Pontikes & Hannan, 2014) , the
revised theory of density dependence and category contrast (Bogaert et al., 2016; Negro et al.,
7

2010), and the network brokerage theory from a collective perspective (Bizzi, 2013; Burt, 2004).
The rationale to combine niche theory and network theory is the contingency principle in
hierarchic social systems, or the dependency between an “inner” environment and an “outer”
environment (Simon, 1962, 2019). Specifically, whether organizations can effectively engage
with the niche positions they occupy is contingent upon their internal properties. This study
adopts event history analysis to predict clan mortality hazards. This study hypothesizes that clans
that occupy broad niches have higher mortality hazards. This is expected to happen because a
broad niche creates an ambiguous identity that confuses existing and potential members
(Hannan, 2010). Further, this study proposes that within-group cohesion is needed for better
engagement with a focused niche, because members’ mindsets need to be well aligned for
typically social or competitive groups. The first moderator is group boundary. Players’ mobility
across groups erodes group boundaries. If members of a group on average allocate most of their
group affiliation experience to this group, this group has sharper boundaries, or a higher contrast
(Negro et al., 2010). Such a group is likely to have lower mortality hazards due to a stronger
collective identity. The second moderator is group-level network brokerage. Groups with a larger
mean and standard deviation of network brokerage have higher mortality hazards, because the
presence of too many brokers creates distrust and conflict (Bizzi, 2013). This study proposes that
groups occupying focused niches would be better off if they have clearer boundaries and a more
equal distribution of relational resources.
With the large picture laid out in the first study, the second study zooms in to model the
micro-mechanisms of multilevel organizational networks. This study seeks to understand if the
boundaries of virtual organizations and the relations among them could structure networks at
meso and micro levels. In other words, this study examines the extent to which interpersonal
8

networks have the autonomy outside organizational structures in the computer-mediated context.
This study follows the neostructuralist framework of multilevel agency that norms and
hierarchies emerge from bottom-up and top-down processes that are intertwined and coevolving
(Lazega, 2016). This framework is appropriate for examining multilevel dependencies because it
considers the interconnectedness between micro-level networks and macro-level networks given
cross-level clustered affiliations (Lazega & Snijders, 2016). In other words, it can be applied to
examine multiple networks simultaneously. This framework can be combined with other theories
to examine a variety of within-level networks such as similarity networks (Hollway & Koskinen,
2016), friendship networks (P. Wang et al., 2016), etc. This study combines this framework with
the niche overlap theory to leverage the competitive inter-group relations in the empirical context
(Bruggeman et al., 2012; Dobrev & Kim, 2006). Such combination seeks to address the research
gap that existing studies on multilevel organizational networks are more of intra-organizational
settings, e.g., collaborative inter-unit workflow networks at the macro-level (Zappa & Lomi,
2016). The inter-group niche overlap network at the macro-level can strengthen existing
multilevel network theories by testing how competitive relations at the macro-level structure
individuals’ boundary spanning behaviors and their collaborative ties. The multilevel network
model is organized as follows: a niche overlap network of clans at the macro-level (network A),
an individual-clan affiliation network at the meso-level (network X), and a co-play network
among individuals at the micro-level (network B). Following the three-level setup introduced
earlier, the macro-level network of inter-clan niche overlap is equivalent as projecting cross-level
ties between clans and clan categories to one-mode ties among clans. That is, a tie between two
clans indicates that they occupy similar positions on the social-competitive spectrum and they
semantically express social and/or competitive aspects in similar ways. This study employs
9

multilevel exponential random graph models (MERGM) to test four micro-mechanisms:
individuals affiliated to the same groups are more likely to build co-play network ties; boundary
spanners are more likely to be affiliated to similar groups; individuals are more likely to build
boundary-spanning ties with other individuals affiliated to similar groups; popular individuals
with more interpersonal ties are less likely to join groups with many niche overlap ties.
The third study adopts the relational event model (REM) (Lerner & Lomi, 2020b) to
study clan joining as a function of multirole network personality (Burt, 2012). If building
interpersonal networks and joining groups are viewed as two ways of accessing resources, will
the patterns be consistent or complementary? Both the network literature and the categorization
literature focus on the tradeoff between diversification and focus. That is, one either builds
closed and dense networks with strong ties or spans structural holes for resource breadth and
arbitrage (Burt, 2004). Likewise, one can concentrate resource allocation in a few categories for
narrow-deep resources. Alternatively, one can claim membership to many categories in a
diversified pattern to access broad-shallow resources (Hannan et al., 2007). This study integrates
the two theories and hypothesizes that the two ways of accessing resources through interpersonal
networks and category membership are consistent. To make such a connection, individual
agency plays a critical role. By following Burt’s (2012) construct of multirole network
personality, this study examines individuals’ consistent preference to build open networks across
roles. This study argues that “brokerage-prone” individuals are more likely to span group
boundaries.
These studies on the multilevel networks of individuals, organizations and organizational
niches make several theoretical and practical contributions. First, these studies seek to explore
the multilevel interdependencies within the theory of niche within the ecology framework.
10

Following the hierarchic principle (Simon, 2019) and multiple interconnected networks (Padgett
& Powell, 2012), these studies argue that the extent to which organizations’ niche occupation
affects organizational outcomes is contingent on constituent individuals’ membership mobility
and relational properties. Thus, organizational networks at different levels do not operate
independently but influence each other. Secondly, these studies try to understand the meaning of
boundaries of virtual organizations. Social relations are largely transient in virtual communities
(Shen et al., 2014), but organizational boundaries are still critical for efficient organizing in
virtual communities (Dahlander & O’Mahony, 2010; O’Mahony & Ferraro, 2007). The current
studies show that, similar to offline organizing, clear boundaries for virtual organizations benefit
their longevity, and these boundaries are powerful in providing opportunities and constraints for
individuals to build social networks and span organizational boundaries. Empirically, this finding
is meaningful for effective virtual collaboration for competitive tasks. Virtual group leaders may
pay attention to group boundaries, which are likely to affect within-group social relations, the
effectiveness of collective achievement, and group turnover. Theoretically, this finding echoes
recent arguments on the parallel between governance in virtual and offline organizations (Vesa et
al., 2017). Since group boundaries matter both online and offline, virtual organizing can be an
empirical context with generalizability. Thirdly, these studies explore the agency problem in
individuals’ network building and categorization. Following the sociological concept of
personality as an outcome of social networks, these studies try to understand if there is
consistency in the pattern that individuals access resources from interpersonal networks and
group affiliations. Finally, from a methodology perspective, these studies extend the application
of the most recent network modeling techniques like multilevel ERGM and large-scale REM.

11

Chapter 2: Literature Overview
Ecology and Evolutionary Theories
Basic Concepts of Ecology Theories
Organizational ecologists challenge the rational adaptation view and instead suggest that
the relationship between entities and their environment is governed by the mechanisms of
evolutionary selection (Aldrich & Ruef, 2006; J. A. Baum & Amburgey, 2002; Hannan &
Freeman, 1984). Organizational ecologists argue that organization changes are subject to the
ecological mechanisms of variation, selection and retention (V-S-R), so that the capacity to adapt
itself is subject to the selective mechanism (Campbell, 1965; Hannan & Freeman, 1977).
Variations are departure from routines and could be random or intentional. Although entities
have the agency to proactively make changes in response to conditions to improve survival
chances, the conditions also change because of such variations, resulting in possible
misalignment between adaptation and outcomes (Aldrich & Ruef, 2006). An example is the
“innovator’s dilemma”: some companies may aggressively invest in new technologies that their
customers want, which leads to a discrepancy between production and market demand, and then
innovation becomes the precise reason for their failure (Bower & Christensen, 1995). Thus, the
ecological perspective posits that most variations are blind. Variations conducive to
environmental fitness are more likely to be selected, although sometimes there is also loose
coupling between selection and fitness, due to reasons like institutional forces (DiMaggio &
Powell, 1983). For example, mimicry is a common practice to address environmental
uncertainty. Firms might follow peer firms to exit market segments without considering that they
could actually benefit from a reduced level of competition after others’ exit (Dobrev, 2007).
12

Selected variations are thus retained, preserved, or reproduced, through diffusion, replication, or
institutionalization (Aldrich & Ruef, 2006).
To survive in competitions, entities need to rely on resource niches, which are restrained
environmental resource spaces to sustain the growth of populations. Campbell (1965) defined an
ecological niche as “a viable mode of living” (p. 39). The classical theory of niche, including
niche width, resource partition and niche overlap, is primarily concerned with how economic
resources affect evolution, although the identity dimension is manifested in mechanisms like
density dependence and mimicry (Carroll & Hannan, 2004; Dobrev, 2007; Hannan & Freeman,
1977; Lander & Heugens, 2017). Scholars proposed a distinction between a fundamental and
realized niche. In the original theory proposed by Hannan and Freeman (1977), a realized niche
is defined as “that area in constraint space (the space whose dimensions are levels of resources,
etc.) in which the population outcompetes all other local populations” (p. 945), or in other words,
a realized niche depicts the restricted environmental space in which a population can be
sustained with growth in the presence of competing populations (Hannan & Carroll, 1992). A
fundamental niche then consists of the full range of conditions under which a population can be
sustained, or a region of resource space that traces the fit between entities and their environment
(Hannan & Freeman, 1977). A population of organizations occupies the same niche as they
depend on the same resource space (Hannan & Freeman, 1977).
Niches not only sort organizations into populations but also communities composed of
interrelated populations (Astley, 1985). Organizational community is defined as “co-evolving
organizational populations joined by ties of commensalism and symbiosis through their
orientation to a common technology, normative order, or legal-regulatory regime” (Aldrich &
Ruef, 2006, p. 243). For example, the life science community consists of interconnected
13

populations like universities, government labs, hospitals, and research institutes (Powell et al.,
2005). The community concept argues that attention should not just be given to how established
and stable populations withstand changes but also how diversity comes into existence (Astley,
1985). In a community, a niche is not a predetermined resource space that only supports the
fittest form; rather, it tolerates a variety of forms, which in turn imposes changes on the
environment as well (Astley, 1985). The argument goes: “organizations do not, in other words,
fortuitously fit into predefined sets of niche constraints; rather, they opportunistically enact their
own operating domains” (Astley, 1985, p. 234). In other words, niches are not stable or static,
but involve a complete evolutionary trajectory.
The larger open environment of organizational community is not solely restricted to a
single dimensional space of economic resources. (J. H. Freeman & Audia, 2006) argued that
organizational fields or communities are multidimensional entities that include
sociodemographic space, technology space, ideology space and identity space. This
multidimensional perspective of community space enables scholars to explore the relationship
between economic niches and identity ones. For instance, to understand how material resource
niches and ideology niches together influence comparison among organizations, Simons and
Ingram (2004) argued that ideological similarity will lead to either mutualism or competition
contingent upon resource overlap, that is, if organizations share similar key resources,
ideological similarity would evoke competition. Similarly, Ingram and Yue (2008) argued that
identity triggered by ingroup-outgroup identification could determine commensalism dynamics.
The debate on the definition of organizational form is a manifestation of this
multidimensional perspective of niches. Central to ecology theories, variations are introduced by
the birth of new organization populations and organizational forms, and death of old ones
14

(Hannan and Freeman, 1984). An organizational form embodies a range of features that allow for
identifying if an organization belongs to a population (Aldrich & Ruef, 2006). As summarized by
Aldrich and Ruef (2006), an organizational form has been defined in terms of a resource niche
(Hannan & Carroll, 1992) and cultural codes (Pólos et al., 2002). This summary clearly indicates
that the emergence, survival and prosperity of an organizational form is not only contingent on
actual capabilities indicated in resource niches, but also on social acknowledgement and
recognition. Organizations with economic proximity over time evolve into a collective with
socially acknowledged identity and symbolic meanings. In other words, both economic resources
and identity are essential for organizations to survive and grow.
The Ecology of Categories
Different from the resource niche concept of organizational form, the cognitive aspect of
organizational form is defined as collective identities indicated in patterns of social codes (Pólos
et al., 2002). Categories refer to the full trajectories of identities from inception to demise
(Hannan et al., 2007; Hsu & Hannan, 2005). This perspective gives rise to a fast-growing body
of literature called categorization theory, which posits that organizational identities come from
membership to categories (Durand et al., 2017; Hannan, 2010; Zuckerman, 1999, 2017).
Hannan et al. (2007) explained this departure from classical ecology theory by arguing
that “organizations should not just be analyzed objectively in terms of their patterned activities,
functions, and external ties—they must also be considered in terms of their social meanings and
interpretation given to them by contemporaneous actors” (p. 31). The categorization theory
begins from a subjective-external definition of organizational form as “codes” (Aldrich & Ruef,
2006; Pólos et al. 2002). Carroll and Hannan (2000) used the term “codes” to denote “both
cognitive recognition and imperative standing”, that is, “a set of signals” and “a set of rules of
15

conduct” (p. 67; also see Pólos et al. 2002). An identity is formed as a composition or pattern of
social codes, and when an identity applies to many entities it becomes a form (Pólos et al. 2002).
A major difference from classical ecology theory is that they define populations in terms
of minimal identities rather than forms (Pólos et al. 2002). Their rationale is that defining a
population in terms of an organizational form prevents researchers from evaluating the
theoretical status of populations before a form emerges. As their argument goes, “[i]f we define
populations in terms of forms, we lack a warrant for extending the definition of the population
back to the origin” (Pólos et al. 2002, p. 106). Tracing back the evolution history of a population
is theoretically crucial because legitimation happens between its inception and reaching a
threshold of taken-for-grantedness. But the classical theory does not provide a theoretical status
for this legitimation stage before reaching the form status (Pólos et al. 2002). Their definition of
population in terms of external identities, under comparison, provides a solution to trace the
complete trajectories of populations including those that possibly never achieve form status
(Pólos et al. 2002).
This is also where the abstract concept of category comes in, which is defined as “a
placeholder for the cultural object that can potentially become a form” (Hsu & Hannan 2005, p.
477). A category emerges when audience members have reached a consensus about the schemata
or defaults of patterns associated with a label and can use such schemata to test the membership
of objects within a category (Hannan et al., 2007). The category concept reflects a change of
assumption to the classical theory of niches. In other words, when ecologists define a population
in this theory, they draw a crisp line between population members and non-members. Within a
population under crisply determined sets, entities are not viewed as different in terms of
typicality in that population because organizational forms are highly legitimated and
16

institutionalized. An entity is either a member or a non-member of a population. The revised
theory of niche introduces the logic of “fuzzy sets” (Hannan et al., 2007). Under this assumption,
organizations’ membership to a category can be partial on a spectrum of zero membership to
full-fledged membership. Such partial membership is operationalized as the grade of
membership (GoM), defined as “the expected actual appeal of an offering in a category to an
audience member at a social position” (Hannan, 2010, p. 165). Grade of membership thus
reflects an entity’s degree of typicality in a category. Collectively, the boundary of a category
can vary on a spectrum of fuzziness. The fuzzier the category boundary, the less clear the
collective social identity. Thus, allowing the category boundary to be fluid potentially captures
the process of identity crystallization. This is because categories usually have fuzzy boundaries
during states of emergence and demise, and some categories are never legitimated enough to
reach form status. This generalizes the classical theory to cover the full trajectories of form
emergence rather than only focusing on states where forms are highly institutionalized (Pólos et
al., 2002).
The revised theory of niche with the fuzzy logic defines the fundamental niche in terms
of “intrinsic appeal and engagement” (Hannan, 2010, p. 116). With these two aspects of niche, it
is essential to differentiate between producers/products that are categorized and audiences that
make evaluations (Hannan et al., 2007). In existing empirical studies, producers can be
organizations such as wine producers (Negro et al., 2010) and law firms (Paolella & Durand,
2016), and individuals such as research scientists (Wry & Castor, 2017). Products can be books
(Kovács & Hannan, 2015), films (Hsu, 2006) and songs (Hannan et al., 2019). Audience
members may include industry analysts, critics, employees, enthusiasts, customers, etc. Intrinsic
appeal is the extent to which a producer’s offering fits the prototype taste of target audiences.
17

Thus, intrinsic appeal focuses on how audiences make sense of producers’ positions in a
categorized social space. Engagement refers to producers’ efforts to learn about the needs of
local audiences and design products to satisfy the typical taste of local audiences. Thus,
engagement emphasizes on producers’ capacity and/or efforts of improving skills to be more
successful in a niche. Intrinsic appeal and engagement together determine actual appeal. In other
words, the actual appeal of a producer to an audience member of a certain social position is
contingent upon the producer’s capability to satisfy a typical taste in that social position
(Hannan, 2010).
The existing literature largely focuses on “intrinsic appeal”. Since audience evaluation
plays an inherent role in determining intrinsic appeal, categories are defined from a cognitive
perspective. For example, Negro et al. (2010) defined categories as “conceptual tools for
understanding organization-environment relationships” (p. 4). Durand and Paolella (2013)
defined categories as “a cognitive infrastructure that enables evaluations of organizations and
their products, drives expectations, and leads to material and symbolic exchanges” (p. 1102).
Durand et al. (2017) defined categories as “groupings of entities that simplify our apprehension
of what surrounds us, focusing attention on a limited number of dimensions or features, enabling
recognition and action” (p.4). A common theme from these definitions is that categories are
cognitive shortcuts to set expectations and facilitate evaluations between producers and
audiences. Such interaction between producers and audiences is central to categorization theory
because the direct effect of categorization on performance is contingent on an audience’s
evaluation. In other words, audience evaluation mediates between organizations’ category
properties and performance (Paolella & Durrand, 2016). For example, in their study of corporate
law market, professional ratings of law firms were found to partially mediate firms’ category
18

spanning and revenue (Paolella & Durrand, 2016). Given the essential role of audience
evaluation, many studies focus on audience appeal as the outcome variable (Hsu, 2006; Negro et
al., 2010). With this definition, an organizational category can be viewed as the material and
symbolic resources to assess category membership agreed upon by a group of similar
organizations and audiences, and a product category can be viewed as a collection of similar
sociotechnical artefacts that producers, customers and critics have reached a certain degree pf
consensus (Vergne & Wry, 2014).
The Social Approach of Categorization
The social approach of categorization views the boundaries of categories not as static but
dynamically negotiated by actors with various goals and interests (Kennedy & Fiss, 2013;
Pontikes & kim, 2017). According to the social perspective, categorization is viewed as “a
collective negotiation of meanings, actions, and artifacts among varying actors” (Glynn & Navis,
2013, p. 1125). This perspective emphasizes on “how market categories are produced and
enacted differently in various social situations providing distinct cues and norms” (Durrand et
al., 2017, p. 9). Assuming category boundaries are permeable and negotiable, the social approach
of categorization “embraces the multiplicity and coexistence of both material and abstract
features of market categories” (Durand et al., 2017, p. 11). That is, actors’ categorization
behaviors are constantly shaping and shaped by the categories, and both the categories and actors
are embedded in a larger system (Glynn & Navis, 2013).
Since actors are agentic with goals and interests and actors are embedded in a broader
system, it becomes important to consider why actors claim membership to certain categories or
why some movements across categories happen. This perspective gives rise to a few studies
focusing on the antecedents of categorization. It becomes evident that inter-category relations
19

and inter-actor relations provide constraints and opportunities. For inter-category relations, the
level of rigidness of category boundaries offers actors different degrees of strategic opportunities
to selectively present category-related attributes (Hsu & Grodal, 2015). Compared to other
categories, more constrained resources and more fierce competition in the home category
motivate actors to span boundaries (Pontikes & Kim, 2017; Wry & Castor, 2017). When
choosing the target category, actors are more likely to span into similar categories. For example,
research scientists were more likely to span into categories of relational proximity, which was
operationalized as co-citation similarities from a patent citation network (Wry & Castor, 2017).
Producers of the software industry would enter a similar one that is covered by market analysts
for legitimacy reasons, and analysts were more likely to initiate a report on a category that was
growing and receiving investment and clearly defined with clear boundaries. For inter-actor
relations, research scientists were more likely to span categories when other scientists of high
status spanned patent categories because they wanted to follow the more experienced and
renowned peers (Wry & Castor, 2017).
Furthermore, scholars pointed out the importance of narratives for the legitimation of a
form (Glynn & Navis, 2013; Navis & Glynn, 2010). Wry et al. (2011) argued that articulation of
consistent, coherent and inviting stories would facilitate the legitimation of a nascent collective
identity. Khaire and Wadhwani (2010) used discourse analysis to examine the process of how the
emergence of a new market category was negotiated by a diverse range of actors including
historians, critics, etc. They particularly pointed out the importance of recognizing how category
meanings and values were embedded in a broader cultural field (Khaire & Wadhwani, 2010).
Navis and Glynn (2010) studied how the emergence of a new category shifted the focus of
producers and audiences. As a category emerges, the producers shift their identity claims,
20

framing and affiliations from stressing the collective identity to stressing distinctiveness within
the category. Audiences also shift attention from the collective identity to individual
organization’s identity. Grodal and Kahl (2017) called for attention to the interaction and power
dynamics in participants, and the material and cultural context that shape categories. They
viewed communicative text as active construction rather than passive representation of
categories, and thus “market categories are created and evolved through negotiations and
contestations between diverse market participants” (Grodal & Kahl, 2017, p. 153). Grodal (2018)
gave a detailed account of how core and peripheral communities strategically manage the social
and symbolic boundaries towards their interest to strengthen their identities. In the case of the
emerging nanotechnology field, periphery communities wanted to broaden the symbolic
boundary of the field to be more inclusive, but the core communities wanted to keep the
boundary narrow to discipline membership (Grodal, 2018).
A noticeable research gap is interdependencies within categorization mechanisms.
Producers, usually in the form of organizations, are not a homogenous unitary entity but
composed of heterogenous individuals who build interpersonal networks and change affiliations
among organizations. For instance, while studying how films positioned in genre categories
appeal to audiences, Hsu (2006) pointed out that “genres facilitate communication and
coordination among project personnel and provide clear frameworks for selecting film projects”
(p. 427). When examining the relationship between organizations’ occupation of niches and
performances, it is equally important to account for the personnel composition of organizations.
Films are produced by collaborating studios, and the network properties of project-based
collaboration have been found to be influential factors for the success of creative projects (Uzzi
& Spiro, 2005). Another example is from Paolella and Durrand’s (2016) study of law firms, in
21

which they found firms spanning broad legal categories were viewed as capable of attending to
complex requirements from clients, as clients of corporate law increasingly seek a full package
of legal services. How law firms span specialty categories is highly contingent upon the lawyers,
who build within-firm and cross-firm interpersonal networks and change jobs. As an example of
research on online communities, Lerner and Lomi's (2018) study on Wikipedia articles revealed
that coarse-grained articles in the hierarchical categorization system attracted many contributors
but received lower evaluations. Wikipedia articles are written by groups of contributors, who
form within-article editing networks and unevenly distribute attention and efforts across various
articles. Wry and Castor (2017) specifically pointed out the importance of considering the
intersection between institution and networks, that is, how category spanning associates with
cross-category networks and career moves.
Considering organizations as hierarchical composed of heterogenous individuals, in
connection with the necessity to consider the embeddedness of actors and categories, reveals the
importance to consider several forms of interdependencies. First, since intrinsic appeal and
engagement together determine actual appeal (Hannan, 2010), how successfully organizations
can engage with certain category patterns they occupy depends on organization-level properties.
Individuals’ membership mobility to organizations erodes boundaries, and they form social
networks that affect the cohesion of organizations. These factors are likely to affect how
organizations can engage with different niche positions. Second, given the clustered affiliation
between individuals and organizations, individuals’ social networks and boundary spanning
behaviors are contingent upon inter-organizational relations given organizations’ positions in
categories. These interdependencies are best examined in multilevel networks, which are
discussed in the section below.
22

The Evolution of Multilevel Organizational Networks
The Evolution of Networks
The view of social actions as category affiliations is in stark contrast with the alternative
view that “networks lay out the space of social action” (White, 2008, p. 8). For structural
sociologists, social actions can only be found “in the network of interstices that exist outside the
normative constructs and the attribute breakdowns of our everyday categories” (White et al.,
1976, p. 733). Social actions are viewed as “a set of trajectories and movements through space-
time” and they exist in the “compositions and layerings of multiple biographical paths and social
networks” (Padgett, 2018, p. 407). Thus, social structures are not buildings of stones, but created
in the dynamic process of network transformations (Padgett & Powell, 2012).
Traditionally, organizational ecologists focus on evolving populations and communities
of organizations defined by traits, but Monge et al. (2008) argued that examining the evolution of
populations and communities should incorporate the networks connecting them, because
networks are channels for acquiring and exchanging resources. Scholars argue that network
transformation is also subject to the evolutionary mechanisms of V-S-R, where exploring linking
partners entails variation, and links with high fitness are selected and retained (Monge et al.,
2008). Just as resource niches have finite carrying capacity to support a limited number of
entities, networks also can only support a limited number of links. In particular, linkage fitness
could be evaluated in multiple dimensions, i.e. the capacity to provide important resources and
the efforts needed to sustain the link (Monge et al., 2008). Hilbert et al. (2016) compared three
ways to identify populations in assessing evolutionary forces: traditional trait-based, network
positions, and network partitions. Their results showed that populations based on network
metrics are subject to stronger evolutionary forces compared with trait-based populations.
23

Corresponding to the concept of an organizational community composed of
interconnected organizational populations (Astley, 1985), a number of empirical studies
specifically focused on the commensalistic ties connecting similar populations and the symbiotic
ties connecting dissimilar ones. Audia et al. (2006) explored how commensalism and symbiosis
in an organizational community contribute to the founding of new companies. Their analysis of
the U.S. instrument manufacturers showed that the founding of new companies within a
population is facilitated by the existence of other populations that are similar or have business
connections because such communities offer relevant information. In contrast, it would be more
difficult for new companies to form if the local community is dominated by very different
populations. Bryant and Monge (2008) proposed a community evolution model of
communication networks that encompassed four stages: emergence, maintenance, self-
sufficiency and transformation. They focused on the children’s TV community that support
children’s television such as Sesame Street. This community consisted of different populations
like educational content creators, government bodies and content programmers. They divided the
community evolution into stages including emergence, maintenance, self-sufficiency, with
decreasing dependence on environmental resources, and a final stage of transformation. Their
analysis showed that the communication networks linking different populations were
characterized by mutual ties during the emergence stage and self-sufficiency stage, whereas the
proportion of competitive ties increased during the maintenance stage. Studying commensalism
and symbiosis in the media industry under disruption, Weber (2012) found that the symbiotic ties
of existing media with emergent media organizations in the form of hyperlinks would mobilize
online audiences.
24

DiMaggio and Powell (1983) defined the field as "those organizations that, in the
aggregate, constitute a recognized area of institutional life: key suppliers, producers, regulatory
agencies, and other organizations that produce similar services or products" (p. 148). Kenis and
Knoke (2002) built on this concept and defined organizational field network as “the
configuration of interorganizational relations among all the organizations that are members of an
organizational field.” (p. 257). They proposed that network properties such as density,
reciprocity, centralization, multiplexity, hierarchy, and number of cohesive subgroups would
affect subsequent nonlinear changes in interorganizational tie-formation rates. Also focusing on
the organizational field, Powell et al. (2005) examined the network evolution of commercial and
scientific organizations surrounding biotechnology, and found the evolutionary mechanisms
including geological homophily, following the trend and diversifying alliance partners.
Similarly, Lee and Monge (2011) found organizations sharing the same resource spaces were
more likely to form multiplex ties for implementation and knowledge sharing.
Organizational Networks from a Multilevel Perspective
The fundamental logic of hierarchic complex systems lies in bounded rationality, which
posits that “the complexity of the environment is immensely greater than the computational
powers of the adaptive system” (Simon, 2019, p.166). Therefore, an artifact can be viewed as the
intersection “between an ‘inner’ environment, the substance and organization of the artifact
itself, and an ‘outer’ environment, the surroundings in which it operates” (Simon, 2019, p .6).
Specifically, a hierarchic system refers to “a system that is composed of interrelated subsystems,
each of the latter being in turn hierarchic in structure until we reach some lowest level of
elementary subsystem” (Simon, 2019, p. 184).
25

The multilevel perspective is especially important for studying organizations because
organizations are primarily naturally multilevel structures (Hitt et al., 2007; Moliterno &
Mahony, 2011). Despite such theoretical importance, studies of organizational networks from a
multilevel perspective are surprisingly rare (Paruchuri et al., 2019). As Paruchuri et al. (2019)
argued, single-level perspectives of organizations tend to result in both impoverished theory
building and biased model estimation. They pointed to two problems of a single-level
perspective. First, the assumption that the same logic applies to both individual and
organizational levels needs probing. Second, organizations are not simple, unitary entities, but
systems with complex internal structures. Cross-level mechanisms are highly likely to depend on
each other. A single-level perspective tends to overestimate the autonomy of networks (Zappa &
Lomi, 2016).
There are three distinct perspectives that will be employed to study the multilevel
organizational networks in this dissertation: (a) local/global networks. (b) internal/external
networks, and (c) nested three-level networks. The first two perspective are also applicable for
single-level networks. The third one is a unique multilevel perspective involving two levels of
clustered relations.
The first perspective, local/global networks, means alignment or misalignment of
structural properties at local and global levels. At the individual level, relational properties are
associated with individual benefits and individual identity; at the collective level, relational
properties are associated with public goods and collective identity. An ideal state would be
network congruence between the individual-level and collective-level, but such alignment, or
isomorphism, is usually harmed when individual or collective interests override (Ibarra et al.,
2005). For instance, network centrality has been considered as a measure for an advantageous
26

position to access information and influence (Borgatti, 2005; Perry-Smith, 2006). From a
multilevel perspective, network density can be viewed as the group-level mean of degree
centrality, and network centralization can be viewed as the group-variance of degree centrality
(Bizzi, 2013). Network density can be viewed as a rough proxy for the strength of collective
identity. Low density indicates low cohesiveness, but overly high density suppresses individual
identity. A meta-analysis of network density supports the argument that network density in both
instrumental and affective networks positively associates with collective-level performance
(Balkundi & Harrison, 2006). Similarly, Rowley et al. (2005) found the density of a group and
the embeddedness of a firm in that group defer this firm from exiting. Other studies found that
within-group network density has an inverted U-shaped relationship with group effectiveness,
and that group members’ external ties with leaders from other groups increase their group’s
effectiveness (Oh et al., 2004). Network centralization is found to negatively associate with
group performance, because these groups are hierarchical with a clear core-periphery structure
(Cummings & Cross, 2003). The rationale is that a “flat” hierarchy facilitates communication
and timely information transmission among members. In a hierarchical structure, the collective
identity is dominated in the hands of a few. From a multilevel perspective, individual in-degree
centrality and betweenness centrality positively affect individual outcomes, but decentralized
networks at the group level are better for group success (Sasidharan et al., 2012).
The second multilevel perspective, internal/external networks, is through the contingency
between internal and external networks. Tsai (2001) found that the knowledge outcomes of
organizational units depended on both internal capability and external resources, such that
organizational units with high centrality had better performance, but this relationship was
contingent on the units’ absorptive capacity of new knowledge. Connecting intra-organizational
27

networks and inter-organizational networks, T.-Y. Kim et al. (2006) proposed that an
organization’s reluctance to dissolve interorganizational network ties was a function of
intraorganizational network characteristics, that is, the lower-level network structure constrains
the higher-level network structure. In Funk (2013) study on innovation in the nanotechnology
industry, the relationship between the geological proximity to other firms for knowledge
spillover and the ability to produce impactful patents was contingent upon intrafirm network
cohesion (measured as clustering coefficient) and inefficiency (measured as the average length of
path). When technology firms could get more knowledge spillover from peers, cohesive
networks were helpful for generating better innovative outcomes; when proximity to peers
decreased, inefficient networks were more beneficial. The rationale is that intraorganizational
networks are not only important for knowledge generation (Hansen, 1999), but also important for
knowledge processing and internalization (Borgatti & Cross, 2003). At an even broader scale,
Kenis and Knoke (2002) made a few propositions as to how field-level networks influence the
formation of interorganizational networks.
The third perspective is nested multilevel networks. In this perspective, there are
networks at different levels: an inter-individual level, an inter-organizational level, and the
affiliation at a meso level (Lazega, 2013). Scholars argue that such nested networks would
enable researchers to consider how bottom-up and top-down mechanisms are linked together
(Hitt et al., 2007; Lazega & Snijders, 2016; Moliterno & Mahony, 2011). This framework of
multilevel networks not only considers within-level endogenous mechanisms but also cross-level
endogenous mechanisms. This multilevel perspective seeks to “contextualize actors’ behavior by
describing the structures of opportunity and constraints that emerge from regularities in relational
choices by these actors” (Lazega, 2013, p. 169).
28

There are three key concepts in this perspective on nested multilevel networks. As a key
concept in this theory, agency is defined as “individual and collective action by combining
identity, culture and authority in actors’ judgements of appropriateness” (Lazega, 2013, p. 170).
Such agency is valued in poststructuralism such that actors are viewed with a capacity to
perceive and respond to social conditions (Kilduff & Tsai, 2003). Multilevel networks in this
framework refer to “the fact that in a stratified society, there are many superimposed levels of
agency, each of them characterized by horizontal interdependencies that sociologists can
examine as sets of ‘local’ social systems” (Lazega, 2016, p. 48). The emergence of two networks
at different levels is governed by its own logic, i.e., interpersonal networks and inter-
organizational networks are formed and transformed with different logics. The two levels of
agency suggest that new norms and hierarchies emerge from bottom-up and top-down processes
that are intertwined and coevolving, that is, individual members could collectively create new
practices that might challenge existing relational structures, or collective norms and hierarchies
at a higher level are imposed on individuals as opportunities or constraints to reinforce their
existing positions (Lazega, 2013).
Another key aspect is the presence of affiliations to categories at the meso level, or
“social forms” (Lazega, 2016, p. 52). The two-mode network at the meso-level is a network
representation of category membership and boundary spanning. With asymmetrical distribution
of power from higher-level sources of the social stratification, subordinates need to bear higher
costs in coordinating successful collective actions (Lazega, 2016). Within and between levels,
affiliation to powerful stakeholders means the maintenance of existing social forms. The
“synchronization costs” refer to the social costs to carry out collective actions at different levels,
which not only depend on micro and macro networks, but also depend on the distribution of
29

resources at the meso level (Lazega, 2016). From this perspective, networks at micro- and
macro-levels are to erode or reinforce pre-existing formal boundaries of social space (Lazega,
2014).
Thirdly, networks and categories are combined in the same model to construct cultural
meanings and shared representations, which make a third key concept in this framework.
Multilevel relational interdependencies are not only defined in terms of structures but also “in
terms of symbolic and moral commitment, as well as economic sources of resources” (Lazega,
2014, p. 174). Cultural meanings are defined “as a set of languages and norms that help actors
stabilize or destabilize prior structures when trying to give meaning to actions and to defend their
political/regulatory interests” (Lazega, 2014, p. 175). In other words, networks, category
affiliations and symbolic meanings co-constitute each other in the creation, relocation and
erosion of social boundaries and social identities. This emphasis on thematic meanings echoes
the recent call for scholarly attention on the importance of combining structures and meanings as
“mutually constitutive and coevolving” (Pachucki & Breiger's, 2010, p. 206, also see Mützel &
Breiger, 2020). Apart from cosnidering “structural holes” that bridge otherwise disconnnected
actors, it is also essential to consider “culturral holes”, meaning the “contingencies of meaning,
practice, and discourse that enable social structures”(Pachucki & Breiger, 2010, p. 206). For
example, a “cultural hole” in cultural tastes means the pattern of cultural choices where a person
bridges distantly connected cultural items (Lizardo, 2014). The emphasis on cultural meanings is
an important future research area for social networks (Mützel & Breiger, 2020).
These three multilevel network perspectives serve to examine the interdependency
question from the previous section. The first two perspectives can help examine how
organizations can successfully engage with certain niche patterns. To be specific, based on the
30

internal/external perspective, how organizations’ niche occupation can affect outcomes depend
on internal networks (Funk, 2013). From the local/global perspective, local network positions
might benefit the individual but not the collective (Cummings & Cross, 2003). The third
perspective of nested multilevel networks can help examine the interdependencies among
interpersonal networks, individual-group affiliation networks and inter-organizational networks
(Lazega & Snijiders, 2016).
To summarize, the theory of niche and the three multilevel network perspectives together
provide a dynamic theory of social formation. The boundaries of social categories are viewed as
fluid so that the trajectories of category formation can be examined (Pólos et al. 2002).
Furthermore, by considering the multilayeredness of social actions (Padgett & Powell, 2012), it
is possible to examine how social formation arises from the embeddedness within interconnected
multiple networks.
Social Boundaries in Online Communities
Ecology in Online Communities
One observation of the current ecology research is that the conceptual utility of the
revised ecology theories is restricted by the empirical contexts examined. As organizational
ecologists theorize the transition to revised ecology theories, the main argument lies in that the
concept of categories has the flexibility to encompass the complete trajectories of shared
resources and identities, including emergence and non-emergence (Pólos et al. 2002). Although
the revised ecology theories have considerably enlarged the scope and enhanced the nuances of
niches, the categories under examination, including films genres (Hsu, 2006; Hsu et al., 2012;
Keuschnigg & Wimmer, 2017), music genres (Askin & Mauskapf, 2017), and e-commerce
categories (Hsu et al., 2009), are still relatively stable. Although these cultural categories are
31

socially constructed by collective action and subject to interpretations, it is difficult to obtain the
empirical data to include the full evolution of categories to capture the permeability or
persistence of boundaries.
Similar to cultural categories, categories in online communities are usually formed by
unsupervised collective action, which can be viewed as an interactive process of sensegiving and
sensemaking to negotiate categorical meanings (Lin et al., 2015). The boundaries and associated
meanings of these categories are usually highly permeable, constantly shaped and reshaped by
fluid members of online communities. For instance, individuals join groups to socialize with
others with similar interest, but such interest groups might emerge and disappear (Zhu et al.,
2013); groups categorize themselves into broader niche categories, and the meanings of these
niche categories are contingent upon the collective interpretation of the whole community (Vesa
et al., 2017). Therefore, online communities provide rich empirical contexts to inform ecology
theories.
The ecological perspective, especially the theory of niche, could also enrich the
theorization of online communities. In general, the conceptual utility of ecology theories has not
been sufficiently explored in the context of online communities (C.-H. Chen et al., 2008; Shen et
al., 2014). Firstly, the theory of niche offers insights about how individuals and groups in online
communities consume economic and identity resources. Within this ecosystem, entities like
individuals and groups occupy resource spaces not only for material resources like information,
knowledge, and social relations, but also for social identities, affective attachment, and emotional
support (Ren et al., 2007). Secondly, the ecological perspective does not just focus on one group
and its members, but views how such micro-mechanisms are embedded in a broad context with
many other groups and members. Such focus on interconnectedness highlights the abundant
32

opportunities of communication through digital technology. Thirdly, the ecological perspective
also provides a dynamic theory of change to study social formations and movements. For
example, boundaries of groups are simultaneously blurred due to members’ turnover and
mobility, and strengthened by cohesive within-group networks. Thus, this perspective allows us
to examine the formation, growth, retreat and demise of social boundaries and social relations.
Such a dynamic perspective well aligns with the fluidity of virtual communities.
The term “online communities” or “virtual communities” has been defined as “any virtual
social space where people come together to get and give information or support, to learn or to
find company” (Preece, 2001, p. 348), and “groups of people with common interests and
practices that communicate regularly and for some duration in an organized way over the
Internet through a common location or mechanism” (Ridings et al., 2002, p. 273). These
definitions have summarized that people are motivated to join online communities for both
access to information and social activities, or in a broader sense, instrumental and affective
reasons (Howard & Jones, 2004; Preece & Maloney-Krichmar, 2005; Y. Wang & Fesenmaier,
2003). These features of online communities are extensions of scholarship on real life
“community,” which denotes not only an entity but also symbolic social meanings and collective
consciousness constructed through interaction and communication (Cohen, 1985). Just like
communities in real life, people socialize in online communities with shared interests and goals,
and for support and empathy (A. J. Kim, 2000). What makes “online communities” different,
however, is the debate about whether aggregating in virtual space has eroded or reinforced social
boundaries.
Meyrowitz (1986) argued that it is not the material environment that enables or constrains
social behaviors but the information flow or access to social information. Social groups are not
33

defined by place-bound settings but information systems, which refers to “a given pattern of
access to social information, a given pattern of access to the behavior of other people”
(Meyrowitz, 1986, p. 37). It is therefore the information system or shared codes that holds a
group together, forms a social group identity, makes individuals identify with a group, and
separates members from non-members (Meyrowitz, 1986). The more socialized a member is to a
group, the more “backstage” information this member gets (Meyrowitz, 1986). By
conceptualizing social roles as “group feelings (identity), role translations (socialization), and
authority (hierarchy)” (p. 53), Meyrowitz (1986) argued that electronic media blurred the
boundaries of groups as information systems. First, digital technology enables much wider
access to the backstage information of social categories and thus erodes the boundaries that
define a common identity. Groups still exist, but the boundaries of shared meanings and
experiences are blurred by information sharing on digital media. The length and difficulty of the
socialization process used to protect groups from revealing information to individuals without
sufficient commitment, but this staged process of socialization is no longer clear. Social
hierarchies are also affected because the required knowledge is no longer exclusive. Building on
these arguments, Meyrowitz (1986) came to the conclusion that digital technology promotes
egalitarian and democratic participation and blurs social boundaries.
Some recent empirical studies have explored the consequences of digital fluidity, where
individuals are free to come and go, join any groups and quit. Zhu et al. (2014) examined the
relationship between the membership overlap of communities and the survival of these
communities. They found greater membership overlap positively associated with community
activity and such benefits were especially strong when the focal community was young, the
intersecting communities were mature, and shared members were core members in the
34

intersecting communities. Although some studies on community turnover suggests that the
ability to attract and retain members plays an important role in the success of communities (B. S.
Butler, 2001), scholars also found that membership turnover has inverted-U shaped effect with
knowledge creation and retention, that is, turnover improved knowledge outcomes to a threshold
and decreased (Ransbotham & Kane, 2011). This finding reveals the benefits of blurred social
boundaries, as a moderate amount of turnover is beneficial for the sustainability of communities
because it lowers the barriers for new participants with new knowledge to join, and also
potentially filters out ill-fitting members to make room for the most suitable members.
On the other side of the debate, scholars call for attention to how digital technology sets
rules and policies. Infrastructures of cyberspace enable something and disables something.
Behaviors in virtual spaces are not only governed by algorithms but also by laws, markets, and
social norms (Lessig, 1999). Scholars also acknowledge the differences in technology
affordances in different online communities. Some communities are governed by predetermined
codes, whereas communities like Wikipedia allow for the alterations of rules, policies and
guidelines, which support the functioning of a variety of social structures (B. Butler et al., 2008).
Similarly, scholars argue that computer mediated interaction is a double edge sword
(Kollock & Smith, 1996). Based on earlier research, clearly defined boundaries are important for
the successful functioning of communities, because clear boundaries protect the collective public
goods created by group members and encourage interaction within groups to strengthen group
cohesion (Ostrom, 1990). Butler (2001) employed a resource-based view to evaluate what makes
sustainable social structures in online communities. The resource created by community
members, in the form of information and social support, is key to attract and retain members.
Butler’s (2001) study of listservs showed that community size and communication activity both
35

had positive and negative effects. Large communities are better at attracting members but worse
at retaining members. Communication activities increase community resources, but members
could not benefit equally from these resources.
This debate could probably be summarized in the prevalent concept of bridging and
bonding social capital (Putnam, 2002; Williams, 2006). Bridging social capital largely refers to
social networks that bring together different people from disparate communities. Bridging social
capital enables individuals to traverse social boundaries, reduces information exclusion, and
promotes mutual understanding across different social groups. Bonding social capital brings
similar people together, reinforces mutual interests, goals, and beliefs. Online communities
encourage interactions within social groups and interactions across social groups, that is, they
strengthen similarities and close divides at the same time.
Common Bond and Common Identity
The debate on whether digital technology erodes or reinforces social boundaries is
echoed in the scholarship of computer mediated communication (CMC), which examines
whether communication in digital space is primarily an interpersonal or individual-group process
from a social psychology perspective (Walther, 2011). Since people participate in online
communities mainly for information and socialization, computer-mediated communication plays
a key role in the functioning of an online community. The debate in this body of scholarship
centers around how interpersonal relations arise in digital contexts. Interpersonal relations can be
formed on the basis of a person’s attraction to each idiosyncratic individual. Interpersonal
relations can also be formed from shared interest and a collective identity. Walther (2011)
discussed the debate between interpersonal theories of CMC like the hyperpersonal model,
Proteus effect, and social identity model of deindividuation effects (SIDE). Some scholars argue
36

that technology affordances could compensate the lack of non-verbal cues in communication and
facilitate the development of interpersonal bond through mechanisms like hyperpersonal model
(L. C. Jiang et al., 2013; Walther, 2007). Other scholars argue that people’s identification
through shared interest and common group membership precedes the development of
interpersonal bonds under the condition of visual anonymity (Carr et al., 2013; H. Jiang &
Carroll, 2009; T. Postmes et al., 2005).
Postmes et al. (2005) reconciled the debate on interpersonal and individual-group
processes by proposing two paths to social identity formation: either deducted from a collective
group identity or inducted from individual qualities. Thus, whether depersonalization increases
or decreases social influences depends on the inductive or deductive path to social identity.
Following this distinction, Postmes et al. (2006) discussed two types of groups: interpersonal
groups made of identifiable individuals, and categorical groups made up of depersonalized
individuals. They also proposed that the bottom-up inductive process of collective identity
construction and the deductive process of inferring social identity were integrated in a coherent
process. A number of empirical studies have compared the deductive and inductive models of
social identity. Cheng and Guo (2015) tested an inductive model by bringing interpersonal social
networks and group membership esteem together as mediating variables between knowledge
contribution and social identity formation. They found social interaction exerted more influence
on the social identity than group membership, which could be explained by the importance of
social embeddedness to the formation of a sense of belonging to virtual communities (Cheng &
Guo, 2015). Jiang and Carroll (2009) took a deductive approach and argued that shared common
identity preceded interpersonal interactions, which further generated social networks. People first
37

identify with others based on commonality in terms of social categories and the resulting social
interaction makes the condition for interpersonal social networks (H. Jiang & Carroll, 2009).
Whether the formation of social identities follows a deductive or inductive path, scholars
agree that both mechanisms are in practice in online communities. Scholars have proposed two
major types of social interaction: the bond among individual members, and individuals’ sense of
belonging to a group as a whole (Kavanaugh et al., 2005; Ren et al., 2007). Kavanaugh et al.
(2005) explored how online communities could strengthen social interaction and community
attachment. People can be connected by interpersonal ties as well as joint membership to groups.
Ren et al. (2007) proposed a framework of two types of attachment: a common bond theory to
explain people’s attachment to individual group members, and a common identity theory to
explain people’s attachment to the group as a whole.
For interpersonal bond attachments, Ridings and Gefen (2004) found that information
exchange was the most popular motivation for people to hang out online, followed by social
support and friendship. Preece (2001) argued sociability and usability affected the success of
online communities, and that the determinants of sociability included members’ information
exchange, reciprocity and satisfaction.
For social identity attachment, a sense of community is formed when members have
similar interests, have feelings of membership, mutual influence, and affective connection (C.-
W. Chen & Lin, 2014). Group identity, including dimensions like group topic, size, and
commitment, affects individual-group interactions (Arguello et al., 2006). People tend to be less
committed to large groups, and the coherence of discussion topics also matters for community
identity.
38

As a summary, Grabowicz et al. (2013) named groups established on the basis of shared
interests, goals and themes as “topic” groups, and groups established on personal social relations
as “social” groups. They argued that most groups are a mixture of both, but somewhat lean
towards either topicality or sociality. Chung et al. (2016) also brought the two types of
attachment into the same model, and examined the antecedents and consequences of identity-
based and bond-based attachment. Their structural equation model showed that identity-based
attachment had a much stronger impact on information sharing.

39

Chapter 3: Research Context
The Social Architecture of Gaming Communities
With the broad theoretical picture laid out, this chapter introduces the research context,
and explains the empirical representation of multilevel networks. Virtual worlds have been
recognized as important sites of online communities (Steinkuehler & Williams, 2006). Gaming
communities are natural laboratories for studying social networks and group dynamics
(Ducheneaut, 2009), particularly because the gaming infrastructure usually provokes competition
and collaboration for resources, like experience, skills, equipment and ranking. With the need for
competition and collaboration, players form interpersonal networks and join subcommunities, or
in-game organizations. These organizations emerge to cater to more specific interest of users and
organize social activities for players (Kim, 2000).
There is a growing number of studies that examine the “guilds” of a massive multiplayer
online game (MMOG) called World of Warcraft to understand individual-group dynamics
(Alvari et al., 2014; C.-H. Chen et al., 2008). How this community as a channel to form and
sustain human relationships has attracted much research attention (Martončik & Lokša, 2016;
Nardi, 2010). A guild is a relatively formal organization with leadership structures to organize
social activities for players (Williams, 2006). This type of in-game organization is prevalent in
other game communities, like clans in World of Tanks. Guilds with different purposes and sizes
have distinctive formal structures, which greatly affect individuals’ behaviors, attitudes and
social interactions (Williams, 2006). Players join guilds for multiple reasons, including access to
information and knowledge, improving skills, socializing and making friends (Williams, 2006).
Players sacrifice their freedom and voluntariness of play and join virtual organizations
which require some degree of responsibility and commitment for the collective (Nardi, 2010).
40

Individuals with more contribution are usually rewarded with more recognition such as status
and responsibilities. Research on general virtual collaboration for peer production has offered
some insights. For example, in open-source production, an important motivation for contributing
to public goods is personal learning and enjoyment that are not available to free riders (Hippel &
Krogh, 2003). Such intrinsic motivations also include kinship and fun (Von Krogh et al., 2012).
Likewise, enjoyment, fellowship, and a sense of achievement are important motivations for
joining in-game groups (Snodgrass et al., 2017).
The formation and performance of guilds are important research topics because scholars
would like to know if such virtual groups follow the same social dynamics as offline human
organizing. If so, the large-scale behavioral traces could well serve to enhance theory building
with general applicability, especially because virtual groups offer unique multilevel organizing
data with both individuals and groups. For example, Johnson et al. (2009) studied the evolution
of MMOG groups and gangs in real life and found that diversity in skillsets, a form of
heterophily, drove the formation of groups, but they didn’t find evidence of homophily.
However, Ahmad et al. (2011) found homophily played a role in driving the formation of guilds.
Huang et al. (2009) explored the formation of task-oriented team formation in gaming contexts,
and found motivations like maximizing personal achievement and learning, and collective action
for tackling difficult tasks affected the formation of teams. Like other types of organizations, the
evolution of guilds follows life cycles (C.-H. Chen et al., 2008). Scholars have found that guild
features like size, members’ skill levels, and within-guild social network structures affect guild
performance (Ducheneaut, 2009). Guilds made up of members with balanced and
complementary skills have better performance. Group size is an impediment to guild
performance, which might be due to difficulty in forming intimate relationships. A wider level
41

spread (variance of skill levels) indicates vibrant recruitment of new players, which is beneficial
for the fitness of guilds. Controlling for guild size, guilds fragmented into sub-groups perform
better (Ducheneaut, 2009).
Such in-game organizations are appropriate research contexts to study the dual
mechanisms of common bond attachment and common identity attachment in online
communities. Firstly, scholars have argued that in-game organizations play an essential role in
the social life of online gaming communities, and membership gives players a sense of
community (Chen et al., 2008). Drawing insights from social identity theory, Guegan et al.
(2015) argued that being a member of the game and being a member of a group both contribute
to players’ social identity. Qualitative studies confirm that group members express a sense of
community and belonging, and they give and receive in-game social support, which can even be
extended to offline context (O’Connor et al., 2015).
Second, players form interpersonal networks alongside joining and quitting these
organizations. A body of social network literature has focused on the interpersonal networks in
online communities (Faraj & Johnson, 2010; S. Y. Lee, 2015; Panzarasa et al., 2009; Shen et al.,
2014). Scholars have found that there is no strong homophily mechanism in players’ social
networks (Lee, 2015). There is tendency for reciprocity but away from preferential attachment,
which means a tendency to connect with already connected individuals (Faraj & Johnson, 2011).
Specific to gaming communities, players’ social relations are more transient than offline social
relations but become increasingly durable over time (Shen et al., 2014). Moreover, shared group
membership, and similarity in skill levels and geolocation proximity help preserve interpersonal
relations (Shen et al., 2014).
42

Players’ social networks in games constitute important forms of social capital, including
both bridging and bonding social capital (Williams, 2006). Bonding social capital generally
refers to benefits like trust and emotional support derived from network closure and strong ties.
Bridging social capital refers to benefits like access to diverse information and broadening views
derived from open networks and weak ties (Williams, 2006). In gaming contexts, both forms of
social capital are essential. Bonding social capital is important because team-based battles
require coordination and mutual trust. Bridging social capital is important because it’s beneficial
for learning and improving skills (Shen et al., 2014).
In empirical studies, scholars found that playing with different types of co-players, i.e.,
family members, existing offline friends and online friends first met in game, associated with
bonding and bridging social capital in different ways (Meng et al., 2015). Specifically, playing
with offline friends positively associated with bonding social capital, and playing with offline
friends and online friends positively associated with bridging social capital (Meng et al., 2015).
Similarly, playing with families, friends and strangers had different effects on players’ health
disruptions (Shen & Chen, 2015).
Third, individual-group interaction and interpersonal interaction are connected. The
boundaries of in-game organizations are very likely to influence or be influenced by players’
social networks. Scholars have found that members tend to team up with others more and play
longer than non-members (Ducheneaut et al., 2006). Alvari et al. (2014) found the knowledge of
existing members of in-game organizations best predicted membership, more so if the members
have a high centrality in social ties. Hsiao and Chiou (2012) found players’ social capital
enhanced their trust towards other fellow players, which in turn increased the likelihood of
staying in a game. Specifically, a player’s intra-guild network centrality positively correlates
43

with access to intra-guild resources but not directly affect enjoyment, whereas non-guild ties
negatively influence intra-guild resources but positively influence enjoyment (Hsiao & Chiou,
2012).
The Research Context: World of Tanks
The research context is World of Tanks (WoT). This game features mid-20
th
Century
tanks and team-based battles in which each player maneuvers a different type of tank. The social
dynamics in WoT include both interpersonal social networks and in-game organizations. First,
teams are formed with the goal of capturing the opponent team’s base or destroying all enemies.
Average battles usually take six to seven minutes. Headphones are mandatory for voice chat
during battles. Secondly, players also voluntarily join clans as in-game organizations for more
committed socialization.
There are several gaming modes. The most common one is the “random battle,”
where players are randomly assigned to be on one of the two teams by an automatic team
assignment process. Players do not have prior interactions with teammates in random battles.
Players can also play in the “platoon” mode where they are able to form very small teams with
friends before or during the battle. Platoons differ from random battles in the fact that players
who are already friends intentionally choose to play together. Thus, platoons are purposeful
collaboration and are ideal for constructing co-play networks. Another type of purposeful
collaboration is manifested in clan-based teams. Members of the same clan can form a team of
15 in the “skirmish” mode. Players who are not in the same clan can also join clan-based
skirmishes as “legionnaires”. Clan members can also choose a practice mode to play against each
other, especially for training purposes. Global map is a game function only available to clans.
Clans play against each other to take control of provinces on a vast map. Once it captures a
44

province, it can expand to adjacent provinces by battling against the clans controlling those
provinces. Therefore, platoons and clan-based battles constitute players’ co-play networks as
purposeful collaboration.
As team-based battles require each player to complete a different task, coordination and
teamwork are key to success (Shen et al., 2020). For instance, scouting is carried out by light
tanks to spot enemy tanks. With scouts lighting up targets, heavy tanks carrying bigger guns will
get into the enemy base and try to hit priority targets. Frontline tanks equipped with the heaviest
armor are directly positioned in the fire line. A player gets credit for not only damaging targets in
the opponent team but also assisting team members to complete tasks. Having more familiar and
trusted collaborators is essential for in-game performance because coordination, risk-taking and
selflessness are required. Thus, building co-play networks with intentionally chosen partners is
an important way for players to accumulate social capital in game communities (Meng et al.,
2015; Shen & Chen, 2015).
Another form of social activity is joining clans. Clans in WoT are formal organizations
with distinctive identities, including names, profiles, and military-like leadership structures
(Shen et al., 2020). To ensure effective governance, clans assign members with hierarchical roles
like commander, executive officer, personnel officer, combat officer, recruitment officer, junior
officer, etc. Clans usually have specific recruitment requirements in terms of skill levels and
participation, and they assign special recruiters to screen membership applications. Clans also
have within-clan chat channels to discuss and coordinate training, recruitment and battle plans. A
clan can have a maximum of one hundred members, and players are only allowed to join one
clan at a time. Thus, the social organizing of clans adds another layer of rules to the program-
code embedded rules of virtual worlds (Vesa et al., 2017).
45

Corresponding to different motivations of game playing, the basic agreement between
clan recruiters and clan seekers gives rise to two niche categories of clans: competitive and social
(Williams et al., 2006; Vesa et al., 2017; Snodgrass et al., 2017). These two categories broadly
organize players’ mindsets and goals for playing. A competitive clan’s main goal is to win
matches against other clans, which results in more in-game resources and a higher position on
the public leaderboards. All competitive matches begin with a display of these leaderboards,
keeping the public ranking highly visible. In contrast, a social clan’s main goal is to have fun
while playing. Winning is still the goal, but the banter and fun of the experience is more
important. Other studies also found there could be a middle ground category that combine both
social organizing and competing (Williams et al., 2006). Clans usually self-categorize among
these types and state who they are and what kind of players they look for in their public profiles.
Thus, players, clans and clan categories are the empirical representation of the multilevel
structure of individuals, organizations, and niche categories. Players are clustered in clans, and
clans are clustered in clan categories. Interpersonal networks are represented as inter-player co-
play networks. Connecting back to the interdependency questions, the intrinsic appeal of clans is
determined by clans’ positioning either as typically competitive/social clans or a blending of the
two on the social competitive spectrum. The extent to which intrinsic appeal affects clan
mortality is likely to be affected by clan boundaries and within-clan co-play networks.
Furthermore, inter-player co-play networks, player-clan affiliation networks and inter-clan
networks are interdependent in their evolution. Below I present qualitative results from
participation observation.
One Day as a Clan Member
46

To know more about how clans operate, I randomly joined the clan Flash Raiders (a
pseudonym) in March 2021 for participant observation. Before sending an application, the only
information about the clan available was on their in-game webpage. Flash Raiders was a large
clan with nighty-eight players. There were quite a few personnel joining and quitting within a
short period of time.
In their public profile, they made it clear that they were family-first and playing for fun.
But they did have skill requirements such as having at least tier 6 tanks and having played at
least 2000 battles. New members would have to go through a two-week probation period. Active
participation in clan activity was necessary and inactivity for three months with no reason would
lead to removal from the clan.
Based on the introduction, Flash Raiders is more of a social clan, because hard-core
competitive clans can require members to make enormous devotion to the extent of
compromising offline lives (Snodgrass et al., 2016). The fact that they are family first shows that
they are more casual about playing. Such organizing culture, based on the recent research, helps
enhance players’ well-being by reducing the stress caused by playing and regulating their
impulse of over playing (Snodgrass et al., 2017).
I sent an application explaining that I would like to join for research purposes, and my
application was quickly approved. As a non-participant observer, I did not perturb the culture by
my very presence (Nardi, 2010). I did not have the necessary gaming skills to join their
skirmishes, so I typically listened to their chat on TeamSpeak (a voice chat application for team-
based gaming) and watched a member’s live streaming in Discord (another voice chat and video
sharing application for communities).
47

Although they positioned themselves as a social clan, in their opinion, each clan was a
unique system, with different mindsets, governance styles and decision-making processes. Such
positioning as a social clan influenced how they recruited new members. Their standards for skill
levels were comparatively low. They said knowing something and not knowing something was
not an issue, because they enjoyed helping and teaching people. Alongside the large number of
new recruits, the central circle of members had been together for several years and they knew
each other personally offline.
Although the clan leaders would also post notices and messages in chat channels in the
game, their TeamSpeak channel was where most communication happened. The channel was
broken into different sections including clan leaders’ offices, rooms for different game modes.
and rooms simply for hanging out. Alongside battle-relevant communication, they also discussed
other game-related topics such as tanks. WoT features over 800 tanks with historically accurate
functional details. These tanks are of different tiers ranging from one to ten. The higher the tier,
the greater its power. Tanks are also of different types, i.e., light, medium and heavy. Usually, a
team needs a combination of different types of tanks to perform different tasks. Since these tanks
are unlocked through both time and purchase by real money, understanding a tank’s details is a
big part of the WoT experience.
Clan-based battles composed of fifteen members reveal how much coordination goes into
the effective functioning of team-based battles. The team leader, named as the caller, has the
authority of leading the team. They do take suggestions from team members, yet they are still the
decision makers. In this clan, the combat officer has played as the caller most often, because the
responsibility of combat officers is to organize clan-based battles. Other higher-ranking members
also play as callers.
48

If the turnout is more than fifteen people, the number require for clan-based skirmishes,
the caller decides who joins the battle. Typically, the caller needs to balance between
performance and fairness. Even though they are a typical social clan, they still want victory as
they can earn gold and other rewards. Since each member of the team is assigned to a different
task, the leader needs to make sure that this member is able to perform a specific task. There are
two concerns. First, a member selected needs to have an appropriate tank to perform that task. As
discussed earlier, tanks are categorized into different types, and within each type, tanks have
different features and functions. Thus, there are cases where the caller is unwilling to select
someone because this player doesn’t possess a good tank to perform a required task. The second
concern is team coordination. Players who don’t show up in the clan very frequently would not
have the level of familiarity and trust developed among players who have been together every
night. Although these reasons for member selection are justified, the caller also needs to consider
how to strengthen the cohesion of their clan as a whole. That means creating opportunities for
members to play with each other, especially for those who have less participation in clan
activities in general. This is a way to equalize the distribution of relational resources within the
clan. What I observed in this social clan is a high respect for fairness. Even when the caller
would make decisions that are the best for team performance and let some members sit aside for
a while, the clan leaders would explain to those members why such decisions are made. In most
cases, they would ask if anyone hasn’t played and make sure everyone who wants to participate
gets a chance. Competitive clans are entirely different. For them, everything needs to make way
for better performance. This accords with studies on the governance within in-game groups.
Competitive groups are characterized by compliance to rules and top-down hierarchy, and social
groups are more democratic (Snodgrass et al., 2017).
49

After forming a team, the caller would discuss strategies with a terrain map for the next
battle. The caller would explain each team member’s position on the map and task in the battle.
In such a strategic plan, some members operating certain tanks might be put into positions that
are easily spotted by the opponent team and they need to be prepared to die. But sacrifice is
required for teamwork, because victory and defeat are earned for the team rather than for any
individual. Naturally, the caller needs to be highly experienced to take such a leadership role, and
it also takes time to build a reputation as a convincing leader. Otherwise, members are likely to
disagree with the caller’s plans, or they simply do not follow the plan in the battle without
voicing disagreement.
An Organizing Context with Generalizability
Previously, I summarized how social interactions inform the theorization of interpersonal
and individual-group dynamics. With contextual details laid out in previous sections, I argue that
WoT is an ideal context to inform the theorization of organizing with wider generalizability. To
establish such external validity, there needs to be sufficient relatedness and similarity between
online and offline organizing (Williams, 2010). The parallel between the two contexts serves two
theoretical purposes. First, theories developed not for virtual organizing can be translated to test
social mechanisms in the virtual context, and those that survive such adaptation are more robust.
Second, given the scale of unobtrusive digital data, virtual contexts can be examined to answer
questions that are not well understood in the offline world, which in turn enrich those existing
theories (Williams, 2010). Following this “mapping” principle, I present the following arguments
to establish the claim that the social architecture of WoT makes the “petri dish” for understanding
human organizing (Williams, 2010).
50

In online gaming communities, the game mechanics and rules incentivize certain social
interactions and constrain others (Lessig, 2006). This is called “social architecture”, where “the
design and system code that shapes player behaviors—filtered through our human social
psychology” (Williams, 2018, p. 177). The social architecture in WoT, especially the design of
clan-based social activities and team-based battles, purposefully constructs individual-group
identity-based attachment and interpersonal bond-based attachment (Ren et al., 2007). Such
social architecture offers opportunities for the application and extension of theories of social
formation discussed here, i.e., the dynamics of category boundaries through multilevel
organizational networks.
Clans are human groups of shared interest, similar-mindedness and compatible functional
skills. Clans build up performance by competing with others through team-based collaboration.
Clan members fulfill different tasks in these collaborative experiences, through which they
develop social bond and comraderies. The competitive organizational dynamics map well on to a
wide range of organizations, which can be divided into project teams for competitive tasks.
These project teams must collaborate to win something, e.g., two teams from two companies
vying for a contract. The theory of niche is applicable to understand how individuals interact
with group resources and how group boundaries are collectively constructed (Sun et al., 2021).
Theories of networks can well examine how individuals build within-/cross-group networks and
how group-level structural properties affect group effectiveness (Monge & Contractor, 2003).
Clans are characterized by formal hierarchy, with eleven roles ranking from the lowest,
i.e., reservist, to the highest, i.e., commander. As a clan increases in size, when the number of
members exceeds the required number of fifteen players in a clan-based battle, hierarchy plays a
more important role in deciding who should be in what team. Although these hierarchical roles
51

are part of the social architecture design, the actual degree of hierarchy depends on the social
norm of specific clans. As clans are categorized towards functional and social goals, they have
bureaucratic or democratic governance styles, different routes of progression to leadership, and
different leadership styles (Snodgrass et al., 2017). Such within-clan hierarchy and leadership
parallel offline formal organizations, and also echo the emergence of lateral leadership in crowd-
based collaboration (Dahlander & O’Mahony, 2011; O’Mahony & Ferraro, 2007).
The social dynamics can also inform questions that are not well understood in the offline
world due to data restrictions. The multilevel and evolutionary perspectives of human organizing
can be extensively explored in the current context. For example, the life cycle of virtual groups is
a much faster evolutionary process with founding, spinoffs, merges, size growth/shrinking, and
disbanding. More importantly, with all the individual-group affiliation data available, it is
possible to identify who founded a clan, who were the first batch of members, who left a clan to
start his or her own clan and who followed, who joined and left a clan, etc. Such multilevel and
evolutionary organizing at a large scale maps well on to the coevolution of companies and
employees, but this problem is not well understood due to data availability. The social dynamics
of virtual groups can thus be leveraged to enrich theories of multilevel networks (Lazega, 2008;
Mützel & Breiger, 2020) and network evolution (Monge et al., 2008).
The Clan Community as an Ecological System
So far, I have provided qualitative understandings of the context. Before deriving
hypotheses following ecology and network theories, two pilot quantitative studies were
conducted to solve two questions essential to the setup of all the hypotheses. Before applying the
ecological framework to the networked organizing in WoT, it is necessary to show evidence that
this is an environment with finite resources such that the number of individuals and groups do
52

not grow infinitely. The first pilot study employs the theory of density dependence to test if the
clan community is competitive.
Previous research has shown that game worlds could be viewed as ecosystems where the
evolution of in-game groups follows life cycles (Chen et al., 2008). In order to apply the ecology
framework to clans, it is important to show that clans compete with finite resources within an
ecological environment. The ecological assumption is that an environment is characterized by
competition for finite resources. Thus, an increasing number of entities would intensify the
competition and reduces the likelihood of the birth of new entities (Carroll & Hannan, 2000).
Such competition is called diffuse competition, because a population of actors doesn’t directly
compete but competes for the same pool of scarce resources in the niche (Carroll & Hannan,
2000).
As explained by Burt and Talmud (1993), such competition due to interdependence in
resources is a form of structural equivalence in network language. One form of inter-
organizational competition is the interdependence on the same pool of members (McPherson,
1983). In other words, organizations are considered as occupying similar niches if they intend to
recruit similar kinds of people. This is applicable for explaining the finite resources in the current
gaming context, because only a limited fraction of players (e.g., 2.74% of total players for the
North American server) are willing to join virtual groups and take on the responsibility of
playing a set number of times per week. And these players are stratified with attributes like skill
levels, ownership of tanks, gender, ethnicity, etc. In other words, in an ecological environment, if
the number of players does not grow infinitely, then the number of clans will not grow infinitely
either. But no studies have tested the assumption of competition for finite resources because an
online gaming community is a novel context in comparison to market competition from the
53

ecological perspective (Shen et al., 2014). According to WoT statistics, the number of players
that played per week has been stable with a slight decreasing trend since 2014 up until now. With
the finite pool of players, how does the number of clans change over time? It is necessary to test
this ecological assumption.
Following the logic of diffuse competition, the intensity of competition increases with
density, or the number of entities, at an increasing rate. Thus, in an ecological environment,
density negatively correlates with founding rate and positively correlates with mortality rate
(Carroll & Hannan, 2000). In the current context, if there’s diffuse competition among clans, the
number of existing clans, or the density of clans, will negatively predict the founding rate.
To test this density dependence assumption, I obtained the clan-level data from July 2016
to December 2020. For each month, a list of newly founded clans was identified if they did not
belong to existing clans from the prior month and the clan age was within one month. The
number of newly founded clans measured the new clan founding rate for each month, and this is
the dependent variable. The founding rate is a time series (Figure 2). The density of clans for
each month was measured as the number of existing clans at that month, and this is the
independent variable.
Figure 2. Founding rate of clans from 2016-07 to 2020-12.

54

Before testing the hypothesis, the times series of new clan founding rate was first
performed unit root test. As shown from the plot, 2017 January is clearly an outlier. I created a
dummy variable for this month and regressed this dummy variable on the time series to control
for this outlier. The residuals were formed into a new time series for unit root test, which was
performed by the Augmented Dickey-Fuller test using the R package of “urca”. In the first step,
none of the three null hypotheses of no unit root, no trend, no drift was rejected. In the second
step, the drift null hypothesis could not be rejected. In the third step, the unit root null hypothesis
was rejected, as the test statistic -3.06 was outside of tau’s one percent range of -2.6. Thus, it is
concluded that this new time series after controlling the outlier contains no unit root.
Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) show that
AR(1) has the strongest correlation (Figure 3). Table 1 shows the autoregressive moving average
(ARMA) model diagnostics. AR (1) has the lowest AIC and higher log likelihood, which
indicate better model fit. Thus, AR (1) model was chosen as the model for further test.
Figure 3. Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) of clan
founding time series.

55

Table 1. Autoregressive moving average (ARMA) diagnostics of clan founding rate from 2016-
2020.

Model 1 Model 2 Model 3 Model 4 Model 5
AR(1) 0.6063

0.6092 0.6036

0.1097

0.1607 0.1431

AR(2)

0.0042

0.1426

MA(1)

0.5039 -0.0047

0.5331

0.1059 0.1939

0.1344
MA(2)

0.2954

0.1296
α 230.898 227.005 230.914 230.921 228.333

25.3805 16.2623 25.448 25.4795 18.8639
AIC 626.08 632.68 628.08 628.08 630.18
Log Likelihood -310.04 -313.34 -310.04 -310.04 -311.09

Negative binomial models are appropriate for analyzing the current data because this time
series is a type of count data and the problem of overdispersion could be taken care of. The sum
of clan battles was treated as a control covariate because this indicates the scale of clan-based
activities in the game. As shown in Table 2, the explanatory variable of the number of clans, or
density, is negatively significant, B = -.191, SE = .079, p < .05. Inclusion of this independent
variable results in higher log likelihood and lower AIC. Thus, the current data supports the
hypothesis of density dependence of clans.
Table 2. Negative binomial times series model predicting the founding rate of clans using tscount
package (N = 54)
Model 1 Model 2
Estimate SE Estimate SE
(Intercept) 1.847 .716 1.883 .671
AR(1) .343*** .109 .429*** .102
Density/10
-4
-.191* .079
Sum of battles/10
-7
.051 .044 .108* .045
Outlier 1.569*** .434 1.523*** .381
Over dispersion .169 .129
Log Likelihood -312.306 -304.691
AIC 634.612 621.381
Note. *p < .05; **p < .01; ***p < .001.
56

Clans Categories—A Semantic Analysis of Clan Profiles
The second pilot study uses semantic analysis to examine clans’ public profiles. It is
necessary to know how such categories are socially constructed at the semantic level within this
community and what each category means. A straightforward way to get to know a clan is to
browse its in-game webpage. There’s abundant information on clan profiles including
performance statistics, complete lists of clan members, acceptance of new members and the loss
of old members. In their profiles, clans usually state who they are and who they want to recruit.
Although the two niche categories of clans are widely acknowledged in the gaming literature
(Williams et al., 2006; Snodgrass et al., 2017; Vesa et al., 2017), how are they semantically
constructed in the current community and how clans articulate these aspects in their profiles?
The semantic approach addresses the recent call for more research attention on how meanings of
categories are socially constructed (Durand et al., 2017). Discourse analysis can uncover the
semantic nuances of collective construction of meanings, especially how clans articulate their
identities in relation to others in the community (Khaire & Wadhwani, 2010; Wry et al., 2011).
The competitive-social category scheme is widely documented in the gaming literature
(Williams et al., 2006; Shen. 2014; Snodgrass et al., 2017; Vesa et al., 2017), A competitive
group, or a “hard-core” group, strives for higher winning rate, superior performance, and
recognition for being the elite of game players. Thus, the competitive aspect focuses on
instrumental elements for playing, e.g., game mode, difficulty level, equipment, character class.
By comparison, a social group, or a “casual” group, values social interactions, friendship,
enjoyment, and helping new players (Williams et al., 2006; Vesa et al., 2017). Thus, the social
aspect focuses on community and social interaction.
57

Due to the different motivations and mindsets of competitive and social groups, they have
distinctive features. Competitive groups are characterized by compliance to rules, reward for
success and punishment for failure (Snodgrass et al., 2017); top rankings, superior performance
over other groups (Vesa et al., 2017); strict playing schedule and selective recruitment
requirements (Shen, 2014); overcoming challenges, necessary equipment to perform better
(Malone, 2009); exclusion towards functional goals (Williams et al., 2006).
Social groups are characterized by real life first, collaboration and camaraderie, collective
achievement, support and tolerance, friends and family (Snodgrass et al., 2017); social harmony,
friendship, leeway for participation (Vesa et al., 2017); playful engagement (Shen, 2014); casual
and togetherness (Malone, 2009); blending of online and offline social interactions and bonds
(Williams et al., 2006).
Clans’ public profiles were quite stable and obtained from the data warehouse in July
2019. This study employs structural topic modeling (STM) to understand the semantic elements
of clan profiles. STM is an increasingly popular topic modeling approach based on the algorithm
Latent Dirichlet allocation (LDA) (K. H. Kwon et al., 2019). Compared to other topic modeling
approaches, STM assumes the presence of multiple topics in one single document and provides a
probability score on a [0,1] interval. Data analysis was conducted with the R package of “stm”
(Roberts et al., 2019). To prepare for topic modeling, language detection was first performed to
remove non-English profiles. The English profiles were cleaned following standard natural
language processing procedure including removing punctuation, numbers, stop words and
stemming.
The next step was to detect the optimal number of topics in the cleaned corpus of text.
Existing studies of topic modeling have demonstrated that statistical measures can be used as a
58

reference but not as an exclusive justification for model selection as they can sometimes lead to
less meaningful model choice (Roberts et al., 2019). I followed the procedure stated by X. Chen
et al. (2020) to compare several models with different numbers of topics. The built-in function
showed that the models with 8, 9, 10, 11 topics had relatively high held-out log likelihood, low
residuals, and high semantic coherence (Figure 4). I invited a PhD student whose main research
area is gaming studies and who has done research in the current research context. We
independently evaluated the four models, including top keywords and the five most probable
profiles. It turned out that we both chose the model with 10 topics. We then independently
labeled these topics and reached an intercoder reliability of .84 (Cohen’s Kappa). The human
coding is consistent with the correlation among 10 topics, such that competitive and social topics
are grouped in separate clusters (Figure 5).
Figure 4. Optimal topic number detection of structural topic modeling.

59

Figure 5: visualization of the correlation matrix of ten topics using the stm build-in function.

60

Table 3. Ten topics and associated keywords and most probable clan profile.
Topics Top keywords and sample clan profile with highest probability
Topic 1: social Keywords: game, friend, people, enjoy, together, make, can, work, old,
everyone, family, start, great, world, community, year, focus, mature,
much, gamer.
Sample: We are looking to expand our little community into various
games to find gamers who enjoy gaming as much as we do. We are
looking to recruit mature active players looking to have fun and help out
their fellow clan mates.

Topic 9: social Keywords: respect, member, rule, other, learn, well, help, welcome,
teamwork, always, tolerate, battlefield, may, mean, import, know,
language, expect, strive, small.
Sample: Honor. Conduct yourself in manner representing VetS and
yourself that brings respect admiration and honour from Clanmates and
others within the WOT community at all times. Integrity. Never lie cheat
or steal and do not tolerate this behavior from others. Maintain your
impartiality and self-control and others will follow. Honesty. Be truthful
with yourself and others.

Topic 10: social Keywords: fun, want, look, like, just, good, player, new, join, come, tank,
win, casual, back, now, take, serious, state, love, life.
Sample: Looking to grow as a WoT player. Look no further. Looking for
an unoppressive casual place to tank. Looking for somewhere where
XvM Rank and Stats don’t matter. Looking for a place to place in
tournaments and become competitive. We are growing. We want you.
We started casual and are feeling the burn to grow and become Great
tankers without becoming Elitist Stat Mongers. Come in and join the fun
help us prove casual is the new competitive.

Topic 8: social Keywords: play, player, time, use, group, need, tanker, welcome, wot,
communication, chat, level, run, skill, voice, social, part, drama, without,
provide.
Sample: A social group we have no tank or tier requirements. No
minimum WR PR or wn either. Haters and those with attitudes need not
apply. We play because we enjoy the game. We focus our efforts on team
building platooning SHing and doing the daily weekly WoT missions.
We encourage players of all skill levels to apply.

Topic 2:
competitive
Keywords: clan, team, war, require, tournament, skirmish, also, better,
improv, build, learn, stronghold, grow, strong, goal, even, experience,
attitude, dedication, commitment.
Sample: Building our team. Training our team. Battling our team.
Improving ourselves. Strongholds. Advances. Clan wars. Tourneys.
Platoons.
61

Topic 3:
competitive
Keywords: tier, require, recruit, stronghold, active, player, week, contact,
recent, look, war, rate, night, current, except, advance, interest, higher,
overall, made.
Sample: SIMP is a top clan seeking active competitive players. We
participate nightly in clan wars advances and skirmishes supplemented
by tier X credit boosters. We have fun hone our skills and regularly seek
fights with other top clans. We compete at a top level while keeping the
good community aspect.

Topic 4:
competitive
Keywords: member, active, platoon, speak, participate, train, office,
work, server, accept, English, stronghold, gold, check, event, ask, base,
address, open, regular.
Sample: We Offer Personal Public and Platoon WOT Training with Top
Commander formerly from RU Server on US server to US an UK
players. We specialize in training those that are willing to be trained and
offer individualized or team clan training sessions. We also train in
public training rooms. Besides training our own clan members we train
non clan players of any sort.

Topic 6:
competitive
Keywords: will, must, join, can, please, get, discord, apply, one,
application, command, free, help, interest, able, see, website, feel, talk,
know.
Sample: This is our Elite Tournament Division. This is the primary focus
of this Division. 18+ age. Tier 6, 8 and 10 tanks, 50%+ win rate, 5 ,000
Player Rating 1,200+ wn8 or 1,500+ recent wn8, 5+ nights each week for
Tournaments.

Topic 7:
Competitive
Keywords: TeamSpeak, tank, battle, least, com, day, minimum,
competition, win, mature, mic, prefer, participate, info, avail, encourage,
expect, age, global, type.
Sample: Looking for a clan? Wanting to improve your experience? 101
Broken Arrow is a new competitive clan founded by experienced,
grounded, and active players. If interested\, you must be willing to
commit to a clan that has a drive to win! TeamSpeak is required.

Topic 5:
Undistinguishable
based on historical
facts.
Keywords: fight, tank, division, world, one, unit, enemy, force, military,
never, armor, war, will, victory, brother, stand, group, honor, army, die.
Sample: Sounds of sporadic rifle fire break across the ridge line. In the
distance the deep rumble of American heavy artillery echoes up and
down the valley. The Allied offensive is about to begin.

62

As can be seen from Table 3, the social aspect of clans is centered around concepts such
as community, friendship, family, togetherness, respect, and enjoyment. The social aspect is
more about a group of like-minded people having fun together. By comparison, the competitive
aspect of clans is explicit about active contribution to clans’ competitive performance. For
instance, clans require members to reach certain skill levels, have certain tiers of tanks, can
commit to certain number of days per week, have audio speaking software for coordination
during battles, and follow clan rules. Topic 5 didn’t follow the conventional language widely
accepted by the community but used the description of troops that existed in history.
It is worthy of pointing out the contextual basis of the keywords shown above. The
keywords associated with the social aspect of in-game groups, e.g., friend, enjoyment, together,
family, community, respect, tolerance, fun, casual, communication, can be generic to different
games regardless of game themes. The keywords associated with the devotion aspect of “hard-
core” playing, e.g., dedication, commitment, improvement, requirement, minimum of
participation, can also be applied to all games. However, the language that describes the
functional and instrumental specifics of competition can be highly dependent on the game
design. In other words, players in different games will use different words to describe the
functional details of team-based competition.
The current research context, WoT, features mid-20th century historically realistic
warfare and tank-themed combats. As an integral part of an authentic military experience, clans
mimic military organizations where members hold roles with different ranks, e.g., commander,
combat officer, intelligence officer, recruitment officer etc. Thus, the functional design of WoT is
characterized with military-style terms. By comparison, another game, World of Warcraft
(WoW), is a fantasy-themed role-playing game featuring magic, dragons, monsters, alien worlds,
63

etc. Much research has been done on the in-game groups in WoW, which are called guilds
(Nardi, 2010; Snodgrass et al., 2017). Although these in-game groups in WoT and WoW are very
similar in organizing players who share interests and mindsets, they use completely different
language systems to describe the functional aspects of playing. For example, in WoT, a game
match is called “a battle”, and it’s called “raiding” in WoW. Game modes in WoT include
“skirmishes”, “platoons”, “strongholds”, “advances”, “global maps”, “tournaments”, so clans
usually state what game modes they play. Game difficulty settings in WoW are called “heroic”
and “mythic”, and some guilds state what level they play. WoT requires a player to maneuver a
tank in team-based battles, so clans usually state that they want to recruit players with certain
“tiers” of tanks. As a role-playing game, WoW has 12 character classes, e.g., “warrior”, “hunter”,
“priest”, so it is common for guilds to state which character classes they look for. Finally, voice
chat tools like “TeamSpeak” and “Discord” are required for all collaborative playing. As shown
in this comparison, although many games are essentially about team-based collaboration and
competition, the actual language to describe the functional specifics, i.e., game modes, difficulty
levels, skill levels and equipment, largely depends on different game themes.
To summarize, the current empirical context is valuable for understanding the dynamics
of social formation. Clan categories are socially constructed with meanings open to changing
interpretations. Clans organize players’ interests, skill levels, motivations and mindsets in a more
refined way. Players’ mobility among clans erodes clan boundaries, and some clans form
stronger collective identities than others. The importance of collaboration in the current context
makes social networks among players a highly integral part of social formations.

64

Chapter 4: Hypotheses Development
The previous two chapters have introduced the broad theoretical picture and the empirical
context. This chapter focuses on hypotheses development under the guidance of specific theories.
As introduced in previous chapters, players, clans and clan categories are positioned at three
levels, and there are co-play networks among players, players’ membership to clans and clans’
membership to clan categories. This chapter includes three studies to understand the evolution of
multilevel organizational networks. The first study on clan mortality is situated in the existing
research on the association between organizational niche width and performance, and takes into
account the fact that organizations are not homogenous entities but they have fluid boundaries
and they are networks structures. Although this study considers the interconnectedness among
organizations by incorporating their fuzzy boundaries, the use of survival models still assumes
the independence of observations. To address this weakness, the second study seeks to directly
model the multilevel interdependency mechanisms using Multilevel Exponential Random Graph
Models (MERGM). However, MERGM has restrictions on sample size and cannot provide a
holistic picture. The third study adopts the Relational Event Models (REM) to examine
interdependency at a large scale. Approaching interpersonal networks and individual-group
affiliation as two ways to access resources, this study highlights the role of individual agency in
social actions.
Study 1: The Survival of Clans
This study follows existing research on the relationship between organizations’ niche
width and performance (Durand et al., 2017) and proposes that the mortality of clans is
influenced by their occupation in niche categories. However, niche occupation only indicates
intrinsic appeal, which focuses on how audiences make sense of producers’ positions in a
65

categorized social space. Actual appeal is determined by both intrinsic appeal and engagement
together (Hannan, 2010). Organizational-level properties would determine how well they could
engage with certain social positions. Echoing the second multilevel perspective of organization
studies as discussed in Chapter 2, how organizations respond to and process external resources is
contingent upon their internal properties (Funk, 2013; Kim et al., 2006; Paruchuri et al., 2019).
This study introduces two organizational-level features. The first one is the fuzziness of
organizational boundaries as a form of cross-organizational dependency, which is introduced by
boundary-spanning individuals affiliated to multiple organizations. The second feature is the
interpersonal network structure in organizations as a form of within-organization dependency.
This study thus proposes that the effect of organizations’ niche occupation on survivability
cannot be examined in isolation of organizations’ boundaries and internal network properties.
Organizational mortality
Organizational mortality is one of the most important fitness indicators (Hannan et al.,
2007). It is defined as “dissolution or loss of identity in a merger with other organizations
(absorption)” (J. Freeman et al., 1983, p. 694). Organizational mortality is different from spin-
offs, where an organizational entity is separated from its parent (Corley & Gioia, 2004), because
for organizational mortality, the original organization no longer exists. Organizational mortality
can happen in several forms such as disbanding (J. A. Baum & Singh, 1994; Hannan & Freeman,
1988), bankruptcy (Levinthal, 1991), merger & acquisition (J. Freeman et al., 1983), and exit the
market (Agarwal et al., 2002). Organizational mortality has been studied in connection with
many ecology theories. It is the key outcome for age dependence theory (Bakker & Josefy, 2018;
Stinchcombe, 1965). For example, following the logic of liability of newness, labor unions that
lasted for ten years had a disbanding rate of only 5% as large as the rate at their founding stages
66

(Hannan & Freeman, 1988). A different perspective is liability of ageing, which suggests that
older organizations are more susceptible to failure because they become less agile in responding
to changes and new competitors (Barron et al., 1994). In the theory of density dependence,
organizational mortality is driven by diffuse competition, and such effects are moderated by
institutional factors (Baum & Singh, 1994; Carroll & Hannan, 2000; Lander & Heugens, 2017).
Organizational survival is not necessarily determined by performance (Meyer & Zucker,
1989). The same is true in the current empirical context. Given social and competitive clans’
distinctive motivations, social clans do not pursue superior high rankings and win rates as the
main objective. Thus, using competitive performance measures tend to be biased in the current
context. Some social clans do not have outstanding rankings, but they may survive for a long
time. Just like offline organizations, virtual groups go through lifecycles, and their disbanding
and founding are common in gaming communities (Chen et al., 2008).
A randomly chosen sample of 200 clans that disbanded before 2019 April reveals what
happened before and after clans ceased to be active. These clans on average had 14.28 members
in the last of month of their survival. And this number was 18.59 for those clans with more than
one member in their last month. For those clans with more than one member in their last month,
during the next month of their clans’ disbanding, on average 33.3% of the original members
joined new clans, the rest were not in any clans. These members with new clan affiliations on
average showed an unequal distribution of new clan memberships (M = 2.00, SD = 1.85, range =
2.85). That is, some members joined new clans together whereas others joined new clans on their
own. On average 61.3% of those members with new clan affiliations joined the same new clans.
This number was 45.9% for those clans with more than the mean number of members in their
last month before disbanding. To summarize, most members did not join any clans right after
67

their own clans disbanded. If they did join new clans, some members were likely to form small
clusters and joined new clans together.
The Classical and Revised Theory of Niche Width
Classical niche theory concerns both fundamental and realized niches. As introduced in
Chapter 2, a fundamental niche refers to the full range of resource spaces needed to sustain a
population, whereas a realized niche refers a population’s resource space in the presence of
competing populations (Hannan & Carroll, 1992). The original niche width theory focuses on the
fundamental niche. J. Freeman and Hannan (1984) interpret niche width as “a population's
tolerance for changing levels of resources, its ability to resist competitors, and its response to
other factors that inhibit growth” (p. 1118). Niche width distinguishes generalists who occupy a
wide range of resources, and specialists who occupy a narrow set of resources (J. A. C. Baum &
Singh, 1994). Hannan and Freeman (1977) discussed the paradox of being a generalist. On one
hand, drawing a wide variety of resources from the environment gives generalist organizations
excessive capacity to change but environmental turbulence would render additional difficulties
for generalists to coordinate changes across sectors. By examining both spatial and temporal
environmental variations, scholars found generalists have a higher mortality rate in fine-grained
environments, but they are better off when environmental variation is both coarse grained and
large (Hannan & Freeman, 1983).
In contrast, the resource-partitioning theory focuses on realized niche. Carroll (1985)
proposed a resource-partitioning model that shows when generalists adopt concentration
strategies, generalists compete among themselves and their mortality rate increases. The winning
generalists are unable to absorb all the resources released by disbanding generalists and the
remaining resources are very likely to be absorbed by specialists to increase their survival rate.
68

Carroll and Hannan (2000) argued that resource-partitioning theory could explain why the
market entry rate increases when the market is dominated by a monopoly. When the generalist
tries to appeal to a diverse range of customers, it loses attraction to small niche market needs,
which releases untapped resource spaces for specialist entrepreneurs to occupy (Carroll &
Hannan, 2000). Carroll & Swaminathan (2000) built on the previous resource-partitioning model
to include scale-based competition and the normative effects of identity to generalize the model.
According to them, specialists choose narrow homogenous targets, while generalists choose
heterogeneous segments (Carroll & Swaminathan, 2000). In the competition among generalists,
larger generalists outcompete smaller generalists, but it is difficult for the winning generalists to
occupy the entire resource space. As a result, the size and breath of generalists increase but the
combined resources they occupy decline, which releases some distant and specialized resource
space that can support specialists (Carroll & Swaminathan, 2000).
The revised theory of niche width takes into account the fuzzy sets logic. The extent to
which a producer crosses cognitive boundaries is measured by the construct of niche width,
operationalized as the unevenness of the distribution of grade of membership (GoM) across
categories. A broader niche means a lack of focus, which is likely to confuse audiences and
lower appeal (Negro et al., 2010). The resulting ambiguity disrupts heuristics for understanding
entities’ positions in the categorized space and so creates a cost to the process (Paolella &
Sharkey, 2017). The rationale behind this logic is a trade-off between width and depth both on
the side of producers and on the side of audiences. To be more specific, on one hand, having a
broad niche width reduces producers’ chance of skill specialization and harms their capacity to
engage with target audiences; on the other hand, claiming membership to multiple categories
confuses audiences who rely on category boundaries to make sense of producers and their
69

products, and therefore, category spanning lowers appeal (Negro et al., 2010; Hsu et al., 2009).
For example, on the producer side, sellers on eBay need to formally declare the categories they
engage with, which define the target consumers. Hsu et al. (2009) found sellers that engage with
more than one category are less likely to sell well. On the audience side, films with larger niche
width in genres received lower ratings from critics and fans, because audiences find it difficult to
make sense of products that claim multiple memberships in established categories (Hsu et al.,
2009).
Consistent with this view, scholars of categorization argue that audiences have limited
cognitive capacity, so they compare objects with typical categories during sensemaking, which
makes categories resemble disciplinary and normative mechanisms (Durand et al., 2017).
Comparison with typical members of a parent population is a type of representativeness heuristic
when evaluating uncertain events (Grodal & Kahl, 2017). Zuckerman (2000) adopts a
neoinstitutional view that companies have to conform to a stock market’s segmentation to gain
endorsement from industry specialists. Following that, abundant studies have supported the
“typicality judgment” perspective (Paolella & Sharkey, 2017). Hsu (2006) found films with a
large niche width of genres could attract many viewers, measured as the number of critics’
reviews on a film rating website, but had low audience appeal, measured as the average ratings
from the film review website. The negative relation between niche width and audience appeal
was found to be mediated by audience consensus (Hsu, 2006). Negro et al. (2010) found a wine’s
appeal to critics declined with the niche width of the producer’ wine style profiles. Lerner and
Lomi (2018) studied the hierarchical categorization of Wikipedia articles, and found the coarse-
grained articles could attract a large number of contributors but it was difficult to reach
70

consensus on the quality. The fine-grained articles, on the other hand, attracted less attention but
were easier to evaluate and received higher ratings (Lerner & Lomi, 2018).
Another stream of research calls for reflection on this view of disciplinary categories and
argues that the imperative pressure of categories changes with “the social challenges and
opportunities for engaging in, and learning from, experiments with unconventionality”
(Zuckerman, 2017, p. 31). That is, audiences might tolerate category spanning more than
previously thought (Durand et al., 2017; Durand & Paolella, 2013). Zuckerman (2017) proposed
“categorization as a theoretical tool” by arguing that imposing imperative and exceptional
performance for unconventionality do not necessarily contradict each other: “[t]he fact that an
audience imposes a categorical imperative does not mean that it is closed to the fact that better
performance can come from unconventionality” (p. 59). The imperative pressure changes with
“the social challenges and opportunities for engaging in, and learning from, experiments with
unconventionality” (Zuckerman, 2017, p. 31).
There are mainly two reasons for this view. First, audiences have different expectations
for category spanners and conformists, and they have more tolerance for spanners. For instance,
Smith (2011) found that the nonconformist identity had both “amplification” and “buffering”
effects in the hedge fund industry (p. 61). That is, atypicality with good performance could
generate additional reward whereas atypicality with bad performance entailed less punishment
from evaluators. Similarly, Hsu et al. (2012) argued for a trade-off between certainty and success
such that although production companies of films combining genres overall had higher mortality
risks, they also had a higher potential for exceptional success.
A second reason is that some audience members constantly learn and upgrade their
evaluation (Durand & Paolella, 2013), and appeal are likely to differ for different types of
71

audience members (Pontikes, 2012). “Market-takers” are consumers who use existing categories
as ways to assess products, and ambiguity makes their assessment task more difficult, and hence
they prefer products that stick to established categories. By contrast, “market makers” seek to
develop new niches, and thus ambiguity means flexibility and the possibility of innovation.
Pontikes’ (2012) research on the software industry found that organizations with ambiguous
categories were less attractive to consumers but more attractive to venture capitalists as market
makers. Similarly, Paolella and Durand (2016) found law firms spanning categories were
evaluated positively because they were viewed by clients as capable of handling complex
situations. Building on this study, Paolella and Sharkey (2017) study of corporate law firms
showed a U-shaped relationship between category spanning and perceived identity clarity. That
is, evaluators of law firms were more likely to reach consensus on the specialized boutique
identity of firms that span a few categories, and the full-service identity of firms that span many
categories. By contrast, those law firms spanning a middling number of categories were more
likely to confuse evaluators. Lo and Kennedy (2014) study of the nanotechnology industry
showed that patents blending multiple classes took longer to approve but had higher citations,
although both effects diminished with time. In the innovative industry of nanotechnology, an
institutional logic encouraging interdisciplinary research also alleviated the negative effect of
category blending.
Other studies found memberships to multiple categories might not be harmful if the
classification boundaries were not well defined (Ruef & Patterson, 2009), or during stages of
category emergence (Alexy & George, 2013). Alexy and George (2013) differentiated between
affiliation to illegitimate categories as “divergence” and attempts to blur category boundaries as
“straddling”. Their study of the software industry showed a negative effect of divergence and a
72

positive effect of straddling. Although category straddling creates less meaningful identities and
thus less appealing, it may also bring positive effects for emergent categories (Alexy & George,
2013). This is because by associating with established categories, an emergent category can
obtain legitimacy spillover from established ones and such association also helps audiences to
make analogies to facilitate the understanding of the new category.
Most studies following the revised theory of niche width focus on either audience appeal
(Hsu et al., 2009; Negro et al., 2010; Paolella & Durand, 2016) and performance (Alexy &
George, 2013) as the dependent variable. Organizational mortality receives relatively less
attention. Nevertheless, organizational mortality is closely connected with organizational niche
width. This relationship has received robust support in various empirical contexts. For example,
Dobrev et al. (2001) argued that the effect of niche width on organizational mortality cannot be
considered in isolation from niche overlap density. This is because as an entity broadens its
niche, the extent to which its overlap density with others also inevitably increases, which is
likely to result in more fierce competition. Their study of automobile manufacturing firms
showed that expansion in niche width increased firms’ mortality hazards (Dobrev et al., 2001).
Films spanning genre categories receive lower average ratings (Hsu, 2006), and film
producers with large niche width are more likely to exit from the market (Cattani et al., 2008). In
their analysis, American film production companies’ full life histories were tracked, and
organizational mortality was indicated as ceasing to release any additional films (Hsu et al.,
2012). The higher the proportion of films with multiple genres produced by a company, the
higher the mortality hazards of that company (Hsu et al., 2012). Cattani et al. (2008)
operationalized film production companies’ niche width as the number of film genres they
73

operated each year, and they also found companies with large niche width were more likely to
exit the market.
Tracking the breweries’ life histories in Germany, Beck et al. (2019) showed that
breweries that deviated from prototypes suffered higher failure rates as audiences penalized
noncompliance to their expectations. However, this relationship was contingent upon
organizations’ geographical locations in relation to the center of the industry cluster. The closer
to the industry center, the stricter the audiences’ expectations of adherence. In other words,
penalty for deviations diminishes with distance from the central cluster, and these distant
breweries were more likely to survive (Beck et al., 2019).
Apart from corporations, how niche width affects organizational survival has also been
recently studied in social movement organizations (SMO) (K. H. Kwon et al., 2019). SMOs’
niches lie in the issues they advocate. A focused niche helps convey an organization’s advocacy
purposes more clearly and thus communicate a more recognizable identity (Olzak & Johnson,
2019). Since SMOs’ audiences consist of supporters, donors, volunteers and the general public, a
higher audience appeal indicates a higher survivability. In their empirical analysis of
environmental protest SMOs, they argued that a clear issue focus signals high legitimacy as an
authentic member of an issue category, and such signal of commitment helps attract and
maintain supports that are also committed to the area. Such mutual commitment from both
organizations’ and adherents’ sides creates a clear objective that prevents internal conflict (Olzak
& Johnson, 2019). These rationales explain why SMOs with a narrow niche have lower
disbanding hazards (Olzak & Johnson, 2019).
In the current case, the two broad niche categories of social and competitive clans are
well defined, as shown in existing research on in-game organizations (Vesa et al., 2017). That
74

means audiences, including current members and potential members, have clear expectations on
the two categories of clan, and they look for clans that align with their interests and mindsets.
Thus, this study proposes that clans with ambiguous self-description to categorize into the two
niche categories are unlikely to establish a rigorous position in the ecological space. As shown in
the findings from the participant observation, social and competitive clans have very different
mindsets. The commander of the social clan I was observing put it in this way: “we try to be
competitive without paying the price”. Competitive clans would do whatever they can to
improve performance. Different mindsets lead to different governance and leadership styles.
Social clans value interpersonal bond and respect equality, whereas competitive clans require
obeying to rules. Given this distinction and incompatibility of the two categories, a broad niche
is likely to confuse their members and potential membership applicants when players evaluate
the alignment of interests. Such violation of expectation reduces the intrinsic appeal of clans with
a broad niche. Given the relationship between niche width and organizational mortality as
discussed above, it is proposed:
H1: Clans with narrower niche width have lower mortality hazards.
The Theory of Density Dependence and Category Contrast
The second ecological factor predicting clan mortality is the sharpness of the
organizational boundary, which is built upon the theory of density dependence and contrast.
Density dependence theory posits that the number of organizations in a population, or density, is
a proxy for its legitimacy or social acceptance of an organizational form, which has a positive
effect on the emergence of that form. Crossing a threshold, density would lead to competition
(Carroll & Hannan, 2000). This theory posits that the two evolutionary processes of legitimation
and competition shape the founding and disbanding of organizations (Hannan & Freeman, 1989).
75

At low levels of density, legitimation dominates, and at higher levels of density, competition
dominates. Because the mechanisms of legitimation and competition function within an
ecological niche at the population level rather than through direct collaboration or competition of
two entities in the niche, they are called “constitutive legitimation” and “diffuse competition”
(Carroll & Hannan, 2000). Lander and Heugens (2017) used a meta-analysis to incorporate
institutional factors into the baseline density dependence model in an effort to reconcile ecology
and institutional theories. They pointed out that both theories prioritize legitimacy, but ecologists
focused on constitutive or cognitive legitimacy, while institutionalists give special attention to
sociopolitical legitimacy, that is, acceptance by key institutional stakeholders (Lander &
Heugens, 2017). They explored how these two types of legitimacy work together to influence the
mechanisms of density dependence (Lander & Heugens, 2017).
Under the fuzzy set assumption, Hannan et al (2007) defined cardinality of a fuzzy set, or
fuzzy density, as the sum of GoM of all members in that category. Bogaert et al. (2010) found
fuzzy density increased the legitimacy of an emergent category. In other words, low fuzziness
and spillover from increasing density facilitated the legitimation of a new category. Using a
meta-analysis, Bogaert et al. (2016) showed the average effect of density dependence on
legitimation is positive, and they proposed that two population-level characteristics: simplicity of
organizational goals and tangibility of offerings, could explain the large variance across
populations. These two constructs are closely related to fuzzy categories, and their conclusion is
that cognitive fuzziness dampens the legitimation effect of density dependence (Bogaert et al.,
2016). In other words, even with a large number of members in a category, if the category
fuzziness is high, the legitimation of this category is still low.
76

A closely related concept to fuzzy density is the construct of category contrast. It is
defined as the extent to which categories demarcate sharp boundaries, and it is measured by the
average GoM of all members in that category (Negro et al., 2010). Bogaert, Boone, and Carroll
(2010) put fuzzy density and contrast in the same study and found both decreased the hazard of
firms exit during the category emergence. The difference between fuzzy density and contrast, as
reflected in their operationalization, is that contrast only considers the sharpness of boundaries
whereas fuzzy density considers both boundaries and density (Bogaert et al., 2010).
When a category has high fuzziness, it means many members of this category have low
GoM, and subsequently there’s little consensus on the attributes that define the membership to
this category (Negro et al., 2010; Vergne & Wry, 2014). When category contrast is low, it cannot
stand out from the background and it is difficult to differentiate it from other categories.
Therefore, membership does not bear much meaning and spanning categories does not create
problems (Kovács & Hannan, 2010). Low contrast also harms the collective identity of members,
such that even loyal members claiming this category have low appeal (Negro et al., 2010). On
the other hand, the higher the average contrast of categories that an actor claims, the more
difficult for competitors to imitate (Carroll et al., 2010). Following that, scholars have found that
the negative effect of boundary spanning is more severe if the categories crossed are distant with
high contrast (Kovács & Hannan, 2015), but less severe if the categories crossed are similar
(Wry & Lounsbury, 2013). From another perspective, the fuzziness of categories even
incentivizes members to span categories because there’s little downside (Hsu et al., 2012).
Categories with low contrast face survival difficulty essentially because categories with
high fuzziness entail disagreement on their meanings and taken-for-grantedness (Hannan, 2010).
Such low consensus prevents a category from emerging and thriving (Hannan, 2010). Category
77

boundary has been conceptualized as internal coherence in terms of members’ homogeneity,
which is conducive to category viability and prevents category failure (J. Y. Lo et al., 2020). For
example, a music genre category failed to be legitimated because of contestations about its
meaning (Boone et al., 2012). Similarly, bitcoin was labeled with over one hundred terms in
British media and such inconsistencies prohibited its emergence (Vergne & Swain, 2017).
Another mechanism that explains the relationship between category contrast and
category-level survival is through how contrast affects individual member’s survival. A clearly
demarcating boundary serves as a basis for enduring expectations and evaluations (Negro et al.,
2015), even for small populations (Navis & Glynn, 2010). A greater category contrast provides
legitimacy to member entities, defines them more clearly, strengthens their identification, and
thus increases the survival chance of individual members (Kuilman & Wezel, 2013). For
example, the contrast of UK airline industry lowers the mortality rates of individual companies
in this category (Kuilman & Wezel, 2013). Similarly, in the tape driver industry, firms producing
formats with high contrast had lower mortality hazards (Carroll et al., 2010).
However, how different subpopulations of a category behave differently and how do
those behaviors in turn influence the overall category contrast and category survival is less clear.
Since contrast represents the average GoM of all member entities, members with different GoM
are likely to benefit differently from category contrast. Some scholars argue that members with
low GoM are more likely to have higher marginal benefits, because they are otherwise difficult
to be identified by audience members (Bogaert et al., 2010). Thus, members with low GoM are
more motivated to stay in the category as free riders of legitimacy spillover (Kuilman & Wezel,
2013). In this case, if there’s little sanctioning force to prohibit entities with low GoM from
claiming the label, such practice is likely to persist. With an increasing number of entities of low
78

GoM, the schema of the category becomes blurred until category contrast fades permanently and
disband (Kuilman & Wezel, 2013). Some other scholars argue that a greater category contrast
actually triggers stronger expectations for members and imposes more penalty on those with low
GoM. That is, a sharp category boundary highlights the mismatch between members with low
GoM that distant from typical members of a category, so members with low GoM might have to
leave the category because of high sanctioning pressure. This alternative reasoning might explain
the finding that firms with low and high GoM did not differ in lower mortality hazards from a
great category contrast (Kuilman & Wezel, 2013). These studies strengthen the importance to
consider how members of low GoM affect category survival.
In the current context, this factor is group boundary. In the hierarchical categorization
system, given the two broad group categories of social and competitive clans, groups are more
fine-grained social categories that coordinate individuals’ interests, mindsets, skill levels and
motivations. Furthermore, as a special type of social category, group categories are also defined
by other factors including gender, ethnicity, values or choices (Postmes & Spears, 2000). Group
boundaries are eroded by boundary-spanning individuals that are affiliated to multiple groups.
For individuals, given one’s complete group affiliation history, the more attention and effort one
allocates to a group, the higher the GoM in that group. Groups with high contrast have more full-
fledged members who have allocated more attention to these groups. Therefore, groups with
sharper boundaries have distinctive collective identities that differentiate them from other
alternative groups. In other words, members from groups with high contrast are more likely to
share similar features. Groups with low contrast, however, are likely to face contestations and
disagreement on their interests, objectives, mindsets and values.
79

For virtual groups like clans, since social relations are especially transient and group
boundaries are more permeable than offline groups, a virtual group would easily disband without
enough members or these members diversify their engagement with other groups. For a clan to
survive, there needs to be a proper boundary and a socialization process to govern the collective
public goods (Ransbotham & Kane, 2011) and for effective leadership to rise (Dahlander &
O’Mahony, 2010; O’Mahony & Ferraro, 2007). For example, for competitive clans, it takes
repeated battles to develop group cohesion and establish authority in leadership, which are
essential for better performance. For social clans, it also takes time to build trust and emotional
support. Effective governance is also critical for social clans, because if they do not have any in-
game achievement, e.g., training opportunities, skirmish victories, they can hardly retain
members. Thus, it is proposed:
H2: Clans with sharp boundaries have lower mortality hazards.
As discussed above, groups occupying broad niches are more likely to disband. When the
categories of groups have distinctive objectives and cultures that are widely accepted in the
community, “jack of all trades” violates expectations of existing and potential members. Given
the rationale established between niche width and organizational mortality (Beck et al., 2019;
Hsu et al., 2012; Olzak & Johnson, 2019), groups occupying a clearly focused niche have lower
mortality hazards. At the group level, groups with clear boundaries are less likely to disband,
because they have a collective identity and within-group consensus on group values and
mindsets.
Since group niche width only captures the intrinsic appeal, how such a niche occupation
pattern affects organizational outcomes is also contingent upon the engagement with that niche
position (Hannan, 2010). Given such contingency, this study argues for an interaction effect
80

between group niche width and group boundary on mortality hazards. Specifically, groups with
sharp boundaries are characterized with homogenous members with shared values and clear
objectives. Based on such homophily, members are more likely to develop a strong social bond,
because identity-based attachment to a community and social embeddedness are closely
connected (Cheng & Guo, 2015; H. Jiang & Carroll, 2009). And stronger social relations in turn
strengthen individual-group attachment (Sun et al., 2021). Thus, groups with sharp boundaries
have higher cohesion, which enables them to reach a greater degree of consensus on group
governance. As discussed earlier, competitive and social clans differ distinctively in
bureaucratic/democratic governance (Snodgrass et al., 2017). Groups with members’ mindsets
better aligned can effectively develop a governance structure with clear prioritization to engage
with a focused niche.
By comparison, if a group has fuzzy boundaries, meaning most members allocate limited
group interaction experience to this group, then this group does not represent many members’
mindsets and interests. In other words, groups with low contrast are likely to face contestations
and disagreement on their interests, objectives, mindsets and values (Olzak & Johnson, 2019).
Fluid boundary also makes it more difficult to develop interpersonal bond and social
embeddedness. Without strong social relations, it is even more difficult to address possible group
conflicts or bridge gaps (Curşeu et al., 2012). Even if a group occupies a narrow niche position,
such misalignment of members’ interest and objectives makes it difficult to perform well in that
narrow niche position. Thus, groups with fuzzy boundaries would be better off to engage with a
broader niche position.
Hence, for clans that occupy narrow niche positions in their self-descriptions for either
social or competitive goals, it would be easier for them to survive if their members allocate more
81

attention and effort in these clans compared to other clans. A strong collective identity will
strengthen this clan’s ability to engage with a clear social or competitive objective. For clans that
occupy broader niche positions, it is less harmful if members’ interests and goals are not
perfectly aligned. Thus, this study proposes an interaction effect between clan niche width and
clan contrast in membership.
(H3) When clans occupy narrow niche positions, those with sharp boundaries have lower
mortality hazards.
Network Brokerage from a Collective View
Following the first multilevel network perspective as discussed in Chapter 2, i.e., the
dependency between local and global network properties, this study proposes the third factor
conducive to clan survival is group-level network brokerage, which indicates the distribution of
relational resources within a group. Structural hole, or network brokerage, means a node bridging
two partners unconnected themselves (S.-W. Kwon et al., 2020; Monge & Contractor, 2003).
Individuals who span structural holes, or connect nodes that are unconnected to each other, have
competitive advantage in the control and access of information and resources (Burt, 2004). The
broker is positioned to access non-redundant information with little overlap, and the broker is
usually unsanctioned from commitment pressures (Burt, 2005; Clement et al., 2018).
Network brokerage is viewed as a structural mechanism that generates social capital
(Burt, 2005). Abundant research has shown that brokers themselves tend to benefit from their
advantageous positions. Burt (2004) found that network brokerage was more influential
compared to human capital in affecting the creation of good ideas, and it positively correlated
with salary, performance and promotion. Kumar and Zaheer (2019) study of U.S. firms in the
biopharmaceutical industry showed that focal firms’ ego-network stability, operationalized as the
82

ratio of ties lost and added, negatively affected innovation outcomes, because lack of network
change resulted in rigidness and information redundancy. Network brokerage could reduce the
harm of network stability on organizational information. Through a meta-analysis, Stam et al.
(2014) found that network brokerage in entrepreneurs' personal networks positively contributed
to small firm performance. Hite and Hesterly (2001) proposed an evolutionary framework of
firm networks. As firms grow, they require different forms of resources and hence different types
of networks. New firms would benefit more from embedded ties and cohesive networks to tackle
initial uncertainties and challenges, which they labeled as identity networks. As firms grow,
firms branch out with bridging weak ties to gain more resources (Hite & Hesterly, 2001).
While brokers may benefit themselves, how they would influence others or the collective
is less clear and generally receives much less scholarly attention (Kwon et al., 2020). As
summarized by Clement et al. (2018), the advantage of network brokerage mainly comes from
two sources: information access and less normative constraint. Therefore, network brokerage’s
outcome for others is contingent upon different circumstances. When information diversity is the
goal, network brokers can produce positive externalities to the group by enriching the pool of
ideas for recombination and creativity (Clement et al., 2018; Schilling & Fang, 2014). By
contrast, when group cohesion and trust is the goal, network brokers will bring negative
externalities because they lack commitment to either side and are usually free from sanctions on
either side. Similarly, Barnes et al. (2016) found brokerage ties were detrimental in an ethnically
diverse and competitive environment because they worsen the lack of trust among diverse ethnic
groups. These studies all suggest that brokers may benefit others by providing diverse
information, but they bring negative externalities when commitment is required.
83

From a collective perspective, the self-oriented nature of brokers might be beneficial for
themselves but detrimental to the group as a whole when cohesion and trust are important for
group success (Bizzi, 2013). When group cohesion is necessary, brokers are viewed as
“individualistic, competitive, manipulative and power-oriented”, without much consideration for
the collective interest (Bizzi, 2013, p. 1554). To examine the role of group-level brokerage, Bizzi
(2013) examined group-level aggregation of individual members’ structural hole positions, and
found it negatively affected individuals’ perceived autonomy, job satisfaction and performance.
The rationales are explained as follows.
The presence of many brokers in a group negatively affects individual outcomes by
lowering members’ perceived autonomy, trust and compatibility (Bizzi, 2013). First, an
independent broker gains information advantage by exploiting the dependence of the alters, who
have to depend on the broker for information. A group with many brokers faces high dependence
among members, which leads to a constraining force on the perception of individual autonomy
(Brass, 1981). Second, an independent broker is free from normative monitoring. In a group,
brokers may feel the necessity to observe others’ behaviors, yet a lack of trust might turn such
observation into a deleterious monitoring climate in the group (Bizzi, 2013). Finally, if many
members of a group are power-seeking for structural hole positions, there will be little marginal
resources to exploit because there’s little information asymmetry, and brokers are likely to
compete and hinder each other’s opportunities (Buskens & van de Rijt, 2008).
Group-level heterogeneity in brokerage is likely to create group friction and block the
formation of shared group values (Bizzi, 2013). This is because brokers and non-brokers
embrace opposite views on the normative expectations of appropriate social behaviors (Buskens
& van de Rijt, 2008). Within-group heterogeneity in brokerage also means inequality in the
84

distribution of social resources. As suggested by Rowley et al. (2005), the greater a group has
access to structural holes as indication of resources, the less likely a firm will exit that group. But
if there’s inequality in distributing these resources and a firm occupies a disadvantageous
position in that group, this firm is more incentivized to leave because it cannot benefit from the
collective resources.
To summarize, these rationales explain that a group with high average and heterogeneity
in brokerage is unlikely be cohesive, due to a lack of discretion and trust, and a high chance of
conflict and inequality. Trust is referred as the positive expectation of individual and
organizational behaviors and relationships (Shockley-Zalabak et al., 2000). Organizational trust
can “minimize the potential for destructive and litigated conflict”, and it directly correlates with
organizational effectiveness (Shockley-Zalabak, et al., 2000, p. 36).
In virtual communities, trust and conflict are important for group survival because these
voluntary systems naturally have porous boundaries. For example, trust has been considered
essential for open-source software (OSS) communities to guard against opportunistic behaviors
of developers because they are from diverse backgrounds (Antikainen et al., 2007). Filippova
and Cho’s (2016) study of an OSS community shows that conflict generally had a negative
impact on developer retention. Specifically, normative conflict, meaning disagreement or
discrepancy on expected practices and actual behaviors, negatively affected developers’ intention
to remain in the community (Filippova & Cho, 2016). Such normative conflict is quite similar to
the concept of group heterogeneity in brokerage, as they both highlight the dissonance between
subcommunities of a group. Such friction is likely to reduce group cohesion and identification,
and ultimately cause members turnover and group disbandment (Filippova & Cho, 2016).
Following Bizzi (2013), this study uses two measures to indicate the group composition of
85

network brokerage: the mean and standard deviation of individual network brokerage positions
for all members in a clan. In game contexts, network brokers have superior performance because
they could improve and learn skill by collaborating with different people (Shen et al., 2014).
However, in-game organizations are social structures that provide opportunities for developing
social relations for better teamwork, and group cohesion has been found to be critical for group
performance in games (Benefield et al., 2016). Thus, if many players in a clan are individualistic
brokers seeking personal benefits, indicated as a large group mean in network brokerage, this
clan is unlikely to build the level of cohesiveness required for better performance. If there’s high
inequality in the distribution of relational resource in a clan, indicated as a large standard
deviation of network brokerage, the clan is unlikely to have a trustful climate because relational
resources are in the hands of a few.
H4: Clans with a larger mean of network brokerage have higher mortality hazards.
H4b: Clans with a larger standard deviation of network brokerage have higher mortality
hazards.
The logic for the interaction effect is very similar to the previous one. Group cohesion is
necessary to align members’ mindsets to agree upon a clear governance structure required of a
focused niche. As discussed above, groups occupying broad niches are more likely to disband
(Olzak & Johnson, 2019). Groups with many brokers are likely to face problems of distrust and
detrimental competition (Bizzi, 2013). Group cohesion is low because of a toxic climate of
suspicion and monitoring of members’ behaviors and motivations (Bizzi, 2013). It would be
difficult for such a group to build shared vision, values and mindsets, because members don’t
trust each other’s motivations and try to undo each other’s work (Somech et al., 2009). Thus,
groups with many brokers are unlikely to engage well with a narrow niche position.
86

Take the player rotation example discussed in participant observation. Typical social
clans are more democratic as they usually respect every member’s right to participate in team-
based battles regardless of one’s equipment and skill levels. If a clan suffers from a monitoring
climate, such a governance style might be perceived by some members as opportunistic with
unrevealed motivations, even it’s consistent with a social clan’ intended values. An even worse
case would be that different members of a clan try to execute contradictory governance styles,
e.g., prioritizing performance or equality. Such distrust, friction and conflict prohibit a clan with
a focused niche to engage well with that position. Thus, this study proposes an interaction effect
between clan niche width and within-clan average brokerage.
(H5) When clans occupy narrow niche positions, those with fewer brokers have lower mortality
hazards.
Study 2: The Formation of Individual-Group Multilevel Networks
The first study sets a broad picture of clan mortality driven by clan types, clan boundaries
and within-clan social networks The second study zooms in to examine the structural
dependency of multilevel organizational networks. Although study 1 takes into account that
organizations are internally complex collectives, and there is dependency between internal and
external mechanisms, the analytical procedure is still based on the assumption of independent
observations. Study 2 therefore adopts network modeling to examine the interdependency micro-
mechanisms that govern the formation of multilevel nested organizational networks.
Conceptually, this study follows the third multilevel network perspective of nested
interdependency as discussed in Chapter 2 (Lazega, 2016). Methodologically, this study follows
the multilevel ERGM paradigm (Wang et al., 2016). Multilevel ERGM is based on the
exponential random graph (p*) models (ERGM), which can capture both endogenous and
87

exogenous social mechanisms (Robins et al., 2007; Snijders et al., 2006). Monge and Contractor
(2003) proposed the Multi-theoretical Multilevel (MTML) framework for empirical analysis of
networks emergence at actor, dyadic, triadic, and global levels, driven by hybrid theories. The
interdependency assumption of ERGM naturally speaks to the “trajectories” concept, as social
actions happen and leave traces waiting to be captured by structural configurations (Padgett,
2018, p. 407).
Extending to multilevel ERGM, multilevel networks include macro-level and micro-level
networks with a clustering structure (P. Wang et al., 2013). For instance, Brailly et al. (2016)
studied an inter-organizational (109 companies) deal network and an interpersonal (128
individuals) information exchange network at a trade fair for television programs in Eastern
Europe. They confirmed a multilevel embeddedness mechanism by fitting two ERGMs using the
other level as the predictor. They also found networks at different levels do not evolve with the
same path, such that the interpersonal network changes faster than the inter-organizational
network (Brailly et al., 2016). A more integrated multilevel ERGM model includes the macro-
level network, micro-level network and meso-level affiliation network. This MERGM model
could simultaneously test within-level and cross-level dependencies. Most empirical studies of
multilevel networks focused on nested interdependencies between individuals and organizations,
although Wang et al. (2016) argued the framework is also generalizable to any two-mode
network structures. This model could reveal ‘‘how an observed network structure at one level of
the system of organizational networks relates to network structures and effects at higher or lower
levels of the system’’ (Moliterno & Mahony, 2011, p. 443).
Study 2 proposes to represent such complex relations among players, clans and clan
niches in multilevel networks. Following the nested multilevel network structure, the three
88

networks in the current case are organized as follows: a niche overlap network of clans at the
macro-level (network A), an individual-clan affiliation network at the meso level (network X),
and a co-play network among individuals at the micro-level (network B). As shown in the pilot
study on the density dependence of clan founding, there’s diffuse competition among clans such
that similar clans depend on the same pool of members. Thus, the similarity tie among clans can
be understood as niche overlap. The individual-clan affiliation is a bipartite network that
indicates players’ membership to clans. The co-play network is a one-mode network that
indicates intentional collaboration among players. Players can change clan affiliation, and
players can also choose to collaborate with others from the same clans or different clans.
The affiliation network and interpersonal social network correspond to the classical
dichotomy concept of “formal” and “informal” organizational structures (McEvily et al., 2014).
Formal organization refers to organizational boundaries, division of labor, set of rules,
concentration of decision rights, that are designed for organizational activities; it also includes
coordinating and integrating systems such as boundary spanning roles, reporting relations,
mentoring and evaluating procedures (McEvily et al., 2014). Informal structure refers to the
social networks that individuals voluntarily build for both instrumental and affective purposes
(Brennecke & Rank, 2016a). The distinction between formal and informal structure is nothing
new. However, only until recently organizational scholars begin to emphasize the interplay
between the two as a holistic picture of organizing (Hunter et al., 2020; McEvily et al., 2014;
Soda & Zaheer, 2012).
Multilevel organizational networks clearly echo this interactionist perspective because
they can examine boundary-spanning behaviors and within-/cross-boundary relations. In the
current context, players can switch clan affiliation as boundary spanners, and they can also build
89

within-clan and cross-clan co-play relations. Following the multilevel perspective, such meso-
level and micro-level ties can be examined in connection with macro-level networks. Due to data
collection constraints in offline organizations, macro-level networks have been mainly studied in
intraorganizational settings, e.g., inter-unit and inter-team relations (Brennecke & Rank, 2016;
Zappa & Lomi, 2016). Within one organization, the formal coordinating relations among units or
project teams are largely collaborative. Inter-organizational relations, however, are much more
complex with both collaborative and competitive aspects (Gnyawali & Park, 2011). The inter-
clan niche overlap network is an appropriate example of inter-organizational relations.
Therefore, the current multilevel networks of interpersonal co-play networks, player-clan
affiliation networks and inter-clan niche overlap networks seek to fulfill two objectives. First,
this study extends the concepts of formal/informal organizational structure to virtual organizing.
Second, by leveraging the niche overlap relations among virtual groups, this study seeks to
understand how informal structures coevolve with competitive formal structures.
Since these virtual organizations vary considerably in terms of life expectancy, size,
collective identity, and group social capital (Chen et al., 2008; Williams et al., 2006; Benefield et
al., 2016), study 2 focuses on the most legitimate organizations with a self-sufficient size and
their most legitimate members. In the current virtual context, group size is critical because it
requires at least fifteen members to field a complete team in clan-based skirmishes. Without a
sufficient size, a clan needs to rely on outside players to form a team. A decent size also makes
other in-group social activities like training and meeting friends possible. Group size is likely to
be self-reinforcing, such that larger groups are better at retaining members (Sun et al., 2021).
Virtual groups with a sufficient size are better parallels and mappings of real-life organizations
(Cohen, 1985; Williams, 2010). Moreover, since digital relations can be more transient than in
90

real life, individual-group affiliation highly varies in tenure length. By focusing on enduring
affiliations, the multilevel mechanisms examined in this study can be useful for informing human
organizing in offline contexts.
The Theory of Niche Overlap
The theory of niche overlap draws heavily from density dependence theory, which is
discussed in study 1. Similarly, the theory of niche overlap also involves the two ecological
processes of competition and legitimation. The difference lies in the application of the two
mechanisms on direct comparison between two entities, rather than constitutive legitimacy and
diffuse competition. Baum and Singh (1994) view a population as comprising multiple niches in
a multidimensional resource space and define such within-population niche as “variation in
productive capacities and resource requirements among organizations within a population” (p.
485). They argued that the tendency for two organizations from the same population to compete
is a function of niche overlap density, that is, niche overlap density negatively affects the
founding of new organizations. On the other hand, niche non-overlap density leads to
complementarity in functions and creates mutualistic interdependence.
Dobrev and Kim (2006) built on the original density dependence theory and made two
modifications. First, they argued that legitimation and competition could also be applied to direct
comparison between individual organizations, rather than two populations. Second, they
challenged the traditional view of legitimation with a ceiling effect and argued that mutualism as
a cognitive and institutional process would exert broader impact than competition. That is,
legitimation as a form of collective action would continue to take effect even beyond the
threshold. Their analysis of the automobile industry showed that the niche overlap density led to
firms deserting their segments because of competition, but at the same time high density from
91

niche overlap made firms less likely to desert their segments because it signaled the pursuit of a
collective identity. Dobrev (2007) argued that both resource dependence and mimicry are in
operation when firms exit their market segments. The exit of similar firms occupying the same
resource space triggers imitative behaviors from the remaining firms, because mimicry is a
strategy to deal with environmental uncertainty. But as the number of exiting firms increases, the
collective identity collapsed due to lower density and more resources are released from the niche,
which reduces the remaining firms’ incentive to exit.
Niche overlap also plays a role in the emergence of a new population. Dobrev et al.
(2006) found that a new population sharing overlap with an established population in resource
and identity niches would face both resource competition and identity. In their empirical case,
the established population is commercial banks, and the emergent population is financial
cooperatives. That is, the emergent population benefits from the proliferation of the established
population at a lower level to allow for legitimacy transfer, but the proliferation of established
population at a higher level intensifies the resource competition between the two overlapping
populations. At the same time, because the established population and the emergent population
also have overlap in the identity space, the proliferation of the emergent population triggers the
established population to hinder its development by questioning its legitimacy. But when the
emergent population grows and becomes more legitimate, such identity comparison diminishes.
As a summary of previous findings, Bruggeman et al. (2012) argued niche overlap affects
organizational fitness non-monotonically, such that a moderate amount of niche overlap
positively associates with fitness by bringing legitimacy spillover.
In the current context, without a clear-cut segmentation of clans, niche overlap is
represented as the semantic similarity of clan profiles. A tie indicates that two clans are more
92

likely to describe themselves in similar ways, and they are more likely to occupy similar resource
positions. A similar study is Hollway and Koskinen’s (2016) examination of the multilevel
interdependencies of bilateral fishery agreements among states as nested in joint membership to
multilateral fishery agreements (MFA). Their multilevel network structure is the semantic
similarity network of 220 multilateral fishery agreements at the macro level, bilateral fishery
agreements of 195 states at the micro level, and the states’ affiliation to MFA at the meso level.
The relation between semantic meanings and organizational networks has recently been
studied in organizational vocabulary (Lomi et al., 2017; Tasselli et al., 2020). Organizational
vocabulary is defined as “the structure of conventional word use captured by the combination of
word frequencies, word-to-word-relationships, and word-to-example relationships—that together
demarcate a system of cultural categories” (Loewenstein et al., 2012, p. 3). For instance, Lomi et
al. (2017) built a bipartite network between 42 managers and 30 organization vocabularies. They
found that members of the same subsidiary tended to be convergent in the vocabulary to describe
their unit but divergent in the vocabulary to describe the company. Incorporating temporal
change, Tasselli et al. (2020) showed that members of the same unit who had social ties
gradually developed similar vocabulary and members of different units who used similar
vocabulary gradually developed social ties.
Affiliation-based Closure - Boundary Constraint
The first multilevel structural mechanism is the dependency between category affiliations
at the meso level, and micro-level network formation. In nested individual-organizational
multilevel networks, the closure between meso-level affiliation networks and micro-level
interpersonal networks indicates if interpersonal ties are constrained within organizational
boundaries. How formal structure constrains or enables informal structure has been examined
93

through within-boundary and cross-boundary social networks. A recurring theme is the higher
probability of within-boundary ties (Lomi et al., 2014). Scholars have long acknowledged that
the boundary of organizations to facilitate tie formation within but limit tie formation across is an
important element of organizational effectiveness and competitive advantage, because it protects
the reservoir of knowledge by making it difficult to imitate for outsiders (Argote & Ingram,
2000). This is also supported in recent MERGM studies. For example, Brennecke and Rank
(2016b) studied the multilevel structures of 434 R&D workers’ membership to 218 formal
project teams and their informal communication network. They found knowledge workers were
more likely to seek advice from colleagues from the same team, but colleagues from the same
team were less likely to reciprocate advice seeking. That is, cross-team advice seeking is largely
reciprocal, but within-team advice seeking is not. To explain the second finding, they gave three
arguments. First, joint membership to knowledge intensive teams means more trust, which
reduces the need to rely on informal advice exchange. Second, being in the same team means
collaboration towards the same goal, which might motivate members to give advice freely.
Thirdly, the collective identity might be more salient than individual identities in a knowledge
collaboration team, which creates a climate against exchange (Brennecke & Rank, 2016b).
Similarly, Zappa and Robins (2016) studied the knowledge transfer network of 196 individuals
affiliated to 10 units with interunit flow ties, and found a positive affiliation-based closure
mechanism and a negative affiliation based reciprocity mechanism. Zappa and Lomi (2016)
argued multilevel networks could address the problem of autonomy of social networks in relation
to formal organizational structure, because boundary spanning behaviors are important for
organizational outcomes. They studied the informal relations among 47 managers affiliated to 6
formal organizational units connected by formal reporting ties, and found the affiliation-based
94

closure arc was positive, supporting that formal organizational boundaries are reinforced by
interpersonal relations.
However, the constrain of formal structure is not conducive to coordination and
communication across formal boundaries. Abundant research has shown that cross-boundary ties
positively affect organizational innovation (Hargadon, 2003; Perry-Smith & Mannucci, 2017),
and individual performance (Cross & Cummings, 2004). Recent research pointed out the
formation of within-boundary and cross-boundary ties is contingent upon local or global
identification (Lomi et al., 2014). Specifically, employees’ identification with subunits within a
company constrains crossing-boundary tie formation and limit knowledge transfer across
boundaries. Employees who are more strongly identified with the company are more likely to
form cross-boundary ties (Lomi et al., 2014). Furthermore, although knowledge transfer across
boundaries is generally difficult, network properties of cross-boundary ties, e.g., tie strength,
network cohesion, and network range, can positively affect the amount of knowledge acquired
through cross-boundary ties (Tortoriello et al., 2012).
Although this mechanism of “formal” affiliation boundaries constraining “informal”
interpersonal relations is widely supported in real-life organizations, it has yet to be tested for in
virtual organizations. If confirmed, this mechanism offers evidence for the parallel between
virtual organizing and offline organizations. Virtual organizational boundaries are fluid, because
virtual groups are usually multi-purposed, self-organized, with low entry/exit costs, and with
abundant boundary-spanning behaviors (Benefield et al., 2016; Gilson et al., 2014). Similar to
offline organizations, within-boundary ties in virtual contexts strengthen coordination and
enhance group effectiveness (Benefield et al., 2016), However, studies have also shown that trust
was not related to individual performance in virtual groups (Goh & Wasko, 2012). On the other
95

hand, cross-boundary ties are conduits from outside information. Individuals exposed to outside
knowledge can improve their own skills, which in turn benefit the group (Benefield et al., 2016).
Thus, it is theoretically important to know whether formal boundaries in virtual contexts can
structure social networks.
In the current context of clans, regardless of players’ social or competitive motivations,
developing social relations is a key component of joining a clan. Clan affiliation enables players
to find co-play collaborators, because clans not only have sub-channels for organizing battles but
also sub-channels simply for socialization. Scholars found that players affiliated to the same in-
game groups are less likely to dissolve their social ties, because being a stable member already
indicates compatibility (Shen et al., 2014). Moreover, interpersonal toxic behaviors are less
likely for players who belong to the same groups than for strangers, because the chance of
defecting from behavioral norms is higher for those who expect no future interactions (Shen et
al., 2020). Thus, it is proposed:
H6: Co-play ties are more likely to be observed between players affiliated to the same clans
(ATXBX).

Affiliation-based Closure-Boundary Spanners
Another type of affiliation-based closure in nested multilevel networks is the closure
between meso-level affiliation networks and macro-level interorganizational networks. If
positively significant, this mechanism indicates that individuals’ boundary spanning behavior, or
common membership to multiple organizations, is influenced by the higher-level
96

interorganizational relations. Specifically, for individuals with multiple affiliations, they are
more likely to join organizations that are similar to those they have joined before.
Boundary spanners play an important role in information transfer across organizational
boundaries (Tushman & Scanlan, 1981) and providing nonredundant information (Benefield et
al., 2016). Boundary spanners usually have access to knowledge across technological boundaries
that may stimulate innovation and also have possible experience in resolving interorganizational
conflicts. Thus, they are more likely to progress to leadership (Fleming & Waguespack, 2007),
and may positively affect team performance (Marrone et al., 2007).
Contextual factors are important antecedents for boundary spanning (Joshi et al., 2009).
Intraorganizational inter-unit boundary spanning is constrained by the degree of centralization of
the organization (Tsai, 2001). In other words, hierarchical differences reduce boundary spanning.
Likewise, inter-group conflict reduces the chance of boundary spanning to coordinate tasks
(Joshi et al., 2009). Inter-group compatibility and similarity, however, facilitate boundary
spanning. When organizations share some similarity, it is comparatively less difficult to transfer
knowledge across organizational boundaries (Argote & Ingram, 2000). This is because for novel
information to be effectively integrated into existing knowledge repositories, there needs to be a
moderate amount of knowledge overlap (Lane et al., 2006). In multilevel ERGM studies,
following the previously discussed example of bilateral and multilateral fishery agreements by
Hollway and Koskinen’s (2016), their affiliation-based homophily suggests that states join
MFAs that were both similar and socially confirmed. This is because these similar MFAs are
more salient to states as they are in line with their existing preferences and would require little
costs or additional commitment.
97

Similar findings are also supported in studies of category spanning. The “category
imperative” view follows the logic that spanning category boundaries violates audience
expectation and therefore negatively evaluated (Zuckerman, 1999). But such penalties cannot be
examined in isolation of the contextualization of categories, e.g., how categories are associated
horizontally as similarities (Wry & Lounsbury, 2013). This is consistent with the social
perspective of categorization such that category boundaries are not predetermined or static but
dynamically negotiated by social actors (Durand et al., 2017). Although certain category
spanning pattern might be criticized initially, with the increasing diffusion and borrowing by
many peers, the boundaries of these categories become less potent (Rao et al., 2005). In other
words, the sanction on category spanning is likely to be attenuated when social actors agree upon
the similarities among some categories (Kovács and Hannan, 2010). For example, in their study
on entrepreneurial firms of nanotechnology and patent categories, Wry and Lounsbury (2013)
showed that the detrimental effect of category spanning was attenuated if the categories were
similar, which was operationalized as co-citation of patents. In a similar study, Wry and Castor
(2017) found similarity among patents motivates scientists to span patent categories because they
would encounter fewer social sanctions.
In the current context, from the perspective of players, as their gaming motivations don’t
change drastically (Schultheiss, 2012), they are mostly bounded within the two broad categories:
competitive and social. Since similar clans are more likely to have similar mindsets, players
would find similar clans more compatible. For example, competitive and social clans have very
different ideas about team member rotations. For social clans, since the objective is to have fun,
when possible, clan leaders would make sure everyone has a chance to participate. This is
especially the case for clan-exclusive game modes like Global Maps, because they have
98

predetermined schedules and usually there’s a turnout large enough to field several teams. By
comparison, competitive clans would make a fixed team play repeatedly to improve
performance. Moreover, the focus of this study is stable organizations and consistent affiliations
to best parallel real-life organizations. With the exclusion of trials and errors, players would
encounter the least efforts in shifting between similar clans. Clans are also more likely to benefit
from new resources brought by these boundary spanners because similar clans are likely to be at
comparable competitive levels. Such mutually beneficial affiliation is more likely to sustain.
H7: When players span boundaries to other clans, they are more likely to join clans that are
similar to their original ones (ATXAX).

Cross-level Alignment
Cross-level alignment represents the association between higher-level and lower-level
ties. It means that “nodes with connected affiliations are also connected” (Wang et al., 2016). In
the current context of hierarchical organizational networks, this mechanism means interpersonal
boundary spanning relations are sustained by formal hierarchical structure (Zappa & Lomi,
2015). In other words, this mechanism indicates that lower-level individuals are likely to form
boundary-spanning relations if their affiliated organizations have ties. This mechanism can be
considered as a form of network governance (Lane et al., 2006). The rationale is that hierarchical
authority in formal organizations usually constrains the formation of lower-level interpersonal
networks. This is relevant to the concept of “synchronization costs” (Lazega, 2016). The fact that
lower-level relations are subordinate to higher-level relations means that they might need to bear
more costs to carry out collective actions to challenge higher-level boundaries (Lazega, 2016).
99

This mechanism has been tested in a few multilevel ERGM studies. Brennecke & Rank
(2016a) fit multilevel networks of 51 researchers from 26 organizations in the metrology
industry and 71 researchers from 54 organizations in the photonics industry. Their results
showed the evidence of cross-level alignment with the same direction, indicating individuals are
more likely to connect if their respective organizations collaborate. Zappa and Lomi’s (2016)
study on 47 managers and 6 organizational units revealed a positive cross-level alignment
reciprocal mechanism, which means that individuals are more likely to reciprocate interpersonal
relations if their affiliated formal units have ties. Zappa and Robins (2016) found a combination
of positively significant alignment and non-significant exchange, meaning the directions of ties
are unlikely to be reversed across levels. This result signifies the alignment of tie direction-based
hierarchy across levels. In the bilateral and multilateral fishery agreements study, Hollway and
Koskinen’s (2016) did not find a significant cross-level alignment mechanism, showing that
bilaterally connected states did not prefer similar multilateral fishery agreements.
In the current context, this mechanism means that individuals are likely to form cross-
boundary relations with other individuals affiliated to similar groups. An example is inter-clan
alliances. Clans form alliances on the basis of similar mindsets about playing. Similar clans do
not necessarily form alliances, but alliance clans are likely to describe themselves in similar
semantic patterns. Alliances clans have agreement on helping each other and not taking each
other’s members. Alliance clans can be quite useful when a clan itself has a low turnout and
cannot field a complete team. Alliance clans also considerably expands the pool of potential
collaborators and friends. Thus, such friendly relations between clans facilitate the formation of
cross-clan collaboration in battles. Members of alliance clans hold the role of “diplomat” and
they frequently play together as “game nights”. “Diplomats” sometimes even play as team
100

leaders to design strategies and give orders to members from the other clan. Such alliances are
especially useful for the game mode of Global Maps. As mentioned earlier, in battles for global
maps, clans compete on a map to take control of territories. Once they seize a section of territory,
they can only expand to adjacent territories. If two clans have alliances, they would have
agreement on not attacking each other to keep certain sections of the map safe. This is an
example of how inter-clan relations enable members from these clans to collaborate. Thus, it is
proposed:
H8: Co-play ties are more likely to be observed between players affiliated to different clans
when these clans are similar
.
Cross-level Assortativity
The last hypothesis tests the transfer of degree-based hierarchy across levels (Hollway &
Koskinen, 2016). Using centrality scores, Lazega (2016) proposed a metaphor of individual and
collective hierarchy: big fish in a big pond, big fish in a small pond, little fish in a big pond and
little fish in a small pond. One could use the name of the collective to obtain resources for
personal interest but not share personal prestige, as an individualist strategy; one could also share
personal prestige for collective interest, as a collectivist strategy. They found the bigger the fish
the more alignment of interest across levels.
Brennecke and Rank’s (2016b) analysis of workers and teams tested if popular advisors
work on many projects, and active advice seekers are less likely to work on many projects.
Contradictory to propositions, both effects were negative, indicating a negative relationship
between working on multiple teams and advice seeking, which might be due to workers’ limited
101

cognitive load. Brennecke & Rank (2016a) found positive popularity assortativity in individual-
organization networks, meaning individuals as popular providers of knowledge are affiliated to
popular organizations with many partners. They also found negative activity assortativity,
indicating the less organizations were nominated as partners, the more their managers sought
knowledge through interpersonal networks. Zappa and Lomi (2016) found positive in-degree
assortativity, which indicates popular managers are members of popular subunits, that is,
hierarchy aligns across two levels. Hollway & Koskinen (2016) didn’t find significant multilevel
alignment between bilateral fishery agreement and multilateral fishery agreement, controlling for
3-path. Wang et al. (2016) studied the multilevel networks of 87 entrepreneurial Ethiopian
farmers and 178 non-entrepreneurial farmers, and found there was a positive tendency for cross-
level assortativity, showing that within-level popular entrepreneurial farmers and popular non-
entrepreneurial farmers are likely to be friends.
In the current case, the macro-level network is based on the niche overlap relations
among clans, thus degree centrality cannot easily equate with popularity. Drawn from the niche
overlap literature discussed above, niche overlap indicates both competition and legitimacy
(Gruggerman et al., 2012). In this case, a popular clan has high legitimacy and faces high
competition. The good side is that such a clan uses conventional language to describe itself and
may be viewed as reliable. The downside is that such a clan has too many substitutes. Given that
niche overlap relations among clans are likely to facilitate individuals to span group boundaries
(H7), a clan with high centrality in the niche overlap network is likely to have fluid boundaries
because there are many alternatives. Clans with fluid boundaries have abundant bridging social
capital but not bonding social capital, because many members have allocated group engagement
elsewhere. Popular players who have abundant co-play ties already have access to diverse
102

information. Given that both bonding social capital and bridging social capital are essential,
popular players are less likely to find fluid clans attractive. Thus, it is proposed:
H9: Popular players with more interpersonal ties are less likely to join clans with more niche
overlap with other clans (ASAXASB).

Study 3: Diversification or Focus? An Agency Perspective
Although study 2 examined closely the relational interdependence of multilevel
networks, its shortcoming is evident in its restriction of data size, due to the fragility of
MERGM. Study 3 seeks to address the problem by applying a cutting-edge network modeling
technique, that is, treating the meso-level affiliation ties as relational events. The application of
relational events models (REM) has been increasingly common in a variety of contexts,
including real-life interorganizational networks (Amati et al., 2019) and virtual social
interactions such as Wikipedia (Lerner & Lomi, 2020a). The availability of automatically
documented large-scale digital traces with fine-grained time stamps has enabled a wider
application of such network modeling techniques. Building upon the original modeling
developed by Butts (2008), Lerner and Lomi (2020b) extended this technique to the examination
of hundreds of millions of dyads.
The theoretical question explored in the third study is the consistency of two resource
access mechanisms: social networks and social categories. As summarized by (Burt, 2012), “the
gist of the network story is that information becomes homogeneous, tacit, and therefore sticky
within clusters of densely connected people such that clusters disconnect, buffered from one
another by the structural holes between them, which gives information breadth, timing, and
103

arbitrage advantages to people whose networks span the structural holes” (p. 544).
Corresponding to this cluster/breadth distinction, the principle of allocation in organizational
categories posits that “the greater the diversity in regions of resource space targeted by an
organization, the lower the organization’s capacity to perform well within them” (Hsu, 2006, p.
420). Since both resource access mechanisms face the tradeoff between diversification and focus,
this study seeks to understand if individuals consistently building diversified or focused social
networks are also more likely to join organizations.
To connect the two resource access mechanisms, attention needs to be brought to the
agency of individuals at the center who make choices of social interactions. This study
approaches such agency by following Burt’s (2012) construct of multirole network personality,
which captures a consistent personal preference for building certain type of social networks
across multiple roles. Correspondingly, the role of agency has also been increasingly important
in studies of social categories, as abundant studies have recognized social categories no longer as
disciplines but as tools to use (Kennedy & Fiss, 2013; Wry & Castor, 2017). This study follows
Burt (2012) and operationalizes personality with social networks in digital traces. In order to
ensure the accuracy of different networks across different roles, Burt (2012) examined social
networks in massively multiplayer online role-playing games (MMORPG), where players
simultaneously play multiple avatars. There’s an increasing research interest to examine
personality from digital traces in social media. For example, Facebook likes can accurately
predict individual attributes such as ethnicity and personality (Kosinski et al., 2013). Such
prediction was found to be more accurate than friends’ evaluation of personality (Youyou et al.,
2015). Personality can also be revealed from social media language use (Schwartz et al., 2013).

104

Personality and Networks
The role of agency in social networks has attracted research attention recently (Tasselli &
Kilduff, 2021). The agency question concerns how much the ego’s psychological characteristics
matters in building social networks. However, this agency question has been largely treated as
equivalent to exposure opportunities by assuming all people will fully act upon their available
opportunities (Burt et al., 2013). Recently, there are some efforts to integrate personality and
networks. For example, taking a longitudinal perspective, Selfhout et al. (2010) traced five
waves of adolescence friendship formation and explored how the Big Five personality types
coevolved with friendship networks. They found that extraversion significantly contributed to
friendship formation, and agreeableness was the most sought-after feature. They also found
homophily effects in three traits: extraction, agreeableness and openness. Through a meta-
analysis, Fang et al. (2015) found openness and neuroticism negatively associated with centrality
and openness positively associated with brokerage.
Among all personality dimensions, self-monitoring is an especially relevant antecedent
for network building (Fang et al., 2015). Self-monitoring is defined as “active construction of
public selves to achieve social ends” (Gangestad & Snyder, 2000, p. 546). It is considered as the
“psychological analogue to bridging structural holes” (Burt, 2012, p. 548), as being a broker is
associated with entrepreneurial and authoritative personality (Landis, 2016). Mehra et al. (2001)
explicitly studied the relationship among self-monitoring personality, network structures and
performance. Within the context of a high-technology firm, they found that high self-monitors
were more likely to occupy central positions in social networks. They also found that self-
monitoring and network centrality independently predicted individuals' workplace performance.
Similarly, Oh and Kilduff (2008) found within the context of a Korean company that high self-
105

monitors were more likely to be brokers. Fang et al. (2015) showed in a meta-analysis that self-
monitoring had stronger predictive power of network centrality and brokerage than Big Five
personality measures. They also found that both personality and network positions associated
with job performance, and network positions partially mediated the effects of self-monitoring,
openness and conscientiousness on performance.
Fang et al. (2015) concluded by saying their findings challenged Burt (2012)’s claim that
personality didn’t predict performance. However, this conflict could be due to the different
assumptions of disposition from psychological and sociological perspectives. Much of
personality psychology is based on the assumption that there’s a stable trait that governs
individual behaviors. However, Landis (2016) pointed out the necessity to deviate from this
tradition and emphasize within-individual variability because “people have distinctive
interpersonal styles that arise in interactions with different people and in different situations” (p.
118). A recent review on personality change also illustrates the value of this perspective in
studying organizational behavior (Tasselli & Kilduff, 2018). In their summary, personality
change is possible as a result of self-development, organizational events and processes including
relationships with co-workers, and external events and processes such as education (Tasselli et
al., 2018). This dynamic view of personality highly aligns with the sociological assumption of
personality, “where the prevailing assumption is that the dispositions of individuals reflect the
structural positions that they occupy” (Burt et al., 2013, p. 537). Personality as a network
construct clearly follows this sociology track, where personality is approached as an outcome,
rather than an antecedent, of social actions. The recent micro-foundation view of agency extends
this view by arguing that “a reciprocal influence between the actors and the situations they
structurally occupy in the network” (Tasselli & Kilduff, 2021, p. 77).
106

Multirole Network Personality
As defined by Burt (2012), network-relevant personality is “the network advantage that
can be attributed to personality manifest in consistent network behavior across roles” (p. 552).
This construct is not strictly personality in a psychological sense, but reveals an average pattern
of a person’s way of engaging with other people. Such a pattern of social engagement manifests
repetitively in every network this person builds. To capture such social preference, it is necessary
to examine networks across multiple roles. A multirole network is defined as “a social system
composed of two or more of a person’s role-specific networks” (Burt, 2012, p. 551). Since
different roles are associated with different expectations or pressures in building certain
networks, such role-specific variations need to be teased out. “Closure-prone” people prefer to
focus role performance on densely connected clusters with high levels of trust, cohesion and
norm constraint; “brokerage-prone” people prefer to focus their role performance on connecting
otherwise separate groups and developing new contacts with high coordination cost and little
normative constraint.
Such multirole network data is difficult to obtain and highly subject to human memory
errors. However, virtual worlds with role playing are appropriate parallel contexts to avoid data
cost and quality problems, because behaviors under each role are recorded completely and
accurately. Burt’s (2012) research context is EQ II, where people play in the roles of 16 avatars.
Using nonredundant contacts and network constraint, Burt found about a third of variance in
role-specific networks could be explained by the average network personality, that is, people do
create similar networks from one situation to another. However, the results didn’t support a
significant correlation between network personality and role performance. In other words, the
result contradicted the hypothesis that “brokerage-prone” individuals are more likely to have
107

higher achievement because they are emotionally and behaviorally equipped to identify and
develop network advantage to their own benefits. Moreover, role performance was also not
predicted by the consistency between network personality and role-specific networks. That is,
even taking into account conditions where “closure-prone” people might not feel comfortable if
they are forced to take up roles that require bridging structural holes, the association between
network personality and performance was still negligible.
Agentic Categorization
Broadly situated in the social perspective of categorization as reviewed in the last
chapter, in contrast with the prototype model, the role of agency is increasingly valued by
scholars of social categories. Actors are not simply constrained by categories to convey
information but also strategically use framing to describe categories to their advantage. As
argued by Pontikes and Kim (2017), “[t]he fact that market categories are not simply a reflection
of objective distinctions, but result from social negotiation, allows for such a framing” (p. 75).
This focus on agency in categorization is echoed in a number of recent works. For example,
Kennedy and Fiss (2013) pointed out the importance of examining the motivations associated
with using categories. As stated by Durand et al. (2017), actors have interests and goals, so that
they could activate cognitive processes to elaborate categories and coordinate with peers to co-
construct categories. Similarly, Durand and Khaire (2017) also called for attention on
categorization as an intentional process of all involved actors with possibly competing interests.
Antecedents of categorization largely remain a research gap. The limited number of
studies mainly focus on resource change and competition intensity in related categories as the
motivation for strategic moves among categories. Wry and Castor’s (2017) study on nano-
technology scientists showed high-status scientists were more likely to span boundaries, and they
108

could serve as a catalyst for peers of lower status to mimic. High-status scientists are more
experienced professionals with sophisticated knowledge on categories, so that they could span
boundaries more skillfully. The high status of experts could also be a buffer against negative
sanctions, because they have already established reputation. Negro et al. (2015) showed category
membership can serve as market signals, especially for categories with high contrast.
These arguments are useful for explaining why “brokerage prone” people might be more
likely to span organizational boundaries. First, compared to “closure prone” people who prefer to
be “one of the guys”, “brokerage prone” people enjoy being “center of attention” (Burt, 2012, p.
552). They are more likely to be visible entrepreneurs and stars of a community (Landis, 2016;
Mehra et al., 2001b). Such status might be a source of halo effect so that organizational
boundaries are less constrictive to them. Second, with superior information access, “brokerage-
prone” people are more knowledgeable of inter-category structures, which could help them
choose right categories to span. As mentioned earlier, there is backstage information about clan
leaders’ secret alliances in clan wars. Officially declared alliances are not part of the social
architecture of the game. But sometimes leaders would publicly announce such collaboration to
their members, but some would keep this as secrets for leaders. Without such knowledge,
switching clans might result in unpleasant experiences. Third, “brokerage-prone” people are
more skillful at building rapport with different people, so that they would encounter fewer
obstacles when being assimilated into groups with different norms, identities and cultures. As
shown in an early piece, there is association between self-monitoring and boundary spanners,
because they could adjust self-representation to different norms of multiple groups (Caldwell &
O’Reilly, 1982).
109

H10: “Brokerage prone” players as manifested in the multirole personality in co-play networks
are more likely to join clans.

110

Chapter 5: Data and Method
Data for Study 1
The game publisher Wargaming helped retrieve all of the player-level records from the
WoT PC data warehouse for the North America server spanning approximately three years. The
complete dataset included clan-level variables, player-level variables, co-play networks, and
player-clan affiliation history for thirty-two months from September 2016 to April 2019. The
data were aggregated at the monthly level to achieve a balance between a general picture and
specific details. All data were anonymized. Wargaming created a unique key for each player, so
that different behaviors of the same player could be properly merged.
I first identified the persistently active players who were consistently active in the game
during the observation period. They are the most committed participants in the online
community, and their affiliated clans could be viewed as the most mainstream clans. Since it is
common for users of online communities to have multiple accounts, and duplicate accounts
could harm the accuracy of results, duplicated accounts were removed and 18,585 persistent
players were found. Thus, the sample of clans were those with at least one persistently active
player for at least one month. Although a player cannot be a member of two clans at the same
time, it is possible to have more than one clan in the monthly aggregated data because they might
switch clan mid-month. To simplify the problem, for all the players, I obtained the clan in which
a player spent most of the time for each month. After excluding duplicate accounts and selecting
the single clan affiliation in each month for each player, 207,354 players affiliated to 5,791 clans
made the data for study one. Clan-level data include clans’ public profiles, number of members,
win rate and ratings. Player-level data include interpersonal co-play networks.
Measures for study 1
111

Dependent variables. Cox’s proportional hazards models were used with clan mortality
as the event occurrence. The time duration of clans’ survival time was measured as the number
of active months during the observation period. The right censoring variable was created to
indicate clan mortality, which means ceasing to have observations before the end of the
observation period.
Clan niche width. Clans’ GoM in the two broad categories were measured as
probabilities to the two categories. Clan niche width was calculated as the unevenness of the
distribution of GoM in the two broad categories. Given the GoM vector μ (#,%,&) , where # ∈ )
(categories), the niche width of entity x at time t, as defined by Hannan et al. (2007), is given by:
*+,&ℎ {µ (#,%,&)} = 1− 4µ
!
"∈$
(#,%,&)
Clan contrast. To measure clan contrast, all clan members’ grade of membership in their
affiliated clans were first measured. For each player, a vector of his or her complete clan
affiliation history in each month as a clan member was created, from which a list of clans and
corresponding tenure were obtained. A player’s grade of membership in each clan was calculated
as one divided by the number of clans weighted on tenure in that clan. A clan’s contrast equals
the mean of its members’ grade of membership. Thus, clan contrast measures the average
resource allocation given by all members.
Within-clan network brokerage. For each of the thirty-two months, members were
identified for each clan and their co-play network was built. Those clan members who did not
participate in within-clan co-play networks were treated as network isolates. The number of
times two individuals played together was set as tie weights. At the local level, player’s network
brokerage was measured by one minus Burt’s constraint (2004) using the igraph package in R
for weighted within-clan co-play networks. At the global level, the distribution of network
112

brokerage was calculated as the mean and standard deviation of all members’ network brokerage.
Both local and global network measures were time-variant variables. To organize the data for
Cox’ proportional hazards modeling, monthly-level within-clan network measures were averaged
across each clan’s active months.
Control variables. Clan-level control variables included clans’ number of members, clan
rating, and win rate. Member count is critical to the performance of clans because clans need a
critical mass to regularly field a full team in battles. Clan win rate and clan rating are indicative
of general clan performance.
Table 4. Descriptive statistics for Cox’s proportional hazards modeling predicting clan mortality.
M SD
member
count 15.20 17.56
clan rating 3725.31 1374.77 .29***
win rate 45.76 9.07 .18*** .41***
niche width .45 .04 -.13*** -.16*** -.03
contrast .71 .22 -.26*** -.29*** .22*** .18***
brokerage
mean .17 .22 .74*** .27*** .04 -.17*** -.52***
brokerage sd .23 .06 -.20*** -.05 .03 .03 .22*** -.46***
Note. *p < .05; **p < .01; ***p < .001.
Data for Study 2
Study 2 builds on the dataset of study 1. Different from study 1’s focus on the general
picture, study 2 shifts attention to the multilevel mechanisms of individuals and groups from a
micro-perspective. Apart from examining the networked organizing in online communities,
another important purpose is to broadly inform human organizing because it parallels real-life
team-based organizing. Considering the fluidity of online organizing structures, the best parallel
of real-life organizing should have the following features. First, the groups need to be stable and
have a decent number of members. As shown in study 1, clan size is likely to reduce the
113

mortality hazard of clans, because larger clans would have the critical mass to regularly support a
full team in battles. Second, individual-group affiliation is also more transient than real life. To
make the case comparable to real-life individual-group attachment, there needs to be a
considerable tenure to constitute group affiliation.
As argued above, starting from 18,585 persistently active players and their affiliated
5,797 clans, several conditions were used to create the subsample for hypothesis testing. First,
clans with complete 32 observations from September 2016 to April 2019, so that clan founding
or mortality was excluded as an influencing factor. To make sure the clans selected were stable
and effective organizational structures, clans with more than the median number of members (19
players) were selected.
Second, to exclude transient affiliation, player-clan affiliation was defined as above the
median of tenure, which was ten months for persistent players. That is, only affiliation relations
longer than ten months were included in the current sample. These conditions generated 9743
players affiliated to 1602 clans with 10,427 dyads, including 684 boundary spanners who had
been affiliated with more than one clan.
Construction of Multilevel Networks
The bipartite network of player-clan affiliation, or network X, was the affiliation network
between 9743 players and 1602 clans from previous data reduction. Inter-clan niche overlap
network of 1602 clans was built on the textual similarity of clans’ public profiles. Similar to the
textual preprocessing in the measurement of clan niche width, the profiles were preprocessed
through stemming, removal of stop words, numbers and punctuation. A similarity matrix was
built using the cosine method in the R package of quanteda. ERGM modeling relies on a specific
range of network density. Overly high or low network density would prevent model
114

convergence. The cosine similarity matrix was dichotomized on the cutoff point of the mean plus
one standard deviation of the matrix. This dichotomization generates a network with density
of .21, which is ideal for ERGM modeling. The niche overlap network of clans is network A at
the macro level. For the 9703 players, their complete co-play networks across 32 months were
aggregated and dichotomized on the median of tie weights, which was two. The process of
dichotomization resulted in a network of 9147 players, because the remaining players either had
no co-play ties or weak co-play ties, but these players needed to be added back to the network as
isolates to make the relational structure correct. The co-play network of players makes the
network B at the micro level. The density of network B is .013. Both network A and B are
undirected. All three networks were transformed into adjacency matrices for sampling and
modeling.
Extraction of Random Samples
Since this study employs MERGM, which has strict restriction on sample size, random
samples need to be drawn for model estimation. The number of boundary spanners was much
smaller than non-spanners, and both were important for hypothesis testing. The affiliation data of
boundary spanners includes 684 players affiliated to 522 clans, and the affiliation data of non-
spanners includes 9059 players affiliated to 1564 clans. All added up to 9743 players and 1602
clans.
Four random samples were drawn from the player-clan bipartite network. First, a
stratified sample of 50 clans were randomly selected based on the quantiles of clan age. Second,
all affiliated players were extracted. Third, all their affiliated clans were selected to keep the
multiple clan affiliations of boundary spanners. That is, all selected players’ affiliated clans were
included in the samples. The four samples include: 70 clans and 250 players including 31
115

spanners, 65 clans and 294 players including 18 spanners, 70 clans and 270 players including 41
spanners, 72 clans and 283 players including 25 spanners. Relevant inter-clan networks and
interpersonal networks were extracted for MERGM modeling. Descriptive network statistics of
the four samples are shown in Table 5.
Table 5. Descriptive network statistics of five samples for MERGM.
Sample 1 Sample 2 Sample 3 Sample 4
Nodes
Network A 70 65 70 72
Network B 250 294 270 283
Network X 70 x 250 65 x 294 70 x 270 72 x 283
Edges
Network A 548 462 759 488
Network B 1617 1234 1151 863
Network X 281 312 311 308
Density
Network A .227 .222 .274 .191
Network B .052 .029 .031 .022
Network X .006 .005 .005 .005

MERGM modeling was conducted in MPnet and the model building process followed the
MPnet manual (P. Wang et al., 2014) and empirical studies using MPnet (Wang et al., 2016).
Multilevel mechanisms with corresponding hypotheses are shown below in Table 6. Other
structural parameters were also added into the model to improve model fit.
Table 6. Visual illustration of MERGM parameters (Adapted from MPnet manual).
Hypotheses Parameter Illustration Mechanism Description
H6 ATXBX

Affiliation-
based
closure
Co-play ties are more likely to be
observed between players
affiliated to the same clans
H7 ATXAX

Affiliation-
based
closure
When players span boundaries to
other clans, they are more likely
to join clans that are similar to
the original ones.
H8 AC4AXB

Cross-level
alignment
Co-play ties are more likely to be
observed between players
affiliated to different clans when
these clans are similar
116

H9 ASAXASB

Cross-level
assortativity
Popular players with more
interpersonal ties are less likely
to join clans with more niche
overlap with other clans

This study chooses relatively large clans for hypothesis testing, because clan size is
critical for the self-sufficiency of clans in the current context. To field a complete team for clan-
based battles, it requires 15 members. Without a critical mass, a clan is unable to participate in
clan-based battles on its own. Previous research of guilds in World of Warcraft (WoW) shows
that a guild needs to have at least 16 players in order to form a core group of over 5 players,
which is the team size requirement for WoW (Ducheneaut et al., 2006). In other words, a group
needs to have more players than the required team size in order to carry out activities regularly
due to unstable participation of periphery members.
As a direct consequence of not reaching the 15-member threshold in the current context,
organizational boundary does not necessarily facilitate social network formation within groups.
The group boundary of smaller clans is inherently more fluid than larger ones, because cross-
boundary relations are needed for sustaining clan-based matches. Thus, due to such differences,
it is unclear if we can make the argument that formal organizational boundaries of small clans
can structure lower-level social networks. Another possibility is that such smaller clans do not
aim for clan-based matches. Just like guilds in WoW, there are one-person clans (1.4%) and two-
people clans (1.6%) in the current sample. Including these clans for hypothesis testing would be
problematic because they have fundamentally different organizational purposes from those
aiming for clan-based matches. Thus, mixing large and small clans in the current context is likely
to introduce heterogeneity.
117

To support this claim, I also collected the other half of the sample for comparison, which
includes clans with fewer than median number of members. Following the exact same procedure
as described above, 545 small clans and 1292 players were collected. I constructed their
respective inter-clan network, player-clan affiliation network and inter-player network. Using the
same sampling strategy, I extracted four samples. I will compare the models of small clans with
those of large ones in the result section.
Data for Study 3
Within-clan Roles
Study 3 examines player-clan joining as a function of multirole network personality. In
clans, there are eleven roles ranking from lowest to highest: reservist, recruit, private, junior
officer, recruitment officer, quartermaster, intelligence officer, combat officer, personnel officer,
executive officer, commander. The higher the ranking, the more clan management functions to
which this player has access. Roles from quartermaster upward have access to clan treasury
management. The Recruitment officer and the personnel officer are responsible for reviewing
applications from new members and managing existing members. The combat officer is of
essential importance because s/he organizes clan battles and makes decisions on Global Map
battles. Only the commander can disband a clan.
Clans differ in terms of the norm of moving up the rank. Some would prioritize battle
performance whereas others would also value back-stage participation. Some clans have a clear
structure of measuring contribution for ranking up, whereas others’ rules are not explicit and
tenure seems to be the only criteria. One said based on his observation that it normally took at
least six months or even a year with constant participation to move up. The combat officer just
mentioned was hanging out at the sideline as a low-rank member for several years before taking
118

up that position. Another said tenure was less a factor compared to devotion, meaning investing
several hours daily rather than only appearing at weekends.
The Role Data
From the company’s data warehouse, I obtained each player’s role at each month as a
clan member. If they switched roles mid-month, the role they held for the most time within that
month was set as that player’s role for that month. I first compared these roles in terms of
performance and social resources. Mixed effects regression models were adopted to see how
players’ performance and network properties differed across different roles (Table 6). Since
quartermaster is in the middle of the role hierarchy, it was set as the baseline for comparison.
The dataset was structured as monthly observations clustered under players. The results show
that high and low ranking roles do not seem to differ much in personal win rate. Combat officer,
intelligent officer, junior officer and private seem to have more ties and be more likely to occupy
brokerage positions. As can be seen from the amount of variance explained by these three
models, the model on win rate explained 17% of the variance, whereas the model on network
degree and network brokerage explained 64% and 62% of the variance. That is to say, role is a
more important factor to predict social resources than combating performance. This is possibility
because progression to higher-ranking roles not only takes time but also social relations. This
might also indicate that when players become leaders in clans, their technical contribution by
helping clans win more battles is less important than their coordination contribution. This is
consistent with the emergence of leadership in open-source software communities (O'mahony
and Ferraro, 2007).

119

Table 7. Mixed effects linear regression models of roles predicting win rate and network patterns
(compared to quartermaster).
Win rate Network degree Network brokerage
Estimate SE Estimate SE Estimate SE
(Intercept) .474*** .001 -.798*** .097 *** .188*** .004
personnel_officer .002 .001 .302** .107 * -.001 .004
private .004*** .001 1.572*** .096 *** .078*** .004
recruitment_officer .001 .001 .116 .101 .006 .004
reservist .001 .001 -1.513*** .105 *** .040*** .004
combat_officer .003* .001 1.564*** .101 *** .033*** .004
commander .005*** .001 -3.246*** .112 *** -.111*** .004
executive_officer .004*** .001 -1.227*** .101 *** -.047*** .004
intelligence_officer .004** .002 .709*** .116 *** .024*** .004
junior_officer .003* .001 .697*** .098 *** .041*** .004
recruit -.002 .001 -.690*** .096 *** .012*** .004
Gold bought/10
7
.101 .130 642.3*** 9.32 5.939*** .319
Monthly active days/10
3
.875*** .011 433.489 .791 *** 9.208*** .036
Obs 2216700 2216700 1064204
Players 201633 201633 140562
R
2
.17 .64 .62
Note. *p < .05; **p < .01; ***p < .001.
The Multi-role Population
There are cases where it is impossible to determine the role a player had for the most time
during a month. Out of 207,355 players, 5721 players had at least one month where it was
impossible to determine the single role this person held for the majority of the time for that
month. These players had multiple roles but had to be removed because data accuracy could not
be guaranteed. For the remaining 207,204 players, 111,001 players had only one role as clan
members, and 96,203 players had more than one role as clan members. It is worth pointing out
that these roles are not restricted within the boundary of clans. That is, it is possible to hold the
same role across different clans, and it is also possible to hold different roles within the same
clan. When determining whether a player has single role or multiple roles, I took a summary of
this individual’s role history without distinguishing if the roles belong to the same or different
clans.
120

Following Burt’s multirole study (2012), I first compared players with a single role and
multiple roles in terms of their in-game performance and network building patterns. It is
important to understand the difference between the two kinds of populations before removing
those with single roles to compute multirole network personality. In Burt’s study, the context is
an MMORPG where players play with different avatars and single-role players make a minority.
Under comparison, half of the clan members only had single roles. I used win rate to indicate
player-level performance, which was computed as the count of won battles divided by total battle
count and aggregated across months to the individual level. Another indicator of performance is
player-level global rating. Descriptive statistics are shown in Table 8. Mixed effect regression
model was adopted and when players’ monthly time and financial investment and co-play
resources were controlled (Table 9). The second model shows that players holding single roles
had lower win rate compared to those holding multiple roles, B = -.007, SE = .0004, p < .001.
Inclusion of the differentiation indicator results in lower AIC, and significant improvement of
model fit, Δ LR χ 2 = 503.35, p < .001, and model 2 explained 17% of the variance. This is
consistent with Burt’s finding that players holding single roles had relatively low achievement.
Table 8. Correlation statistics of mixed effects model predicting player’s monthly win rate (N =
2,309,035)
Mean SD
Win rate .49 .13
Co-play network degree 7.00 13.43 .08
Gold bought 2249.39 7752.32 .02 .16
Monthly active days 14.59 9.89 .07 .41 .23

121

Table 9. Mixed effects linear regression model predicting players’ win rate.
Model 1 Model 2
Estimate SE Estimate SE
Fixed effects

(Intercept) .476*** .000
.479 *** .000
Single role
-.007*** .000
Network degree *10
-2
.020*** .000
.018*** .001
Gold bought *10
-7
-.065 .125
-.078 .125
Monthly active days * 10
-2
.078*** .001
.075*** .001
Random effects

Variance .002 .002
SD .045 .045
Obs 2309035 2309035
No. of players 207204 207204
AIC -3025898 -302638
BIC -3025822 -3026296

Log likelihood 1512955 1513199
Δ LR χ 2 503.35
R
2
.17 .17
Note. *p < .05; **p < .01; ***p < .001.
Players with single/multiple roles also differ in network building patterns within their
respective clans. Along with the calculation of clan-level network properties for each month in
study 1, I also computed the individual-level within-clan network properties. Aggregating
monthly network measures to individual level, players with single roles had lower degree
centrality, higher eigen centrality, lower brokerage, and lower coreness. Again, consistent with
Burt (2012), players with multiple roles had higher network advantage. Thus, the focus of this
study is the population of multi-role players with performance and network advantages.
Measure of Multi-role Network Personality
For the 96,203 players holding multiple roles, cross-role network personality was
computed. Although these players all held more than one role, some of them did not have co-
play network measures for one or more roles because they didn’t participate in the within-clan
co-play networks under that role. It is only possible to compute the network personality if a
122

player has network measures for at least two roles. Thus, 67,785 players with at least two
network measures associated with different roles were kept (Table 10). For those removed
players, the roles they held where their network measures were unavailable were primarily low-
ranking roles. 84.7% of these roles were the four lowest-rank roles: reservist, recruit, private,
junior officer. That is to say, these low-ranking players didn’t have access to much relational
resource as a clan member. The remaining players had comparatively more abundant social
resources and higher-rank roles. Similar to Burt’s study, the final population is a biased one with
higher rankings and more social resources. The 67,785 players’ network personality had a mean
of .47 (SD = .25). A higher score indicates consistently building diverse ties with structural holes
across roles.
Table 10. Comparison of roles held by 29.6% removed players (who had fewer than two network
measures across roles) and 70.4% kept players (who had at least two network measures across
roles).
Removed players Kept players
Roles Roles
reservist 5923 reservist 7971
recruit 21424 recruit 51513
private 20307 private 52455
junior_officer 5472 junior_officer 20608
recruitment_officer 3042 recruitment_officer 13871
quartermaster 451 quartermaster 1921
intelligence_officer 734 intelligence_officer 4031
combat_officer 1773 combat_officer 12836
personnel_officer 893 personnel_officer 7506
executive_officer 1306 executive_officer 11555
commander 628 commander 3642

123

The multi-role network personality was tested for its association with performance and
network formation. If following the monthly level data structure, with the large data size and
three-level clustering, the mixed effects modeling is overly computationally intensive. Thus, the
data of 67,785 players were aggregated to role level, making the final dataset a two-level
clustering structure. I included the following controls: players’ number of roles, number of
months in a specific role, a categorical variable of eleven roles, and purchased gold and active
days. These controls were chosen by following Burt’s (2012) study. A player’s number of roles
and the tenure in a specific role would affect the amount of devotion in each role and the extent
to which social relation patterns can be fully developed in each role. Gold amount and active
days represent a player’s interest and investment in the game in general, which would also affect
whether they fully develop social relations in this context. In table 11, mixed effects regression
models show that “brokerage-prone” personality was positively associated with win rate in
model 2, B = 2.624, SE = .089, p < .001. Model 2 explained 11% of the variance. Compared to
model 1 with only control variables, AIC for model 2 is lower, as shown in the lower part of the
table. AIC is the application to an information theoretic distance between the current model and
the “true” model, so smaller AIC indicates better model fit (Akaike, 1974). AIC can help rank
the model from the worst to best, but it does not provide statistical significance test between
models. As a statistical test, likelihood ratio test shows that inclusion of multirole network
personality significantly improved the model fit, Δ LR χ 2 = 858.64, p < .001.

124

Table 11. Mixed effects regression models of network personality predicting role performance.
Model 1 Model 2
Win rate/100 Win rate/100
Estimate SE Estimate SE
(Intercept) 47.396*** .087 46.631*** .091
Fixed effects:
Personality brokerage 2.624*** .089
role_num .257*** .018 .137*** .019
role_executive_officer .028 .087 .161 .087
role_quartermaster -.597*** .156 -.414** .156
role_combat_officer
-.210* .084 -.128 .084
role_commander .120 .122 .286** .122
role_intelligence_officer .082 .116 .191 .116
role_junior_officer -.135 .075 -.047 .075
role_personnel_officer -.207* .096 -.079 .096
role_private -.038 .068 .011 .068
role_recruit -.174** .067 -.089 .067
role_recruitment_officer -.184* .081 -.070 .081
time_in_role
.026*** .003 .024*** .003
gold*10
-7
80.605*** 21.676 60.472** 21.624
active_days*10
-3
68.199*** 2.240 57.741*** 2.261
Random effects
Variance 17.130 16.740
SD 4.138 4.092
No. of obs 201143 201143
No. of players 67785 67785
AIC 1348975

1348121

Loglikelihood -674470.5 -674042.7
Δ LR χ 2 858.64
R
2
.11 .11
Note. *p < .05; **p < .01; ***p < .001.
Why is personality relevant to performance in the current context while irrelevant in
Burt’s study? A considerable difference between characters in EQ II and WoT lies in the
simultaneity of roles. In EQ II, players could play multiple avatars at the same time, and this is
analogous to being an employee and a mother at the same time, thus difficult combinations could
strain the individual. Whereas in WoT, one can only hold a single role at a time within a clan, and
125

this is more analogous to employment in a professional domain. As argued by Burt, having too
many roles at the same time strains the player and prioritization on primary roles dwarfs
demands from secondary roles. In MMORPGs, it is common to use different strategies across
different characters in order to complement each other. In real life, being an employee and being
a mother are roles in different domains that imply fundamentally different goals and tactics. It is
possible that an individual is highly social and aggressive in building open and diverse networks
to gather resources for higher professional achievement, but feels more comfortable embedded
with close friends who share similar experience in child raising. Such different goals and tactics
would dilute the network personality if the weights of different roles are not taken into
consideration. In the context of WoT, however, game motivation stays relatively consistent. What
multi-role network personality captures in the current context is the network building preference
in a specific domain. As shown from above, the role of combat officer associates with higher
social network advantage because this role requires the player to coordinate among clan
members and organize battles, and the role of intelligence officer does not have such social
liaison responsibilities. But the personal and collective objectives stay the same, so a brokerage-
prone player is unlikely to reduce too many structural holes transferring from a combat officer to
an intelligence officer. Similarly, in an organization, roles differ in the requirements to liaison
with other departments, but the network agency stays comparatively consistent.
The Relational Event of Clan Joining
To test network personality’s impact on clan joining behavior, I used Relational Event
Modeling (REM). REM treats a sequence of interactions as discrete relational events and
examines the temporal dependencies among these events (Butts, 2008; Schecter & Quintane,
2020). Different from other network modeling methods, REM treats edges as sequences of
126

discrete events rather than durable states, and edges are recorded with accurate time stamps
rather than aggregated into fixed time intervals (Vu et al., 2017). It allows for testing network
dependencies such as preferential attachment and triadic effects from a time-ordered perspective
(Butts, 2008). REM’s emphasis on time scale makes it an appropriate strategy for analyzing the
unfolding of dynamic interactions and modeling how current events are influenced by historical
events. With the current extension to examining large datasets (Lerner & Lomi, 2020), REM has
considerable potential for wider applicability, because interdependencies of such scales would be
computationally impossible for simulation-based ERGM.
I obtained the exact time stamp of all players’ clan joining. Among 67,785 players with
the network personality measure, 60,121 players had clan joining data, because some of them
stayed in the same clan throughout the observation period. I followed the modeling procedure of
two-mode REM presented in Lerner and Lomi (2020). The two-step modeling process includes
first setting up the configuration of network statistics using eventnet and then evaluating these
network statistics using the Cox proportional hazard model. The most common network statistics
include degree-based and clustering mechanisms. In the current context of player-clan
interaction, players are set as sources that initiate the behavioral event of joining, and clans are
set as targets that receive the event of joining (Lerner & Lomi, 2020). This is a setup in the
program, rather than indicating tie direction of the two-mode network. Player activity indicates
that a player joins many clans. Clan popularity indicates that a clan is joined by many players.
Four-cycle is to model the clustering mechanism as in two-mode networks.. Players sometimes
can’t maintain their clan membership because they are unable to fulfill the obligation of playing
a certain amount of time, but as long as they leave on good terms, they can rejoin the clans they
left. Thus, join repetition is also included as a statistic.
127

REM has special consideration in terms of exogenous variables (Lerner, 2016).
Specifically, if covariates are assumed to interact with previous events to influence future events
of the same dyad or a different dyad, then these covariates need to be included in the set-up of
network configurations; if covariates are assumed to only directly influence the same dyad, then
they don’t have to go through the network configuration process. For example, Lerner (2016)
focused on the modeling of positive (cooperative/friendly) and negative (hostile/conflictive)
interaction events among countries. In the context of international relations, existing formal
alliances or trade relations among countries as an exogenous covariate will interact with previous
events to influence future events (Brandes et al., 2009). Therefore, these covariates need to be
modeled into network configurations. Such a condition is not applicable in the current case.
Personality is assumed to only affect the dyads for corresponding individuals. Thus, the
personality variable was appended after the network configurations were generated. The
combined data was evaluated by a survival model.

128

Chapter 6: Results
Results for Study 1
Hypotheses in study 1 are tested by Cox’s proportional hazards models. Clans with
complete observations were included in the final dataset for modeling. Although not a hypothesis
to test, the survival probability decreased with time (Figure 6), showing that the survival of clans
followed the logic of “liability of oldness” (Bakker & Josefy, 2018). This also indicates the
difficulty of keeping the activeness and vitality of virtual organizations. This is echoed by an
interviewee. “Clans are all in a life cycle…They were all moving on. There’s just so much
novelty. Almost no matter what you do, it just eventually fizzles out. People move on to other
clans or different things.”
Figure 6. Kaplan-Meier survival estimate of clan mortality.

For hypothesis testing, as shown in Table 12, controlling for clan size and performance,
clan niche width had a positive association with mortality hazards in model 2, B = .09, SE = .04,
p < .05. Clan niche width was marginally significant after clan contrast was controlled in model
3, B = .07, SE = .04, p = .07. H1 was partially supported. Clans with a larger niche width were
129

more likely to disband. Compared to model 1, model 2 indicated lower AIC, but slightly higher
BIC. This is because compared to AIC, BIC penalizes more on model complexity. Inclusion of
clan niche width significantly improved the model, Δ LR χ 2 = 4.34, p < .05. Model 2 explained
4% of the variance.
Clan contrast in model 3 and model 4 were both negatively associated with mortality
hazards, B = -6.21, SE = .23, p < .001, offering support to H2. Clans with sharper boundaries and
clearer collective identities were more likely to survive. Inclusion of clan boundary highly
increased the amount of variance explained (13%), and improved the model fit as indicated in
lower AIC and lower BIC. Compared to model 1, adding two main effects significantly
improved the model, Δ LR χ 2 = 726.84, p < .001.
As a comparison, another measure of inter-clan membership overlap was computed to
validate this finding. If a player was affiliated to multiple clans across the observation period,
these clans were set a clique and ties were recorded among these clans (Benefield et al., 2016).
Each clan’s network centrality was calculated within this inter-clan membership overlap
network. Higher centrality means more membership overlap with other clans and therefore fuzzy
boundaries. With the same control variables, this measure was found to positively associate with
clan mortality hazards, B = .100, SE = .044, p < .05. This model explained 5.03% of the variance.
This measure had a weaker effect than the clan contrast measure, possibly because this measure
can only capture membership mobility but not varied tenures in different clans.
The interaction effect between clan niche width and clan contrast was positive and
significant in model 4, B = .41, SE = .16, p < .05. That is to say, the positive effect of clan niche
width on mortality hazard was stronger when clan contrast was higher. If occupying a narrow
niche, clans would be better off with lower mortality hazards by having clear boundaries; if
130

occupying a wide niche, clans would be better off by having fuzzy boundaries. H3 was
supported. Compared to model 3, inclusion of the interaction effect resulted in lower AIC.
Compared to model 1, inclusion of the two main effects and the interaction effect significantly
improved the model, Δ LR χ 2 = 733.17, p < .001. Model 4 explained 13% of the variance.
Since average within-clan network brokerage was highly correlated with member count,
member count was not controlled in model 5-7. Average within-clan network brokerage was
found to have a positive association with mortality hazards in model 7, B = .60, SE = .21, p < .01.
H4a is supported. The existence of more brokers within a group is not conducive to the survival
of the group. Model 7 explained 4% of the variance. Compared to model 5, inclusion of the two
main effects resulted in lower AIC but slightly higher BIC, and significantly improved the
model, Δ LR χ 2 = 11.55, p < .01.
The standard deviation of within-clan brokerage was found to have a curvilinear
relationship with mortality hazards. The linear item is negatively significant, B = -9.67, SE =
2.34, p < .001, and the quadratic idem is positively significant, B = 12.93, SE = 4.48, p < .01.
That is to say, a moderate amount of dispersion of brokerage benefited clan survival, but too
much dispersion was not good. H4b was not supported. Model 9 explained 5% of the variance.
Compared to model 5, inclusion of brokerage heterogeneity resulted in lower AIC and BIC, and
significantly improved the model, Δ LR χ 2 = 37.24, p < .01. A possible explanation is that clan
longevity might depend on the most active members who have the most relational resources.
The interaction effect between brokerage mean and niche width was negatively
significant in model 8, B = -.45, SE = .20, p < .01. The positive effect of clan niche width on clan
mortality hazards was weaker when the brokerage mean was large. In other words, if occupying
a narrow niche, clans would be better off with lower mortality hazards by having fewer brokers;
131

if occupying a wide niche, clans would be better off by having more brokers. H5 was supported.
Model 8 explained 5% of variance. Inclusion of the two main effects and the interaction effect
resulted in lower AIC, and significantly improved the model, Δ LR χ 2 = 17.40, p < .001.
Table 12. Results for Cox’s proportional hazards modeling predicting clan mortality.

Model 1

Model 2

Model 3

Model 4

B SE B SE B SE B SE
niche width

.09* .04 .07† .04 -.15 .10
contrast

-6.21*** .23 -6.70*** .30
niche width x contrast

.41* .16
member count .00† .00 .00 .00 -.02*** .00 -.02*** .00
rating .13*** .03 .13*** .03 -.30*** .04 -.30*** .04
win rate -.54 .48 -.54 .48 7.64*** .59 7.72*** .60
observations 2692

2692

2692

2692

events 564

564

564

564

AIC 8771.13

8768.79

8048.29

8043.97

BIC 8784.14

8786.13

8069.97

8069.98

Loglikelihood -4382.57

-4380.40

-4019.15

-4015.98

Δ LR χ 2

4.34

726.84

733.17

Pseudo R
2
.03 .04 .13 .13
Note. †p < .10; *p < .05; **p < .01; ***p < .001.

Model 5

Model 6

Model 7

Model 8

Model 9

B SE B SE B SE B SE B SE
niche width

.08† .04 .09* .04 .17** .05 .08† .04
brokerage mean

.60** .21 1.06*** .28

niche width x brokerage mean

-.45* .20

brokerage sd

-9.67*** 2.34
brokerage sd sqr

12.93** 4.48
member count

-.01** .00
rating .08* .03 .08* .03 .05 .04 .06 .04 .11** .04
win rate -1.69** .55 -1.67** .55 -1.54** .55 -1.60** .56 -1.15* .56
observations 2134

2134

2134

2134

2134

events 494

494

494

494

494

AIC 7443.74

7442.38

7436.19

7432.34

7414.50

BIC 7452.15

7454.99

7453.00

7453.36

7439.72

Loglikelihood -3719.87

-3718.19

-3714.10

-3711.17

-3701.25

c

3.36

11.55

17.40

37.24

Pseudo R
2
.02 .03 .04 .05 .05
Note. †p < .10; *p < .05; **p < .01; ***p < .001.
132

Results for Study 2
All four MERGM models were converged with t-ratios below .1. Results are shown in
Table 13. Apart from parameters in hypotheses, other meso-level and multilevel configurations
were added to improve model fit. Meso-level signatures included players sharing membership to
clans, and clans sharing the same members. Multilevel signatures included the mechanism
indicating that clans with niche overlap ties are also more likely to have more members. These
structural controls were selected after repeated trials for model convergence and better GoF.
The parameter indicating affiliation-based closure at the micro-level (ATXBX) was
consistently positive across four samples, showing robustness of this mechanism. Players
belonging to the same clans are more likely to build co-play network ties. H6 was supported.
The parameter indicating affiliation-based closure at the macro-level (ATXAX) was
positively significant across three samples, but was not significant in the third sample. This result
offers partial support to H7. Boundary-spanning players are more likely to join two similar clans,
but this mechanism is not as robust as the former one.
The parameter indicating cross-level network alignment was positively significant in two
samples, not significant in one sample, and caused model degeneracy in the second sample. This
mechanism is not robust across samples, showing players’ cross-clan co-play relations are not
necessarily governed by clan similarity. In other words, players would collaborate with others
from very different clans. H8 is not supported.
The parameter indicating cross-network assortativity was consistently negatively
significant across four samples, offering robust support to H9. Popular players with more
interpersonal ties are less likely to join clans with more niche overlap with other clans, because
they are already endowed with more social capital.
133

Table 13. Estimates of MERGMs for the multilevel networks of players and clans across four
random samples.
Sample 1 Sample 2 Sample 3 Sample 4
Effects Parameter SE Parameter SE Parameter SE Parameter SE
EdgeA (clan-blue
square) -0.809* 0.060

-0.542*

0.075 0.147* 0.069 -1.104* 0.083
EdgeB (player-red
circle) -3.156* 0.052

-3.735*

0.064 -4.023* 0.059 -4.339* 0.084
XEdge -0.896* 0.341
-0.985* 0.302
-2.594* 0.522 -4.736* 0.500
XACA

-1.016* 0.064

-0.829*

0.070 -0.724* 0.093 -0.590* 0.080
XACB

-1.561* 0.367

-2.169*

0.473 -0.930* 0.413 -2.414* 0.440
StarAX1A

0.152* 0.011

0.199

0.016 0.066* 0.010 0.139* 0.017
ATXBX (H6)

3.789* 0.077

4.138*

0.074
4.025* 0.090 4.903* 0.181
ATXAX (H7)

1.649* 0.443

1.077*

0.440 0.611 0.384 3.468* 0.492
AC4AXB (H8)

0.000 0.023 0.174* 0.033 0.548* 0.071
ASAXASB (H9)

-0.143* 0.013

-0.196*

0.014 -0.139* 0.016 -0.391* 0.035
Note. Parameters with * indicates that the estimate divided by its standard error is larger than 2,
which means that the estimate is statistically significant at a minimum of p < .05.

134

Goodness of fit (GoF) (Table 14) tests how closely the estimated model matches the
observed networks, both for parameters included in the model and for additional network
dimensions (Wang et al., 2016). In the case of MERGM, the upper half of the table shows
multilevel parameters included in the model, and the lower half of the table shows global
network statistics for individual networks. Ideal t-ratios should be below 2 to indicate very good
fit. With the large size and complexity of networks in MERGM, absolute values of t-ratios
between 2 and 6 were also observed in other MERGM studies (Brennecke & Rank, 2016). GoF
diagnostic showed that the degree distribution and clustering in network A and network B
themselves were underrepresented as some of them were well above 2. However, in the current
model building process, inclusion of any within-level structures in network A or B resulted in
model degeneracy after repeated trials.
Table 14. Goodness of fit for MERGMs for the multilevel networks of players and clans across
four random samples.
Sample 1 Sample 2
Statistics Observed Mean SD t-ratio
Observed
Mean SD
t-ratio
EdgeA 548 545.551 32.006 0.077
462 462.045 19.354 -0.002
EdgeB 1617 1399.485 114.589 1.898
1234 962.918 79.85 3.395 #
XEdge 281 205.294 24.961 3.033 #
312 216.44 21.653 4.413 #
XACA 1735 1162.942 319.034 1.793
1626 897.132 279.067 2.612 #
XACB 23.9688 20.7344 5.467 0.592
15.875 10.2131 4.017 1.409
StarAX1A 6753.2031 4263.4182 656.831 3.791 #
5492.4941 3198.6044 423.289 5.419 #
ATXAX 14.2188 11.2445 3.478 0.855
6.875 3.5832 2.686 1.225
ATXBX 1321 600.432 168.273 4.282 #
1055 371.889 118.792 5.75 #
ASAXASB 16080.9489 9712.2477 1551.432 4.105 #
11661.1169 5966.511 860.57 6.617 #
AC4AXB 5618.957 4309.846 448.126 2.921 #
stddev_degreeA 11.3097 4.177 0.396 18.014 #
10.9851 4.8245 0.372 16.573 #
skew_degreeA 0.1558 0.1921 0.25 -0.145
0.1832 0.3209 0.197 -0.699
clusteringA 0.5549 0.2361 0.016 19.432 #
0.5743 0.2467 0.013 25.125 #
stddev_degreeX_A 7.3947 6.0911 0.84 1.552
7.4687 5.5525 0.795 2.409 #
skew_degreeX_A 0.5801 1.2216 0.408 -1.574
-0.4417 -0.3863 0.42 -0.132
stddev_degreeX_B 1.1737 1.0483 0.074 1.693
1.0898 0.9014 0.056 3.358 #
skew_degreeX_B -1.0544 -1.3041 0.028 8.803 #
-1.0257 -1.3878 0.044 8.223 #
clusteringX 0.1144 0.3645 0.04 -6.224 #
0.1053 0.0757 0.091 0.325
135

stddev_degreeB 16.1883 6.3663 1.274 7.712 #
9.4505 3.2097 0.626 9.967 #
skew_degreeB 1.2347 1.3189 0.174 -0.484
1.1671 0.759 0.243 1.679
clusteringB 0.7499 0.1972 0.048 11.416 #
0.7358 0.1235 0.044 13.813 #

Sample 3 Sample 4
Statistics Observed Mean SD t-ratio Observed Mean SD t-ratio
EdgeA 1144 1156.646 25.855 -0.489 488 454.855 28.285 1.172
EdgeB 1151 819.569 68.846 4.814 # 863 610.812 32.161 7.841 #
XEdge 287 254.035 8.645 3.813 # 308 261.605 13.258 3.499 #
XACA 1510.5 829.4145 217.562 3.131 # 1241 680.198 108.857 5.152 #
XACB 29.4375 32.3952 5.513 -0.537 23.5 30.4239 7.284 -0.951
StarAX1A 12089.3251 10874.5827 950.566 1.278 6444.8928 4023.874 447.163 5.414 #
ATXAX 13.6875 15.969 3.604 -0.633 12 18.1914 4.755 -1.302
ATXBX 928.5 323.5805 76.532 7.904 # 741.5 254.9845 35.406 13.741 #
ASAXASB 22159.1037 17896.1272 1097.515 3.884 # 11155.0446 6639.5083 521.673 8.656 #
AC4AXB 9785.4129 9173.1666 443.516 1.38 5285.6529 4040.5757 228.794 5.442 #
stddev_degreeA 8.4725 4.4522 0.369 10.89 # 10.6426 3.5121 0.29 24.568 #
skew_degreeA -0.1005 0.0448 0.253 -0.576 0.686 0.3482 0.235 1.44
clusteringA 0.5189 0.4812 0.011 3.349 # 0.4981 0.18 0.015 21.012 #
stddev_degreeX_A 6.9522 5.236 0.581 2.952 # 6.2754 4.8082 0.291 5.037 #
skew_degreeX_A 0.1797 -0.6252 0.502 1.603 -0.7303 -1.1547 0.191 2.227 #
stddev_degreeX_B 1.1649 1.0914 0.03 2.479 # 1.1267 1.0801 0.051 0.921
skew_degreeX_B -1.1945 -1.2892 0.023 4.187 # -1.0387 -1.3028 0.022 12.094 #
clusteringX 0.1949 0.0234 0.025 6.912 # 0.0515 0.1365 0.108 -0.785
stddev_degreeB 9.4926 2.9769 0.544 11.98 # 5.4918 1.9883 0.218 16.068 #
skew_degreeB 1.4364 0.9138 0.318 1.643 0.9726 0.5569 0.206 2.023 #
clusteringB 0.7004 0.096 0.031 19.619 # 0.6612 0.07 0.021 28.443 #
Note. Parameters with # indicates the t-ratio is large than 2, meaning this statistic is not well
represented.
As a comparison, four samples were drawn from small clans with fewer than the median
number of members (Table 15). Control configurations were selected after repeated trials for
model convergence and better GoF. The purpose is to see how the models perform when the
configurations from the hypotheses were added. In general, cross-level alignment can be added
and the models converged, but this mechanism was not significant in any sample. Group
affiliation-based homophily caused most models to degenerate except sample 2, but it was not
significant in sample 2. As shown in the previous results, this mechanism was consistently robust
136

across samples for larger clans. This is probably an evidence that small clans do not establish
formal organizational boundaries to organize homogeneity within. Cross-level assortativity
caused all models to degenerate, but it was also consistently robust for larger clans. To
summarize, smaller clans follow different social mechanisms as compared to larger clans.
Table 15. Estimates of MERGMs for the multilevel networks of small clans.

Sample 1

Sample 2

Sample 3

Sample 4

75 x 106

75 x 110

75 x 112

75 x 106

Effects Estimate SE Estimate SE Estimate SE Estimate SE
EdgeA -2.796* .455 -2.169* .178
-3.050* .296 -1.886* .265
EdgeB -5.236* .331 -4.507* .203
-4.904* .304 -4.807* .323
XEdge -3.669* .972 -4.599* .615
-3.974* .761 -1.102 .949
XASA -.084 .829

-1.605*

.672

-1.842*

.795
XACA

-.419 .428

StarAX1A -.500* .213 .012 .075

.048

.116

-.067

.131
Star2BX -.104 .280 .123 .101

-.186

.243

.230

.237
L3XAX .178 .160 .033 .044

L3XBX

.010

.160

-.006

.160
AC4AXB .109 .258 .009 .094

.186

.143

-.236

.137
ATXBX degenerate .000 .403 degenerate degenerate
ASAXASB degenerate degenerate degenerate degenerate
Note. Parameters with * indicates that the estimate divided by its standard error is larger than 2,
which means that the estimate is statistically significant at a minimum of p < .05.

137

Results for Study 3
Study 3 examines how multirole network personality effects of clan joining behaviors. I
followed the two-step procedure of evaluating a relational event model. In the first step of setting
up the network configuration, I set the time unit as one calendar month to be consistent with the
monthly-aggregated data. I then set a time decay parameter with a halflife of one time unit,
meaning that the influence of past events on the current event gets halved when time increases by
one month. Apart from being consistent with the time unit, selection of time decay parameters
also accounts for common probationary period in clans, which usually last for a couple of weeks.
After setting up the network configurations, the next step is to find the best fitting model
that can best capture the temporal endogenous dependencies. To ensure the goodness of fit (GoF)
of the chosen network statistics, I followed the model selection process introduced by Schecter et
al. (2018). I assessed the GoF of each model by computing the log likelihood, Akaike
information criterion (AIC), and Bayesian information criterion (BIC). I compared each model
with the null model to see the change in model fit. Based on the model selection process, if
including a set of endogenous parameters results in a significant improvement in model fit
compared to the null model, the data have a higher likelihood of occurring given that set of
parameters, so that set of structural parameters can better capture the endogenous dependencies
(Schecter et al., 2018) As shown in Table 16, when adding each of the parameter, AIC, BIC and
likelihood tests all show that it independently improved the model fitting. Finally, the full model
with all the network statistics had the best fit with lowest AIC, BIC and highest log likelihood. It
also resulted in the highest improvement of model fit, Δ χ
2
= 263418, p < .001. Thus, the full
model was considered the model that best represented the observed relational events, and it was
chosen to test the hypothesis on multirole network personality.
138

Table 16: Model selection based on goodness of fit (GoF) statistics.
Models Log likelihood AIC BIC
Δ χ
2
p value
Null -290654.3 581308.5 581308.5

Null + repetition -258523.9 517049.7 517059.7
64261 ***
Null + popularity -238316.4 476634.7 476644.7
104676 ***
Null + activity -290358.4 580718.8 580728.8
591.68 ***
Null + popularity x activity -232315.3 464636.6 464666.6
116678 ***
Null + four cycle -214824.2 429650.5 429660.5
151660 ***
Null + all -158945.3 317900.5 317950.5
263418 ***

Figure 7 shows Schoenfeld residuals plots of network statistics and the independent
variable for the proportional hazards assumption (Vu et al., 2017). Join repetition and activity
deviate from this time-invariant assumption. Thus, robust standard errors were computed in
Table 17. All the degree and clustering network statistics were significant. The interaction effect
between player activity and clan popularity was positively significant, indicating the tendency
for active players to join popular clans. In the current case, player activity means quickly shifting
clans, and clan popularity means members quickly coming and going. Such assortativity
indicates a match between uncommitted members and fluid groups. The covariate in question,
multirole network personality, is positive and significant, Estimate = .017, Robust SE = .004, p
< .001. This result shows that “brokerage-prone” players are more likely to span clan boundaries.
Compared to the full structural model shown above, inclusion of the multirole network
personality significantly improved the model fit, Δ χ
2
= 23.308, p < .001. This model explains
80% of the variance. By accounting for the structural interdependency of player-clan interaction,
players who are consistently brokers across multiple roles are more likely to be boundary
spanners. This result is robust after controlling for clan age and clan rating. H10 was supported.
Clans with higher ratings are more popular with players, which is consistent with existing
139

findings that actors evaluate and choose to span to categories with abundant resources (Pontikes
& Kim, 2017; Wry & Castor, 2017).
Table 17. Relational Event Model (REM) of multirole network personality to predict the event of
players joining clans.
Model 1 Model 2

Estimate Robust SE Estimate
Robust SE
Join repetition .919*** .016 .913***
.016
Join popularity .653*** .007 .600***
.007
Join activity -1.533*** .033 -1.520***
.032
Four cycle 1.881*** .019 1.820***
.019
Join popularity x Join activity .138*** .009 .147***
.009
Personality brokerage (H10) .017*** .004 .013***
.004
Clan age -.382***
.004
Clan rating .177***
.003
No. of the risk set 969,449 950,427

No. of relational events 161,576 161,441

AIC 317879.2 304467.3

BIC 317939.2 304547.3

Log likelihood -158934 -152225.7

Pseudo R
2
.80 .81

Note: The first model only contains structural parameters and the independent variable. The
second model also includes controls of clan age and clan rating.

Figure 7: Model checking results based on Schoenfeld residuals.

140

141

Chapter 7: Discussion
Summary of Findings
Theoretical Framework
This dissertation proposes a multilevel network framework of organizing to theorize
around the interdependencies in revised theory of niche. It makes two propositions for the
multilevel network framework. First, it proposes that categories at multiple levels are nested in a
hierarchical pattern. The recent decade has witnessed a considerable growth of the categorization
literature, but this stream of scholarship has only just begun to formally consider the hierarchical
structure of categories (Hannan et al., 2019). This proposition follows the theoretical foundation
of hierarchic social systems that considers the contingency between adaptation and environment
(Simons, 1962). The rationale is bounded rationality, which posits that “the complexity of the
environment is immensely greater than the computational powers of the adaptive system”
(Simon, 2019, p. 166), and hierarchic system has evolutionary advantage in problem solving. A
common type of hierarchical category system is an organizational structure, i.e., individuals are
clustered in organizations, and organizations are clustered in organizational categories.
Organizations are naturally composed of interconnected subsystems until the division reaches the
lowest level of elementary subsystems (Simon, 1962). There are abundant examples. Law firms
with legal specialty areas are formed by lawyers (Paolella & Durand, 2016); films labeled with
different genres are produced by teams with personnel overlaps (Hsu et al., 2009);
nanotechnology companies clustered in patent categories are formed by research scientists (Lo &
Kennedy, 2014); Wikipedia articles embedded in the hierarchical knowledge system are written
by contributors (Lerner & Lomi, 2018). These existing studies focus on how organizations are
142

categorized in higher-level categories but fail to consider that organizations are not unitary
entities but with complex internal systems.
Secondly, the multilevel network framework proposes that category membership can be
reframed as a bipartite affiliation network that connects lower-level and higher-level entities. For
the above examples, there could be a bipartite network of lawyers and law firms and a bipartite
network of law firms and legal specialties; similarly, there could be a bipartite network of films
and genres and a bipartite network of films and production teams. The purpose of positioning
category affiliations as bipartite networks is to connect networks at different horizontal levels
and account for how entities are embedded in their respective horizontal levels. For instance,
lawyers build friendship and advice networks, law firms may form strategic alliances, and legal
specialty areas may have different degrees of overlap. In some cases, such within-level networks
are available and play important roles; in other cases, bipartite networks can also be projected to
one-mode networks. The multilevel network framework thus creates a conceptual perspective to
consider complex interdependencies of within-level and cross-level networks. It also highlights a
dynamic view of how different systems coevolve in response to each other. This proposition
accords with the multiple-network framework of newness emergence (Padgett & Powell, 2012).
The reality can be highly complex involving many networks within different domains, but a
single research design is usually limited to a couple of networks (Padgett & Powell, 2012).
Adopting this multilevel network framework, the empirical context of this dissertation is
virtual organizing in an online gaming community. Following the first proposition on
hierarchical categorization, players are members to clans, and clans are members to clan
categories. Following the second proposition on within-level networks, players build
collaborative networks. With such a multilevel network setup, three empirical studies were
143

conducted to theorize around multilevel organizing. Before discussing the theoretical and
practical implications, findings from these three studies are summarized below.
Study 1
The first study examined the mortality of virtual groups. This study confirmed the logic
of “typicality judgement” such that groups with larger niche width have higher mortality hazards
(Paolella & Sharkey, 2017). Competitive and social groups represent players’ distinctive
motivations for game playing, and they have different decision-making structures and
organizational cultures. Given the breadth and depth tradeoff, a group that evenly allocates
resources in these two broad niche categories is difficult to survive. A large niche width also
creates an ambiguous identity, which confuses existing and prospective members and lowers
appeal to them. Thus, groups occupying a broad niche are more likely to disband.
Following the multilevel framework, the positive relationship between niche width and
mortality hazards is contingent upon group boundaries and within-group networks, because
groups are formed by heterogenous individuals that change group membership and build
interpersonal networks. The first moderator, group boundary, is operationalized as a category
contrast based on individuals’ membership mobility. Sharp group boundary means many
members in a group have high GoM. Category contrast has been conceptualized as internal
coherence, which means a category is concrete and specific with an effective boundary, and such
internal coherence is conducive to category viability (Lo et al., 2020). Although I agree with
them on the connection between coherence and effective boundary, I disagree that contrast is
internal to categories. Category boundaries become fuzzy because members are exposed to the
opportunities of claiming membership to other categories. In other words, category boundary
represents inter-category interconnectedness. Thus, by integrating group boundary into the
144

model, this study accounts for inter-group dependency. Aligning with the coherence view,
groups with clear boundaries have lower mortality hazard. Following the multilevel perspective,
group boundary also plays a moderating role such that higher coherence enables a group to better
engage with a focused niche. This is because in the current case, the group level categories and
the group niche level categories together make a hierarchical category system, such that the
group level categories are fine-grained categories under the broad group niche level categories.
To better engage with a focused niche, a group needs to be a cohesive community with similar
mindsets and interests. A sharp group boundary means members hold similar interests because
they allocate more group experience with a focal group.
The second moderator is the within-group network structure, and this study focuses on
network brokerage from a collective perspective. Group cohesion has been approached as
network density in many studies (Benefield et al., 2016). Network brokerage from a group
perspective can better capture the unequal distribution of relational advantages within a group
(Bizzi, 2013). If the construct of group boundary represents group coherence from an inter-group
perspective, the distribution of network brokerage represents group coherence from a within-
group perspective. The logic is the same that coherence is beneficial for groups with narrow
niche positions, because members’ mindsets, interests, motivations, and goals could be better
aligned. Findings from this study show that the existence of more brokers in a group makes the
group more likely to disband. A smaller mean of brokerage within a group, however, enhances a
group’s ability to engage with a narrow niche and makes it even less likely to disband.
Furthermore, following the multilevel perspective, a smaller group mean of brokerage enhances
a group’s ability to engage with a narrow niche and makes it even less likely to disband.
Study 2
145

The second study adopted MERGM to examine the micro mechanisms of multilevel
networks at all three levels. Consistent with the well-established fact in offline organizations, the
first finding confirmed a higher likelihood of within-group ties than cross-group ties. That is to
say, just like offline organizations, virtual group boundaries can provide opportunities for
members to build interpersonal networks.
The other three hypotheses bring in the macro-level inter-group niche overlap networks.
The emergence and evolution of networks at lower levels cannot be viewed in isolation from
macro-level networks. In the current context, how individuals join groups and how they
collaborate are largely dependent on groups’ positioning in the ecological space. For example, do
they strive for superior performance or do they simply want to socialize? Given the competitive
relations among groups in the current context shown in the pilot study, the inter-group network is
conceptualized as a niche overlap network. In network terms, this is basically a projection of the
bipartite affiliation network between groups and group categories onto a one-mode similarity
network among groups.
The MERGM findings show that boundary-spanning individuals affiliated to multiple
groups are more likely to join similar groups, and this result is robust across three samples. This
finding shows evidence that inter-group niche-overlap provides opportunities for individuals to
span group boundaries. This is probably because individuals are more likely to find similar
groups compatible if their motivations for game playing stay consistent. Since this study focuses
on durable affiliations, these boundary spanners have actually stayed in multiple groups for a fair
amount of time. Boundary spanners can bring novel information, create bridging social capital,
and promote information flow across sub-communities. Niche overlap between groups helps
146

reduce the additional investment and cost associated with joining and getting socialized into a
new community.
MERGM findings offer limited support to cross-level alignment, which means that
individuals do not necessarily form boundary spanning relations with others who are affiliated to
similar groups. In other words, individuals may collaborate with others from different groups.
There could be two reasons for this finding. First, cross-level alignment in offline organizations
represents authority. Such hierarchy might not be evident in virtual groups. Testing this claim
would be a worthwhile future direction. As discussed in the literature on online communities,
virtual groups might be formed from a collective group identity, which is a top-down process;
groups might also be formed from identifiable individuals, which is a bottom-up process (Cheng
& Guo, 2015; Postmes, 2006). Thus, in computer-mediated contexts, lower-level networks might
not necessarily be shadows of higher-level networks. Second, niche overlap ties among groups in
the current context are mainly competitive. Although collaboration between individuals from
similar groups is more likely to be smooth, there’s also potential risk of knowledge leakage. For
example, since similar groups are likely to be close on the level of competitiveness, they might
adopt similar strategies in team battles. Such direct competition between groups might prevent
their members from collaborating.
Finally, the findings show that popular individuals with more collaborative ties are less
likely to join groups with more niche overlap with other groups, and this result is robust across
four samples. Groups with much niche overlap face a lot of competitors. As shown in the second
finding of this study, niche overlap facilitates individuals to span boundaries. Thus, groups with
much niche overlap are likely to have fluid boundaries. This is confirmed in the data by a
negative correlation between group network centrality in the niche overlap network and group
147

contrast, r = -.11, p < .001. Popular individuals with many social ties have limited capacity to
invest and maintain these ties, so they are more likely to access bridging social capital rather than
bonding social capital. Since popular individuals are already exposed to diverse information,
they are less motived to join and stay in groups with fuzzy boundaries.
Study 3
From a multilevel perspective, the third study seeks to examine the interdependency
between a micro-level interpersonal network and a meso-level affiliation network. As discussed,
the multilevel network framework is an integration of social categories and social networks. The
theoretical motivation is to examine the role of agency in engaging with the two forms of social
interactions. That is, whether an individual has the consistency in accessing resources through
social networks and joining groups, given both social categories and social networks differentiate
between diversification and concentration. To capture such agency, this study first followed the
construct of multirole network personality to capture individual qualities and preferences that
reflect in consistent network building patterns across different roles (Burt, 2012). Multirole
network personality is not a personality construct in the psychological sense. It involves many
dimensions of an individual’s social preference, temperament, perception of one’s capability,
which associate with network advantage and can be attributed to personality.
Different from Burt (2012), this study found that “brokerage-prone” individuals tend to
have better performance. My explanation is that the current gaming context and the role-playing
context studied by Burt have different affordances and map onto different multi-role problems in
real life. Burt seeks to examine roles serving entirely different objectives in different domains
(e.g., professional roles and family roles). There is a possible discrepancy between such real-life
condition and MMORPGs, because some in-game roles are purposefully not performed to their
148

full potential when multiple roles are played simultaneously due to prioritization strategy in
games. By comparison, the current study examines roles serving similar objectives but having
different requirements (e.g., different professional roles like a marketing executive and a human
resource executive). Multirole personality in this case captures the core personal preference in
one specific domain (e.g., work). In the current context, players can only hold one role at a time
and different roles require the player to perform different social functions. The current context
offers a good parallel to examine domain-specific cross-role network personality. Relational
event modeling shows that the multirole network personality significantly affects group joining
behavior. “Brokerage-prone” individuals are more likely to join groups, which indicates a
consistency of resource access (diversification or concentration) through social categories and
social networks.
Theoretical Contribution
Meanings of Categories
These three studies make several theoretical contributions. First, when examining groups’
niche occupation in the two broad niche categories, study 1 considers how the two categories are
semantically constructed within the community through a discourse analysis of group public
profiles. Such discourse analysis uncovers the semantic nuances of how groups articulate their
identities in the community. Consistent with the view that categories are cultural objects (Hsu &
Hannan, 2005), group categories in the current context represent the interests, mindsets, and
motivations for gaming. Thus, group categories are not only material based on observable
attributes (e.g., gender, ethnicity, skill level, membership mobility) but also conceptual based on
abstract meanings (e.g., rules and ordering, enjoyment and fellowship) (Durand et al., 2017).
149

This discursive approach responds to the recent critique on predetermined categories, and
instead views categorization as collective negotiation of meanings (Glynn & Navis, 2013). This
approach also echoes Pachucki and Breiger’s (2010) view on the mutually constitutive nature of
social networks and cultural meanings. Methodologically, this discursive approach corresponds
to Grodal and Kahl’s (2017) emphasis on communicative text as active construction rather than
passive representation of categories. According to these authors, communicative text could
reveal negotiations and contestations among actors embedded in material and cultural contexts
(Grodal & Kahl, 2017).
Despite scholars’ call for more research attention on the semantic dimension of
categories, it has largely been an under-studied area (Durand et al., 2017). Some research on
category emergence and mortality have adopted this perspective. For example, Khaire and
Wadhwani (2010) examined how Modern Indian art as an emerging category was discursively
constructed by a diverse range of actors including historians and critics. In a conceptual article,
Wry et al. (2011) argued the legitimation of nascent collective identities was culturally mediated
by storytelling or discourse. In another conceptual piece, Lo et al. (2020) argued that the
distinctiveness of categories in relation to a broad meaning system is essential for category
viability. This semantic perspective echoes the organizational vocabulary literature that
combines network structure and meanings (Loewenstein et al., 2012; Lomi et al., 2017; Tasselli
et al., 2020). To summarize, despite some conceptualization on the semantic and discursive
perspective, empirical analysis has been rare. Study 1 helps address this research gap by
analyzing a large-scale textual dataset as collective construction of categorical meanings. I will
further discuss more recent semantic analysis approaches to empirically explore this perspective
in the following section on future directions.
150

Multilevel Categorization
Following the proposed multilevel network framework, these studies treat collectives as
heterogenous individuals rather than unitary entities. Thus, how well can organizations engage
with a particular niche occupation pattern depends on their constitutive members. Study 1 has
empirically shown that a clear organizational boundary can help a group effectively engage with
a focused niche position. That is, category membership and their consequences at different
hierarchical levels are interdependent.
This finding is generalizable to many hierarchical categorization structures. For example,
in the case of corporate law industry, law firms with a wide niche in legal specialties areas are
considered as more competent because client companies usually seek one-stop full services, e.g.,
mergers and acquisitions, litigation, tax, intellectual property, real estate, employment (Paolella
& Durand, 2016). Law firms operating in such diverse areas are evaluated positively and better
at maintaining their clients. Such full-range service might require lawyers with different
specialties to collaborate on the same client case, and thus coordination would be essential. Thus,
a fuzzy boundary of law firm, meaning lawyers’ high employment mobility, might reduce its
ability to effectively engage with a broad niche. By contrast, boutique law firms that only operate
in specialized areas might not be as affected by organizational boundaries because lawyers are
likely to work independently.
Another example is the much-studied case of films, as an example of cultural products.
Films labeled with a broad niche in genres receive lower ratings from audiences because of the
difficulty to make sense of (Hsu et al., 2009). For film production companies, category straddling
is also risky and might cause organizational mortality (Cattani et al., 2008). But at the same time,
it could also be a form of an innovation, and hybrid firms might have chances for exceptional
151

succuss (Hsu et al., 2012). This is because artistic success requires a balance of conventionality
and novelty (Askin & Mauskapf, 2017; Godart & Galunic, 2019). For the same reason, artistic
production teams need to have sufficient familiarity to allow for experimentation, and novelty to
avoid information redundancy (Uzzi & Spiro, 2005). Thus, when a film is more conventional by
occupying a narrow niche in genres, probably a more innovative team with a fuzzier boundary
could add novel spirit to the production; whereas a film with a broad niche might be better
reviewed if some convention is maintained by a cohesive production team (Hsu et al., 2012; Uzzi
& Spiro, 2005).
Cross-level Network Dependency
The third theoretical contribution lies in the emphasis on cross-level dependency, which
shows that a single-level approach is likely to overestimate the autonomy of network evolution.
Study 1 focuses on the dependency between organizations’ external niche occupation and
internal network properties. Such multilevel dependency follows the logic that different patterns
of external resource access need to be complemented by appropriate internal processing
capability to achieve the best outcome (Funk, 2013). When organizations can access abundant
resources from external networks, they are better off by having cohesive internal networks for
higher absorptive capacity (Tsai, 2001). When organizations face the challenge of depleted
resource access from external channels, less cohesive internal networks can allow diverse
resource access from interpersonal channels (Funk, 2013). The same logic can be applied to
niche occupations. As shown in the findings of study 1, the presence of fewer brokers within a
group indicates a higher likelihood of group cohesion, and such groups are more likely to have
internal cohesion and align members’ motivations and mindsets. A cohesive group can better
engage with a focused niche compared to a fragmented group. Such a finding extends existing
152

literature on niche width by simultaneously considering both intrinsic appeal (Negro et al., 2010)
and engagement (Hsu et al., 2009). Existing studies either focus on intrinsic appeal or
engagement, but the breadth-depth tradeoff is the result of both a lack of skill specialization and
an ambiguous identity. Thus, it is necessary to probe the dependency between intrinsic appeal
and engagement in evaluating the consequence of niche width.
Also, by considering brokerage from a collective perspective, study 1 contributes to the
dependency between local and global network positions. Although there’s abundant research on
the private goods perspective of network brokerage, how brokers can influence others has
received limited support (Burt, 2007; Clement et al., 2018). Existing research on second-order
brokerage has showed that relational advantage primarily benefits the broker and it is difficult to
spill over to neighbors (Burt, 2007), or such positive externalities are only possible when brokers
are leaders (Galunic et al., 2012). Study 1 shows that a low mean and a moderate standard
deviation of network brokerage within a group benefits its survival. In other words, the existence
of some brokers with relatively low level of brokering benefits a group. An example might be a
group with a central circle of committed members who are willing to interact with new and
peripheral members. There’s a moderate dispersion of network brokerage, but since the brokers
are not individualistic but caring for the group cohesion, the average brokerage can be kept at a
low level.
Study 2 considers cross-level dependencies at all three levels. One important contribution
of this study is the competitive niche overlap network at the macro-level and its dependency with
lower-level networks. Rarely can a researcher find a multilevel network dataset including inter-
company networks, their affiliated employees, and the networks among them. The virtual
organizing context provides parallels that researchers cannot afford to ignore. The findings show
153

that macro-level networks provide opportunities and constraints for micro-level and meso-level
networks to form. In particular, niche overlap relations make individuals’ boundary spanning
behaviors easier. Precisely because of this reason, organizational boundaries are eroded, and
collective identities are likely to be weak. Since organizations with fuzzy boundaries are unlikely
to collectively create resources critical to their survival, they are less attractive to popular
individuals who already have abundant relational resources.
Study 3 focuses on the dependency between micro-level interpersonal networks and
meso-level affiliation networks. Different from the way dependencies are examined in above
studies, this study views dependencies through the perspective of agency. If social categories and
social networks are viewed as two mechanisms for accessing resources, and both focus on
diversification and concentration as distinctive resource access patterns, individual agency
should play an important role. This study captures such agency through multirole network
personality extracted from micro-level networks and examines how it affects the formation of
meso-level networks. The two forms of resource access patterns are found to be consistent for a
focal individual. Bringing attention to the role of agency does not mean deviating away from the
fundamental assumption of the ecology framework. The question here is whether affiliation
behaviors could be explained by a consistent personal preference. Whether such consistency
results in better survival or performance in an ecological environment is a different question.
Practical Implication
Organizational Niche
The findings from these studies have practical implications for virtual organizing, and
they are also generalizable to wider contexts of human organizing. The results show that groups
with large niche widths are more likely to disband. For virtual group organizing, when niche
154

categories are clearly defined and well mapped onto community members’ perception, and when
niche categories involve distinctive group governance styles, “jack of all trades” is difficult and
harmful. Recent studies also show that it is important to differentiate between team diversity and
members’ diverse interests (Lerner & Lomi, 2018b). Team diversity, that is the extent to which a
team’s members edit different articles, benefits team performance in producing high-quality
articles. However, team performance is worse for a team consisting of members who are “jack of
all trades” to the extent that they edit atypical combinations of articles that are rarely co-edited
by others (Lerner & Lomi, 2018b).
For general organizing, the effect of organizations’ niche width on survivability is also
contingent upon a series of other factors including institutional logic (Lo & Kennedy, 2014),
audience members (Pontikes, 2012), industry characteristics (Paolella & Durand, 2016), the
evolutionary stage of categories (Alexy & George, 2013), etc. These influential factors need to
be taken into account for evaluating the consequence of “jack of all trades” in different contexts.
How macro-level niche overlap relations affect micro- and meso-level relations has
implications for virtual and offline organizations alike. Managers of organizations need to find a
strategic niche position. A moderate amount of niche overlap brings legitimacy spillover from
the community, and facilitates some boundary spanning that may potentially bring novel
information. Too much niche overlap is likely to erode organizational boundaries to a
detrimental extent, because there will be many boundary spanning opportunities that require little
additional effort. Since organizational boundaries are essential for survivability, failing
organizations are unlikely to attract popular members, which makes it even more difficult for
these organizations to retain members.
Organizational boundary
155

The negative relation between group contrast and mortality hazards shows that virtual
group boundaries matter. Although digital relations are transient and organizational boundaries
are more blurred than in offline organizations (Shen et al., 2014; Sun et al., 2021), the efficiency
and longevity of virtual groups still depend on their collective identity. However, for virtual
groups that choose the risky route of a broad niche, fuzzy boundaries might reduce their chance
of mortality.
The importance of organizational boundary is also reflected in the finding that fellow
members of the same virtual groups are more likely to collaborate. At least for those above-
medium sized groups, their boundary can powerfully structure interpersonal relations. This is
more evidence that virtual groups can serve as parallels of offline organizing and help answer
theoretical questions that are difficult to measure in offline contexts. Offline data are usually
only available as a group of organizations, or one organization and its members, and this is why
multilevel organizational studies have been limited. Results from these studies show the potential
of virtual organizing as empirical contexts for multilevel theorization.
Whom to Recruit?
Study 3 finds that multirole network personality positively associated with performance.
For possible extension of this finding, there are two major considerations. First, as argued in
studies on the motivations for contributing to peer production, intrinsic motivations of
enjoyment, community, personal fulfillment are fundamental drivers (Von Krogh et al., 2012).
Similarly, intrinsic motivations also explain why players would like to turn the voluntariness of
play into commitment by joining clans and even take on responsibilities in running clans
(Snodgrass et al., 2017). Due to such intrinsic motivations, individuals in virtual organizations
are more likely to self-select into certain roles out of the congruence with personal preferences.
156

By comparison, functional roles in the offline professional organizations are more infused by
exogenous motivations such as career and financial factors. This probably explains why Burt
(2012) didn’t find any changes to the results even after accounting for the inconsistency between
network personality and role-specific network behaviors. The effect of inconsistency might play
a bigger role in offline network behaviors and performance. Second, this study examined how
multirole network personality affected performance and boundary spanning. In studies of offline
organizations, individual performance is usually measured as evaluation by superior managers,
which might involve multidimensional qualities (Rodan & Galunic, 2004). The objective
measure of win rate in the current context is unidimensional and restrictive in functional
performance. It is likely a “brokerage-prone” individual might be individualistic in improving
their own skills without contributing to the collective. Moreover, with low entry and exit cost,
the risk and obstacles associated with boundary spanning are also comparatively lower in the
current context. For boundary spanning in offline organizations, however, knowledge transfer
and integration might be more important than simply building social rapport (Flemming &
Waguespack, 2007),
Given the consistency of diversification or concentration in building social networks and
joining groups, organizational managers may predict members’ potential group engagement
based on their pattern of building interpersonal relations. Referring to the rich body of literature
on organizational turnover, voluntary turnover means depletion of human capital, because it
takes time for replacement to reach the original level and turnover is also likely to make
incumbent employees perform less well (Shaw, 2011). However, low rate of turnover increases
organizational rigidness and workforce can become close-minded, which is detrimental to
organizational innovation (Shaw, 2011).
157

The optimal level of turnover may be different for different organizations, and
organizations may want to recruit different types of members. Brokerage-prone members can
bring new ideas but they are unlikely to commit in the long term; closure-prone members are
willing to invest in maintaining strong ties and building trust, but over time relational
embeddedness within groups may lead to homogeneity and rigidness. Furthermore, since intra-
and cross-organizational social relations are likely to affect organizational turnover (Moynihan &
Pandey, 2008), how brokerage-prone members and closure-prone members interact may be also
considered to induce an ideal level of turnover.
Limitations and Future Research
Advanced Semantic Analysis
Study 1 adopts structural topic modeling to analyze the semantic construction of
categories. Askin and Mauskapf (2017) discussed the limitation of using categories and proposed
to use features instead. They argued that existing research on how categories affect audience
evaluation only provides a relatively coarse and static picture, which runs counter the view that
meanings and boundaries of categories are constantly contested and under negotiation. By
comparison, features provide more fine-grained information about those categories. Hannan et al.
(2019) identified semantic features from rap songs lyrics and positioned rap genres in a high
dimensional feature space to understand meanings and relations of genre categories (Hannan et
al., 2019). Apart from the feature space approach, future studies could also use word vectors to
replace the “bag of words” method in topic modeling. It can help increase the accuracy of
clustering by accounting for word meanings (Bojanowski et al., 2017).
Additional Within-level Networks
158

Following the multilevel network framework, the current empirical context has two
meso-level affiliation networks and one within-level network. Future studies could explore
empirical contexts with additional within-level networks. In addition to the law firm and film
examples mentioned above, for the case of nanotechnology companies (Lo & Kennedy, 2014),
research scientists may form advice and collaboration networks, companies may form
collaborative alliance networks, and patent categories have citation networks. For the case of
popular songs (Askin & Mauskapf, 2017), genres are embedded in a high-dimensional feature
space, and songs are recorded in different cities as musicians recorded their music (Phillips,
2011). These networks can greatly enrich the theorization on multilevel interdependencies.
Another important reason for examining a variety type of within-level networks is that tie
direction can be modeled. In the current case, interpersonal collaborative ties are not directional,
which only allows for testing degree-based hierarchy across levels, i.e., cross-level assortativity.
However, direction-based hierarchy is also essential for multilevel theorization. For example, the
cross-level alignment exchange mechanism indicates that the lower-level tie between individuals
is in the opposite direction as the higher-level tie between their respective organizations. This
mechanism means a change of hierarchy across levels. If this parameter is found to be non-
significant, then it means hierarchy in the form of tie directions is maintained across levels
(Zappa & Robins, 2016). Thus, incorporating directed networks would open more theorization
opportunities for multilevel dependency. Futures studies can leverage different affordances of
empirical contexts to further explore the multilevel network perspective.
Alternative Factors from Micro-perspectives
These studies focus on ecological factors that drive group mortality and network
evolution, but there could be many alternative explanations, possibly from micro-perspectives.
159

For instance, since these virtual groups follow similar hierarchical decision-making structures,
leadership plays an essential role in the functioning of virtual groups. In my participant
observation, I heard stories of clans disbanding due to horrible clan commanders. Future research
may explore how leadership styles, i.e., transformational, transactional, and laissez-faire, affect
virtual group performance and viability (Khajeh, 2018). Leaders in battles, or callers, also play
important roles, not only in team-based battles but also in group management in general. Callers
need sufficient practice and experience to establish reputation and authority. Novice callers
usually face resistance to leadership, which might lead to ineffective coordination.
Other studies of virtual communities have shown that effective boundary and hierarchy is
essential for governance. As group size grows, classical management reasoning argues that
central leadership is usually required for effective governance of large groups (Von Hippel &
von Krogh, 2003). Different from offline organizations, formal leadership structure is not the
default for online communities. Governance structures in virtual communities usually emerge
with features of both bureaucratic authority and democratic mode (O'Mahony and Ferraro, 2007).
Dahlannder and O’Mahony (2011) defined lateral authority as “task-based authority”, where
individuals “gain responsibility and decision rights over a greater proportion of collective work
without supervising others”. The leadership progression stage includes two lateral roles: project
members and board directors. Technical contributions affected the progression to project
members but didn’t affect the progression to board directors. Both technical and coordinative
contribution explain progression to leadership (O'Mahony and Ferraro, 2007). Future studies can
extend this line of research on leadership emergence and evolution in online communities.
Moreover, study 1 shows that the presence of many brokers in a group is harmful for the
survival of that group, but it is also important to differentiate brokers’ motivations rather than
160

viewing them all as individualistic and exploitative. Research has shown that if brokers try to
control the transfer of information and exploit the information asymmetry to maximize private
benefits, they could transfer (Spiro et al., 2013) and arbitrage (Soda et al., 2018). When brokers
seek to openly share and actively integrate information with neighbors and mediate between
them, they could coordinate (Spiro et al., 2013) and collaborate (Soda et al., 2018). For example,
in the current context, callers are more likely to be brokers within virtual groups, because they
need to collaborate with many different members. They are essentially contributing to group
cohesion rather than seeking individualistic purposes. Future research could take different
brokering motivations into account when assessing the collective effects of network brokerage.
As shown in study 3, brokerage-prone individuals are more likely to be boundary
spanners joining multiple groups. That is, individuals might span group boundaries because they
have a personal preference to meet new people rather than being motivated by the niche overlap
of groups. Since self-monitors are good at constructing public selves to different audiences, they
may have a propensity for being affiliated to multiple groups. Such psychological factors are
worthy of future research attention.
Alternative Operationalization and Modeling
In study 1, when measuring group contrast and within-group networks at each month, I
included every member affiliated to a clan for each month. However, only 9 percent players were
persistently active in the game for 32 months. The majority of players had some time gaps when
they didn’t play any games and didn’t participate in clan activities. The current study only
examines their activities when they were clan members. If a player is inactive for a considerable
amount of time, according to common clan rules, this player will be removed from the clan.
Then when this player comes back to the game, he or she has to join another clan. If this player
161

has a reason for being away and explains it to the clan commander, the commander may be
willing to keep the members. But even though this player can maintain their membership, a
period of inactivity still creates difficulty for resuming social relations because social ties need
maintenance efforts or they decay swiftly. Such periodic inactivity is likely to affect the
formation of social networks within clans and players’ clan membership mobility. In game
contexts, unpredictable periodic inactivity is likely to be confounded with predictable seasonal
inactivity, because some players would leave and come back when game features update. It is
necessary to distinguish these two cases. Future research could examine how such periodic
inactivity affect individual development and organizational effectiveness. This can serve as a
parallel to real life employment cases like gap time between jobs, and resuming one’s job after
sick or maternity leaves.
In study 2, niche overlap networks among groups are constructed through textual
similarity of group profiles. This may not be the most appropriate way to operationalize niche
overlap. It can only capture the extent to which two groups publicly express their mindsets of
playing in similar ways. Alternative operationalization in other dimensions of niche overlap is
possible. Previous studies on niche overlap took advantage of fine-grained market segmentation
to approach niche overlap, e.g., automobile engine horsepower (Dobrev & Kim, 2006). Another
example is audience overlap of media (Reis et al., 2013). Future studies could explore innovative
ways to capture niche overlap relations or similarity relations in general.
Two concepts from the most recent theorization of hierarchical categorization can be
useful (Hannan et al., 2019). Consider a two-level hierarchy of concepts, including a root level of
concepts and cohorts of first-level subconcepts. A root concept and its direct subconceps are
located in a domain. Hannan et al. (2019) extended category contrast to informativeness,
162

referring to the amount of “information a concept adds to what is already presumed or known by
virtue of its domain” (p. 80). The construct of informativeness specifically considers the different
amount of information that concepts at sub-levels add to the inherited information from a root
concept. Another concept is distinctiveness, meaning “how much the information in a concept
differs from that conveyed by the other concepts in its cohort of concepts” (p. 80). These ideas
are useful to operationalize relations among categories at different levels, especially how
category meanings are embedded in a broad semantic system.
Study 3 follows the construct of multirole network personality. Although it’s not a
substitute for psychological personality of any particular dimension, it captures personal
preference that are exhibited in network building and can be attributed to personality. Future
studies could use surveys to measure different dimensions of personality and examine how they
affect group outcomes. Recent studies have explored how personality composition affects group
effectiveness (Lykourentzou et al., 2016). Future studies could also take into account the fluid
boundaries of groups when examine how personality affects individual-group interaction and
group outcomes.
To conclude, this dissertation examines the multilevel organizational networks of an
online gaming community by following the theory of niche (Durand et al., 2017; Hannan et al.,
2007) and the theory of multilevel networks (Lazega, 2016; Padgett & Powell, 2012; Simon,
1962). With online gaming becomes an increasingly popular form of digital media, how people
establish, maintain and dissolve social relations has attracted abundant scholarly attention (Shen
et al., 2014). At the same time, virtual group dynamics can be leveraged to serve as “petri dish”
for understanding general human organizing. When the social architecture of a community
enables multiple forms of social interactions, i.e., interpersonal and individual-group
163

interactions, it becomes important to consider the multilevel dependencies that govern the
coevolution of different forms of social interactions (Ren et al., 2007; Paruchuri et al., 2019).
Interpersonal collaboration cannot be appropriately examined without referring to individual-
group interactions. Likewise, how within-group dynamics affect group effectiveness should not
be viewed in isolation of the extent to which group boundaries are strengthened or eroded by
membership mobility and interpersonal networks. This multilevel perspective can enrich the
theorization of organizing by focusing on intertwined coevolution of different social systems.

164

References
Agarwal, R., Sarkar, M., & Echambadi, R. (2002). The conditioning effect of time on firm
survival: An industry life cycle approach. Academy of Management Journal, 45(5), 971–
994. https://doi.org/10.5465/3069325
Ahmad, M. A., Borbora, Z., Shen, C., Srivastava, J., & Williams, D. (2011). Guild play in
MMOGs: Rethinking common group dynamics models. In A. Datta, S. Shulman, B.
Zheng, S.-D. Lin, A. Sun, & E.-P. Lim (Eds.), Social Informatics (pp. 145–152). Springer
Berlin Heidelberg.
Aldrich, H. E., & Ruef, M. (2006). Organizations evolving (2nd edition). SAGE Publications
Ltd.
Alexy, O., & George, G. (2013). Category divergence, straddling, and currency: Open innovation
and the legitimation of illegitimate categories. Journal of Management Studies, 50(2),
173–203. https://doi.org/10.1111/joms.12000
Alvari, H., Lakkaraju, K., Sukthankar, G., & Whetzel, J. (2014). Predicting guild membership in
massively multiplayer online games. In W. G. Kennedy, N. Agarwal, & S. J. Yang (Eds.),
Social Computing, Behavioral-Cultural Modeling and Prediction (pp. 215–222).
Springer International Publishing.
Amati, V., Lomi, A., & Mascia, D. (2019). Some days are better than others: Examining time-
specific variation in the structuring of interorganizational relations. Social Networks, 57,
18–33. https://doi.org/10.1016/j.socnet.2018.10.001

165

Antikainen, M., Aaltonen, T., & Väisänen, J. (2007). The role of trust in OSS communities—
Case Linux Kernel community. In J. Feller, B. Fitzgerald, W. Scacchi, & A. Sillitti
(Eds.), Open Source Development, Adoption and Innovation (pp. 223–228). Springer US.
https://doi.org/10.1007/978-0-387-72486-7_19
Arazy, O., Morgan, W., & Patterson, R. (2006). Wisdom of the crowds: Decentralized
knowledge construction in Wikipedia. In 16th Annual Workshop on Information
Technologies & Systems (WITS) Paper.
Argote, L., & Ingram, P. (2000). Knowledge transfer: A basis for competitive advantage in
firms. Organizational Behavior and Human Decision Processes, 82(1), 150–169.
https://doi.org/10.1006/obhd.2000.2893
Arguello, J., Butler, B. S., Joyce, E., Kraut, R., Ling, K. S., Rosé, C., & Wang, X. (2006). Talk to
me: Foundations for successful individual-group interactions in online communities.
Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 959–
968.
Askin, N., & Mauskapf, M. (2017). What makes popular culture popular? Product features and
optimal differentiation in music. American Sociological Review, 82(5), 910–944.
https://doi.org/10.1177/0003122417728662
Astley, W. G. (1985). The two ecologies: Population and community perspectives on
organizational evolution. Administrative Science Quarterly, 224–241.
Audia, P. G., Freeman, J. H., & Reynolds, P. D. (2006). Organizational foundings in community
context: Instruments manufacturers and their interrelationship with other organizations.
Administrative Science Quarterly, 51(3), 381–419. https://doi.org/10.2189/asqu.51.3.381
166

Bakker, R. M., & Josefy, M. (2018). More than just a number? The conceptualization and
measurement of firm age in an era of temporary organizations. Academy of Management
Annals, 12(2), 510–536. https://doi.org/10.5465/annals.2017.0035
Balkundi, P., & Harrison, D. A. (2006). Ties, leaders, and time in teams: Strong inference about
network structure’s effects on team viability and performance. Academy of Management
Journal, 49(1), 49–68.
Barnes, M., Kalberg, K., Pan, M., & Leung, P. (2016). When is brokerage negatively associated
with economic benefits? Ethnic diversity, competition, and common-pool resources.
Social Networks, 45, 55–65. https://doi.org/10.1016/j.socnet.2015.11.004
Barron, D. N., West, E., & Hannan, M. T. (1994). A time to grow and a time to die: Growth and
mortality of credit unions in New York City, 1914-1990. American Journal of Sociology,
100(2), 381–421. https://doi.org/10.1086/230541
Baum, J. A., & Amburgey, T. L. (2002). Organizational ecology. In J. A. Baum (Ed.), The
Blackwell companion to organizations. Oxford: Blackwell Publishers.
Baum, J. A. C., & Singh, J. V. (1994). Organizational niches and the dynamics of organizational
founding. Organization Science, 5(4), 483–501. https://doi.org/10.1287/orsc.5.4.483
Baum, J. A., & Singh, J. V. (1994). Evolutionary dynamics of organizations. Oxford University
Press.
Beck, N., Swaminathan, A., Wade, J. B., & Wezel, F. C. (2019). Industry clusters and
organizational prototypes: Evidence from the Franconian brewing industry. Journal of
Management, 45(7), 2978–3008.

167

Benefield, G. A., Shen, C., & Leavitt, A. (2016). Virtual team networks: How group social
capital affects team success in a massively multiplayer online game. Proceedings of the
19th ACM Conference on Computer-Supported Cooperative Work & Social Computing,
679–690.
Bizzi, L. (2013). The dark side of structural holes: A multilevel investigation. Journal of
Management, 39(6), 1554–1578. https://doi.org/10.1177/0149206312471385
Boda, Z. (2019). Friendship Bias in Ethnic Categorization. European Sociological Review, 35(4),
567–581. https://doi.org/10.1093/esr/jcz019
Bogaert, S., Boone, C., & Carroll, G. R. (2010). Organizational form emergence and competing
professional schemata of Dutch accounting, 1884–1939. In Categories in Markets:
Origins and Evolution (pp. 115–150). Emerald Group Publishing Limited.
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.
Bojanowski, P., Grave, E., Joulin, A., & Mikolov, T. (2017). Enriching word vectors with
subword information. Transactions of the Association for Computational Linguistics, 5,
135–146. https://doi.org/10.1162/tacl_a_00051
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.
https://doi.org/10.1016/j.poetic.2011.12.003
Borgatti, S. P. (2005). Centrality and network flow. Social Networks, 27(1), 55–71.
https://doi.org/10.1016/j.socnet.2004.11.008
168

Borgatti, S. P., & Cross, R. (2003). A relational view of information seeking and learning in
social networks. Management Science, 49(4), 432–445.
Bower, J. L., & Christensen, C. M. (1995). Disruptive technologies: Catching the wave. Harvard
Business Review, 73(1), 43–53.
Brailly, J., Favre, G., Chatellet, J., & Lazega, E. (2016). Embeddedness as a multilevel problem:
A case study in economic sociology. Social Networks, 44, 319–333.
https://doi.org/10.1016/j.socnet.2015.03.005
Brandes, U., Lerner, J., & Snijders, T. A. (2009, July). Networks evolving step by step: Statistical
analysis of dyadic event data. In 2009 international conference on advances in Social
network analysis and mining (pp. 200-205). IEEE.
Brass, D. J. (1981). Structural relationships, job characteristics, and worker satisfaction and
performance. Administrative Science Quarterly, 26(3), 331–348.
https://doi.org/10.2307/2392511
Breiger, R. L. (1974). The duality of persons and groups. Social Forces, 53(2), 181–190.
Brennecke, J., & Rank, O. N. (2016a). Knowledge networks in high-tech clusters: A multilevel
perspective on interpersonal and inter-organizational collaboration. In E. Lazega & T. A.
B. Snijders (Eds.), Multilevel Network Analysis for the Social Sciences: Theory, Methods
and Applications (pp. 273–293). Springer International Publishing.
https://doi.org/10.1007/978-3-319-24520-1_11
Brennecke, J., & Rank, O. N. (2016b). The interplay between formal project memberships and
informal advice seeking in knowledge-intensive firms: A multilevel network approach.
Social Networks, 44, 307–318. https://doi.org/10.1016/j.socnet.2015.02.004
169

Bruggeman, J., Grunow, D., Leenders, M. A. A. M., Vermeulen, I., & Kuilman, J. G. (2012).
Market positioning: The shifting effects of niche overlap. Industrial and Corporate
Change, 21(6), 1451–1477. https://doi.org/10.1093/icc/dts009
Bryant, J. A., & Monge, P. R. (2008). The evolution of the children’s television community,
1953-2003. International Journal of Communication, 2, 33.
Burt, R. S. (2004). Structural holes and good ideas. American Journal of Sociology, 110(2), 349–
399. https://doi.org/10.1086/421787
Burt, R. S. (2005). Brokerage and closure: An introduction to social capital. Oxford university
press.
Burt, R. S. (2012). Network-related personality and the agency question: Multirole Evidence
from a Virtual World. American Journal of Sociology, 118(3), 543–591.
https://doi.org/10.1086/667856
Burt, R. S., Kilduff, M., & Tasselli, S. (2013). Social network analysis: Foundations and frontiers
on advantage. Annual Review of Psychology, 64(1), 527–547.
https://doi.org/10.1146/annurev-psych-113011-143828
Burt, R. S., & Talmud, I. (1993). Market niche. Social Networks, 15(2), 133–149.
Buskens, V., & van de Rijt, A. (2008). Dynamics of networks if everyone strives for structural
holes. American Journal of Sociology, 114(2), 371–407. https://doi.org/10.1086/590674
Butler, B., Joyce, E., & Pike, J. (2008). Don’T look now, but we’ve created a bureaucracy: The
nature and roles of policies and rules in Wikipedia. Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems, 1101–1110.
https://doi.org/10.1145/1357054.1357227
170

Butler, B. S. (2001). Membership size, communication activity, and sustainability: A resource-
based model of online social structures. Information Systems Research, 12(4), 346–362.
Butts, C. T. (2008). A relational event framework for social action. Sociological Methodology,
38(1), 155–200. https://doi.org/10.1111/j.1467-9531.2008.00203.x
Caldwell, D. F., & O’Reilly, C. A. (1982). Boundary spanning and individual performance: The
impact of self-monitoring. Journal of Applied Psychology, 67(1), 124–127.
https://doi.org/10.1037/0021-9010.67.1.124
Campbell, D. T. (1965). Variation and selective retention in socio-cultural evolution. Social
Change in Developing Area. https://ci.nii.ac.jp/naid/10018132722/
Carr, C. T., Vitak, J., & McLaughlin, C. (2013). Strength of social cues in online impression
formation: Expanding SIDE research. Communication Research, 40(2), 261–281.
Carroll, G. R. (1985). Concentration and specialization: Dynamics of niche width in populations
of organizations. American Journal of Sociology, 90(6), 1262–1283.
https://doi.org/10.1086/228210
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.
In Categories in markets: Origins and evolution (pp. 203–233). Emerald Group
Publishing Limited.
Carroll, G. R., & Hannan, M. T. (2000). The demography of corporations and industries.
Princeton University Press.
Carroll, G. R., & Hannan, M. T. (2004). Why corporate demography matters: Policy implications
of organizational diversity. California Management Review, 42(3), 148–163.
171

Carroll, G. R., & Swaminathan, A. (2000). Why the microbrewery movement? Organizational
dynamics of resource partitioning in the U.S. brewing industry. American Journal of
Sociology, 106(3), 715–762. https://doi.org/10.1086/318962
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. https://doi.org/10.2189/asqu.53.1.145
Chen, C.-H., Sun, C.-T., & Hsieh, J. (2008). Player guild dynamics and evolution in massively
multiplayer online games. CyberPsychology & Behavior, 11(3), 293–301.
https://doi.org/10.1089/cpb.2007.0066
Chen, C.-W., & Lin, C.-S. (2014). Building a sense of virtual community: The role of the
features of social networking sites. Cyberpsychology, Behavior, and Social Networking,
17(7), 460–465. https://doi.org/10.1089/cyber.2013.0530
Chen, X., Zou, D., Cheng, G., & Xie, H. (2020). Detecting latent topics and trends in educational
technologies over four decades using structural topic modeling: A retrospective of all
volumes of Computers & Education. Computers & Education, 151, 103855.
https://doi.org/10.1016/j.compedu.2020.103855
Cheng, Z., & Guo, T. (2015). The formation of social identity and self-identity based on
knowledge contribution in virtual communities: An inductive route model. Computers in
Human Behavior, 43, 229–241. https://doi.org/10.1016/j.chb.2014.10.056
Chung, N., Nam, K., & Koo, C. (2016). Examining information sharing in social networking
communities: Applying theories of social capital and attachment. Telematics and
Informatics, 33(1), 77–91. https://doi.org/10.1016/j.tele.2015.05.005
172

Clement, J., Shipilov, A., & Galunic, C. (2018). Brokerage as a public good: The externalities of
network hubs for different formal roles in creative organizations. Administrative Science
Quarterly, 63(2), 251–286. https://doi.org/10.1177/0001839217708984
Cohen, A. P. (1985). The Symbolic Construction of Community. Routledge.
Contractor, N. (2009). The emergence of multidimensional networks. Journal of Computer-
Mediated Communication, 14(3), 743–747. https://doi.org/10.1111/j.1083-
6101.2009.01465.x
Corley, K. G., & Gioia, D. A. (2004). Identity ambiguity and change in the wake of a corporate
spin-off. Administrative Science Quarterly, 49(2), 173–208.
https://doi.org/10.2307/4131471
Cummings, J. N., & Cross, R. (2003). Structural properties of work groups and their
consequences for performance. Social Networks, 25(3), 197–210.
https://doi.org/10.1016/S0378-8733(02)00049-7
Curşeu, P. L., Janssen, S. E., & Raab, J. (2012). Connecting the dots: Social network structure,
conflict, and group cognitive complexity. Higher Education, 63(5), 621–629.
Dahlander, L., & O’Mahony, S. (2010). Progressing to the center: Coordinating project work.
Organization Science, 22(4), 961–979. https://doi.org/10.1287/orsc.1100.0571
DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and
collective rationality in organizational fields. American Sociological Review, 147–160.
Dobrev, S. D. (2007). Competing in the looking-glass market: Imitation, resources, and
crowding. Strategic Management Journal, 28(13), 1267–1289.
https://doi.org/10.1002/smj.651
173

Dobrev, S. D., Kim, T., & Hannan, M. T. (2001). Dynamics of niche width and resource
partitioning. American Journal of Sociology, 106(5), 1299–1337.
https://doi.org/10.1086/320821
Dobrev, S. D., & Kim, T.-Y. (2006). Positioning among organizations in a population: Moves
between market segments and the evolution of industry structure. Administrative Science
Quarterly, 51(2), 230–261. https://doi.org/10.2189/asqu.51.2.230
Dobrev, S. D., Ozdemir, S. Z., & Teo, A. C. (2006). The ecological interdependence of emergent
and established organizational populations: Legitimacy transfer, violation by comparison,
and unstable identities. Organization Science, 17(5), 577–597.
https://doi.org/10.1287/orsc.1060.0209
Ducheneaut, N. (2009). Collective solitude and social networks in World of Warcraft. In Social
networking communities and e-dating services: Concepts and implications (pp. 78–100).
IGI Global.
Ducheneaut, N., Yee, N., Nickell, E., & Moore, R. J. (2006). “Alone together?”: Exploring the
social dynamics of massively multiplayer online games. Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems, 407–416.
https://doi.org/10.1145/1124772.1124834
Durand, R., Granqvist, N., & Tyllström, A. (2017). From categories to categorization: A social
perspective on market categorization. In From categories to categorization: Studies in
sociology, organizations and strategy at the crossroads (Research in sociology of
organizations) (Vol. 51, pp. 3–30). Emerald Group Publishing.
174

Durand, R., & Khaire, M. (2017). Where do market categories come from and how?
Distinguishing category creation from category emergence. Journal of Management,
43(1), 87–110. https://doi.org/10.1177/0149206316669812
Durand, R., & Paolella, L. (2013). Category stretching: Reorienting research on categories in
strategy, entrepreneurship, and organization theory. Journal of Management Studies,
50(6), 1100–1123. https://doi.org/10.1111/j.1467-6486.2011.01039.x
Fang, R., Landis, B., Zhang, Z., Anderson, M. H., Shaw, J. D., & Kilduff, M. (2015). Integrating
personality and social networks: A meta-analysis of personality, network position, and
work outcomes in organizations. Organization Science, 26(4), 1243–1260.
https://doi.org/10.1287/orsc.2015.0972
Faraj, S., & Johnson, S. L. (2010). Network exchange patterns in online communities.
Organization Science, 22(6), 1464–1480. https://doi.org/10.1287/orsc.1100.0600
Filippova, A., & Cho, H. (2016). The effects and antecedents of conflict in free and open source
software development. Proceedings of the 19th ACM Conference on Computer-
Supported Cooperative Work & Social Computing, 705–716.
https://doi.org/10.1145/2818048.2820018
Fleming, L., & Waguespack, D. M. (2007). Brokerage, boundary spanning, and leadership in
open innovation communities. Organization Science, 18(2), 165–180.
https://doi.org/10.1287/orsc.1060.0242
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. JSTOR.
https://doi.org/10.2307/2094928
175

Freeman, J. H., & Audia, P. G. (2006). Community ecology and the sociology of organizations.
Annu. Rev. Sociol., 32, 145–169.
Freeman, J., & Hannan, M. T. (1984). Niche width and the dynamics of organizational
populations. American Journal of Sociology, 88(6), 1116–1145.
Funk, R. J. (2013). Making the most of where you are: Geography, networks, and innovation in
organizations. Academy of Management Journal, 57(1), 193–222.
https://doi.org/10.5465/amj.2012.0585
Galunic, C., Ertug, G., & Gargiulo, M. (2012). The positive externalities of social capital:
Benefiting from senior brokers. Academy of Management Journal, 55(5), 1213–1231.
https://doi.org/10.5465/amj.2010.0827
Gangestad, S. W., & Snyder, M. (2000). Self-monitoring: Appraisal and reappraisal.
Psychological Bulletin, 126(4), 530–555. https://doi.org/10.1037/0033-2909.126.4.530
Glynn, M. A., & Navis, C. (2013). Categories, identities, and cultural classification: Moving
beyond a model of categorical constraint. Journal of Management Studies, 50(6), 1124–
1137. https://doi.org/10.1111/joms.12023
Gnyawali, D. R., & Park, B.-J. R. (2011). Co-opetition between giants: Collaboration with
competitors for technological innovation. Research Policy, 40(5), 650–663.
Goh, S., & Wasko, M. (2012). The effects of leader-member exchange on member performance
in virtual world teams. Journal of the Association for Information Systems, 13(10).
https://doi.org/10.17705/1jais.00308
Grabowicz, P. A., Aiello, L. M., Eguiluz, V. M., & Jaimes, A. (2013). Distinguishing topical and
social groups based on common identity and bond theory. Proceedings of the Sixth ACM
International Conference on Web Search and Data Mining, 627–636.
176

Grodal, S. (2018). Field expansion and contraction: How communities shape social and symbolic
boundaries. Administrative Science Quarterly, 63(4), 783–818.
https://doi.org/10.1177/0001839217744555
Grodal, S., & Kahl, S. J. (2017). The discursive perspective of market categorization: Interaction,
power, and context. In From categories to categorization: Studies in sociology,
organizations and strategy at the crossroads (pp. 151–184). Emerald Publishing Limited.
Guegan, J., Moliner, P., & Buisine, S. (2015). Why are online games so self-involving: A social
identity analysis of massively multiplayer online role-playing games. European Journal
of Social Psychology, 45(3), 349–355. https://doi.org/10.1002/ejsp.2103
Hannan, M. T. (2010). Partiality of memberships in categories and audiences. Annual Review of
Sociology, 36(1), 159–181. https://doi.org/10.1146/annurev-soc-021610-092336
Hannan, M. T., & Carroll, G. R. (1992). Dynamics of organizational populations: Density,
legitimation, and competition. Oxford University Press.
Hannan, M. T., & Freeman, J. (1977). The population ecology of organizations. American
Journal of Sociology, 82(5), 929–964. https://doi.org/10.1086/226424
Hannan, M. T., & Freeman, J. (1984). Structural inertia and organizational change. American
Sociological Review, 149–164.
Hannan, M. T., & Freeman, J. (1988). The ecology of organizational mortality: American labor
unions, 1836-1985. American Journal of Sociology, 94(1), 25–52.
Hannan, M. T., Mens, G. L., Hsu, G., Kovács, B., Negro, G., Pólos, L., Pontikes, E., & Sharkey,
A. J. (2019). Concepts and categories: Foundations for sociological and cultural
analysis. Columbia University Press.
177

Hannan, M. T., Pólos, L., & Carroll, G. R. (2007). Logics of organization theory: Audiences,
codes, and ecologies. Princeton University Press.
Hansen, M. T. (1999). The search-transfer problem: The role of weak ties in sharing knowledge
across organization subunits. Administrative Science Quarterly, 44(1), 82–111.
Hargadon, A. (2003). How breakthroughs happen: The surprising truth about how companies
innovate. Harvard Business Press.
Hilbert, M., Oh, P., & Monge, P. (2016). Evolution of what? A network approach for the
detection of evolutionary forces. Social Networks, 47, 38–46.
Hippel, E. von, & Krogh, G. von. (2003). Open source software and the “private-collective”
innovation model: Issues for organization science. Organization Science, 14(2), 209–223.
https://doi.org/10.1287/orsc.14.2.209.14992
Hite, J. M., & Hesterly, W. S. (2001). The evolution of firm networks: From emergence to early
growth of the firm. Strategic Management Journal, 22(3), 275–286.
Hitt, M. A., Beamish, P. W., Jackson, S. E., & Mathieu, J. E. (2007). Building theoretical and
empirical bridges across levels: Multilevel research in management. Academy of
Management Journal, 50(6), 1385–1399.
Hollway, J., & Koskinen, J. (2016). Multilevel embeddedness: The case of the global fisheries
governance complex. Social Networks, 44, 281–294.
https://doi.org/10.1016/j.socnet.2015.03.001
Howard, P. N., & Jones, S. (2004). Society online: The Internet in context. SAGE.
Hsiao, C.-C., & Chiou, J.-S. (2012). The effect of social capital on community loyalty in a virtual
community: Test of a tripartite-process model. Decision Support Systems, 54(1), 750–
757. https://doi.org/10.1016/j.dss.2012.09.003
178

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.
https://doi.org/10.2189/asqu.51.3.420
Hsu, G., & Grodal, S. (2015). Category taken-for-grantedness as a strategic opportunity: The
case of light cigarettes, 1964 to 1993. American Sociological Review, 80(1), 28–62.
https://doi.org/10.1177/0003122414565391
Hsu, G., & Hannan, M. T. (2005). Identities, genres, and organizational forms. Organization
Science, 16(5), 474–490. https://doi.org/10.1287/orsc.1050.0151
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. https://doi.org/10.1177/000312240907400108
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. https://doi.org/10.1093/icc/dts011
Huang, Y., Zhu, M., Wang, J., Pathak, N., Shen, C., Keegan, B., Williams, D., & Contractor, N.
(2009). The formation of task-oriented groups: Exploring combat activities in online
games. 2009 International Conference on Computational Science and Engineering, 4,
122–127.
Hunter, S. D., Bentzen, H., & Taug, J. (2020). On the “missing link” between formal
organization and informal social structure. Journal of Organization Design, 9(1), 13.
https://doi.org/10.1186/s41469-020-00076-x
179

Ibarra, H., Kilduff, M., & Tsai, W. (2005). Zooming in and out: Connecting individuals and
collectivities at the frontiers of organizational network research. Organization Science,
16(4), 359–371.
Ingram, P., & Yue, Q. Y. (2008). Structure, affect and identity as bases of organizational
competition and cooperation. Academy of Management Annals, 2(1), 275–303.
https://doi.org/10.5465/19416520802211578
Jiang, H., & Carroll, J. M. (2009). Social capital, social network and identity bonds: A
reconceptualization. Proceedings of the Fourth International Conference on
Communities and Technologies, 51–60.
Jiang, L. C., Bazarova, N. N., & Hancock, J. T. (2013). From perception to behavior: Disclosure
reciprocity and the intensification of intimacy in computer-mediated communication.
Communication Research, 40(1), 125–143.
Johnson, N. F., Xu, C., Zhao, Z., Ducheneaut, N., Yee, N., Tita, G., & Hui, P. M. (2009). Human
group formation in online guilds and offline gangs driven by a common team dynamic.
Physical Review E, 79(6), 066117. https://doi.org/10.1103/PhysRevE.79.066117
Joshi, A., Pandey, N., & Han, G. (Helen). (2009). Bracketing team boundary spanning: An
examination of task-based, team-level, and contextual antecedents. Journal of
Organizational Behavior, 30(6), 731–759. https://doi.org/10.1002/job.567
Kavanaugh, A., Carroll, J. M., Rosson, M. B., Zin, T. T., & Reese, D. D. (2005). Community
networks: Where offline communities meet online. Journal of Computer-Mediated
Communication, 10(4). https://doi.org/10.1111/j.1083-6101.2005.tb00266.x
Kenis, P., & Knoke, D. (2002). How organizational field networks shape interorganizational tie-
formation rates. Academy of Management Review, 27(2), 275–293.
180

Kennedy, M. T., & Fiss, P. C. (2013). An ontological turn in categories research: From standards
of legitimacy to evidence of actuality. Journal of Management Studies, 50(6), 1138–
1154. https://doi.org/10.1111/joms.12031
Keuschnigg, M., & Wimmer, T. (2017). Is category spanning truly disadvantageous? New
evidence from primary and secondary movie markets. Social Forces, 96(1), 449–479.
https://doi.org/10.1093/sf/sox043
Khaire, M., & Wadhwani, R. D. (2010). Changing landscapes: The construction of meaning and
value in a new market category—Modern Indian art. Academy of Management Journal,
53(6), 1281–1304. https://doi.org/10.5465/amj.2010.57317861
Kilduff, M., & Tsai, W. (2003). Social Networks and Organizations. SAGE.
Kim, A. J. (2000). Community building on the Web: Secret strategies for successful online
communities (1st ed.). Addison-Wesley Longman Publishing Co., Inc.
Kim, T.-Y., Oh, H., & Swaminathan, A. (2006). Framing interorganizational network change: A
network inertia perspective. Academy of Management Review, 31(3), 704–720.
Kollock, P., & Smith, M. (1996). Managing the virtual commons: Cooperation and conflict in
computer communities. In S. C. Herring (Ed.), Computer-mediated Communication:
Linguistic, Social, and Cross-cultural Perspectives. John Benjamins Publishing.
Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits and attributes are predictable
from digital records of human behavior. Proceedings of the National Academy of
Sciences, 110(15), 5802–5805.
Kovács, B., & Hannan, M. (2010). The consequences of category spanning depend on contrast.
In G. Hsu, Ö. Koçak, & G. Negro (Eds.), Categories in markets: Origins and evolution
(Vol. 31, pp. 175–201). Emerald Group Publishing.
181

Kovács, B., & Hannan, M. (2015). Conceptual spaces and the consequences of category
spanning. Sociological Science, 2, 252–286. https://doi.org/10.15195/v2.a13
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.
https://doi.org/10.1177/1476127012460945
Kumar, P., & Zaheer, A. (2019). Ego-network stability and innovation in alliances. Academy of
Management Journal, 62(3), 691–716. https://doi.org/10.5465/amj.2016.0819
Kwon, K. H., Chadha, M., & Wang, F. (2019). Proximity and networked news public: Structural
topic modeling of global Twitter conversations about the 2017 Quebec Mosque Shooting.
International Journal of Communication, 13(0), 24.
Kwon, S.-W., Rondi, E., Levin, D. Z., De Massis, A., & Brass, D. J. (2020). Network brokerage:
An integrative review and future research agenda. Journal of Management, 46(6), 1092–
1120. https://doi.org/10.1177/0149206320914694
Lander, M. W., & Heugens, P. P. (2017). Better together: Using meta-analysis to explore
complementarities between ecological and institutional theories of organization.
Organization Studies, 38(11), 1573–1601.
Landis, B. (2016). Personality and social networks in organizations: A review and future
directions. Journal of Organizational Behavior, 37(S1), S107–S121.
https://doi.org/10.1002/job.2004
Lane, P. J., Koka, B. R., & Pathak, S. (2006). The reification of absorptive capacity: A critical
review and rejuvenation of the construct. Academy of Management Review, 31(4), 833–
863. https://doi.org/10.5465/amr.2006.22527456
182

Lazega, E. (2013). Network analysis and morphogenesis: A neo-structural exploration and
illustration. In Social morphogenesis (pp. 167–185). Springer.
Lazega, E. (2014). ‘Morphogenesis unbound’ from the dynamics of multilevel networks: A neo-
structural perspective. In Late Modernity (pp. 173–191). Springer.
Lazega, E. (2016). Synchronization costs in the organizational society: Intermediary relational
infrastructures in the dynamics of multilevel networks. In E. Lazega & T. A. B. Snijders
(Eds.), Multilevel Network Analysis for the Social Sciences: Theory, Methods and
Applications (pp. 47–77). Springer International Publishing. https://doi.org/10.1007/978-
3-319-24520-1_3
Lazega, E., & Snijders, T. A. B. (2016). Introduction. In E. Lazega & T. A. B. Snijders (Eds.),
Multilevel Network Analysis for the Social Sciences: Theory, Methods and Applications
(pp. 1–12). Springer International Publishing. https://doi.org/10.1007/978-3-319-24520-
1_1
Lee, S., & Monge, P. (2011). The coevolution of multiplex communication networks in
organizational communities. Journal of Communication, 61(4), 758–779.
Lee, S. Y. (2015). Homophily and social influence among online casual game players.
Telematics and Informatics, 32(4), 656–666. https://doi.org/10.1016/j.tele.2015.02.007
Lerner, J., & Lomi, A. (2018a). Knowledge categorization affects popularity and quality of
Wikipedia articles. PLOS ONE, 13(1), e0190674.
https://doi.org/10.1371/journal.pone.0190674
Lerner, J., & Lomi, A. (2020a). The free encyclopedia that anyone can dispute: An analysis of
the micro-structural dynamics of positive and negative relations in the production of
183

contentious Wikipedia articles. Social Networks, 60, 11–25.
https://doi.org/10.1016/j.socnet.2018.12.003
Lerner, J., & Lomi, A. (2020b). Reliability of relational event model estimates under sampling:
How to fit a relational event model to 360 million dyadic events. Network Science, 8(1),
97–135. https://doi.org/10.1017/nws.2019.57
Lerner, J., & Lomi, A. (2018b). Diverse teams tend to do good work in Wikipedia (but jacks of
all trades don’t). 2018 IEEE/ACM International Conference on Advances in Social
Networks Analysis and Mining (ASONAM), 214–221.
https://doi.org/10.1109/ASONAM.2018.8508597
Lerner, J., & Tirole, J. (2001). The open source movement: Key research questions. European
Economic Review, 45(4–6), 819–826.
Lessig, L. (2009). Code: And Other Laws of Cyberspace. ReadHowYouWant.com.
Levinthal, D. A. (1991). Random walks and organizational mortality. Administrative Science
Quarterly, 36(3), 397–420. https://doi.org/10.2307/2393202
Lizardo, O. (2014). Omnivorousness as the bridging of cultural holes: A measurement strategy.
Theory and Society, 43(3), 395–419. https://doi.org/10.1007/s11186-014-9220-9
Lo, J. Y., Fiss, P. C., Rhee, E. Y., & Kennedy, M. T. (2020). Category viability: Balanced levels
of coherence and distinctiveness. Academy of Management Review, 45(1), 85–108.
Lo, J. Y.-C., & Kennedy, M. T. (2014). Approval in nanotechnology patents: Micro and macro
factors that affect reactions to category blending. Organization Science, 26(1), 119–139.
https://doi.org/10.1287/orsc.2014.0933
184

Loewenstein, J., Ocasio, W., & Jones, C. (2012). Vocabularies and vocabulary structure: A new
approach linking categories, practices, and institutions. Academy of Management Annals,
6(1), 41–86.
Lomi, A., Lusher, D., Pattison, P. E., & Robins, G. (2014). The focused organization of advice
relations: A study in boundary crossing. Organization Science, 25(2), 438–457.
https://doi.org/10.1287/orsc.2013.0850
Lomi, A., Tasselli, S., & Zappa, P. (2017). The network structure of organizational vocabularies.
In Structure, Content and Meaning of Organizational Networks: Extending Network
Thinking (pp. 65–85). Emerald Publishing Limited.
Lykourentzou, I., Antoniou, A., Naudet, Y., & Dow, S. P. (2016). Personality matters: Balancing
for personality types leads to better outcomes for crowd teams. Proceedings of the 19th
ACM Conference on Computer-Supported Cooperative Work & Social Computing, 260–
273. https://doi.org/10.1145/2818048.2819979
Malinen, S. (2015). Understanding user participation in online communities: A systematic
literature review of empirical studies. Computers in Human Behavior, 46, 228–238.
https://doi.org/10.1016/j.chb.2015.01.004
Marrone, J. A., Tesluk, P. E., & Carson, J. B. (2007). A Multilevel investigation of antecedents
and consequences of team member boundary-spanning behavior. Academy of
Management Journal, 50(6), 1423–1439. https://doi.org/10.5465/amj.2007.28225967
Martončik, M., & Lokša, J. (2016). Do World of Warcraft (MMORPG) players experience less
loneliness and social anxiety in online world (virtual environment) than in real world
(offline)? Computers in Human Behavior, 56, 127–134.
185

McEvily, B., Soda, G., & Tortoriello, M. (2014). More formally: Rediscovering the missing link
between formal organization and informal social structure. Academy of Management
Annals, 8(1), 299–345. https://doi.org/10.5465/19416520.2014.885252
McPherson, M. (1983). An ecology of affiliation. American Sociological Review, 48(4), 519–
532. https://doi.org/10.2307/2117719
Mehra, A., Kilduff, M., & Brass, D. J. (2001). The social networks of high and low self-
monitors: Implications for workplace performance. Administrative Science Quarterly,
46(1), 121–146. https://doi.org/10.2307/2667127
Meng, J., Williams, D., & Shen, C. (2015). Channels matter: Multimodal connectedness, types of
co-players and social capital for Multiplayer Online Battle Arena gamers. Computers in
Human Behavior, 52, 190–199. https://doi.org/10.1016/j.chb.2015.06.007
Meyer, M. W., & Zucker, L. G. (1989). Permanently failing organizations. SAGE Publications,
Incorporated.
Meyrowitz, J. (1986). No sense of place: The impact of electronic media on social behavior.
Oxford University Press.
Moliterno, T. P., & Mahony, D. M. (2011). Network theory of organization: A multilevel
approach. Journal of Management, 37(2), 443–467.
https://doi.org/10.1177/0149206310371692
Monge, P., & Contractor, P. S. (2003). Theories of communication networks. Oxford University
Press, USA.
Monge, P., Heiss, B. M., & Margolin, D. B. (2008). Communication network evolution in
organizational communities. Communication Theory, 18(4), 449–477.
https://doi.org/10.1111/j.1468-2885.2008.00330.x
186

Mützel, S., & Breiger, R. (2020). Duality beyond persons and groups. The Oxford Handbook of
Social Networks, 392.
Nardi, B. (2010). My life as a night elf priest: An anthropological account of World of Warcraft.
University of Michigan Press.
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. https://doi.org/10.2189/asqu.2010.55.3.439
Negro, G., Hannan, M. T., & Fassiotto, M. (2015). Category signaling and reputation.
Organization Science, 26(2), 584–600. https://doi.org/10.1287/orsc.2014.0935
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. https://doi.org/10.1093/icc/dtq003
O’Connor, E. L., Longman, H., White, K. M., & Obst, P. L. (2015). Sense of community, social
Identity and social support among [layers of Massively Multiplayer Online Games
(MMOGs): A qualitative analysis. Journal of Community & Applied Social Psychology,
25(6), 459–473. https://doi.org/10.1002/casp.2224
Oh, H., Chung, M.-H., & Labianca, G. (2004). Group social capital and group effectiveness: The
role of informal socializing ties. Academy of Management Journal, 47(6), 860–875.
https://doi.org/10.5465/20159627
Oh, H., & Kilduff, M. (2008). The ripple effect of personality on social structure: Self-
monitoring origins of network brokerage. Journal of Applied Psychology, 93(5), 1155.
187

Olzak, S., & Johnson, E. W. (2019). The risk of occupying a broad niche for environmental
social movement organizations. Mobilization: An International Quarterly, 24(2), 177–
198.
O’Mahony, S., & Ferraro, F. (2007). The emergence of governance in an open source
community. Academy of Management Journal, 50(5), 1079–1106.
https://doi.org/10.5465/amj.2007.27169153
Ostrom, E. (1990). Governing the commons: The evolution of institutions for collective action.
Cambridge University Press.
Pachucki, M. A., & Breiger, R. L. (2010). Cultural holes: Beyond relationality in social networks
and culture. Annual Review of Sociology, 36(1), 205–224.
https://doi.org/10.1146/annurev.soc.012809.102615
Padgett, J. F. (2018). Faulkner’s assembly of memories into history: Narrative networks in
multiple times. American Journal of Sociology, 124(2), 406–478.
https://doi.org/10.1086/700831
Padgett, J. F., & Powell, W. W. (2012). The emergence of organizations and markets. Princeton
University Press.
Panzarasa, P., Opsahl, T., & Carley, K. M. (2009). Patterns and dynamics of users’ behavior and
interaction: Network analysis of an online community. Journal of the American Society
for Information Science and Technology, 60(5), 911–932.
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. https://doi.org/10.5465/amj.2013.0651
188

Paolella, L., & Sharkey, A. (2017). Forging consensus: An integrated view of how categories
shape the perception of organizational identity. In R. Durand, N. Granqvist, & A.
Tyllström (Eds.), From categories to categorization: Studies in sociology, organizations
and strategy at the crossroads (Research in sociology of organizations) (Vol. 51, pp.
327–353). Emerald Group Publishing.
Paruchuri, S., Goossen, M. C., & Phelps, C. (2019). Conceptual foundations of multilevel social
networks. In The handbook of multilevel theory, measurement, and analysis (pp. 201–
221). American Psychological Association. https://doi.org/10.1037/0000115-010
Perry-Smith, J. E. (2006). Social yet creative: The role of social relationships in facilitating
individual creativity. Academy of Management Journal, 49(1), 85–101.
https://doi.org/10.5465/amj.2006.20785503
Perry-Smith, J. E., & Mannucci, P. V. (2017). From creativity to innovation: The social network
drivers of the four phases of the idea journey. Academy of Management Review, 42(1),
53–79. https://doi.org/10.5465/amr.2014.0462
Phillips, D. J. (2011). Jazz and the disconnected: City structural disconnectedness and the
emergence of a Jazz canon, 1897–1933. American Journal of Sociology, 117(2), 420–
483. https://doi.org/10.1086/661757
Pólos, L., Hannan, M. T., & Carroll, G. R. (2002). Foundations of a theory of social forms.
Industrial and Corporate Change, 11(1), 85–115. https://doi.org/10.1093/icc/11.1.85
Pontikes, E. G. (2012). Two sides of the same coin: How ambiguous classification affects
multiple audiences’ evaluations. Administrative Science Quarterly, 57(1), 81–118.
https://doi.org/10.1177/0001839212446689
189

Pontikes, E. G., & Hannan, M. T. (2014). An ecology of social categories. Sociological Science,
1, 311–343. https://doi.org/10.15195/v1.a20
Pontikes, E. G., & kim, R. (2017). Strategic categorization. In R. Durand, N. Granqvist, & A.
Tyllström (Eds.), From Categories to Categorization: Studies in Sociology,
Organizations and Strategy at the Crossroads (Research in the Sociology of
Organizations) (Vol. 51, pp. 71–111). Emerald Group Publishing.
Postmes, T., Spears, R., Lee, A. T., & Novak, R. J. (2005). Individuality and social influence in
groups: Inductive and deductive routes to group identity. Journal of Personality and
Social Psychology, 89(5), 747.
Postmes, T. T., & Spears, R. (2000). Refining the cognitive redefinition of the group:
Deindividuation effects in common bond vs. common identity groups.
Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network dynamics and
field evolution: The growth of interorganizational collaboration in the life sciences.
American Journal of Sociology, 110(4), 1132–1205.
Preece, J. (2001). Sociability and usability in online communities: Determining and measuring
success. Behaviour & Information Technology, 20(5), 347–356.
https://doi.org/10.1080/01449290110084683
Preece, J., & Maloney-Krichmar, D. (2005). Online communities: Design, theory, and practice.
Journal of Computer-Mediated Communication, 10(4). https://doi.org/10.1111/j.1083-
6101.2005.tb00264.x
Ransbotham, S., & Kane, G. C. (Jerry). (2011). Membership turnover and collaboration success
in online communities: Explaining rises and falls from grace in Wikipedia. MIS
Quarterly, 35(3), 613–627. JSTOR. https://doi.org/10.2307/23042799
190

Rao, H., Monin, P., & Durand, R. (2005). Border crossing: Bricolage and the erosion of
categorical boundaries in French gastronomy. American Sociological Review, 70(6), 968–
991. https://doi.org/10.1177/000312240507000605
Reis, S., Negro, G., Sorenson, O., Perretti, F., & Lomi, A. (2013). Resource partitioning
revisited: Evidence from Italian television broadcasting. Industrial and Corporate
Change, 22(2), 459–487. https://doi.org/10.1093/icc/dts020
Ren, Y., Kraut, R., & Kiesler, S. (2007). Applying common identity and bond theory to design of
online communities. Organization Studies, 28(3), 377–408.
https://doi.org/10.1177/0170840607076007
Ridings, C. M., & Gefen, D. (2004). Virtual community attraction: Why people hang out online.
Journal of Computer-Mediated Communication, 10(1). https://doi.org/10.1111/j.1083-
6101.2004.tb00229.x
Ridings, C. M., Gefen, D., & Arinze, B. (2002). Some antecedents and effects of trust in virtual
communities. The Journal of Strategic Information Systems, 11(3), 271–295.
https://doi.org/10.1016/S0963-8687(02)00021-5
Roberts, M. E., Stewart, B. M., & Tingley, D. (2019). stm: An R package for structural topic
models. Journal of Statistical Software, 91(1), 1–40.
https://doi.org/10.18637/jss.v091.i02
Robins, G., Snijders, T., Wang, P., Handcock, M., & Pattison, P. (2007). Recent developments in
exponential random graph (p*) models for social networks. Social Networks, 29(2), 192–
215.
191

Rowley, T. J., Greve, H. R., Rao, H., Baum, J. A., & Shipilov, A. V. (2005). Time to break up:
Social and instrumental antecedents of firm exits from exchange cliques. Academy of
Management Journal, 48(3), 499–520.
Ruef, M., & Patterson, K. (2009). Credit and classification: The impact of industry boundaries in
nineteenth-century America. Administrative Science Quarterly, 54(3), 486–520.
https://doi.org/10.2189/asqu.2009.54.3.486
Sasidharan, S., Santhanam, R., Brass, D. J., & Sambamurthy, V. (2012). The effects of social
network structure on enterprise systems success: A longitudinal multilevel analysis.
Information Systems Research, 23(3-part-1), 658–678.
Schecter, A., & Quintane, E. (2020). The power, accuracy, and precision of the relational event
model. Organizational Research Methods, 1094428120963830.
https://doi.org/10.1177/1094428120963830
Schilling, M. A., & Fang, C. (2014). When hubs forget, lie, and play favorites: Interpersonal
network structure, information distortion, and organizational learning. Strategic
Management Journal, 35(7), 974–994. https://doi.org/10.1002/smj.2142
Schultheiss, D. (2012). Long-term motivations to play MMOGs: A longitudinal study on
motivations, experience and behavior. https://www.db-
thueringen.de/receive/dbt_mods_00021226
Schwartz, H. A., Eichstaedt, J. C., Kern, M. L., Dziurzynski, L., Ramones, S. M., Agrawal, M.,
Shah, A., Kosinski, M., Stillwell, D., & Seligman, M. E. (2013). Personality, gender, and
age in the language of social media: The open-vocabulary approach. PloS One, 8(9),
e73791.
192

Selfhout, M., Burk, W., Branje, S., Denissen, J., Aken, M. V., & Meeus, W. (2010). Emerging
late adolescent friendship networks and big five personality traits: A social network
approach. Journal of Personality, 78(2), 509–538. https://doi.org/10.1111/j.1467-
6494.2010.00625.x
Shaw, J. D. (2011). Turnover rates and organizational performance: Review, critique, and
research agenda. Organizational Psychology Review, 1(3), 187–213.
https://doi.org/10.1177/2041386610382152
Shen, C., & Chen, W. (2015). Social capital, coplaying patterns, and health disruptions: A survey
of Massively Multiplayer Online Game participants in China. Computers in Human
Behavior, 52, 243–249.
Shen, C., Monge, P., & Williams, D. (2014). The evolution of social ties online: A longitudinal
study in a massively multiplayer online game. Journal of the Association for Information
Science and Technology, 65(10), 2127–2137.
Shen, C., Sun, Q., Kim, T., Wolff, G., Ratan, R., & Williams, D. (2020). Viral vitriol: Predictors
and contagion of online toxicity in World of Tanks. Computers in Human Behavior, 108,
106343.
Shockley-Zalabak, P., Ellis, K., & Winograd, G. (2000). Organizational trust: What it means,
why it matters. Organization Development Journal, 18(4), 35–48.
Simon, H. A. (2019). The Sciences of the Artificial. MIT Press.
Simons, T., & Ingram, P. (2004). An ecology of ideology: Theory and evidence from four
populations. Industrial and Corporate Change, 13(1), 33–59.
193

Smith, E. B. (2011). Identities as lenses: How organizational identity affects audiences’
evaluation of organizational performance. Administrative Science Quarterly, 56(1), 61–
94. https://doi.org/10.2189/asqu.2011.56.1.061
Snijders, T. A., Pattison, P. E., Robins, G. L., & Handcock, M. S. (2006). New specifications for
exponential random graph models. Sociological Methodology, 36(1), 99–153.
Snodgrass, J. G., Batchelder, G., Eisenhauer, S., Howard, L., Dengah, H. F., Thompson, R. S.,
Bassarear, J., Cookson, R. J., Defouw, P. D., Matteliano, M., & Powell, C. (2017). A
guild culture of ‘casual raiding’ enhances its members’ online gaming experiences: A
cognitive anthropological and ethnographic approach to World of Warcraft. New Media
& Society, 19(12), 1927–1944. https://doi.org/10.1177/1461444816644804
Soda, G., & Zaheer, A. (2012). A network perspective on organizational architecture:
Performance effects of the interplay of formal and informal organization. Strategic
Management Journal, 33(6), 751–771. https://doi.org/10.1002/smj.1966
Somech, A., Desivilya, H. S., & Lidogoster, H. (2009). Team conflict management and team
effectiveness: The effects of task interdependence and team identification. Journal of
Organizational Behavior, 30(3), 359–378. https://doi.org/10.1002/job.537
Spiro, E. S., Acton, R. M., & Butts, C. T. (2013). Extended structures of mediation: Re-
examining brokerage in dynamic networks. Social Networks, 35(1), 130–143.
https://doi.org/10.1016/j.socnet.2013.02.001
Stam, W., Arzlanian, S., & Elfring, T. (2014). Social capital of entrepreneurs and small firm
performance: A meta-analysis of contextual and methodological moderators. Journal of
Business Venturing, 29(1), 152–173.
194

Steinkuehler, C. A., & Williams, D. (2006). Where everybody knows your (screen) name: Online
games as “third places.” Journal of Computer-Mediated Communication, 11(4), 885–909.
Stinchcombe, A. L. (1965). Social structure and organizations. Handbook of Organizations, 7,
142–193.
Sun, J., Williams, D., & Li, Y. (2021). The evolution of individual-group affiliation in an online
gaming community. Computers in Human Behavior, 106818.
https://doi.org/10.1016/j.chb.2021.106818
Tasselli, S., & Kilduff, M. (2018). When brokerage between friendship cliques endangers trust:
A personality–network fit perspective. Academy of Management Journal, 61(3), 802–
825. https://doi.org/10.5465/amj.2015.0856
Tasselli, S., & Kilduff, M. (2021). Network agency. Academy of Management Annals, 15(1), 68–
110. https://doi.org/10.5465/annals.2019.0037
Tasselli, S., Zappa, P., & Lomi, A. (2020). Bridging cultural holes in organizations: The dynamic
structure of social networks and organizational vocabularies within and across subunits.
Organization Science, 31(5), 1292–1312. https://doi.org/10.1287/orsc.2019.1352
Tortoriello, M., Reagans, R., & McEvily, B. (2012). Bridging the knowledge gap: The influence
of strong ties, network cohesion, and network range on the transfer of knowledge
between organizational units. Organization Science, 23(4), 1024–1039.
https://doi.org/10.1287/orsc.1110.0688
Tsai, W. (2001). Knowledge transfer in intraorganizational networks: Effects of network position
and absorptive capacity on business unit innovation and performance. Academy of
Management Journal, 44(5), 996–1004.
195

Uzzi, B., & Spiro, J. (2005). Collaboration and creativity: The small world problem. American
Journal of Sociology, 111(2), 447–504.
Vergne, J.-P., & Swain, G. (2017). Categorical anarchy in the UK? The British media’s
classification of bitcoin and the limits of categorization. In R. Durand, N. Granqvist, & A.
Tyllström (Eds.), From Categories to Categorization: Studies in Sociology,
Organizations and Strategy at the Crossroads (pp. 185–222). Emerald Publishing
Limited.
Vergne, J.-P., & Wry, T. (2014). Categorizing categorization research: Review, integration, and
future directions. Journal of Management Studies, 51(1), 56–94.
https://doi.org/10.1111/joms.12044
Vesa, M., Hamari, J., Harviainen, J. T., & Warmelink, H. (2017). Computer games and
organization studies. Organization Studies, 38(2), 273–284.
https://doi.org/10.1177/0170840616663242
Von Krogh, G., Haefliger, S., Spaeth, S., & Wallin, M. W. (2012). Carrots and rainbows:
Motivation and social practice in open source software development. MIS Quarterly,
649–676.
Vu, D., Lomi, A., Mascia, D., & Pallotti, F. (2017). Relational event models for longitudinal
network data with an application to interhospital patient transfers. Statistics in Medicine,
36(14), 2265–2287. https://doi.org/10.1002/sim.7247
Walther, J. B. (2007). Selective self-presentation in computer-mediated communication:
Hyperpersonal dimensions of technology, language, and cognition. Computers in Human
Behavior, 23(5), 2538–2557. https://doi.org/10.1016/j.chb.2006.05.002
196

Walther, J. B. (2011). Theories of computer-mediated communication and interpersonal
relations. The Handbook of Interpersonal Communication, 4, 443–479.
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). Springer International
Publishing. https://doi.org/10.1007/978-3-319-24520-1_6
Wang, P., Robins, G., Pattison, P. E., & Koskinen, J. H. (2014). MPNet: Program for the
simulation and estimation of (p*) exponential random graph models for multilevel
networks. Melbourne: Melbourne School of Psychological Sciences, The University of
Melbourne.
Wang, P., Robins, G., Pattison, P., & Lazega, E. (2013). Exponential random graph models for
multilevel networks. Social Networks, 35(1), 96–115.
Wang, Y., & Fesenmaier, D. R. (2003). Assessing motivation of contribution in online
communities: An empirical investigation of an online travel community. Electronic
Markets, 13(1), 33–45. https://doi.org/10.1080/1019678032000052934
Weber, M. S. (2012). Newspapers and the long-term implications of hyperlinking. Journal of
Computer-Mediated Communication, 17(2), 187–201. https://doi.org/10.1111/j.1083-
6101.2011.01563.x
White, H. C. (2008). Identity and Control: How Social Formations Emerge. Princeton University
Press.
White, H. C., Boorman, S. A., & Breiger, R. L. (1976). Social Structure from Multiple Networks.
I. Blockmodels of Roles and Positions. American Journal of Sociology, 81(4), 730–780.
https://doi.org/10.1086/226141
197

Williams, D. (2010). The mapping principle, and a research framework for virtual worlds.
Communication Theory, 20(4), 451–470.
Williams, D. (2018). For better or worse: Game structure and mechanics driving social
interactions and isolation. In C. J. Ferguson (Ed.), Video Game Influences on Aggression,
Cognition, and Attention (pp. 173–183). Springer International Publishing.
https://doi.org/10.1007/978-3-319-95495-0_14
Williams, D., Ducheneaut, N., Xiong, L., Zhang, Y., Yee, N., & Nickell, E. (2006). From tree
house to barracks: The social life of guilds in World of Warcraft. Games and Culture,
1(4), 338–361. https://doi.org/10.1177/1555412006292616
Wry, T., & Castor, A. R. (2017). Opportunity, status, and similarity: Exploring the varied
antecedents and outcomes of category spanning innovation. In R. Durand, N. Granqvist,
& A. Tyllström (Eds.), From categories to categorization: Studies in sociology,
organizations and strategy at the crossroads (Research in the sociology of organizations)
(Vol. 51, pp. 355–389). Emerald Group Publishing.
Wry, T., & Lounsbury, M. (2013). Contextualizing the categorical imperative: Category
linkages, technology focus, and resource acquisition in nanotechnology entrepreneurship.
Journal of Business Venturing, 28(1), 117–133.
https://doi.org/10.1016/j.jbusvent.2012.03.001
Wry, T., Lounsbury, M., & Glynn, M. A. (2011). Legitimating nascent collective identities:
Coordinating cultural entrepreneurship. Organization Science, 22(2), 449–463.
https://doi.org/10.1287/orsc.1100.0613
198

Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are
more accurate than those made by humans. Proceedings of the National Academy of
Sciences, 112(4), 1036–1040.
Zappa, P., & Lomi, A. (2016). Knowledge sharing in organizations: A multilevel network
analysis. In E. Lazega & T. A. B. Snijders (Eds.), Multilevel Network Analysis for the
Social Sciences: Theory, Methods and Applications (pp. 333–353). Springer International
Publishing. https://doi.org/10.1007/978-3-319-24520-1_14
Zappa, P., & Robins, G. (2016). Organizational learning across multi-level networks. Social
Networks, 44, 295–306. https://doi.org/10.1016/j.socnet.2015.03.003
Zhu, H., Kraut, R. E., & Kittur, A. (2013). The impact of membership overlap on the survival of
online communities. ICIS. https://doi.org/10.1145/2556288.2557213
Zuckerman, E. W. (1999). The categorical imperative: Securities analysts and the illegitimacy
discount. American Journal of Sociology, 104(5), 1398–1438.
https://doi.org/10.1086/210178
Zuckerman, E. W. (2017). The categorical imperative revisited: Implications of categorization as
a theoretical tool. In R. Durand, N. Granqvist, & A. Tyllström (Eds.), From Categories to
Categorization: Studies in Sociology, Organizations and Strategy at the Crossroads
(Research in the Sociology of Organizations) (Vol. 51, pp. 31–68). Emerald Group
Publishing.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The evolution of multidimensional and multilevel networks in online crowdsourcing

PDF

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

PDF

Ecology and network evolution in online innovation contest crowdsourcing

PDF

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

PDF

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

PDF

Crisis and stasis: understanding change in online communities

PDF

The evolution of scientific collaboration networks

PDF

Media reinvented: the transformation of news in a networked society

PDF

Benefits of social networking in online social support groups

PDF

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

PDF

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

PDF

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

PDF

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

PDF

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

PDF

Disentangling the network: understanding the interplay of topology and dynamics in network analysis

PDF

The good, the bad, and the longitudinal: testing dynamic prosocial and toxic behaviors in online commercial games

PDF

Contagious: social norms about health in work group networks

PDF

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

PDF

Self-presence: body, emotion, and identity extension into the virtual self

PDF

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

Asset Metadata

Creator Sun, Jingyi (author)

Core Title The evolution of multilevel organizational networks in an online gaming community

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Degree Conferral Date 2021-08

Publication Date 07/12/2021

Defense Date 06/18/2021

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

Tag category,individual-group interaction,massively multiplayer online game (MMOG),multilevel networks,network evolution,OAI-PMH Harvest,organizational networks

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Williams, Dmitri (committee chair), Lomi, Alessandro (committee member), Monge, Peter (committee member)

Creator Email jingyisu@usc.edu,sjingyi@hotmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC16255863

Unique identifier UC16255863

Legacy Identifier etd-SunJingyi-9722

Document Type Dissertation

Format application/pdf (imt)

Rights Sun, Jingyi

Type texts

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

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

Repository Name University of Southern California Digital Library

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

Repository Email cisadmin@lib.usc.edu