University of Southern California Dissertations and Theses

Crisis and stasis: understanding change in online communities

(USC Thesis Other)

Crisis and stasis: understanding change in online communities

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Running head: CRISIS AND STASIS I
CRISIS AND STASIS: UNDERSTANDING CHANGE IN ONLINE COMMUNITIES
by
Joshua Clark

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA

In Partial Fulfillment of the
Requirements for the Degree of
Doctor of Philosophy
In Communication

August 2016
CRISIS AND STASIS II
Acknowledgements
This dissertation is the product of contributions from a large number of people and it is
impossible to thank everyone who helped in its creation.
I am extremely thankful to my advisor Dmitri Williams who helped push me beyond my
comfort zone and into new and rewarding areas of study. I am also thankful to my wonderful
committee members. Peter Monge and Dennis Wixon for their constant support and advice
throughout the writing process.
My colleagues at and around the Annenberg School for Communication, especially Alex
Leavitt for being by sounding board, co-author and friend. Additionally, Tisha Dejamee and
Flemming Rhodes for their friendship throughout the process. Also my former friends and
classmates from Queen’s University: Joshua, Laura, Tory, Stephanie and Eva whose effort and
vast knowledge always inspire me to work just a little bit harder.
I also am grateful to my parents and brother for their love, support and help with this
process. I am also thankful for my new family, namely my fiancé Heather who has had to deal
with the moments of triumph and despair that come with producing a dissertation. Thank you for
sticking with me!
Last but not least a big thank you to my little dog, Macho, for warming my heart and my
lap throughout the writing process.

CRISIS AND STASIS III
Table of Contents
Abstract ............................................................................................................................. xi
CRISIS AND STASIS: UNDERSTANDING CHANGE IN ONLINE COMMUNITIES . 1
Chapter One- Introduction and Motivations ....................................................................... 1
Chapter Two- Theoretical Framework ................................................................................ 8
Network Evolution and Change ...................................................................................... 9
Variation Selection Retention. .................................................................................. 10
Scalable Change ........................................................................................................ 12
Variable Rates of Evolution. ..................................................................................... 12
Affordances of Online Communities ............................................................................ 14
The Appearance of Anonymity. ................................................................................ 15
Malleable Governance. ............................................................................................. 18
Joining and Leaving. ................................................................................................. 20
Punctuated Equilibrium ................................................................................................ 21
Variable Change. ....................................................................................................... 23
Chapter Three- Disruptive Behavior ................................................................................. 26
Magic Circles, frames and keys. ............................................................................... 27
Sources of Order ........................................................................................................... 30
Norms. ....................................................................................................................... 30
Rules. ........................................................................................................................ 31
CRISIS AND STASIS IV
Code. ......................................................................................................................... 32
Why Disrupt? ................................................................................................................ 33
Characterizing Disruptive Behavior. ......................................................................... 35
Disrupting a Network .................................................................................................... 35
Social Capital ................................................................................................................ 37
Disruptive Signatures .................................................................................................... 43
Private-Collaborative. ............................................................................................... 45
Private-Isolated ......................................................................................................... 47
Public-Isolated. ......................................................................................................... 50
Induced Disruption.................................................................................................... 53
Conceptual Summary .................................................................................................... 56
Chapter Four- Methods ..................................................................................................... 57
Data Sources ................................................................................................................. 57
Communities of Interest ................................................................................................ 57
Defining Disruption ...................................................................................................... 61
Common Metrics .......................................................................................................... 62
Local Brokerage and Closure.................................................................................... 62
Methodological Approach ............................................................................................. 64
Learning Disruption. ................................................................................................. 65
Disruptive Quirks. ..................................................................................................... 66
CRISIS AND STASIS V
Ground Truth ............................................................................................................. 70
Change Over Time ........................................................................................................ 74
Moving Forward ........................................................................................................... 76
Chapter Five- Private-Collaborative Disruptive Behavior ............................................... 77
Cheating, Hacking and Collaborative Disruption ......................................................... 78
The Steam Community ................................................................................................. 86
Method and Data ........................................................................................................... 88
Results ........................................................................................................................... 91
Rates of Disruptive Play over Time .......................................................................... 98
Discussion ................................................................................................................... 101
Chapter Six- Private-Isolated Disruptive Behavior ........................................................ 104
Private-Isolated Disruption ......................................................................................... 104
Team Fortress Two ...................................................................................................... 108
Method and Data .......................................................................................................... 111
Results ..........................................................................................................................114
Discussion ................................................................................................................... 129
Chapter Seven – Public-Isolated Disruptive Behavior ................................................... 132
DOTA. ......................................................................................................................... 139
Identifying Disruptive Players .................................................................................... 140
Method and Data ......................................................................................................... 148
CRISIS AND STASIS VI
Results ......................................................................................................................... 150
Discussion ................................................................................................................... 152
Chapter Eight - Induced Disruption ................................................................................ 156
Play and Protest........................................................................................................... 156
Network Flexibility. ................................................................................................ 159
Collective Action Problems. ................................................................................... 161
EVE and Micro-transactions ................................................................................... 165
Method and Data ......................................................................................................... 167
Demographics ......................................................................................................... 168
Results ......................................................................................................................... 176
Statistical Learning Model. ..................................................................................... 176
Rare Events Logistic Regression ............................................................................ 179
Discussion ................................................................................................................... 180
Change Over Time .................................................................................................. 184
Chapter Nine – Conclusion and Limitations ................................................................... 189
Structural Signatures ................................................................................................... 189
Change Over Time ...................................................................................................... 190
Implications of Theory ................................................................................................ 193
Limitations and Future Research ................................................................................ 199
Conclusion .................................................................................................................. 201
CRISIS AND STASIS VII
References ....................................................................................................................... 203

CRISIS AND STASIS VIII
List of Tables
Table 1: Disruptive Behavior Typology ........................................................................................ 44
Table 2: Sample Sizes. Private-Collaborative Model ................................................................... 90
Table 3: Dependent and Independent Variables. Private-Collaborative Model. ........................... 91
Table 4: AUC Scores Private-Collaborative Model. ..................................................................... 92
Table 5: Variable Importance: Private-Collaborative Model. ....................................................... 93
Table 6: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Private-
Collaborative Model. .................................................................................................................... 95
Table 7: Sample Sizes. Private-Isolated Model. ..........................................................................112
Table 8: Dependent and Independent Variables. Private-Isolated Model. ...................................114
Table 9: AUC Scores- Private-Isolated Model. ............................................................................115
Table 10: Variable Importance: Private-Isolated Model. .............................................................117
Table 11: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Private-
Isolated Model. ........................................................................................................................... 121
Table 12: Major events in the TF2 timeline. ............................................................................... 127
Table 13: Shifts in conventionality between February to September, 2014. .............................. 147
Table 14: Sample Sizes: Public-Isolated Disruption Model ....................................................... 149
Table 15: Dependent and Independent Variables: Public-Isolated Disruption Model. ............... 149
Table 16: AUC Scores- Public-Isolated Model. .......................................................................... 150
Table 17: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Public –
Isolated Model. ........................................................................................................................... 151
Table 18: Independent Variables: Private-Collaborative Model ................................................. 169
Table 19: LDA keywords. ........................................................................................................... 173
CRISIS AND STASIS IX
Table 20: Selected posts. ............................................................................................................. 173
Table 21: AUC Scores- Public-Collaborative Model. ................................................................. 176
Table 22: Variable importance: Private-Collaborative Model. ................................................... 177
Table 23: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Public-
Collaborative Model. .................................................................................................................. 179

CRISIS AND STASIS X
List of Figures
Figure 1: Private-collaborative disruptive signature. .................................................................... 47
Figure 2- Private-isolated disruptive signature ............................................................................. 50
Figure 3-- Public-isolated disruptive signature ............................................................................. 53
Figure 4: Comparison of Network Scope of Selected Metrics. . .................................................. 64
Figure 5: Average Edge Age Partial Dependency ......................................................................... 94
Figure 6: SHESD results on bans over time. ................................................................................ 99
Figure 7: Variable prediction plots, network constraint. .............................................................. 118
Figure 8: Variable prediction plots, effective network size. ........................................................119
Figure 9: Variable prediction plot, average edge age. ................................................................. 120
Figure 10: Rates of reported private-isolated disruption,. .......................................................... 128
Figure 11: Change in hero popularity, Feb-Sept 2014. ............................................................... 143
Figure 12: Z-score changes with 95% CI from Feb. to Sept. ..................................................... 147
Figure 13: Z-LDA topic selection results.................................................................................... 172
Figure 14: Partial dependency plots for key variables. ............................................................... 178

CRISIS AND STASIS XI
Abstract
This dissertation explores the interaction between and among players, developers and rule-
breakers in a series of popular online gaming communities. Using a combination of large-scale
network data, statistical learning, data modelling and anomaly detection techniques it examines
the role of “disruptive players” within in-game social networks over time. Covering a wide array
of disruptive behavior, from cheating to in-game protest the results demonstrate that players who
choose to challenge regulatory systems within a community often assume distinctive positions
within their host social networks. These results are contextualized using an evolutionary
framework that posits that changes in a game’s community may influence large scale shifts in the
regulatory structures that govern player behavior. The final results were mixed, while disruptive
players often do find themselves in distinctive positions within their host networks spikes in
disruptive play are not always associated with changes in their associated regulatory structures.
These findings have implications for the fields of organization communication and game study as
they shed light on the complex dynamics that govern the development and evolution of online
rules and regulatory structures.

CRISIS AND STASIS 1
CRISIS AND STASIS: UNDERSTANDING CHANGE IN ONLINE COMMUNITIES
“The future's a weirder place than we thought it would be when we were little kids.”
Cory Doctorow, Down and Out in the Magic Kingdom
“Anyone can tell a crisis when it arrives. The real service is… to detect it in embryo.”
Isaac Asimov, Foundation,
Chapter One- Introduction and Motivations
These two quotes represent the trends at the heart of this dissertation’s analysis.
Doctorow, a commentator and science fiction author closely associated with the free culture and
creative commons movements, states that technology has changed (or evolved depending on
your perspective) in new and unconventional ways (Doctorow, 2003). The new “weird future” is
clearly evident when looking at online communities and social communication through the
Internet. When Howard Rheingold examined the nascent communities being formed through
bulletin board systems in the early 1990s, he envisioned the future as a libertarian, free-wheeling
network of communities (2000). Individuals would come together in un-commercialized
discussion forums based on shared interests and ideas to exchange ideas and communicate.
Clearly online communities and social interaction did not evolve along these lines (Jenkins,
2006a).
Part of the reason why both the social and technological evolution of online communities
has diverged from these early predictions is the emergence of crisis as a driver for change.
Asimov, also a science fiction writer, argued that an understanding of moments of crisis (and
how they are managed) is an essential step to see how a social group changes over time (Asimov,
2010). This view is echoed in the scholarly literature. Both biology and sociology point towards
periods of crisis (either within an ecological niche or society respectively) as critical pivot points
CRISIS AND STASIS 2
within the history of a system (Gould, 1972; Greiner, 1972). Complex systems such as
communities often have periods of relative stability, continuing along an established track until
an event shocks the system and forces change, setting a new path (Bak & Sneppen, 1993;
Gersick, 1991).
A clear example of this within early online communities can be seen in Julian Dibbell’s
piece A Rape in Cyberspace (1994). Dibbell gives an account of a multi-user dungeon (an early
form of online game) where users roleplayed as various characters. Within this small community,
a particularly bad actor known as Mr. Bungle exploited the social and technological architecture
of the community to take control of another player’s character. Using this newfound power, Mr.
Bungle forced other people’s avatars into a series of compromising positions in front of their
peers. This incident represented a crisis within the community. The previous laissez -faire
libertarian structure of self-governance lacked both the authority and coercive power to deal with
Mr. Bungle. Outraged users clashed with the anarchist and libertarian factions within the
community over the proper course of action. Finally, one of the community’s developers took
unilateral action and banned Mr. Bungle, scrubbing the account from the community’s database
and destroying the character. This action prompted a series of more extensive reforms including a
formalized system of rules and punishments and a drift away from the communal, libertarian
system of governance that previously held sway.
The example of Mr. Bungle demonstrates the central thrust of this dissertation. To
understand how online communities and the technical systems which support them have
developed and, where things might go from here, it is important to examine moment of crisis.
When existing systems are thrown into question, communities have to organize a response,
leading to change over time (Baumgartner, Jones, & Mortensen, 2014). This evolution does not
CRISIS AND STASIS 3
come from a single source. As the previous story demonstrates, both participants within the
community and administrators interacted to determine the future course of their social group.
However, unlike Dibbell and Rheingold’s exploration of small bulletin board systems, online
communities have grown into a significant part of our day-to-day social and economic lives
(Castronova, 2008a).
One of the most interesting examples of this process of change and crisis can be seen
within online game communities. These groups, brought together through the shared experience
of a virtual world, feature many of the characteristics that tend to prompt crisis and change
(Bakioğlu, 2012; Lessig, 2006a). Games feature highly codified rule systems embedded in both
the code and rules enforced by the developer (Carter, 2013; Koster, 2013; Lessig, 2006a). These
rules are essential for maintaining the smooth and efficient function of the game. Rules help keep
the game “fair” and fun for the community, but these structures rub against the desire among
some player groups to stand out within the game (Consalvo, 2007; Huizinga, 1971). Scholars of
games and play have repeatedly identified segments of these communities who prioritize
achievement within the game as their primary motivator (Bartle, 1996; Bateman, Lowenhaupt, &
Nacke, 2011; Yee, 2006a). These groups often take this motivation to extreme ends, breaking
rules or code within a game in order to achieve their definition of fun (Bakioglu, 2009; Chesney,
Coyne, Logan, & Madden, 2009; Consalvo, 2005; Foo, 2008). In some instances these incursions
cannot be tolerated by the game’s developers or other members of the community as they may
actively detract from the enjoyment of other players with different goals (Sun, 2005). These
parties are often forced to take steps to reinforce or change existing rule structures in light of the
challenge from the community or else the game risks being “spoiled” for all of those involved
(D. A. Fields & Kafai, 2010). As a result, the game changes over time as the rules or norms that
CRISIS AND STASIS 4
define it are constantly rewritten or reinterpreted in response to both small and large crises. It is
important to note that in this situation crisis is not necessarily a negative event (Schumpeter,
1942). A crisis can help change rules or norms that are unfair or un-fun, making the game better
as a result (Drain, 2011). However, regardless of the direction of these changes, games present an
accelerated case study of evolution and change over time within an online setting.
Games therefore provide a unique window into the broader question of how the social
and technical makeup of online communities changes over time. While constant competition and
experimentation prompt a fast rate of change within games, they still share a common framework
with other online communities. Specifically, online communities, including games and other
forms of association are modes of social interaction where people are brought together by a
common purpose, governed by rules and norms and supported by technology (Preece &
Maloney-Krichmar, 2005; Shen, 2010). These three elements — technology, governance
structures (rules and norms) and a community with a shared interest — are the fundamental
building blocks of online association and the critical moving parts within this dissertation.
As A Rape in Cyberspace demonstrates, the various elements of change over time are not
independent processes. Individual-level choices such as a character choosing to assault one of
their compatriots cascade through a community creating change and prompting a shift in the
much slower moving set of policies which govern a community. These alterations also feed back
into the community, changing how people interact. Therefore, in order to understand the process
of change over time any attempt needs to take into account processes at multiple levels, and
attempt to link changes between these different tiers of analysis.

.
CRISIS AND STASIS 5
Given the complexity of understanding change in online communities across multiple
levels of analysis it is important to establish why this exercise is worthwhile. Simply put, it is
important to understand the evolution of online communities for both their social and economic
impact on our day-to-day lives. Since this examination is focusing in on the specific case study
of games, it is appropriate to use these communities as a justification for why understanding
change over time is important. From a crude population metric, four in ten Americans actively
play online games at least once a week (Anderson, 2015). A significant subset of these
individuals choose to participate in gaming communities, creating a market valued at 26.6 billion
dollars in 2016. This means that a significant amount of time and energy is spent within these
communities both contributing and reacting to changes over time. Beyond the sheer weight of
numbers, games are becoming an increasingly important part of the global economy
(Castronova, 2008b). Above and beyond the considerable revenues that these products create for
their developers and publishers, innovations from gaming are diffusing into other areas of the
economy. The advent of “gamification” means that concepts pulled from the gaming world are
finding their way into business communities, health networks and other organizations
(Deterding, Dixon, Khaled, & Nacke, 2011; Deterding, Sicart, Nacke, O’Hara, & Dixon, 2011).
Therefore, understanding how change occurs within games specifically, and online communities
in general, is extremely important.
Online communities can provide insights regarding offline social phenomena. Given
sufficient parallels between a game and offline social process, the two may “map” onto each
other so that insights gained online can inform research in offline contexts. For the purposes of
this dissertation, any discussion of results or findings will be limited to online communities.
Change over time online is governed by significantly more flexible structures than offline social
CRISIS AND STASIS 6
evolution. So while the two venues may share processes and mechanisms of change, developers
and other powerful groups in an online setting are able to respond quickly and with more
granularity than those in charge of offline spaces. As a result, change may be faster in online
communities as the (relative) lack of institutional momentum lowers the threshold for making
broad alterations to a given space. That being said, these caveats do not exclude results from this
examination from being applied to offline spaces, but the focus of this dissertation will be on
examining online communities and their evolution on their own terms.
In addition to the numerical argument presented the results from this examination may
have bearing on the current state of theory within the field of communication. . Scholars across
the field of communication have been calling for an increased focus on understanding
communication processes longitudinally (Ellison, Steinfield, & Lampe, 2007; Shen, 2010). Due
to the rapid evolution of culture and technology within online communities static studies can
rapidly become outdated. Understanding how change works allows cross-sectional findings to be
placed in the proper context (Williams, 2007). It also enables industry professionals to engage in
interventions to steer the course of future changes in a favorable direction (El-Nasr, Drachen, &
Canossa, 2013). This dissertation aims to answer several research questions that can support
these efforts. Namely, how does change emerge out of the interaction between users, developers
and the technology that supports them? When and why do crises emerge within a community?
How does a community change in response? What is the relationship between small-scale
changes at the personal level and larger shifts in a community’s rules and policies?
Answering these questions should help continue the process of equipping both academics
and professionals with the tools to understand why, when, and how change occurs within various
online communities. Hopefully, these findings provide both a template and series of new
CRISIS AND STASIS 7
questions for the former group to examine, while readers who belong to the latter category may
find some insights that can help them steer the course of the communities under their care.
To achieve this goal, this dissertation is divided into several sections. The first
summarizes and synthesizes the existing literature surrounding social and policy evolution. In
addition, it also imports insights from the more specialized fields of network analysis and game
studies to create a typology of various paths that online communities follow when undergoing a
period of dramatic change. The second section addresses the data which support this dissertation,
introducing the various communities that serve as test beds for the arguments laid out in the
preceding section. Additionally, this section introduces a general methodology that combines
network analysis with statistical learning techniques. The subsequent sections tackle a distinct
online community and examine how change occurs within each setting, exploring how the
technical infrastructure, user base and developer motivations interact to create shifts in policy or
player behavior.

CRISIS AND STASIS 8

Chapter Two- Theoretical Framework
While games and gaming communities are often studied in insolation they represent a
specific case study of online communities. Specifically, online communities are defined as a
collection of people interacting through technologically mediated forms of communication and
governed by both formal rules and informal norms (Preece & Maloney-Krichmar, 2005).
Gaming communities fit all three elements of this definition. Players are brought together out of
their shared enjoyment of the game the question (a collection of people), the game itself and its
associated forms or para-texts technologically afford interaction (technologically mediated
communication), and these communities are rich in both formal and informal rules that govern
behavior (formal and informal rules) (Cole & Griffiths, 2007; Consalvo, 2005; J. P. Williams &
Kirschner, 2012).
Given these commonalities, in order to answer the research questions put forward earlier
in this dissertation, it is important to consider not only the existing literature on change over time
and games, but also studies that tackle the broader topic of online communities.
This chapter is an attempt to draw these three bodies of literature together in order to
synthesize a framework for understanding how gaming communities change and the role of crisis
in this process. To achieve this goal, each element of these related theories will be addressed in
turn. First the literature surrounding evolution and change over time will be summarized. This
provides a common set of assumptions and concepts for tackling change in a wide variety of
settings. Having established these processes, a discussion will follow about how the introduction
of technology as a mediating variable both affords and closes off certain courses of action. These
incentives and disincentives help influence the process of evolution by defining what variations
CRISIS AND STASIS 9
are easier or more difficult within a given community (Hutchby, 2001). Finally, existing
knowledge surrounding the specific topic of games will be examined. This provides a structure
for grounding the knowledge of affordances and evolution established earlier in this chapter in
the concrete studies that comprise the back half of this dissertation. Bringing these three
elements together helps construct a framework for understanding the relationship between
change, technology and crisis within games and emphasizes the high-level hypotheses, which
will be operationalized and tested later in this paper.
With this general outline in mind, it is time to turn towards the existing literature
surrounding social evolution which will provide the basic vocabulary needed for the rest of this
discussion.
Network Evolution and Change
Evolutionary theories of all stripes trace their origin back to Charles Darwin’s seminal
work On the Origin of Species (1859). Darwin described a framework for understanding how
species emerge and differentiate themselves from each other. He demonstrated that variations
among animals were subject to the constant pressure of survival. Advantageous changes aided in
survival were retained and passed down to subsequent generations(Weiner, 1995). The
accumulation of these changes over time leads the fission of groups of animals into separate
species as each group adapts to the specific selection pressures inside its environment. While
social systems such as online communities are quite different from the finches in the Galapagos
studied by Darwin, the basic framework of variation-selection-retention (VSR) has been
extended to domains outside of biology (Campbell, 1965, 1982). The VSR framework works
across many levels, individuals, organizations or entire communities can be the basic unit of
analysis. The important point is that evolutionary theories offer a flexible metaphor for
CRISIS AND STASIS 10
considering change over time in a variety of settings (Astley, 1985). Theories drawing inspiration
from evolution have been successfully applied in topics ranging from policy studies to sociology
and communication (H. Aldrich, 1999, 2008; Baum & Singh, 1994; Bryant & Monge, 2008;
Shen, Monge, & Williams, 2011). This success suggests that utilizing the evolutionary
framework is a key step in answering the research questions at the heart of this research.
Variation Selection Retention. The central concept that links various theories of
evolution together across fields is the idea variation-selection-retention framework (VSR). This
model provides a generic tool for understanding change over time in multiple settings. The
concept of VSR begins with the development of differences between members of a population
(such as players within a game). These changes can be either intentional or stochastic, or some
mix of the two (Shen, 2014). Additionally, the scope and import of these variations range from
small, seemingly meaningless choices, to large issues that can influence a person’s status or
success for years to come. Variation springs from a number of different sources. As an example,
the choice to pursue a specific strategy within a game may emerge out of personal preference,
exposure to certain ideas while playing or a combination of seemingly random exogenous factors
(Shen, Monge, & Williams, 2014). The vast majority of variations are relatively innocuous and
drift silently into the past as a person navigates a given community. However, within most social
situations there is some form of selection pressure. This generally takes the form of some type of
feedback that grants differential rewards based on certain patterns of variation within the
community. As an example, adopting a specific strategy may lead to more or less success within
a game, with winners being encouraged by their good fortune, while losers are pressured to adapt
in order to better their lot (Andrews, 2010). This selection process feeds back into the community
as certain variations are retained due to their success. Players with winning strategies or tactics
CRISIS AND STASIS 11
gain the esteem of their colleagues and are emulated, while their compatriots with unsuccessful
approaches have a higher likelihood of leaving the game, as factors related to achievement are
some of the best predictors of churn (exiting a community) available (Z. Borbora, Srivastava,
Hsu, & Williams, 2011). This is not to say that unsuccessful players will automatically leave a
community, rather players who underperform are have a higher probability of leaving. .
An important supplemental element of the VSR approach is some form of scarcity or
pressure within a given environment (H. E. Aldrich & Pfeffer, 1976). Without competition or the
unequal division of finite resources there is no way for selection pressure to be brought to bear
within a given system. Each variation in a competition-free area can still obtain the necessary
resources or goods to be considered successful. Within the specific context of gaming
communities this competitive pressure is generally instituted by the game’s developer as a way
to make the game fun or interesting by requiring a level of skill from participants (Koster, 2013).
Additionally, competition can serve to drive cooperation and friendship within a game by
providing a shared incentive for players to work together. In-game selection pressures can be
internal to the player community as different members of the game compete with each other for
resources with varying levels of skill. Alternatively, pressure can exogenous as developers create
scarcity within the game by manipulating its code (Holin Lin & Sun, 2010). However, not all
games have scarcity built in as a central force. Some “sandbox” games offer players the ability to
do almost anything without limitations, making creativity the driving factor for player
engagement. This alters a core dynamic of the VSR process and renders the framework less
useful. As a result, this study will primarily focus on non-sandbox games, which have a level of
scarcity and/or competition between members of the community. Fortunately, these mechanisms
are often associated with a game being “fun” as they provide countless opportunities for
CRISIS AND STASIS 12
emergent behavior and new experiences. As a result, true sandbox games are relatively rare, with
oft-cited examples such as Second Life still retaining a level of competition through its
integrated in game economy where players compete for sales and attention.
Scalable Change. The generic nature of the VSR approach means that it can be scaled up
or down to suit different units of analysis (Astley, 1985). Keeping with the example of a gaming
community, an evolutionary perspective can be applied at numerous levels within a given space.
Most of the examples thus far have focused on players, who vary based on their choices and
strategies within a game with the game’s rules and competition providing a mechanism for
selection and retention. This dynamic can be scaled upwards to examine groups of players within
a game. These “guilds” or “corporations” compete over membership and in-game rewards, with
many of the same selection pressures brought to bear against players also influencing groups
(Chen, Sun, & Hsieh, 2008). At an even higher level, the game’s developer can be seen as being
in an evolutionary struggle against competitors within a crowded marketplace. Variations at this
level are the result of design choices and community management, with selection and retention
taking the form of marketplace success and the associated revenues (H. Aldrich, 1999; Baum &
Singh, 1994).
Variable Rates of Evolution. Each level previously detailed operates at the same time
and in parallel. However, the rate of evolution is not consistent across different sectors.
Biologists refer to a unit of evolutionary time as a Darwin or Haldane unit (Haldane, 1949). One
Haldane unit represents the rate of change over the elapsed time for a different species. Different
organisms move through the VSR cycle at different rates and will have a higher or lower number
of Darwin units. As an example, consider the difference between bacteria and human beings. As
single cell organisms, bacteria replicate at a much higher rate than humans. As a great deal of
CRISIS AND STASIS 13
variation within biology comes from the replication phase, bacteria also experience a much
higher rate of change within the same period of time. One human generation is approximately
thirty years long. In this time frame there will be billions of generations of bacteria, each of
which has the potential to express variations to fit their environment. Furthermore, the process
for bacteria takes place across a broader population as most species outnumber the humans by a
significant margin. Bacteria therefore change more rapidly than the human beings who host
them, which is why beneficial traits for bacteria, such as antibiotic resistance, can proliferate
quickly (Davies & Davies, 2010). In contrast, retained adaptations to diseases among humans
(such as the sickle cell trait which protects against Malaria) can take hundreds of years for
selection to take place (Wiesenfeld, 1967).
The varying scale of evolution can also be seen when the VSR approach is applied to
social settings. The three major levels operating within gaming communities (player, group,
developer) all iterate at a different speed. Players can adapt their strategies within game relatively
quickly depending on the genre and affordances of the game world. Furthermore, new strategies
are often distributed within the game’s community through associated para-texts such as guides
or forums, which increases the likelihood of change within the player base (Consalvo, 2007).
This process is further supplemented by the constant churn of people joining and leaving the
community and this provides another sources for new ideas. In-game groups move at a slightly
slower pace. Many groups have been around for months or years within a specific community
and have developed entrenched institutions and norms over this time (Pisan, 2007). Furthermore,
recruitment can often be time consuming as new members need to be brought up to speed and
integrated into the existing social network within the group. As a result, the relative pace of
change for groups is generally much slower, spread out over many months or years (Chen et al.,
CRISIS AND STASIS 14
2008). Finally, developers experience the slowest rate of change. Any variations that they take
with regard to managing their constituent communities requires extensive testing, engineering
and deployment. Even slower is change at the developer level. Developers are constrained by the
need to turn a profit and deal with multiple overlapping technical systems, which adds a further
limitation on their potential rate of change (Blow, 2004).
The discrepancy between how fast the various levels of a gaming community cycle
through the VSR process is important given that each actor is operating interdependently with
each other (Baum, 1999). This places developers in the unenviable position of having to manage
and keep track of a significantly more nimble player base, one which has strong incentives to
explore all of the possible variations within a game (Consalvo, 2007; Moeller, Esplin, &
Conway, 2009). Before exploring how this disjuncture can lead to crisis and shift the entire
makeup of a community, it is important to contextualize the evolutionary process that has just
been described. The social and technological boundaries that make online communities possible
also influence every step of the variation-selection-retention framework. In order to explore the
process of change over time, it is important to step back and establish the other factors playing
into the process.
Affordances of Online Communities
As mentioned earlier in this chapter, games are often seen as specific subset of the
broader category of online communities. Like most forms of computer-mediated communication
this means that certain actions are afforded or constrained by the social and technical
infrastructure that supports the community (Carr, 2000). Affordances and constraints represent
the built-in incentives and disincentives that are integrated in the design of a technological
system (Carr, 2000; Hutchby, 2001). The presence of these structures makes certain patterns of
CRISIS AND STASIS 15
behavior significantly more or less likely than would be otherwise expected. That being said,
affordances and constraints rarely prohibit a certain course of action.
With time, dedication and more often than not a degree of programming skill, the barriers
created by affordances can be bypassed. Nevertheless, understanding the behaviors afforded to
players and developers within a community is an extremely important aspect of understanding
change over time (Faraj, Jarvenpaa, & Majchrzak, 2011). Affordances and constraints represent
strong influences on variation, selection and retention by making specific strategies more or less
open to members of the community.
This dissertation will focus on affordances that previous literature has identified within
games and their associated communities. Additionally, to be included, the affordance or
constraint in question should have a clear connection to the process of evolution within an online
community. Specifically, an affordance should open up or close off strategies within the VSR
process. As an example, anonymity within an online community enables strategies around having
alternative accounts or “sock puppets” that would otherwise not be available. Similarly, some
affordances may close off strategies, the presence of a flexible governing system instituted
through code means that developers can restrict specific playstyles or strategies if they feel that
they threaten the game’s entertainment value or profit margin.
The Appearance of Anonymity. The first major affordance offered within many online
communities is the appearance of anonymity. The scope and durability of this anonymity differs
from community to community. Generally speaking, members of an online community can
choose to conceal essential information such as location, age or legal name directly or indirectly
(Preece, Nonnecke, & Andrews, 2004). Direct obfuscation emerges when a community does not
ask for, or support the sharing of this information between its members. Many forums, message
CRISIS AND STASIS 16
boards and in-game communities often have players identify themselves with a “handle” or
“screen name” that is exposed to other players (Steinkuehler & Williams, 2006). Other
communities may require more or less information from their participants, but these demands
can still be circumvented through indirect obfuscation by simply making up information (Gross
& Acquisti, 2005; Leavitt, 2015). The ability to manipulate identity within online communities
has important implications for the networks that form within these spaces. Participants can create
multiple accounts, recover from shaming or other forms of social sanctioning by abandoning an
old identity or pretending to be someone who they are not (E. J. Friedman & Resnick, 2001). For
many communities these affordances are not bugs, they are features, as they allow participants to
role-play or explore elements of their identity which would be otherwise closed off (Burn &
Carr, 2003; Corneliussen, Rettberg, MacCallum-Stewart, & Parsler, 2008). Regardless of
whatever stance a community’s population or developer takes towards the issue of anonymity,
often times individuals can achieve a form of ad hoc obfuscation even if personally identifying
information is required to participate by providing false or unrelated information.
While the baseline for association within online communities is often anonymous that
does not mean that all forms of interaction hosted in these venues follow this pattern. Instead,
relative anonymity is a baseline from which an individual can choose to disclose more or less
data depending on the social situation (Huynh, Lim, & Skoric, 2013). Many online communities
mirror and extend relationships formed offline, so by definition these ties are not anonymous
(Yee, 2008). Additionally, many connections formed through online communities evolve to the
point where the involved parties trust each other enough to disclose personal information (Poor
& Skoric, 2014). The important thing to take from this element of online communities is the
existence of a pseudo anonymous baseline of disclosure that represents the minimal level of
CRISIS AND STASIS 17
engagement for most users. However, the existence of this baseline should not be misinterpreted
as the universal rule but rather a violable norm.
It is essential to note that the durability of this appearance of anonymity is variable
depending on the level of data access a particular person has. For the average user it may be
difficult to ascertain if two avatars belong to the same person, but for developers and analysts
with access to harder to obscure data sources, such as credit card records or IP addresses, it is
significantly easier (T. V . Fields, 2013). Labelling this affordance the “appearance of anonymity”
is therefore a recognition of the fact that while avatars may appear to be undifferentiated there
are mechanisms built into the game which allows privileged groups to firmly establish a player’s
identity
With regard to how this affordance influences change over time, the appearance of
anonymity means that it may be possible for a single individual to have multiple personas within
a given community. Exploiting the appearance of anonymity, these avatars can act in concert
with each other in ways that may directly contradict the interests of any one character within the
game’s community. As an example, many players will have a “main” persona representing their
public facing image as well as having side characters designed to support the main persona
(Ducheneaut & Moore, 2004). Examined independently, it appears that the side characters are
sacrificing their utility within the game by constantly supporting the main persona without any
form of reciprocity. This can create contradictory signals (Leavitt, Clark, & Wixon, 2016).
Studies of change over time as members of a community can appear to act directly against their
self-interest, but because both personas are controlled by an individual, the behavior is internally
consistent and self-serving. The studies embedded in this dissertation will take a number of steps
CRISIS AND STASIS 18
to ensure that each individual is represented once within the datasets (see the methods section for
further details).
Malleable Governance. Another important affordance within online communities is the
greater malleability of the rules and norms that give a community structure. Games and other
online communities are governed by an overlapping series of rules and norms, which are rooted
in the technical infrastructure that makes a community possible. These rules also define the
boundaries of permissible behavior within the game. As an example, the immutable laws of
gravity and physics make hiking up a mountain difficult offline. However, within a game the
slope, difficulty and effort requires to climb a peak are baked into the underlying code of the
game (Lessig, 2006a). Because this code is open to developers (and to a lesser extent, players) it
is also changeable. Unless you are Superman it is impossible to leap a tall building in a single
bound, but within a game simply changing the value of gravity or how far a character can jump
within the underlying code makes this feat easy (“Skiing,” 2015). This wide scale malleability
opens up new avenues of change for both developers and players within a given community.
For developers, the ability to use “code as law” means that it is much easier to make,
maintain and enforce rules within the community (Lessig, 2006a, p. 16). As an example, if a
developer is looking to ban a certain item within a game, they can simply change the item’s
permissions so that it is impossible for anyone within the community to obtain (Lehdonvirta &
Castronova, 2014). The powers afforded through control of the game’s code also extend into the
realm of data and surveillance. “Data hooks” built into the game constantly monitor micro level
activities within the game and pull this information down into a database (El-Nasr et al., 2013).
By exploiting this combination of power and data, developers can bring a degree of regulatory
CRISIS AND STASIS 19
force to bear with a degree of accuracy and strength which is impossible within an offline
context.
Within the context of understanding how these communities evolve, the power of
developers to exploit code in a regulatory fashion means that they are a critical factor in
determining the course of change over time. With direct control over what is and is not possible
the developer defines the borders of the space within which players can experiment and engage
with each other (Lehdonvirta & Castronova, 2014; Lessig, 2006a). Additionally, many
communities feature an added level of power through the legal End User Licensing Agreements,
which all players have to sign (Balkin, 2004; Lastowka, 2011). These documents generally give
the developer additional powers to ban, mute and otherwise manage the trajectory of the
communities under their care. Given these powers and the active role that many developers take
within their community, it is important to recognize the power afforded by code when examining
how gaming communities evolve over time.
In addition to the power granted to developers, the malleable nature of the rules within a
game also open up opportunities for players. In most cases, the ability to manipulate the code of
a game is kept away from the average member of the community (Lessig, 2006a). There are two
cases where this power may devolve to the player level. Glitches or bugs are accidental
exposures of the underlying code to the users who should not have access to it (Kabus, Terpstra,
Cilia, & Buchmann, 2005; Yan & Randell, 2005). This gives player access to elements of the
community’s infrastructure which are normally off limits and allows for actions that would not
normally be allowed. Hacks or exploits are incursions into the code that come about due to
direct, proactive measures taken by a person (or group) who normally does not have access
(Doherty et al., 2014; Duh & Chen, 2009). This differentiates them from glitches that open up as
CRISIS AND STASIS 20
an unintended consequence or are due to programmer error (Bainbridge & Bainbridge, 2007a).
In either case, users can benefit from the access to the code to expand their potential sphere of
actions within the game and gain advantage over other players. These advantages are not limited
to technically inclined players; prepackaged cheats are frequently distributed through a variety of
channels and only require a simple installation from the end user. This influences how a
community evolves, as the realm of potential actions open to participants is not static, instead it
grows or shrinks as bugs or hacks surface, are exploited and then eliminated by the developer.
Joining and Leaving. Most gaming communities feature costs associated with joining
and leaving both the game itself and groups hosted within it. As an example, games often
demand significant contributions of time and/or money as a prerequisite for success
(Ducheneaut, Yee, Nickell, & Moore, 2006). The entry costs for popular titles can include an
upfront fee as well as a monthly subscription (Hirschman, 1970). Even so-called “free to play”
titles, which eschew these fees to expand their player base, have a number of incentives that
direct players towards contributing resources to the game (Hamari & Lehdonvirta, 2010;
Hirschman, 1970). The more time or energy a player sinks into a game, the more difficult it is to
break down longstanding ties within the game’s community (Shen et al., 2011).
Within games there are also entry and exit costs associated with movement in and out of
groups. Many games have structures in place that allow players to form associations and work
together in a group (Chen et al., 2008; Ducheneaut, Yee, Nickell, & Moore, 2007). Over time,
these groups can become quite entrenched with well-established traditions, iconography and
norms (D. Williams et al., 2006). This creates both entry and exit costs as new players need to be
socialized and adapt to these traditions while existing members are bound up in the group’s
social network.
CRISIS AND STASIS 21
Entry and exit costs influence evolution within gaming communities if leaving the
community is conceptualized as a strategy. Players who no longer enjoy a game or have other
options competing for the finite resource such as their money and or attention, may choose to
remove themselves from the community all together (Chang et al., 2008). In many ways, this can
be seen as an overarching meta-form of the VSR function where the game and its community
compete for the time and energy of the player base. When placed in this setting, exit from a
community is a rational response to a lack of success or enjoyment (Hirschman, 1970).
The presence of exit as an option means that the evolutionary system within a game
demonstrates a constant rate of churn as players leave or join and time passes (Z. H. Borbora &
Srivastava, 2012) . This churn differs from entry and exit from biological systems (i.e., birth and
death) in so far as it is largely voluntary and therefore is a form of variation as opposed to the
termination of the process. Indeed, exit may even be selected for and retained as a strategy by
users if they feel like their goals or needs are not being met by the community (Hirschman,
1970). However, as exit costs increase users are more and more likely to be unwilling to adopt
this strategy and may turn to alternative approaches to maximize their success. Research suggests
that one popular alternative to exiting a community is to attempt to change it by complaining or
implementing change on one’s own (Kucuk, 2008; H. Lin & Sun, 2007).
Punctuated Equilibrium
Gaming communities are a distinct form of online community that share some basic
features with other forms of online association. The presence of competition and limited
resources within a game helps make it fun by creating a dynamic and challenging environment
for players (Koster, 2013). However, competition also enables evolution at the player level
providing an external form of selection pressure that differentially rewards or punishes variations
CRISIS AND STASIS 22
among users. If a player’s strategy does not succeed within the game’s competitive setting it is
under greater pressure not to be retained and they may choose to exit the community. This
process leads to change within the player base as the community evolves towards better solutions
to the challenges presented within the game.
This in-game evolution is paralleled by change at the developer level. In order to compete
with other members in the market, developers need ensure that their game remains both fun and
novel for their constituent community (Ducheneaut et al., 2006). These pressures encourage
developers to manage their products and constituent communities in order to stay successful
within the market.
Both tracks of evolution take place at variable speeds. The constant churn of players and
strategies usually lends itself to much faster variation than what a developer can achieve due to
fiscal and technological constraints (Blow, 2004; Callele, Neufeld, & Schneider, 2005).
Additionally, evolution at the developer and player levels is influenced by a number of
technological affordances and constraints that constrain what variations are or are not possible,
such as control over code (Lessig, 2006a).
These concepts are the basic building blocks for understanding change over time within
online communities. But as the case study of Mr. Bungle in the introduction demonstrates, the
on-the-ground reality of change is not always a gradual, slow process (Dibbell, 1994). Moments
of crisis and uncertainty can and do occur among players, developers or between the two parties.
In order to answer the research questions put forward earlier in this dissertation, any potential
models need to tackle the presence of these periods of uncertainty and shift.
Once again, contributions from evolutionary biology offer a solution. The concept of
punctuated equilibrium was pioneered by Stephen J. Gould. He and other authors argue that
CRISIS AND STASIS 23
evolution is not a continuous, gradual process (Gould, 1972; Weiner, 1995). Instead, species
undergo extensive periods of stasis where they are relatively unchanged. These periods are
broken up by short bursts of extensive change. As an alternative to the classical Darwinian view,
small changes do not always slowly accrete until species drift apart. Instead, species vary around
a well optimized “phenotypic mean” until conditions change to such a degree that a rapid shift is
called for (Weiner, 1995).
Scholars in political science and organizational communication have extended the
concept of punctuated equilibrium to other systems (Miller et al., 2011). Organizations and
businesses often undergo extensive periods of stasis, puttering along with “business as usual”
until endogenous or exogenous changes cut into their profits and force a spasm of change
(Romanelli & Tushman, 1994; Tushman & Romanelli, 2008). Often these periods of intense
change at the organizational level had little to do with the accumulation of small variations or
shifts over time (Romanelli & Tushman, 1994). Instead, organizations in periods of stasis adhere
to a central strategy, making small tweaks based upon responses from the environment but not
deviating from the central theme of their business unless it faces an existential challenge. These
challenges may be exogenous, such as a change in the marketplace or new competitors drawing
market share away from a business (Mitchell, 2004). But for developers working with online
communities challenges may also be endogenous, emerging out of the constant churn and
development within a constituent community
Variable Change. Within gaming communities the variation and the constant selection
pressure of in-game competition ensures a much faster rate of change than most developers can
achieve (Chen et al., 2008). As a result, the game varies within the constraints created by the
CRISIS AND STASIS 24
technological infrastructure that supports it and the regulations put in place by a developer. From
the perspective of many developers, most of these changes are relatively harmless.
The choice between two strategies or groups within a game happens thousands of times
daily without significantly harming the bottom line for the developer. Both the slower speed of
change within the complex infrastructure of a software developer and the fiscal disincentives
from fixing something which is not broken mean that the vast majority of changes do not prompt
a response from those in charge of the gaming community. (Blow, 2004; Davidovici-Nora,
2009).
It should be noted that a period of stasis is not permanent. Change can originate from
exogenous forces, such as new actors entering the marketplace or shifts in the economy (Paap &
Katz, 2004; Schumpeter, 1942; Yu & Hang, 2010). This challenges the developer’s strategy and
necessitates a radical change, ending the period of stasis. The potential range of exogenous
factors that can prompt a shift in developer strategies is endless.
The pressure to create change can also come from within a developer’s zone of control.
As stated earlier, an online community is a shared collection of people governed by rules and
mediated through technology (Preece & Maloney-Krichmar, 2005). These rules are essential for
maintaining the focus and coherence of a community. When rules are challenged or disrupted,
the coherence of a community is also called into question and leads to dissatisfaction (Koster,
2013). This influences the strategic priorities of the developer running the community and
prompts a break from stasis to address the issue (Koster, 2013). Therefore, in order to understand
how crisis and change emerge endogenously from within a community, it is necessary to
examine situations when rules or regulatory structures are disrupted. The following section lays
CRISIS AND STASIS 25
out both the mechanisms for how this disruptive behavior works and why it prompts change in
relation to the concept of punctuated equilibrium.

CRISIS AND STASIS 26

Chapter Three- Disruptive Behavior
Having established and extended the logic of evolutionary systems to online communities
and explored how change over time tends to occur in fits and starts, it is now time to address one
of the engines behind that change, disruptive behavior. Disruptive behavior refers to variations
within player behavior that actively transgress or challenge existing regulatory structures and
norms. To illustrate, it is helpful to conceptualize a zone bounded on all sides. The boundaries
represent the various regulatory structures that control behavior inside the community. Within
this zone players are free to vary amongst themselves, trying out new patterns of association and
play. Player behaviors often push up against the boundaries, driven by the competitive selection
pressure of looking for advantage over their peers (Meades & Canterbury, 2012). In a large
number of instances they rebound off of these controls as the cost associated with overcoming
them and/or the possible penalties create an effective deterrent (Lastowka & Hunter, 2004).
However, in some cases, either due to sufficient motivation or the presence of an opportunity,
players end up outside of the regulatory structure, calling into question the effectiveness of the
rules by engaging in disruptive play (Consalvo, 2005; Koster, 2013).
In order to comprehend why disruptive behavior poses a threat to the stability of a
developer’s strategy, it is essential to understand both how and why regulatory structures enable
the day-to-day functioning of a community. Additionally, because disruptive behavior exists in
close relation to the rules of a community, it is essential to understand where these sources of
control come from. Some regulatory structures are a direct result of the technological affordances
described earlier in this chapter. Still, a separate class of regulations is not created out of
technological necessity but rather to help the community function. For games, a popular
CRISIS AND STASIS 27
theoretical construct for explaining both the presence and function of rules that go above and
beyond technological constraints is the idea of a magic circle.
Magic Circles, frames and keys. For regulations to be successful they have to offer
some form of protection or incentive to justify their own existence. Without this central animus,
rules may exist “on the book” but lack the will or strength to be exercised (Greene, 1997). With
this in mind, most scholars point to regulatory structures within a community as a means to
helping to preserve the status of this area as a special space, set apart from other areas of the
world (Huizinga, 1971). Within the specific realm of gaming communities this privileged status
is what makes a game fun (Koster, 2013). Johan Huizinga referred to this special area where the
rules of reality are set aside for the purposes of fun as a “magic circle” (Huizinga, 1971). Inside
the circle, normal laws or regulations (be they human or natural) can be set aside. As an example,
a person tackling someone else on the street would be a form of criminal assault, but if the tackle
takes place inside a football stadium, the normal rules are suspended and the tackler is instead
applauded and praised.
In order to maintain this boundary between reality and play, the borders of the circle have
to be guarded to keep external concerns from breaking through.in (Salen & Zimmerman, 2004).
These external threats can take many forms, such as knowing that football players often suffer
from serious brain injuries as a result of gameplay. If these thoughts infiltrate the minds of
participants (be they fans or team members) during a game, it can destroy their shared illusion of
fun and ruin the game. Disruptions challenge the privileged space of the circle by shifting the
rules unexpectedly (Bakioglu, 2009). This in turn alters the boundaries of the demarcated area,
leaving some players outside of it, damaging their immersion in the game process.
CRISIS AND STASIS 28
Huizinga’s conception of the magic circle goes a long way towards explaining why
regulations exist within games (1971). Each rule, law or norm helps prop up the walls that
delineate the game from the outside world. Disruptions knock down the walls by challenging the
specific and special rules within the magic circle.
This strict split between inside and outside the circle has come under fire in recent years
for being too dichotomous. More recent proposals have moved toward a descriptive terminology
centered on frames and keys (Consalvo, 2009). This viewpoint recognizes that players have
different definitions of what may be considered fun, and therefore defines the area inside
Huizinga’s magic circle in varied and non-overlapping ways.
Instead, drawing from the work of Erving Goffman, games can be seen as one of many
frames, shared understandings of a set of rules that a group of people subscribe (Consalvo, 2009;
Goffman, 1974). There can be a frame for home, a frame for work, a frame for going online and
playing a game. Each frame is “keyed” between, with people switching from one to another
quickly (Consalvo, 2009; Goffman, 1974). These constant shifts can make the game more fun by
enabling players to import external knowledge into the game as jokes or for their own advantage
by shifting keys (Fine, 1983). Alternatively, players can immerse themselves in the game,
shifting keys again to suppress the “home life” or “work” frame in favor of a greater focus on an
in-game frame.
Given this newer perception of frame shifting and keying within games, the role of
regulatory structures changes. Instead of being the guardians and boundaries of the delineated
circle, players use rules as a form of synchronizing device (Consalvo, 2009). Just as film reels
have codes for syncing all of the various technical elements of a movie, a shared set of regulatory
structures keys players into the same perception of the game world.
CRISIS AND STASIS 29
From this perspective, preserving the rules and regulations within a community is
extremely important for a developer. When players start to disrupt existing controls it can cause
the community to fall out of sync with their conceptions of what is and is not permitted within
the game itself. To return to the example of A Rape in Cyberspace, the antagonist, Mr. Bungle
was proposing an alternative framing of the community where technological power and control
were used as a means to legitimate his fantasies (Dibbell, 1994). This view contested with the
more communal and egalitarian frame held by other members of the community, leading to
tension and conflict until the administrators brought their power to bear and clarified the
situation.
Disruptive play therefore poses an endogenous threat to the stability of a given
community. Widespread disruption throws existing rules into question and leads to a
desynchronization of the assumptions held by members of the community. Without this common
basis, the social engagement at the heart of any gaming community becomes increasingly
difficult, damaging the health of networks embedded within this space. Players may feel that the
game is unfair or not understand the strategies their compatriots are choosing to pursue (Koster,
2013). Over time this tension may influence the evolution of the player community and damage
its overall health. This endogenous crisis can provide the stimulus needed to tip a developer out
of a period of prolonged stasis and force variation within the game’s community. Therefore, in
order to understand how communities change over time it is necessary to explore moments of
crisis and disruption, as these provide the endogenous stimuli that can push developers to employ
the power afforded to them and make significant changes.
Understanding this process requires a clear outline of the various overlapping regulatory
structures that make gaming communities possible. Only by defining the boundaries surrounding
CRISIS AND STASIS 30
and constraining the process of variation, selection and retention, does it become possible to
explore the mechanisms people use to disrupt these systems.
Sources of Order
Regulatory structures are an essential part of what makes an online community function
(Preece & Maloney-Krichmar, 2005). However, these structures do not all originate from one
source. Instead, control is cobbled together out of a series of partially overlapping but distinct
systems that constrain the potential variations available to members of a community (Lastowka,
2011; Lessig, 2006a). These systems have different strengths and weaknesses; additionally, they
vary in how easy or hard it is for potentially disruptive individuals to bypass them. Generally
speaking, scholars have identified three broad categories of regulatory structure, norms, rules and
code (Eklund, 2010; Koster, 2013; Lessig, 2006a). The following section lays out the
fundamentals of each system and demonstrates how they serve to govern player behavior using
the familiar case study provided by A Rape in Cyberspace.
Norms. Norms refer to shared beliefs and practices within a community about how
individuals should interact both with each other and with the technical systems that support their
engagement. Norms are not usually codified, but develop organically out of the constant churn of
social interaction within a community (Woodford, 2013). Generally speaking, norms are
enforced within the player base through social sanction and incentive, a process which can be
aided by design choices originating from the developer (Bergstrom, Carter, Woodford, & Paul,
2013). Following norms of behavior may make it easier for players to engage with each other,
smooth social interactions and align expectations of behavior.
Developers can also implicitly or explicitly support certain norms within their community
based on how they conceptualize their work and their communication with their constituent
CRISIS AND STASIS 31
community (Moeller et al., 2009). The mission or goal of developers when creating particular
space shapes how those in charge of it function at later points in time. So if a community is
viewed by its developers as being fair and equal they may take steps to preserve this mental
concept. Norms both constrain and afford activity among both parties involved.
Within the case of A Rape in Cyberspace, conflict surrounding norms can be most clearly
seen in the debate about what to do with the disruptive player, Mr. Bungle. Dibbell documents
how members of the community adhered to a certain perception of their shared space as a
relatively anarchic and democratic area (1994). This norm meant that when some members of the
community advocated for new regulatory systems that would prevent future disruptions, some
people disagreed with this plan. Additionally, members of the community held to a strong norm
of democratic deliberation, which was broken when a developer chose to unilaterally ban Mr.
Bungle. In both situations, the shared conceptions of what is good and proper within a
community and the pressures which underlay these concepts came together to regulate how the
entire collection of individuals functioned.
Rules. Unlike norms, rules are specific, well-defined and codified regulations that are
often enforced using the coercive power of the law. The vast majority of online communities
require their participants to agree to a terms of service (TOS) document before they are permitted
to enter (Lastowka, 2011; T Laurie Taylor, 2002). The TOS represents a legally binding contract
between developer and player. The document formally lays out the rights and responsibilities for
both participants and developers within the community. While there has been a tendency among
users to simply agree to the TOS without reading it, this does not change the fact that their
subsequent behaviors are governed by its dictates (Bakos, Marotta-Wurgler, & Trossen, 2014).
Generally the TOS categorizes certain behaviors that challenge the stability of the community as
CRISIS AND STASIS 32
illicit (Lastowka, 2011). As an example, individuals may not be allowed to modify the
community’s code without permission or use it as a place to conduct economic transactions.
While the legal status of concepts such as property or ownership within online games is still a
matter of debate among scholars and professionals, the contractual nature of rules means that
they tend to carry more robust penalties and enforcement mechanisms (Kennedy, 2008). Players
can be banned, prevented from communicating, fined or penalized in other ways for violating the
terms of their agreement with the developer. However, like norms these rules are flexible. Many
in-game rules are ignored by the players due to a lack of enforcement or willpower from the
developer (Carter, 2013). As an example, it can be difficult to enforce rules around foul language
or abuses within a game (V oulgari & Komis, 2011). Conversely, many developers take a strong
stance against fraud or other illicit activities within in-game economies (Valve, 2015b).
Within the case study of LamdaMOO and Mr. Bungle, rules can be seen emerging
towards the end of the narrative in response to the events chronicled within the piece (Dibbell,
1994). In order to ensure that other players would not be victimized, the administrators of
LamdaMOO instituted a number of rules surrounding player conduct and behavior. While
Dibbell laments that these rules made the community less democratic, they did help constrain the
behavior of the player who controlled Mr. Bungle (after he exploited the affordance of
anonymity to make a new account) and relegate him to the status of a curiosity.
Code. Finally, the strongest mode of enforcement is the code of the game itself. Much of
the regulatory function of code has already been summarized in the “Malleable Governance”
section of this chapter. It is sufficient to say that code constrains what is and is not possible
within a game in a way that can be directly manipulated by the developers (Lessig, 2006a). As an
example, many games feature fairly realistic gravity, dropped items will fall, jumping characters
CRISIS AND STASIS 33
will descend, and so forth. Instead of being an immutable law of the universe as it is within the
physical world, gravity is written into the game through code. Since code is law it is both the
most basic and the strongest regulatory system within a game (Lessig, 2006a). But unlike the law
of the universe, code can be manipulated and exploited by players. So, while a person cannot
change physics in order to fall from a great height without hurting themselves, a skilled player
can change these rules within the game (Balkin, 2004).
The use of code as an enforcement mechanism can be seen through the reaction of the
“wizards” (developers) in A Rape in Cyberspace (Dibbell, 1994). In response to the actions of
Mr. Bungle, one developer exploited their control over the community’s code to systematically
annihilate all mention of the offender. All records, logs, markings and messages were deleted and
destroyed and the name Mr. Bungle was banned from ever being used within the LamdaMOO
community, demonstrating the potency of code as a regulatory force.
Why Disrupt?
The strength of these regulatory structures raises the question, why would anyone within
a community choose to disrupt them? There is no single answer to this query. But insights can be
found from previous research in psychology, game studies and online sociology.
Specifically, within the area of online gaming communities there has been a significant
effort to develop a typology of participant engagement with these spaces. Given the diversity of
the subject matter these typologies have also been quite divergent, but several common
motivations have emerged across studies, namely: achievement, socialization and immersion
(Bartle, 1996; Bateman et al., 2011; Yee, 2006a, 2006b). These motivations are not mutually
exclusive and players can be driven by a combination of factors.
CRISIS AND STASIS 34
Players motivated by achievement generally aim to demonstrate that they are able to out-
perform their compatriots within a given gaming community (Yee, 2006a). Finding better
strategies, reaching higher scores, or being the first person to complete a specific feat, are all
examples of achievement based motivations (Yee, 2006b). As a result of these goals the pressure
on achievement motivated players to select the proper strategies within the VSR process in
consistent and intense. Disruptive behavior offers the ability to achieve more, faster, by pushing
out at the boundaries that constrain variation within a game. This opens up fallow ground for
new strategies and may give a disruptive player an edge on their competitors.
Social players view a gaming community as a means to engage with other people. The
community is a platform for shared experiences which help create long term bonds. From this
perspective, the process of challenging and breaking through a regulatory structure can be seen
as a social process (Lowood, 2006). Hacker or exploit communities exist around most major
games that provide a social network, which encourages disruptive behavior as a sign of status
and prestige.
Immersion based players enjoy embedding themselves within their community of choice
(Yee, 2006a, 2006b). Due to the relative anonymity and communicative affordances laid out
earlier in this chapter, it is possible to take on a different identity or perspective when
participating in a gaming community (D. Williams, Kennedy, & Moore, 2011). For many players
this takes the form of roleplaying or acting out fantastic scenarios that would otherwise not be
possible. However, for a smaller subset of “grief” players, immersion means acting out darker
impulses based around victimizing other players (Foo, 2008). For these players role playing
emerges not as a form of collective storytelling but as way to engage in narratives focused on
CRISIS AND STASIS 35
victimization and bad behavior (Bakioglu, 2009). Becoming embedded within the game-world
therefore can entail engaging in disruptive behavior as a prerequisite for immersion
Characterizing Disruptive Behavior. To summarize what has been covered so far,
online communities are rule-governed social spaces mediated through technology. Low level
shifts are constantly occurring through a process of variation, selection and retention at the level
of individuals within the community and the developers who run it. However, large scale
realignments in the trajectory of a community are much less likely to occur given the slower rate
of change at the developer level and institutional baggage. Large scale shifts are more likely to
occur when variations among the player base disrupt regulatory structures such as rules, norms
or code.
The final section of this chapter elaborates on this last phase of the process. How does the
presence and growth of disruptive behavior translate into pressure to create change? Answering
this question requires diving into the literature on social network analysis to ground previously
introduced concepts (disruption, VSR, punctuated equilibrium, affordances and regulatory
structures) in the language of networks and social interaction, before using this framework to
operationalize said concepts.
Disrupting a Network. The first step in this conceptualization is to recast the idea of an
online community (focused on gaming or any other topic) from a single entity into a series of
connected social and associational networks. Networks are comprised of two fundamental parts:
“Nodes” that generally represent people, places, things or concepts and “edges” which capture
relationships between nodes (Wasserman & Faust, 1994). Online communities play host to
multiple overlapping networks of social interaction, communication and exchange (Shen, 2010).
Variations in player strategies due to the constant evolutionary churn within a community are
CRISIS AND STASIS 36
therefore reflected in shifts within the network. As an example, if a member of a community
chooses to join a particular group or association, their personal network will shift to reflect these
new associations and friendships. Social network analysis therefore offers a framework both for
understanding the static characteristics of a community and its evolution through the movement
of edges and nodes (Monge, Heiss, & Margolin, 2008).
The variation, selection and retention at the heart of small scale evolution maps well onto
the network perspective. Variation can be modelled as changes in the node and edge structure
(Shen, 2010; Shen et al., 2014). The choice to align with members of a community is directly
influenced by a given person’s choice of strategy. The constant churn within the community due
to lower entry and exit costs provides a mechanism for selection within the network. Nodes,
which drop out or leave, break their edges and connections may be broken between two
individuals if the relationship is no longer beneficial. Those edges and nodes which do well
within this process are therefore more likely to be retained and propagate their strategies in the
future.
Within this tangle of edges and nodes, distinctive motifs emerge across human interaction
in multiple spheres. A motif is a distinctive pattern or structure of nodes and edges that repeats at
a rate far higher than random chance (Milo et al., 2002). A simple example is triadic closure,
when two individuals share a common friend they are more likely to become friends, the
repetition of this trend over and over again across networks creates a motif (Huang, Tang, Wu, &
Liu, 2014). Individuals generally have an upper limit (or carry capacity) on how many
relationships they can maintain at a given point in time. Additionally, certain structures are
associated with positive outcomes (Monge & Contractor, 2003). Numerous studies have
demonstrated the connection between specific patterns of nodes and edges, often labelled
CRISIS AND STASIS 37
structural signatures, with social status and wellbeing (Burt, 2000). Specifically, a structural
signature represents a motif that previous research has associated with a specific outcome. To
return to the example of triadic closure, this motif is often associated with the theory of bonding
social capital as closed networks encourage reciprocity and trust (Burt, 2001; Shen, 2010).
Therefore, the motif of triadic closure can be conceptualized as a structural signature associated
with bonding capital.
Social Capital
Social capital refers to the “network and the norms of reciprocity and trustworthiness that
emerge from them” (Adler & Kwon, 2002; R. Putnam, 2000, p. 19). Specifically, certain
positions within a network give individuals access to more information or social resources (Burt,
2000; J. S. Coleman, 1988). Social capital therefore reflects both a specific position within a
network and the benefits which result from it. Using the first half of this definition, it becomes
possible to explore a given network and look for specific patterns of association, which previous
research has identified as being correlated with a high stock of social capital (Burt, 2000). These
“structural signatures” can be generalized across various networks and provide a framework for
identifying individuals rich in social capital in other situations (Monge & Contractor, 2003).
Existing research has demonstrated two broad forms of social capital. Bonding social
capital refers to tightly bound networks of exclusive overlapping trust and association (Burt,
2005; J. S. Coleman, 1988). Bonding ties are repeated, strong, and difficult to both create and
dissolve. Generally this form of capital emerges among homogenous sections of a network which
repeatedly engage with each other over an extended period of time.
Alternatively, bridging social capital captures the “strength of weak ties” argument put
forward by Grannovetter (1983). Acquaintances and other “weaker” connections within a
CRISIS AND STASIS 38
network offer an advantageous position for members of a community. Having a large number of
these connections lets a person broker the flow of information between groups and closes
“structural holes,” gaps in a network between densely packed subgroups (Burt, 1995, 2000).
Both varieties of social capital lend themselves to unique structural signatures when
viewed from a network perspective. Because bonding social capital relies on repeated,
overlapping networks it is generally expressed as network closure. Network closure refers to the
process whereby any three nodes in a network (called 1, 2 and 3 for this example) all know each
other. In concrete social situations this is often seen in the “friend of my friend is my friend”
situation or among homogenous groups with long term exposure to each other (Ganley &
Lampe, 2009; Goodreau, Kitts, & Morris, 2009).
In stark contrast to closure and bonding social capital, bridging emerges from a structural
signature generally labelled as brokerage. Brokers are individuals who cross “structural holes”,
areas of low density separating clusters of nodes within a network (Ahuja, 2000). By linking two
otherwise unconnected groups brokers can control the flow of information between otherwise
unrelated parties and extract a number of advantages which manifest as bridging social capital
(Adler & Kwon, 2002).
For a community to be successful, it needs to encourage network structures associated
with both bridging and bonding structures. Bonding facilitates the growth of trust. Reciprocated,
overlapping connections make it easier for small groups to organize and operate. Trust between
the members allows for favors or investments within a group in long term projects (Ratan,
Chung, Shen, Williams, & Poole, 2010; D. Williams et al., 2006). Bridging capital helps keep
information circulating within the community by introducing new sources of information into
otherwise closed groups (Steinkuehler & Williams, 2006). By linking together clusters,
CRISIS AND STASIS 39
information on new strategies, ideas and ways to participate in the community can diffuse from
their origin throughout the remainder of the population.
While previous work has identified a number of positive outcomes associated with social
capital, the network structures associated with it can also lead to negative outcomes. Network
closure facilitates the growth of bonding capital, but in excess it can lead to a group becoming
inward-looking and isolated as the closed borders of their network shut out new information
(Ratan et al., 2010). Similarly, brokers can find themselves pulled in too many directions at once,
creating a bottleneck for future information flow.
Despite these potential negative outcomes it is in the interest of developers and
community managers to encourage the development of network structures that lead to the
formation of bridging and bonding social capital. If managed and kept in balance, social
networks can provide a powerful social “glue” which helps make the game fun and engaging.
However, some varieties of disruptive behavior offer alternative forms of engagement within a
community, which imbalances these structures. Breaking down the concept of disruptive
behavior (i.e., strategies chosen by an individual or group that directly challenge a regulatory
system) the concept of benefit is implicit within this formulation. Why spend the time and effort
to challenge a rule, norm or section of code otherwise? Disruptive behaviors therefore generate
some form of benefit, be it achievement, social status or immersion (Bakioglu, 2009; Chesney et
al., 2009; Consalvo, 2005; Koster, 2013). However, the existing literature on public goods serves
as a reminder that some benefits are internal to a specific group or individual while others have
to be shared (Cornes & Sandler, 1996; Olson, 1965a). Social capital is fundamentally a shared
benefit in that it originates from networks of dependence and reciprocity and is extremely
difficult to constrain to a single person or group (R. D. Putnam, Leonardi, & Nanetti, 1994).
CRISIS AND STASIS 40
Nevertheless, not all forms of disruptive behavior share this characteristic. The benefits from
breaking through a regulatory structure can fall only to a person or group and are successfully
internalized within certain sections of the network.
As an example, consider a well-studied form of disruptive play, cheating. Cheaters are
player who import third-party tools or exploits to give themselves an advantage (Consalvo,
2007). The introduction of external tools is generally strictly prohibited under most games’ TOS
and the presence of these tools challenges the enforcement power of code (Koster, 2013; Lessig,
2006a). Many exploits are developed by communities of disruptive individuals who share tips
and tricks on the best ways to achieve results above and beyond what is available to their non-
disruptive counterparts.
Cheats have a wide variety of functions since they can improve the wealth or status of a
player or give them abilities their opponents may lack. Unlike social capital, a cheat or exploit is
not a public good, instead it represents a form of private, rivalrous good. The private part of this
description comes from the fact that people can be excluded from the benefits generated by the
cheat (by denying access to the tool). Additionally, the benefits are rivalrous as too many people
using a cheat will ruin the game as everyone will gain the same new abilities, negating any
advantage. Therefore, the advantages from disruptive behavior do not necessarily generalize to
the entire community. This creates an unhealthy dynamic by opening up strategies that still result
in success but do not generate the positive network structures associated with fair play.
A brief example helps illustrate this point. Consider a hypothetical gaming community
with various challenges of escalating difficultly laid out for players. In many games, a single
player may have a hard time overcoming a higher level challenge and must team up with other
people (Muhammad Aurangzeb Ahmad, Borbora, Shen, Srivastava, & Williams, 2011; Ang &
CRISIS AND STASIS 41
Zaphiris, 2010). This process helps foster connections within the community, extending the
aggregate bridging social capital (Steinkuehler & Williams, 2006). A few of these ties may also
mature into stable reciprocal links, which also aides in the development of bonding social capital
(Ratan et al., 2010; D. Williams et al., 2006). A disruptive player using a hack, exploit or trick,
bypasses the need to participate in this network. Using their external tool as a crutch they can
push through the challenging content without having to reach out to the surrounding network
more than they want to, while internalizing any rewards or glory from their achievements. As a
result, the aggregate stock of social capital suffers, which in turn damages the long term health of
the community.
In other words, the presence of disruptive play introduces new strategies to the
evolutionary mix within a gaming community. These strategies still offer success and can survive
under selective pressure. However, the benefits from these strategies are not public goods and
can be internalized within and among the disruptive members of a community. This in turn
displaces structures that generate social capital and damages the overall health of the community
(Blackburn et al., 2012).
H1- Disruptive behavior actively displaces structures associated with bridging and
bonding social capital with alternative structural signatures. The exact form of these signatures
depends on the game and disruption in question.
1

This long-term damage presents a threat to the fundamental security of the developer.
Degrading stocks of social capital within a community can damage the social nature of the space,
making it harder for participants to socialize or coordinate effectively. This fulfills the
requirements laid out within punctuated equilibrium theory for a threat that can push an

1
See sub-hypotheses in following section.
CRISIS AND STASIS 42
organization out of a period of stasis and introduce major changes (Romanelli & Tushman,
1994). If a developer does not directly address the disruptive behavior within their constituent
community it may continue to displace helpful network structures. This in turn damages the
social fabric of the game and makes cooperation or competition more difficult. Faced with these
challenges and given the fact that they are afforded a high degree of control over a community
via code and rules the developer is incentivized to step in.
An example helps illustrate this process. Many games promote the creation of bonding
social capital through the creation of teams and groups within the community. These groups
provide structures that help generate trust among their members and give players a ready-made
team for tackling in-game challenges (Ratan et al., 2010). One of the primary incentives driving
team formation is the need to overcome challenges within that are too difficult for a single player
(Ducheneaut et al., 2006). These challenges provide a reason for in-game groups to exist and
motivate participants to work together to overcome a series of escalating goals. The resulting
bonding social capital helps keep players within the game by enabling the formation of long-term
reciprocal relationships.
However, if a player is engaged in disruptive behavior and finds a bug or exploit within
the game, they may have powers or abilities well beyond other members of the community
(Consalvo, 2007). With these powers they can overcome barriers which would normally require
close coordination and teamwork to surmount. This effectively removes one of the major drivers
for group formation and by extension bonding social capital. Without bonding capital, a
community may face issues with trust and reciprocity among its members, degrading an essential
social lubricant and precursor of effective collaboration (J. S. Coleman, 1988; R. D. Putnam et
al., 1994). As a result, the game’s network shifts and as players have fewer ties to the community
CRISIS AND STASIS 43
they are more likely to leave (Ducheneaut et al., 2006). This decrease in the player-base poses a
threat to the developer’s revenue, one of the possible scenarios where corporate stasis is broken
and the business enters a period of crisis and readjustment.
Following this concept through to its conclusion, spikes in disruptive play displace
bridging and/or bonding social capital within a given community. Lower stocks of social capital
influence the overall health of the community by reducing trust or information spread. This in
turn makes the community less appealing and promotes player exit, which represents a serious
threat to a developer. While developers are normally constrained by organization stasis, the threat
posed by disruptive play may be enough to force a period of crisis and realignment, therefore:
H2- Large scale changes by administrators of online communities are correlated with
increases in disruptive behavior.
If disruptive behavior displaces structures associated with social capital by offering an
alternative strategy it should have its own distinctive structural signature(s) (Hu, Kaza, & Chen,
2009; Magaloni, Diaz-Cayeros, Matanock, & Romero, 2011; Xu & Chen, 2008). Unfortunately,
networks are complex, interdependent structures with a theoretically limitless number of forms
and permutations. In order to identify the structural signatures that are most likely to be
correlated with disruptive play it is necessary to draw from existing research on illicit networks.
Disruptive Signatures
The easiest way to interrogate the structural signatures of disruptive behavior is to start
with the most visible aspect of the process – the outcome. Disruptive behavior is an extremely
broad category. Given the wide variety of gaming communities and regulatory structures
available, different patterns will emerge depending on what tools and abilities are afforded by a
CRISIS AND STASIS 44
given setting. Within this diverse range, the outcomes from disruptive play generally break down
along two major axes.
Collaborative versus Isolated: Collaborative disruptions require a group of players to act
in concert with each other in order to be effective. Often times they rely on the volume of
disruption happening all at once to be successful (Groen, 2011; Jenkins, 2006d). Isolated
disruptions find success from the fact that players operate on their own and can successfully
disrupt regulatory structures without the help of others.
Public versus Private results: Public results refers to a reward or outcome from a
disruption that benefits all players whether or not they participated in the disruptive activity
(Olson, 1965a). This could be because players choose to share the results of their actions or
because they cannot deny use of these results to other players. Alternatively, private results go
only to the perpetrators who can deny access to others.
These two axes influence the network structure of the disruption and how developers and players
react to their presence. Specifically, by looking at the interaction between the two factors, a two-
by-two grid of possible outcomes emerges with each representing a different variety of disruptive
behavior.
Table 1: Disruptive Behavior Typology
Public Private
Collaborative Induced Disruption (special case,
see end of chapter)
Hackers, Gold Farmers (Keegan,
Ahmed, Williams, Srivastava, &
Contractor, 2010)
Isolated Meta-breaking, public exploits
(Moeller et al., 2009)
Scammers, frauds (Bardzell et al.,
2007; Dilla, Harrison, Mennecke, &
Janvrin, 2013)

CRISIS AND STASIS 45
Each axis also carries implication for the potential structural signature of a given
disruptive strategy. The following section will briefly explore these predictions and the literature
behind them before combining each prediction
Private-Collaborative. Private-Collaborative disruptions generally represent individuals
or small groups who work together to disrupt the game for their own benefit. This category
includes gold farmers and players who develop exploits or hacks for a game’s code. Gold
farmers are players who work collaboratively to generate in-game currency before selling it for
hard cash (in violation of most terms of use for games) (Muhammad A. Ahmad, Keegan,
Srivastava, Williams, & Contractor, 2009). Hackers usually work in teams to find exploits in the
game’s code or collaborate to create programs that can break down regulatory structures and
benefit the user (S. Coleman & Dyer-Witheford, 2007). In either case, the outcome of the
disruption can be kept internal to the disruptive group, as neither gold farmers nor hackers share
their products or results with the general public.
Structural Signature. Private-Collaborative disruptive behavior requires a fine balance
between communication and security. Cheaters or hackers often have to work together to find
and exploit flaws/develop tools. These tools represent a form of “club good” as the benefits are
exclusive to members of the group, but non-rivalrous (Cornes & Sandler, 1996). Once a cheat
has been developed, its use by one disruptive player does not preclude the use by another.
However, external detection by other members of the community increases the likelihood of
being reported or punished, which removes the benefit for all those involved. These two
pressures, in-group coordination and the danger of out-group contact, have been identified in a
number of different illicit networks in other settings (Bright, Hughes, & Chalmers, 2012; Krebs,
2002b). An interesting point of comparison is drug networks. Dealers, distributors and producers
CRISIS AND STASIS 46
all have to coordinate to achieve their goals but are vulnerable to incursions from external forces
(Bright et al., 2012). Numerous studies have demonstrated that this combination of pressures
results in three distinctive network patterns:
Homophily – due to the need for coordination and the risks of linking with a non-
disruptive partner, this brand of disruption promotes cautious homophily, the tendency of people
to link to others similar to them (McPherson, Smith-Lovin, & Cook, 2001; Raab & Milward,
2003). In other words, disruptive players are more likely to link to others who are similar to
them, especially if they also engage in disruptive behavior. Therefore:
H3a- Individuals engaged in private-collaborative style disruptive behavior will be more
likely to connect with other disruptive players and therefore display a higher than expected rate
of homophily.
Closure – edges between disruptive players are likely to result in triadic closure. Closure
helps cement trust and coordination between disruptive actors and is a safer way to add edges to
a network as the target of the new connection has been pre-vetted by a friend (Krebs, 2002b).
H3b – Private-collaborative disruptive actors will have a higher rate of network closure
than their counterparts within a host network.
Brokerage suppression – conversely, brokerage connections that link two previously
unconnected groups will be less likely as it creates an opportunity for external discovery of the
disruptive network (Krebs, 2002a).
H3c – Private-collaborative disruptive actors will have lower rates of brokerage than
comparable members of their host network.
To tie back to the social capital framework laid out earlier in this chapter private-
collaborative disruptions promote structures associated with bonding social capital. Disruptive
CRISIS AND STASIS 47
actors may grow to trust and rely on each other over time as they develop and experiment with
different types of cheating and exploits. The danger posed by introducing new individuals into
this community means that unlike non-disruptive members of the community there is an interest
to generate structural holes and reduce brokering (Fellman & Strathern, 2007; Krebs, 2002a). For
regular participants these gaps close off information flow and reduce the chances for beneficial
coordination, but since disruptive actors draw their evolutionary success from an external source
this priority is subsumed to the need for security.

Figure 1: Private-collaborative disruptive signature, nodes colored red are disruptive.

Private-Isolated. This variety of disruption represents an individual working on their
own to break down in-game regulatory structures. The actual disruptive behavior only requires
one person in this case and any benefits reaped from these actions only fall to the disruptor. An
excellent example in this case are scammers or other fraudsters who populate many games (Dilla
et al., 2013). These players take advantage of existing norms and rule structures designed to
promote trust between players to increase their own wealth and stature at the expense of others.
Often this takes the form of stealing accounts through social engineering (phishing) or taking
CRISIS AND STASIS 48
currency through fraud (Bardzell et al., 2007). In either case, the value, be it extra money or a
new account, only benefits the disruptive actor and the disruption itself is generally carried out
by an individual who chooses their targets carefully.
Structural Signature. This pattern of networked incentives has parallels to findings from
the criminology literature tackling the issue of fraud. Fraudulent actors tend to operate on the
fringes of a marketplace, taking advantage of low information consumers (Tillman & Indergaard,
1999). Additionally, fraudsters often attempt to interpose themselves between isolated actors and
other sources of knowledge, creating a “bogus brokerage” situation (Tillman, 2003; Tillman &
Indergaard, 1999). Within a network, this manifests as a “heavy star” structure where the fraud is
brokering between numbers of unrelated clients. The clients can exchange information only
through the central disruptive actor, allowing that person to alter the flow of data to their benefit.
A fraudulent car dealer may misrepresent previous customers’ statements to give themselves an
air of legitimacy. If two or more of the fraudster’s clients and communicate and coordinate with
each other, it is possible for them to deduce that their bad treatment may be the rule, not the
exception (Baker & Faulkner, 2003). As a result, the criminology literature suggests that
fraudulent actors keep their marks isolated within their local network. These incentives combine
to suggest that private-isolated disruptors will have the following structural signature:
Brokerage – this brand of disruptive actor will search out local brokerage opportunities.
Being connected to two or more victims with less experience/knowledge but who do not know
each other is the ideal situation (Tillman & Indergaard, 1999). However, this brokerage will not
necessarily result in the closure of large scale structural holes due to the fact that the central
broker is operating in bad faith. . Instead of linking two well clustered groups and creating a
bridge between them, the criminology literature suggests that this variety of disruptive actor will
CRISIS AND STASIS 49
broker between relatively unconnected nodes to exploit their lack of information (Tillman &
Indergaard, 1999).
H4a –Private-isolated disruptive actors will be more likely to broker connections between
members of their host network.
Less closure – alternatively, closure means that the victims of this brand of disruption can
compare notes on the disruptive player and coordinate a response (Baker & Faulkner, 2003).
Therefore, individuals who select this strategy will try to avoid closure between their
connections.
H4b- Private-isolated disruptive actors will have lower levels of closure than equivalent
participants in their host networks.
Edge novelty- Because private-disruptive players often exploit people they are connect to
there is little incentive for either party to maintain a long term reciprocal relationship, therefore:
H4c – Private-isolated disruptive players will have a lower average edge age when
compared to other players in their host network.
Once again, linking back to the social capital framework, private-isolated disruptive
actors no not appear to generate any structure associated with bridging or bonding social capital.
They avoid bridging by linking to low information and isolated marks instead of linking well
connected groups (Tillman & Indergaard, 1999). Bonding is also bypassed due to the dangers
inherent in network closure for this style of disruptive behavior (Baker & Faulkner, 2003). The
result is a series of network structures with few public benefits for the rest of the community.
CRISIS AND STASIS 50

Figure 2- Private-isolated disruptive signature, nodes colored red are disruptive
Public-Isolated. Public-isolated disruptions are generally cases where an individual
discovers a mechanism where every player can easily break one or more regulatory structures in
the game without the use of special tools or skills. If related to code, this represents a bug or an
exploit that does not need any special hacking skill or software to exploit. (Bainbridge &
Bainbridge, 2007a). Because the execution of the disruption is equally available to every player
and there is no way to prevent other people from using it, the person who pioneered the
disruption cannot retain a monopoly on its use. When in-game rules or norms are being broken in
this manner, the disruption is often a new strategy or way of looking at the game, which
overturns previous assumptions of how to play. Often labelled as “cheese” or “unconventional”
play styles, the player engaged in the disruption cannot prevent other players from observing or
adopting their innovations, making the results public (Cheung & Huang, 2011; Moeller et al.,
2009).
Structurally, players who participate in public-isolated disruptive behavior are more
likely to inhabit a brokerage role. Finding an exploit or designing a “cheese” strategy requires an
expansive and well-developed knowledge of the game in order to find the loopholes in existing
CRISIS AND STASIS 51
regulatory structures (Meades & Canterbury, 2012). Brokers have had time to embed themselves
in their game network and can draw information or inspiration from the well-connected players
around them (Burt, 2004). This increases the probability that they will find a naturally occurring
gap in the existing regulatory structures.
Structural Signature. This structure and process shares similarities to the diffusion of
innovations process which has been repeatedly described in the network studies literature
(Everett, 2003a). The growth of disruptive strategies fits all of the parameters of a classical
innovation. A select group pioneers the innovation and introduces it to a marketplace. Through
communication (in this case in-game friendships and co-play) the innovation spreads over time,
with particularly well-connected individuals doing a better job at spreading the behavior.
Brokers; are particularly suited to this role because they have access to information from various
parts of the network, which can serve as the seed for a new (disruptive) strategy (Susarla, Oh, &
Tan, 2012; Valente & Davis, 1999). They also have both the connections and reputation to cause
other members of the network to notice and adopt their strategy.
Using a “cheese” strategy generally means that the disruptive player is able to access
abilities or power that is not normally available to their non-disruptive counterparts (Moeller et
al., 2009). As a result, forming long-term relationships of mutual support and reinforcement
becomes more difficult as “cheesers” often find themselves on the wrong side of local norms
regarding fairness and equality (Moeller et al., 2009). The need for friends and allies decreases if
a bug or exploit can provide all of support needed to achieve a goal. Therefore, the local bonding
social capital structures for public-isolated players may be less developed. To summarize:
CRISIS AND STASIS 52
Edge proliferation – because information regarding this type of disruptive ability can
freely circulate within a community’s network, players who are more tied into these processes
are more likely to discover/be informed of a potential strategy (Hau & Kim, 2011).
H4a- Public-isolated disruptive players will have more connections than their non-
disruptive counterparts in the surrounding network.
High brokerage – disruptive players also benefit from linking numerous different well-
connected communities within a network. This increase the amount of information available to
the disruptive player and provides more opportunities for acquiring a useful exploit.
H4b- Public-isolated disruptive players will be more likely to be in a position of
brokerage in their host network.
Low closure – because their success derives from an unintentional breakdown in
regulatory systems, disruptive players are less dependent on friends and allies within the game’s
network. As a result, these structures may be less well developed non-disruptive players with a
similar level of experience.
H4c- Public-isolated disruptive players will have lower levels of network closer then
equivalent members of the host network.
Within the social capital framework, public-isolated disruptive players can be seen as
over emphasizing bridging social capital at the expense of bonding. By linking together disparate
parts of a network, a disruptive player can stay up to date on the latest exploits. However, these
powerful strategies can degrade the need to collaborate with others over the long term, damaging
bonding capital.
CRISIS AND STASIS 53

Figure 3-- Public-isolated disruptive signature, nodes colored red are disruptive

Induced Disruption. Within the two-by-two framework established earlier in this
chapter, one quadrant remained unfilled, the public-collaborative. This represents a special case,
in many situations public-collaborative actions are explicitly sanctioned and included within the
rules of a gaming community (Ang & Zaphiris, 2010). However, there are cases when these
actions can drift outside of established regulatory structures, not because they break through the
barriers, but because the established boundaries shift to make previously sanction behavior
disruptive.
This dissertation tackles the endogenous forces behind disruption, originating within a
community. However, developers are also embedded within their own networks of competition
and cooperation (Zackariasson & Wilson, 2008). Exogenous shocks from these connections can
lead to shifts in the regulatory structures of a community independent of the level of disruptive
behavior. In some cases, these changes may shift the boundaries in such a way that previously in-
bounds variations are pushed out of bounds, creating a situation of induced disruption.
CRISIS AND STASIS 54
An example of induced disruptive behavior is player protests where individuals act en
masse to break a developer’s rules in order to make their voice heard regarding an element of the
game (Edwards, 2011; Jenkins, 2006b, 2006d; Ward, 2006). There have been numerous recorded
instances where a group of players have banded together and disrupted the game for other users
through collective actions like destroying the game’s economy, blockading critical sections of
the game world or bringing down servers (Carter, Woodford, & Bergstrom, 2013; Jenkins,
2006d). Unlike other disruptive behaviors these actions have generally emerged as a result of
changes, as opposed to prompting them.
The benefits from a protest are public but it requires group coordination, so this form of
disruption faces the classic collective action problem. Individuals who dislike changes to the
rules or have found their existing strategies made illicit by a shift can choose not to participate in
future disruptions, free riding off of the efforts of others while still reaping the benefits
(McAdam, 1999; Olson, 1965a). For any mobilization under these conditions to be successful it
needs to offer strong social incentives to participate and be able to draw in a wide range of
participants. Committed members of a community are the most likely to exercise their voice in
response to changes despite the freeriding incentive due to their sunk costs in the existing
network (Hirschman, 1970)
Social incentives and sanctions often emerge out of the clustered structures associated
with bonding social capital. The overlapping, reciprocal ties formed through network closure
mean that if a person embedded in a protesting network chooses to free ride, they will face
scrutiny and sanction from all corners of their network (Diani & McAdam, 2003). This increases
the likelihood that entire clusters will mobilize in order to prevent allegations of free-riding from
close friends or allies.
CRISIS AND STASIS 55
This cluster-based spread is further supplemented when brokers within a community’s
network choose to participate as well. Brokers enable the diffusion of a message across tightly
clustered groups by assuming leadership roles, further increasing the collective clout of the
disruptive behavior (McAdam, 1999).
Since these incentive structures align with the network patterns that are frequently
promoted within games, this type of disruptive behavior does not induce change by damaging the
overall structure of the community. Instead, powerful, centralized actors have found themselves
in a situation whereby their normal patterns of behavior have become disruptive. Given the exit
cost associated with leaving a community, the first response to this development is to exercise
their voice (protest), but in cases where this does not work, leaving often becomes the attractive
option (Blodgett, 2009; Edwards, 2011). This places the developer in a bind, either re-legitimize
the well-networked core sections of their community or risk having them leave, gutting the social
fabric of the community and further damaging their product.
Given the need for both bridging and bonding, this brand of disruptive player should be
rich in structural signatures associated with both types of social capital:
H5a- Public-collaborative disruptors will be more experienced than non-disruptive
players due to the need for a strong motivating factor to push them to protest.
H5b -Public-collaborative disruptive players will have higher than average levels of
network closure given the size of their local networks due to the in-game incentives for
coordination and to prevent free riding.
H5c- Additionally, public-collaborative disruptive players will assume brokerage
positions both as a result of their play style and the need to internally coordinate their disruptive
behavior.
CRISIS AND STASIS 56
Conceptual Summary
This dissertation set out to examine the endogenous sources of change over time within
online communities (rule-governed, technologically mediated spaces of interaction with a
number of affordances and restrictions). Using gaming communities as an exemplar, and
applying both the variation selection and retention and punctuated equilibrium frameworks, a
picture emerged of a two level system. Within the community, members constantly strive to find
optimal strategies within a space prescribed by rules, norms and code. Developers are more
constrained, operating with institutional momentum and overhead; consequently, they are
reticent to make significant changes without good reason. Among participants, variations
occasionally press up against regulations and a small subset of participants are motivated to
attempt to break through existing controls. If successful, these disruptive behaviors provide
alternative routes to success within the community without providing the positive network
effects associated with a healthy community. This degrades the stocks of social capital within the
community and threatens its long-term health. Consequently, rising levels of disruptive behavior
represent a crisis point which is associated with shifts in policy from the developer.
The next chapter introduces the datasets and methodology which support the
aforementioned hypotheses. In addition, this section will lay out the various demographic and
social network variables common across each study. After laying out these commonalities the
remainder of this dissertation will explore each study and work towards supporting or falsifying
each hypothesis.
CRISIS AND STASIS 57
Chapter Four- Methods
Data Sources
The majority of the data supporting the hypotheses laid out in the first chapter comes
from server logs associated with a wide variety of games and gaming communities. The use of
log data has become increasingly popular within the field of Communication over the past ten
years (D. Williams & Xiong, 2009). Logs offer large-scale, detailed datasets of social
interactions and performance within a community. In addition to their detail and granularity,
server logs also offer a number of advantages specific to the focus of this examination, namely
the ability to gather longitudinal data and identify disruptive players relatively easily. The
following section lays out the host communities for each study in the back half of this
dissertation and what type of disruptive play is prevalent in each community. This is intended as
a high-level overview of the various data sources at work within this examination. Each study
will also contain a detailed methods section that describes specific variables unique to that
setting.
Communities of Interest
Each of the four analysis chapters draws from a distinct gaming community that features
particularly high levels of a specific variety of disruptive play. These communities were selected
based on three criteria: the community must have an active developer and player-base; offer a
meaningful communication and interaction between and among these two parties; and feature
some form of in-game scarcity or profit motivation. These restrictions are necessary given the
theoretical framework laid out in the preceding chapters.
The first criterion comes about due to the evolutionary model set out in Chapter Two. In
order for players and developers to interact and engage with each other, both parties have to be
CRISIS AND STASIS 58
present. There are many developers presiding over games that feature dead or dying
communities, extracting what rents they can from the product without making significant
changes. While rarer, there are also cases where communities exist without developers, such as
“abandonware” projects where the developer has turned over their code to the community and
either folded or moved onto new projects (Khong, 2007). Both of these cases lack the necessary
ingredients to confirm or deny hypothesis two, and are therefore excluded.
Similarly, the second selection criterion comes out of the need to identify the structural
signatures associated with disruptive play. Some gaming communities drastically restrict the
level of communication or social engagement available to their players. Games targeted at young
children or at risk populations often lack the means to form long-term relationships or engage in
any form of communication outside of prearranged canned phrases (Burley, 2010). This reduces
the demand on the developer to maintain and support a communication system while protecting
players from harassment or other forms of predation by their compatriots. Without these
mechanisms it is difficult to construct a meaningful social graph that captures the game’s
community. This in turn makes identifying structural signatures extremely difficult and breaks
down any attempt to support or falsify hypothesis three through six.
Finally, the third criterion reflects the need for some form of competitive pressure behind
player behavior. The majority of games feature some form of competition between members of
the community as it provides a fun and easily implemented way to keep people coming back.
However, there are true “sandbox” games where resources are theoretically infinite. These games
approach fun as a form of creative expression where players use these theoretically infinite
resources to create and explore (Duncan, 2011). This changes the motivation structures at work
within the game and alters the dynamics behind social processes. To avoid confounding the
CRISIS AND STASIS 59
central assumptions of this examination, the communities of interest will be limited to those
games that align with the majority of the industry and feature some form of competition.
In addition to these criteria, certain design elements of a game may make it more or less
likely to feature specific forms of disruptive play. As an example, if rewards or benefits are
distributed on the group level, it is extremely unlikely that private-isolated disruption will occur.
This particular variety requires any benefits from disruptive behavior to fall to an individual who
has exclusive rights over it, this becomes difficult with group level rewards, Similarly, if a game
does not have a robust group system due to the lack of a persistent game world or some other
factor, it is less likely that large teams will organize and engage in either form of public facing
disruptive behavior.
The research setting for public-collaborative disruptive behavior is the Steam community.
The Steam Community is a broad cross-game social network that provides a common framework
for player or social engagement between titles offered by various developers (Valve, 2015b). By
using the Steam platform, developers can piggyback on existing infrastructure to distribute their
game and host a constituent community without the technical overhead of supporting these
systems on their own. Since the Steam Community features both a robust social network
consisting of friendships and gaming groups and numerous overlapping layers of code based
regulation across the hundreds of different titles it hosts, it provides a rich base for public-
collaborative disruptive play.
Within the overarching Steam community, a subset of games feature rich economies
based around trading or selling in-game goods. The earliest example of these marketplaces is
attached to the popular shooter Team Fortress Two (TF2). The TF2 marketplace is comprised of
individual traders who follow trends in the game and the marketplace in an attempt to make a
CRISIS AND STASIS 60
profit or earn desirable in-game items (Peterson, 2013). These benefits fall to the individual
traders who maintain their own private accounts. Within the daily wheeling and dealing, specific
players have taken to leveraging their extensive knowledge of prices and skill in social
engineering to defraud other members of the community, breaking the game’s terms of service
(Valve, 2015a) These scammers generally act alone and any profit from their disruptive behavior
falls directly to them. This combination of private actors working in a complex, individualistic
marketplace means that the TF2 community plays host to private-isolated disruptive behavior,
making it an excellent testbed to explore these activities.
Public-isolated disruptive behavior generally takes the form of using glitches or unfair
“cheese” strategies to get ahead within the game (Moeller et al., 2009). These opportunities
generally come about passively as unexpected consequences of design choices made by a
developer. The more complex a game is, the more opportunities there are for two or more
specific rulesets to interact in an unexpected way and create room for a public-isolated disruption
to emerge. Multiplayer Online Battle Arenas (MOBAs) are well known for their complexity and
usually feature hundreds of distinct characters interacting in a high pressure, time constrained
situation (Yang, Harrison, & Roberts, n.d.). The complex interdependence between these various
factors opens up plenty of room for unintended design consequences and private-isolated
disruption. Therefore, Chapter Seven will focus on the Defense of the Ancients 2 (DOTA) and its
associated social networks.
Finally, public-collaborative disruptive play emerges when disruption is induced through
a series of rule changes or other shifts in a game’s regulatory structure. There have been
numerous examples of player intervention after major changes in a game’s regulatory structure
among communities in Asia, North America, and Europe. One of the most significant cases of
CRISIS AND STASIS 61
public-collaborative disruption was the 2011 EVE Incarna protests. Players revolted against the
introduction of micro-transactions into the community and attacked the technical foundation of
the game (Edwards, 2011). EVE provides an interesting testbed for this form of disruptive
behavior as it is a famously unconstrained game. Players are free to form a variety of
associations and organizations and there are very few restrictions on the development of social
networks (Bergstrom, 2013). This provides a large potential feature space which allows structural
signatures to manifest clearly within the community, instead of being occluded or distorted by
restrictions due to the game’s rules.
Defining Disruption
One of the two research questions central to this examination is how disruptive players
differ from their counterparts within a given community. This bifurcation of a community into
players who have been identified as disruptive and those who have not creates a two-group
classification problem. However, this binary classification can be problematic given the shifting
natures of what constitutes disruptive play (Doherty et al., 2014). Like most other human
behaviors disruptive activity is a gradient, with numerous grey areas surrounding its definition.
Activities explicitly banned within one community may be allowed or even encouraged in
another setting with a different slate of regulatory structures (Bergstrom, 2013). Instead of using
a universalist definition of disruptive play to differentiate the various groups within these
datasets, the following examinations will make use of identifiers originating from the community
in question. Since players and developers are the most intimately acquainted with these
regulatory structures, they are also the best equipped to define who is or is not breaking through
them. The four studies that make up the back half of this dissertation explore four distinct
gaming communities, each with a unique data structure that displays unique disruptive patterns.
CRISIS AND STASIS 62
Common Metrics
Due to the diversity of data behind each study, this section will not layout all the
variables at work in each chapter. Different datasets offer different control variables and metrics
that are best addressed in the context of a discussion of that community. However, since the core
datasets for each study are centered on a social network, there are some common features across
studies that speak to the central hypotheses of this paper.
Given the interaction between disruptive play and network structures associate with
social capital, it is critical to establish features rooted within the various social networks that
assess both brokerage and closure in a given dataset. Fortunately, there has been a significant
amount of effort within network science directed towards these efforts. All of the studies in this
dissertation draw from Burt’s measures of effective network size (ENS) and constraint to assess
brokerage and closure at different levels of the network.
Local Brokerage and Closure. Local in this case refers to nodes immediately connected
to a given person. As an example, one’s immediate friends or followers on a social network like
Facebook or Twitter represent a local network. A given player’s local network is their immediate
social neighborhood– the people who they have chosen to directly associate with or have chosen
to be tied to them. Brokerage and closure represent two poles within a local network (Burt,
2001). Brokerage (and the bridging capital associated with it) occurs when an individual is able
to mediate between members of a local network (Burt, 2004). Being the friend that brings two
groups together or someone who organizes a party that brings together connections from work,
play and family, are both examples of local brokerage in an offline setting. Conversely, network
closure (and bonding social capital) represent a relatively closed local network where each
contact knows each other (J. S. Coleman, 1988). These two processes exist in a state of tension,
CRISIS AND STASIS 63
the more closed a local network is the fewer opportunities there are for brokerage and vice versa.
Therefore, using two metrics to gauge brokerage and closure simply adds confusion to a given
model, high brokerage will be correlated with low closure and low brokerage with high closure.
Fortunately, there are a number of local metrics that can capture both concepts at once. One of
the more popular is Ronald Burt’s measure of effective network size (Burt, 1992). ENS can be
expressed as the total number of edges a node has minus the number of redundant edges. In this
case redundancy refers to links to contacts who otherwise know each other. Networks with high
rates of closure will have a small effective network size, while networks rich in brokerage will
have a larger one.
Indirect Brokerage and Closure. Aside from local measures there is also the matter of a
given node’s indirect ties. Local networks such as the examples described above are also situated
in a broader social setting. An individual can be a broker linking together different groups in their
immediate social circle, but if they are embedded in a relatively close network they are
essentially a big fish in a small pond. Similarly there may be “knots” of relatively tight
association in loose, associational networks rich in brokers (Burt, 2001). Many early network
metrics captured this dynamic by considering the entire network at once. As an example,
betweenness centrality considers the number of shortest paths (i.e. the number of: ”hops” along a
given path) in an entire network that run through a given node (Freeman, 1977). However, when
dealing with large scale network data these metrics become both computationally intensive and
unwieldy. The pathway between two people separated by eight intermediaries is not a realistic
channel for information exchange in many social setting, especially if the path crosses group
boundaries. Fortunately there is a supplement to ENS which considers not only the local network
but the indirect network (friends of friends) of a given node (Burt, 1992). Burt’s constraint
CRISIS AND STASIS 64
generates a numeric metric that captures whether a node’s-indirect network is closed or open
based on hierarchy and density. Hierarchy reflects when everyone in a node’s neighborhood
knows someone, even if they do not know each other. While the node’s local network may be
relatively open, any information flowing into their general proximity has to go through this
choke point, constraining the network. Similarly, if a node is embedded in a dense neighborhood
their contacts may not know each other, but they could be indirectly tied through a friend of a
friend, leading to an indirect form of closure.

Figure 4: Comparison of Network Scope of Selected Metrics. The local network for the red node
has an effective size of 6.42 for both networks. But in the left hand network its constraint score is
0.23 while on the right it is 0.20.

Methodological Approach
The remainder of this section will lay out the common methodological approach across
all four studies. The central research questions of this dissertation can be broken down into two
broad categories, those addressing whether disruptive players have a distinct social position
within a given community and those exploring if changes in the level of disruptive play is related
CRISIS AND STASIS 65
to changes in the community itself. Both of these questions require specific approaches which
the rest of this chapter will explore.
Learning Disruption. Given the binary nature of disruptive behavior in each of the
datasets, learning the distinctive attributes of these players can be recast as a classification
problem. By training a model to parse between the two labelled groups (disruptive/non-
disruptive) it is possible to determine which, if any, of the demographic or network characteristic
available differentiate the behavior of the two groups. There are numerous models available for
binary classification. This plurality of choice means that there needs to be a consistent standard
for choosing one approach out of the multitude of options in each subsequent study.
Choosing model selection parameter is not a trivial task; recent developments in the field
of statistical learning have introduced multiple different techniques for assessing the relative fit
and performance of classifiers. At the most fundamental level, there is a division within the slate
of possible approaches between algorithmic techniques, which sacrifice immediate
interpretability for the potential for increased predictive power, and data modeling approaches
that are more interpretable but occasionally less accurate. The statistician Leo Breiman argued
that the division between these two approaches represented a split into distinct cultures, which
place a different emphasis on the tension between interpretability and predictive accuracy
(Breiman, 2001a). These two goals come into conflict when selecting a model, as newer
statistical learning techniques offer a non-parametric and robust approach by sacrificing off-the-
shelf interpretability. There is no easy solution to this problem, instead the classification models
featured later in this analysis attempt to split the difference and use both categories of techniques
in conjunction. By using complex algorithmic approaches tuned to maximize predictive
accuracy it is possible to determine disruptive players are structurally distinct from their host
CRISIS AND STASIS 66
community. But to unpack how these differences manifest, these models need to be interrogated
using a combination of information loss/variable importance metrics and triangulating the results
with a more interpretable data model, and assessing the relative agreement between the two
approaches. This approach plays the strengths and weaknesses of both techniques at the cost of
increased complexity due to the proliferation of possible models.
Disruptive Quirks. Before laying out this model selection and validation strategy, a short
detour highlights the unique challenges associated with classifying disruptive players. Due to the
social stigma and regulatory structures often arrayed against disruptive play, it is generally not a
very common phenomena (Blackburn et al., 2012). The level of disruption in the various datasets
laid out earlier in this chapter varies, but does not cross five percent of the total sample. The
relative rarity of disruptive play leads to a serious problem of class imbalance that can pose
problems for many forms of statistical analysis.
Class Imbalance. Because disruptive behavior is so rare, a standard statistical classifier
can achieve high levels of predictive accuracy by simply assuming that all players in a
community are non-disruptive. As an example, if 50 players in a hypothetical gaming community
with a total population of 1000 were using exploits to break down the game’s code, a classifier
could achieve 95 percent accuracy simply by stating that no one was cheating (King & Zeng,
2001). The exact route that an unadjusted classifier takes to end up at this conclusion varies,
based on the technique, but the “accuracy paradox” produced with highly unbalanced datasets is
generally a result of there not being enough cases in the minority class to prompt an adjustment
in the fitted solution (Zhu & Davidson, 2007).While this does generate an impressive accuracy
score it is also functionally useless for researchers of any stripe as the model is not actually
providing any information aside from the established fact that disruption is relatively rare.
CRISIS AND STASIS 67
Resampling. There are two popular approaches to dealing with the class imbalance
problem. Resampling takes subsets of an existing dataset and either oversamples the rare class or
down-samples the majority class to create a balanced dataset which is then analyzed (Chawla,
Bowyer, Hall, & Kegelmeyer, 2002). Since the data is balanced the model is forced to consider
difference between the two classes as the baseline accuracy for randomly guessing what is now
50 percent as opposed to 95 percent. This process is not without a downside. For some models,
such as logistic regression, this means excluding data from the analysis to achieve a balanced
sample, not an ideal practice (Ertekin, Huang, Bottou, & Giles, 2007). However, approaches that
make use of bagging (bootstrap aggregation) can create balanced samples without throwing away
data (Galar, Fernandez, Barrenechea, Bustince, & Herrera, 2012). Bagging approaches create
multiple “weak learners” such as decision trees. Each learner is fit with a random subset of the
available variables on a random subset of the dataset. By repeating this process thousands of
times, a large number of relatively weak models are created. For each case, the predictions of this
multitude of weak learners are aggregated and a prediction is made based on the more popular
verdict within this multitude (Breiman, 2001a). Since bagging approaches are already taking
subsamples of the data during the process of fitting the model, making these subsamples
balanced ensures that the results will not simply settle on a conservative estimate without
throwing away data (Blaszczyński, Stefanowski, & Idkowiak, 2013).
The two main statistical learning approaches used in this dissertation are random forests
and stochastic gradient boosting. Both of these techniques are ensemble methods built out of
“weak learners”, in this case classification and regression trees (CART). These decision trees are
non-parametric learners which attempt to find points in the data where a split would increase the
overall purity (i.e. the proportion of cases of the same class) on either side of the rule (Breiman,
CRISIS AND STASIS 68
2001a). These rules can tackle interval or nominal data. As an example, a simple tree fit on
demographic data may first split a dataset based on gender, with one branch of the “tree”
containing women and the other men. Subsequent branches can split on points in other variables
such as age or income until the final “leaves” of the tree contain subsets comprised mainly of one
class or another. Using the Gini coefficient to measure the relative equality or inequality of cases
on each side of a rule, the decision tree algorithm greedily searches through a given problem
space for the optimal split points (Breiman, 2001a).
While decision trees are scale invariant, non para-metric, robust against outliers, and can
handle categorical and continuous data, they are also prone to overfitting. That is to say the tree
will continue to establish rules until it is so closely fit to a given dataset that the delicate balance
of conditions created for that specific case is not generalizable to new datasets. Some approaches
attempt to work around this problem by “pruning” the tree and limiting it to only a handful of
cases (Esposito, Malerba, Semeraro, & Kay, 1997). This decreases the error due to bias within
the learning process but increases the susceptibility to variance in the dataset. Ensemble methods
can control for overfitting without dramatically limiting predictive power (Breiman, 2001a).
Random forests are the simpler of the two algorithms and provide a good demonstration
of the process. Instead of fitting a tree on the entire dataset, the forest will select a random subset
of the data. It is at this point that balanced sampling can take place without any data loss. Using
this constrained subset, a small classification tree is fit, using a random subset of the available
predictors in a given model. The smaller dataset and constrained number of predictors mean that
the tree will not necessarily be a good predictor, but its verdicts for the remaining cases are
stored regardless. After one constrained tree is fit, the process is repeated to create a battery of
“weak” learners, each fit on a different subset of the data with a random selection of predictors.
CRISIS AND STASIS 69
Predictions are made by aggregating the verdicts of each of these weak learners and taking the
most prevalent prediction (Breiman, 2001a). By constraining the features and data available to
each learner, the Random Forest can avoid some issues with overfitting while retaining the
benefits of the decision tree approach. Gradient boosting also uses an ensemble approach based
around decision trees, but before laying out how it works it is important to discuss case
weighting, as it plays an important role in this algorithm.
Weighting. Another approach to dealing with class imbalance is weighting. Weighting
differentiates the cost of misclassifying a specific case within the dataset. By adjusting the cost
function associated with misclassifying a disruptive player it is possible to force a model to
explore differences between the two classes within the dependent variable. By weighting
disruptive players higher than non-disruptive players the model is forced to explore possible
differences between the minority and majority class or risk misclassifying a player (King &
Zeng, 2001). Weighting is more difficult to establish than resampling as each dataset has an
optimal balance. If disruptive players are weighted too high, the model will inflate the number of
false positives (non-disruptive players classified as disruptive). Alternatively, weights that are too
low may not ameliorate the class imbalance problem. Rare-events logistic regression developed
by King and Zeng use weighting and a number of probability correction techniques to extend
logistic regression into cases with high degrees of class imbalance (2001). This provides a route
for applying data models such as logistic regression to the classification problems in the
subsequent chapters without discarding data through balanced subsampling.
Weighting also plays a role in stochastic gradient boosting. This approach uses ensembles
of constrained decision trees with randomly selected predictors??? similar to a random forest.
However, instead of training these weak learners on randomly selected subsets of the data,
CRISIS AND STASIS 70
boosting prioritizes the misclassified cases from an initial classification model and attempts to
append this first attempt with subsequent models that are directed towards initial errors (J. H.
Friedman, 2002; Meir & Rätsch, 2003). The model is built iteratively, with every step providing
a number of misclassified cases that serve as the basis for an add-on model, which attempts to
learn the root cause of the error. This iterative approach allows gradient boosting to focus on
difficult to classify cases and generally leads to better accuracy than random forests. There is an
increased danger of overfitting so it is helpful to fit both models as a safeguard.
Ground Truth. The issues presented by class imbalance are further complicated by the
fact that in all of the datasets within this dissertation the identification of disruptive players may
not be completely accurate. As described earlier in this chapter, each study uses a data source
drawn from the community in question to isolate disruptive players in the dataset. It is extremely
unlikely that a developer or play group can isolate every single actor engaged in disruptive
activities in the dataset. The true “ground truth” of who is and is not disruptive remains
unknown. This makes assessing the relative predictive power of a given model tricky. Each
classifier outputs a relative probability that a player is disruptive, with a certain threshold being
the line past which an individual is flagged by the algorithm. Setting this threshold low means
that there may be a high number of false positives in the dataset (Muhammad A. Ahmad et al.,
2009). Alternatively, a very high bar means that some disruptive players will be categorized as
non-disruptive.
This creates a dilemma for assessing the performances of a predictive model. As the
ground truth is not explicit, setting a classification threshold becomes tricky, as if it is too high or
too low, it may inflate or deflate measures like precision, recall or accuracy. Subsampling and
weighting cannot help as most algorithmic models need to be validated against an unaltered
CRISIS AND STASIS 71
dataset which was held out from the training data in order to prevent overfitting. This means any
performance measures are being assessed in an environment with a high degree of class
imbalance, which artificially deflates a number of performance measures such as F1 score by a
significant amount. The solution is to consider all possible cutoff points, from the most liberal to
the most conservative. By plotting the true positive rate and false positive rate of a model at any
given cutoff point, a curve is created. At one end is the most liberal interpretation where even the
slightest evidence will classify a player as disruptive. On the other end is the most conservative
approach where the data has to be utterly overwhelming before a verdict is handed down. The
area under this curve (called a receiver-operator curve or ROC
2
) represents the predictive gains
of the model when compared to random chance, taking into account all possible classification
thresholds (Bradley, 1997). Area under the curve (AUC or ROC-AUC) provides a flexible
assessment of performance when the ground truth is unknown. Additionally, it is extremely
resilient to skewed test datasets, providing a good all-around testing criteria (Jeni, Cohn, & De
La Torre, 2013).
In order to prevent overfitting and an inflated ROC curve the statistical learning and
logistic regression models will be constructed based on a held out “training” dataset consisting of
80% of the total sample. The remaining 20% will be held out as a test dataset. By blinding the
model to the actual status of a given participant in the test dataset and comparing the predicted
probability generated by the model against the actual status of an individual it is possible to test
the overall predictive capabilities of the model on data which it was not constructed on, partially
controlling for the possibility of overfitting (James, Witten, Hastie, & Tibshirani, 2013).

2
The name “receiver-operator curve” represents the fact that this approach was developed in the Second World War
to judge the predictive capabilities of radar operators who classified contacts as enemies or friendlies.
CRISIS AND STASIS 72
Methodological Protocol. Given the difficulties of working with imbalanced data and
the tension between interpretability and predictive power, the subsequent studies will all take a
compromise approach towards classifying disruptive players. This entails fitting multiple
classifiers per dataset, two based on statistical learning techniques (generally bootstrap-
aggregation and up sampling for class imbalance) and a logistic regression weighted to handle
the fact that the dependent variable is relatively rare (Blaszczyński et al., 2013; King & Zeng,
2001). The performance of the two models will be assessed and compared using the area under
the ROC. Specifically, if the ROC curve for a given technique achieve the highest average score
across all the sampling periods that model will serve as the primary basis for the subsequent
analysis. This does not mean that the contributions from other under-performing models will be
discarded, but these results will be weighted less. If the statistical learning approach functions
significantly better than the logistic regression it becomes the primary model, or vice versa. In
the event of a tie or close score, the interpretability of the rare-event logistic regression means
that it wins in a draw.
In the event that the statistical learning approach achieves a higher area under the curve,
the next step is to assess the relative importance of each variable in the model. Unlike regression,
there are no extractable coefficients that can be examined to determine the magnitude and
direction of a specific variable’s effect. These effects can be approximated using a combination
of variable importance and variable prediction plots.
Variable Importance. This approach produces a ranking of variables fed into a given
model by their relative importance to the final scores. Generating these scores is relatively
simple. Once a model has been fitted, the data it was trained on is fed back into it. If the data are
unchanged, the model will produce the exact same results; however by changing the data, the
CRISIS AND STASIS 73
results will also shift. One by one, each feature in the model is replaced with randomly generated
“garbage” data and the model is refit (Liaw & Wiener, 2002). The relative performance of the
model between the unaltered data and each subsequent iteration is then assessed. If the
performance drops dramatically when a specific variable is replaced with random noise, this
suggests that the variable in question is particularly important to the functioning of the model.
Alternatively, if performance stays stable on both sets of data, the variable is not as important to
the classification process.
Variable Prediction. Variable importance provides a ranking of features within a model
based on their relative importance, but it does not unpack the direction of the feature’s influence.
Determining if a certain score is positively or negatively associated with the likelihood of
engaging in disruptive play requires an extra step. By taking the most important variable
identified within a model and plotting them alongside the predicted probability of engaging in
disruptive behavior, a two dimensional feature space is created. A smoothed regression line can
be fit within this space, capturing how the probability of disruptive behavior changes as the
variable in question goes up or down (Liaw & Wiener, 2002). Since the predicted probabilities
which form the basis of a variable prediction plot are the result of a complete model the resulting
visualization captures how one variable influences the likelihood of falling into a certain
classification category while controlling for the other predictors in the model. This provides
directionality alongside magnitude for the results coming out of the statistical learning process.
These results can be triangulated with the logistic regression model. If the relative
ranking of variable importance and the direction of effects is similar between the two models this
provides further support for the variable importance ranking and variable prediction plot. If the
CRISIS AND STASIS 74
two disagree then it means that an added degree of caution is needed when interpreting the
results and any discussion of the findings needs to be read in light of the disagreement.
Change Over Time
The classification approach helps assess if disruptive players occupy a distinct space
within their host communities. However, it does not directly address how developers and non-
disruptive players respond and change in response to disruptive play.
Assessing changes from the developer’s side is relatively easy due to the need to
document any changes in the legal or technical infrastructure of a community. Whenever a
“patch” (change in the game’s code or rules) is released, a series of notes describing the changes
is bundled with it. By delving into this archive and the various para-texts associated with it,
points of divergence and change can be identified in the history of the game. These points can be
correlated in the relative rates of identified disruptive behavior over time. Using time series
based anomaly detection approaches it is possible to determine if significant spikes or dips in
disruptive behavior are correlated with the introduction of major changes by the developer
(Neuvirth et al., 2015). Specifically, this study draws upon the Seasonal Hybrid Extreme
Studentized Deviation (SHESD) anomaly detection approach. This technique was popularized as
twitter as a mechanism for detecting anomalous bursts of activity in heavily networked areas.
The SHESD approach decomposes a seasonal or non-seasonal time series and establishes an
expected floor and ceiling of activity within a given dataset. These limits take into account any
existing seasonal variations as well as trends away from a stationary time series. Any spikes or
dips that exceed the established boundaries are flagged as anomalous, allowing for further
qualitative investigation of a specific moment within a given time series. The SHESD approach
has found numerous applications within academic publications and among industry practitioners
CRISIS AND STASIS 75
as a simple, robust technique for anomaly detection in large datasets (Aniszczyk, 2015; Kelly &
Ahmad, 2015; Neuvirth et al., 2015). This method does not function in all of the subsequent
studies. The EVE Online case takes in a concentrated time frame without historical data making
anomaly detection fruitless. Nevertheless, it is still possible to qualitatively assess if the
developer faced a critical juncture in their business strategy based on the media, their own
response to the crisis and subsequent actions.
This developer focused strategy has several characteristics which make it preferable over
directly assessing the level of disruptive play within a network. First, much of the information
used throughout this analysis is only available as separated cross sectional datasets with the
timing of collection heavily influenced by data availability and sampling limits. As a result, if the
chunks do not straddle a period of increased disruptive behavior closely it is extremely difficult
to establish a baseline of “health” for a given network and then say that this metric has increased
or decreased. Developer strategy is constantly updated due to the continuous flow of information
provided by modern analytics, allowing them to be much more responsive to the status of a
community. Additionally, because developers have legal obligations to the community and
shareholders major shifts in strategy are generally documented through the media, public
relations and patch notes associated with shifts within the game’s regulatory structure. This
provides a more granular historical record which can be paired up against information on the rate
of disruptive play over time. The downside of this approach is that it is strictly correlational.
While documents and existing interviews can strongly suggest that a shift was influenced by
certain player behaviors further qualitative work is needed to contextualize the results.
CRISIS AND STASIS 76
Moving Forward
Having laid out the datasets, methodological approach and theory behind disruptive play
all that is left is to move onto the studies that comprise the remainder of this examination. Each
of the four studies will briefly expand and situate the theoretical concepts introduced in the first
chapter before laying out the specific operationalization of the relevant variables for the
particular dataset(s) in question. Then the result of the two classification models will be fitted to
determine the structure position of disruptive players within a given community. These results
will help contextualize the anomaly analysis comparing rates of disruptive behavior with the
introduction of significant changes to the structure of a community.

CRISIS AND STASIS 77

Chapter Five- Private-Collaborative Disruptive Behavior
Private-collaborative disruption is one of the most studied elements of the disruptive play
typology advanced in Chapter Two. Generally speaking, this disruption centers around sub-
communities embedded within a game who collaborate to produce cheats, hacks or other third-
party tools in order to breakdown regulations embedded in the game’s code (Consalvo, 2007).
These changes are generally designed to give the user an unfair advantage and do not require the
consent of other players within the game (Consalvo, 2007). As an example, a popular disruption
is the “wall-hack.” This changes the content of the game’s code for a particular player so that
they can see through otherwise opaque barriers within the game (Laurens, Paige, Brooke, &
Chivers, 2007). Wall hacking is unfair as it gives one player an advantage not normally available
to other members of the community. Additionally, the use of a wall hack does not require the
consent of the other players. Indeed, its successful use depends on other players not suspecting
that someone is hacking so that they will continue to play with and lose to the disruptive player
(Consalvo, 2007). Developing a wall hack often requires a team of coders and other individuals
familiar with the game’s code who work together to bypass anti-tampering mechanisms installed
by the developer (De Paoli & Kerr, 2009). However, once a hack has been successfully created it
can be bundled into an executable file and distributed easily to less technically inclined players
seeking an advantage.
This section will start by exploring the social dynamics of hacking and other forms of
private-collaborative disruption. After demonstrating how these processes translate into
distinctive structural signatures which differ from the normal process of social capital
production, these predictions will be situated and contextualized within the cross-game Steam
CRISIS AND STASIS 78
dataset, operationalizing the hypothesis put forward earlier in this dissertation. With this base
established, the remainder of the chapter will explore the relationship between disruptive play,
network position and punctuated equilibrium using a combination of classification analysis and
anomaly detection.
Cheating, Hacking and Collaborative Disruption
In her 2007 work Cheating: Gaining Advantage in Video Games, Mia Consalvo argues
that cheating and other behaviors labelled as disruptive in this dissertation can be understood as a
social process. Consalvo describes how players, who hack or break through the code and rule-
based regulatory structures that bind a gaming community, use these activities as a form of social
activity. The scope and direction of these social ties takes on a two-faced character, with one side
looking inward towards a sub-community of disruptive players, while the other faces outward
towards the host community.
The inward-facing elements of the private-collaborative community come about due to
the dual needs for recognition and collaboration. The need for recognition emerges from the
varying social dynamics that drive players up against a game's rule or code. Like their
counterparts who stay within a game's regulatory system, disruptive players are motivated by a
number of different factors. Qualitative studies of cheaters in various games have argued that
players are often motivated by a need to achieve mastery over their social environment (Ribbens,
Poels, & Lamotte, 2011). While the social signs of mastery can emerge through time, effort and
skill, the ability to break down and manipulate the code or rules that form the basis for the
game's reality offers an alternative route (Consalvo, 2007). For some disruptive players, the
presence of a terms of service contract and restrictions which keep them from accessing the
game’s code represent challenges. The best way to show mastery over these challenges is to
CRISIS AND STASIS 79
understand, manipulate and control them (Ribbens et al., 2011). In many ways, breaking through
regulatory structures becomes a form of game within a game. Disruptive players work to unpack
the various security features integrated into the community while developers or other security
services work to stop them. Rewards come from finding exploits, developing new tools and
demonstrating mastery over the regulatory structures that comprise the community (De Paoli &
Kerr, 2009).
Given the game-like nature of this form of disruptive play it is not altogether surprising
that many of the dynamics surrounding reputation and prestige among disruptive players also
mimic the processes among their more regulated counterparts. Skilled non-disruptive players are
often feted for their achievements with rewards, social esteem and leadership roles within the
game (T. L. Taylor, 2012). Disruptive players who find new ways to cheat or breakthrough
existing regulatory structures can use these innovations to gain status among fellow disruptive
players. Many communities feature “cheat wars” that are illicit competitions where both sides are
fully aware that their competitor within a match is breaking through one or more in-game
regulatory systems (Reddit User tomatocarrotjuice, 2015). The point of the game therefore
transforms from a contest of in-game skill to one of knowledge about the game, with the person
or team with the best means of disruption winning.
The internal dynamics of the private-collaborative disruptive communities are further
enhanced by the fact that the process of developing tools or techniques for breaking through
regulatory systems is often technically intensive. Developers in charge of a community generally
take steps to protect their code against tampering by users in order to maintain the integrity and
uniformity of the in-game experience for all members of the community (Lan, Zhang, & Xu,
2009; Laurens et al., 2007). Breaking through these barriers often requires finding previously
CRISIS AND STASIS 80
unknown vulnerabilities or reverse engineering a significant volume of code. This process
requires a range of technical skills including programming, network administration and computer
exploit analysis, among other talents (De Paoli & Kerr, 2009). The odds of a single disruptive
player having all of the abilities needed is relatively low, although isolated developers do exist.
Instead, disruptive players often gather in forums or other forms of para-association at the fringes
of the community to exchange information and tips. This helps create a collaborative
environment not unlike a software development team, where individuals with different
specializations work in tandem to create new pieces of code.
The technically proficient developers generally make up a minority within a private-
collaborative sub-community. The vast majority of disruptive players are users who take and
apply cheats or hacks developed by these developers to their own games (Castronova, 2007).
This segment of enthusiasts does not directly contribute code or exploits to the community, but
they still play a role in the development of new disruptive techniques through two distinct
processes.
The first is simply by providing a social and fiscal incentive structure that supports
disruptive developers. Many tools require payment or the application of a special key to become
active. Disruptive developers subsidize their efforts by selling exploits on a variety of third-party
sites (“Chod’s Cheats,” n.d.). The enthusiast community who uses these tools enables the further
development of new forms of disruption through their wallets (Ribbens et al., 2011). Even if the
tool in question was released for free, the non-technical enthusiast community can still reward a
developer. Social accolades, praise and admiration on forums or other para-texts mean that even
freeware developers can receive rewards for their products (Scacchi, 2010).
CRISIS AND STASIS 81
In addition to stimulating demand by providing social and fiscal incentives, the enthusiast
community also forms a kind of internal software testing market. As players use the tools
developed by the more technically inclined members of the community they increase the number
of disruptive players in the host community. This has several effects. First, in some situations it
provides social legitimacy to the disruptive behavior. Qualitative studies have demonstrated that
one of the major factors behind the decision to cheat within a game is the sense that “everyone is
doing it” (Consalvo, 2007; Dumitrica, 2011). The perception that cheating is common lowers the
fear of punishment due to the ubiquity of disruptive players, and it incentivizes further disruption
as previously uninvolved players seek out tools to help them level the playing field. Additionally,
the enthusiast community creates a form of technological churn within the disruptive sub-
community. Each time a tool is downloaded and used, the likelihood of it being discovered by
other members of the community or the developer increases (Lan et al., 2009). As tools are
discovered, the regulatory structure of the game is reinforced against that particular flaw or
exploit which made a breakthrough possible, removing the efficacy of that particular form of
disruption. This necessitates the development of new tools to enable disruption, creating new
demand and stimulating further development.
The complex, inward looking social dynamics of the public-collaborative disruptive
community are complemented by the outward facing relationship between disruptive players and
their host community. This relationship is defined by the central tension between the need to be
noticed and the need not to be seen as a disruptive player (Consalvo & Vazquez, 2015). The
advantages offered from tools like wall hacks serve to give a player an unfair edge over their
compatriots in the game. But for that advantage to be meaningful, the non-disruptive players
need to believe that their defeat was due to their opponent’s superior skill, rather than a
CRISIS AND STASIS 82
disruptive technical intervention (Consalvo, 2007). Maintaining this fiction has an advantage for
the disruptive player as well. It makes them much less likely to be reported or banned from the
community for their actions and enhances their image within the community due to the
appearance of skill. This creates a necessary but contentious dynamic between the disruptive
sub-community and their host network. Disruptive players need the host community to witness
them, but cannot be too overt without being detected or banned (Kuo, 2001). Meanwhile, the
host community is generally aware that cheaters exist and in some cases will believe that truly
skilled players are cheating, creating a false positive (“Have you been accused for hacking
before?,” 2011).
In order to untangle how this complex web of inward- and outward-facing social
pressures influences the network position of public-collaborative disruptive players, it is helpful
to situate these processes in existing research on illicit organizations. In recent years, the increase
in data collection by law enforcement has led to a number of collaborations between
criminologists and network scientists to study so-called “dark networks” (Bakker, Raab, &
Milward, 2012; Hu et al., 2009; Raab & Milward, 2003). These social formations represent the
connections between members of an organization who have an interest in keeping their
associations and actions secret. As an example, drug manufacturing and distribution networks are
an oft studied dark network. (Bright et al., 2012). The members of the network have to interact
with a host network (everyday society) while operating within an embedded and secretive sub-
community.
While disrupting an online gaming community is in no way equivalent to the challenge
the illegal drug trade or other criminal networks pose to the law, it does mean that insights from
these studies can be applied to the case of disruptive play. In many ways, there are parallels in
CRISIS AND STASIS 83
the incentive structures and affordances between the two cases. Members of multifaceted dark
networks like drug cartels have to coordinate and work together to produce a complex, illegal
and hard-to-manufacture product (Morselli, Giguère, & Petit, 2007). These demands require
close coordination among members of the sub-community. Additionally, the community needs to
be robust if certain nodes are removed (due to arrest or other circumstances) and adapt to the
changing social situation (Duijn, Kashirin, & Sloot, 2014). As a result, those dark networks that
face complex, high-coordination activities tend to assume a closed structure rich in triadic
closure and mutual connections (Morselli et al., 2007).
Network closure provides several advantages for illicit networks. Closed triads and
groups display a high degree of redundancy. In networks with high levels of closure, removing a
single node is unlikely to fragment the network as the connections between the removed node’s
neighbors assume the communication burden. Network closure also makes close coordination
and cooperation easier as the mutual ties between the members of the sub-network provide a
framework for reciprocity and trust between all members (Morselli et al., 2007). These
incentives mean that unless there is a strong and compelling reason to avoid a closed structure
most illicit networks feature a tightly clustered coordinating central core.
These impulses may also influence private-collaborative disruptive players. For
developers, closure helps facilitate the exchange of information in order to develop new cheats or
hacks. Enthusiasts are also driven towards closure and homophily due to the latent hostility
between disruptive and non-disruptive players within a gaming community (De Paoli & Kerr,
2009). Friendships that cross this boundary run the risk of either exposing a given player’s
disruption or dissolving the friendship once the connection learns of their contact’s behavior
(Morselli, 2010). Ultimately, research suggests that the relationship between disruptive players
CRISIS AND STASIS 84
and the host community is performative. Disruptive players perform by using their technical aids
for the rest of the community instead of exposing their true level of skill and commitment and
forming long-term relationships with their non-disruptive counterparts (Ribbens et al., 2011).
This performative element does not remove the fundamental need for friendship and social
contact, but it restricts the possible outlets for this impulse to other players who understand and
participate in the intricacies of disruptive play.
H3a- Individuals engaged in private-collaborative style disruptive behavior will be more
likely to connect with other disruptive players and therefore display a higher than expected rate
of homophily.
H3b – Private-collaborative disruptive actors will have a higher rate of network closure
than their counterparts within a host network.
In addition to the tendency towards homophily and closure, illicit networks tend to avoid
both direct and indirect brokerage. By occupying a brokerage position, a given node becomes an
important figure within a community, bridging structural holes and knitting together otherwise
disparate aspects of their social circle. While this situation offers a number of advantages to the
various members of the network, it is not very redundant in the face of attack (Holme, Kim,
Yoon, & Han, 2002). If the broker is removed from the network for whatever reason, the two
groups run the risk of drifting apart from each other and fragmenting (Krebs, 2002a).
Additionally, given the tendency of illicit networks to display clustering and redundancy, nodes
in a brokerage position are likely to bridge the structural holes between a tightly clustered illicit
subgraph and the broader network. This path increases direct social contact between the two
sides and increases the likelihood that the disruptive component will be uncovered by unfriendly
members of the host community (Morselli, 2010). As a result, active members of illicit networks
CRISIS AND STASIS 85
tend not to directly engage with both their compatriots and the outside network, mediating any
ties to the latter group through indirect routes. This reduces the likelihood that a member of an
illicit network will hold a brokerage role. If these findings hold true for disruptive players it
follows:
H3c – Private-collaborative disruptive actors will avoid positions of brokerage and have
more network constraint than their host network.
Having established the structural incentives that criminology and network science suggest
may play a role in private-collaborative disruptive play, it is important to ground these insights in
an existing gaming community. By exploring how the rates of disruptive play shift and change
over time within this community provides a mechanism for assessing H2.
H2- Large-scale changes by administrators of online communities will occur following
increases in disruptive behavior.
At this point it is important to note why H2 is focused primarily on developers.
Measuring developer reactions to disruptive play has two advantages. First, developers generally
document their interventions and changes (Meades & Canterbury, 2012). This provides a much
more robust set of data regarding policy shifts over time than what is available through public-
facing APIs. Directly measuring the health of a network before and after a spike in disruptive
behavior requires granular information on both sides of the event and is thus highly contingent
on when data was collected or made available, a factor that does not always align perfectly with
the needs of researchers. If there are cases where network data are available to supplement
network logs, this examination, and any of the subsequent analysis chapters will exploit these
cases if they appear. Developer reactions therefore represent the minimum possible threshold of
regularly available public data needed to evaluate H2. That being said, if it is possible to assess
CRISIS AND STASIS 86
H2 using more granular data regarding network health before and after a spike in disruptive play
the opportunity will be exploited.
The Steam Community
Steam is a multi-title digital distribution platform and social network. Instead of
representing a single title or community, Steam is an overarching series of common networks
that move between games developed by a wide range of developers (Blackburn, Kourtellis,
Skvoretz, Ripeanu, & Iamnitchi, 2014). Individuals use Steam as a tool for purchasing and
playing games, accessing each title and its associated community from a centralized storefront
and library. While the actual content of each game is dependent on its respective developer, titles
integrated into the Steam platform can leverage a common cross community series of
institutions.
As an example, consider a network consisting of three players all using the Steam
platform. At any given time each person may be playing a different game, ranging from simple
puzzle games to elaborate roleplaying titles. During their gameplay experience Steam is running
in the background, providing notifications when friends come online and offering a helpful series
of forums and other para-texts for each hosted title. Players can leverage the social network built
into Steam to make friends through one game and continue to interact with them in a completely
different title.
Steam was released in 2003 by the Valve Corporation as one of the first gaming digital
distribution platforms. Despite a rocky start featuring widespread player skepticism and a
number of bugs, Steam has grown into the dominant digital distribution platform and network
within the personal computer gaming marketplace with more than 125 million active users as of
early 2015 (Saed, 2015).
CRISIS AND STASIS 87
Part of Steam’s popularity stems from the fact that it offers an out-of-the-box solution to
a number of difficult challenges facing developers. Steam institutes a social network, community
forums and storefront for developers who publish their titles on the platform (Valve, 2015b).
These developers can use the services to handle the social and sales aspects of their titles, leaving
them free to focus on building content and regulatory systems that appeal to players.
Additionally, Steam offers a number of optional side benefits, such as the Valve Anti-Cheat
service (V AC) which actively monitors player’s versions of a given game for illicit
modifications. V AC has been applied to more than 300 games over the course of its lifespan and
is actively maintained and updated as new cheats or hacks emerge (Blackburn et al., 2011).
When V AC detects a cheater the player is banned from the entire V AC system. They still own
their games and can play on a small minority of server that have disabled VAC security but are
otherwise restricted from playing with non-cheaters.
Exploring the Steam community offers a chance to study a cross-title gaming community.
Because of Steam’s unified social architecture, players can form friendships between and among
different gaming sub communities allowing for a more free-flowing set of social connections that
are unimpeded by differences among developers. This plurality of titles also offers more
opportunities for private-collaborative disruptive players to ply their trade. Each distinct game
has the potential to be exploited or manipulated for fun or profit. The presence of anti-cheating
systems such as V AC also means that disruptive players have to constantly update their tools or
else lose the ability to break through a given regulatory structure (De Paoli & Kerr, 2009). These
factors make the Steam Community fertile ground for exploring disruptive play.
CRISIS AND STASIS 88
Method and Data
Due to the cross-title nature of Steam, the datasets at the heart of this chapter are both the
largest, but also the shallowest within this dissertation. Different developers expose varying
levels of information about their games on Steam. Many third-party creators share almost no
information, while other studios maintain robust public application programming interfaces
(APIs) which share data with the community (Valve, 2015b). Because of this variability over the
approximately 5,600 distinct titles on Steam, it is impossible to gather reliable, granular player
statistics that accurately cover the entire community. However, data from the unified cross-title
services such as the Steam social network and V AC is consistently and reliably available for
analysis. As a result of this discrepancy, this chapter features a relatively sparse slate of
independent variables when compared to subsequent analyses, consisting primarily of variables
derived from the Steam social network. The bright side of this limitation is that it provides a
robust test for one of the key hypotheses of this paper, which is that disruptive players occupy a
distinct position within their host social networks. If this assertion bears out, then the fact that
this study relies predominantly on features created from Steam’s social network should not
prevent the subsequent classification models from achieving good performance.
The data were collected during four sampling windows at the same time as the Team
Fortress Two data in the following chapter. The network data consist of two paired samples taken
six months apart. The 2014 samples were drawn in two periods from that calendar year, January
and February (Period One) and July and August (Period Two). The 2015 samples were drawn
from June and July (Period Three) and December, into early January of 2016 (Period Four).
Disruptive players were identified with the cross-game Valve Anti-Cheat API. This data
source identifies which players have been banned as well as providing a to-the-day timestamp of
CRISIS AND STASIS 89
when the ban took place (Valve, 2016). This means that it is possible to situate bans on a
timeline, which is useful for the anomaly analysis later in this chapter. Defining what players to
classify as disruptive for the sake of training a model is difficult. Players who were banned well
in the past may have left the community or had their friends ostracize them. Alternatively,
training the model on people who will be banned in the future is troublesome as the person may
not have begun their behavior at a given data window. Specifically, if someone engages in
disruptive play a year from a given sampling window (as an example) their network may not
have shifted to accommodate the different social demands which theory suggests are connected
to cheating. Constraining the test set to just those banned around a given sampling window
exacerbates the class imbalance problem by discarding relatively rare cases of disruption. To
split the difference each classification model was trained on players who were already banned at
the moment of the sampling window, but performance was assessed based on the model’s ability
to predict who was already banned as well as future targets. If the model performs significantly
worse on the second set of cases it is over-fit or capturing distortions due to the banning process.
It is not possible to fit a forward looking model for Period Four as subsequent data have not
become available to validate the predictions, so those cells in the ROC table are left blank.
The data were collected using an iterative snowball sample that was not constrained to
players of a specific game. The complete network therefore traverses different communities
embedded in and around various titles within the steam platform. The total sample size for each
period is described in the table below.

CRISIS AND STASIS 90
Table 2: Sample Sizes. Private-Collaborative Model
Sample Players
Period One, Winter 2014 1,873,808
Period Two, Summer 2014 2,268,117
Period Three, Summer 2015 3,336,994
Period Four, Winter 2015 3,771,831

It is important to note that there was a significant change in the Steam environment
between Period Two and Three. In response to the increased visibility of professional gaming
within the Steam community Valve launched a series of crackdowns on cheating. These “ban-
waves” lead to an increased rate of punishment within the Counter Strike: Global Offensive
community, a major component of the broader Steam network. As a result, the overall level and
tolerance for disruptive play on behalf of the developer shifted between sampling periods. The
changing base conditions within the Steam network can influence the results in the latter two
sampling periods, which should be kept in mind when exploring the results of the classification
models.
As mentioned above, because of the variability in the amount of data made available by
different communities, the only features used in this chapter’s classification model were drawn
from the cross-game Steam social network. As a result of this limitation and the scale of the
networks in question, the number of predictor variables is small, and summarized in the table
below.

CRISIS AND STASIS 91
Table 3: Dependent and Independent Variables. Private-Collaborative Model.
Name Description Category Variable Role
Effective Network
Size
See methods chapter Network Independent
Network Constraint See methods chapter Network Independent
Average Connection
age
Mean duration of a
player’s friendships
Network Independent
Number of banned
contacts
Number of players in a
node’s immediate
network who have been
banned
Network Independent
Disruptive Has a player been
banned for cheating?
User Data Dependent

Results
With the predictors established, it is now time to turn towards the various classification
models that help parse between disruptive players and their host network. A mixture of weighted
and balanced-sample approaches was used. On the statistical learning front, balanced sample
gradient boosting, random forest and bagged trees were used to parse between disruptive and
non-disruptive players. Using the area of the ROC curve performance was assessed for each
classifier based on its ability to predict currently banned players as well as those destined to be
banned after the sampling window (Bradley, 1997).

CRISIS AND STASIS 92
Table 4: AUC Scores, raw scores in brackets with Delong’s D measure of AUC difference at the
intersection of both scores.
3

Period One
Gradient Boost (0.929)
Rare Event (0.916) 9.54***
Period Two
Gradient Boost (0.935)
Rare Event (0.923) 9.3002***
Period Three
Gradient Boost (0.876)
Rare Event (0.726) 43.462***
Period Four
Gradient Boost (0.969)
Rare Event (0,848) 51.523***

The standard interpretation of an AUC score generally uses a 0.1 step to categorize model
performance. Scores in the 0.5 – 0.59 range represent a model that performs just as well as, or
slightly better than random guessing if a player is disruptive or not. In other words, failing
performance. Scores between 0.6-0.69 represent poor performance, better than random guessing
but not a substantial improvement. Results in the range of 0.7-0.79 generally represent adequate
performance, with scores between 0.8-0.89 and 0.9-1.0 coming from good or excellent models
respectively.
The gradient boosting model consistently outperformed the alternatives except in the
fourth sampling period. Additionally, while none of the forward-looking models performed better
than those assessed on currently banned players, there was no significant drop in performance.
Models trained on historical data are therefore not over fit or modelling the social dynamics that

3
Variable importance scores for the random forest model did not function properly given the size of the dataset and
the computational burden so the model is excluded from this chapter.
CRISIS AND STASIS 93
occur after a player has been caught and can be used to predict future bans with a degree of
success.
Seeing as the gradient boosting model performed better than the alternatives in all cases,
the first step in assessing the hypothesis related to the structural signatures of disruptive players
is to assess the variable importance of each predictor in the various models. The scores are scaled
so that 100 represents the most important variable with relative influence of each predictor
decreasing from e. Predictors with scores under one have only a fraction of the overall
importance of the most important predictor, which always receives a scaled score of one
hundred.
Table 5: Variable Importance: Private-Collaborative Model.
Variable
Period One
2014 (Winter)
Period Two
2014 (Summer)
Period Three
2015 (Summer)
Period Four
2015 (Winter)
Average
Average Edge
Duration
100 100 100 4.576 76.144
Number of
Connections to
Banned Players
0 0 0 100 25
Effective Network
Size
0.438 0.177 0.905 10.172 2.923
Network Constraint 0.953 0.485 1.009 0 0.612

These results indicate that the average duration of a player’s edges within the Steam
social network is the most important predictor three of the four sampling periods. The number of
connections that a given player has to other disruptive players was an utterly unimportant
predictor except for the last time period.
While variable importance metrics isolate the critical predictors within a given model,
variable prediction plots are needed to evaluate the scope and direction of the effect. As the
CRISIS AND STASIS 94
average duration of a connection was consistently the only important predictor it is a logical
place to start the examination.

Figure 5: Average Edge Age Variable Predictions
The variable prediction plot shows a strong positive relationship between the probability
of being a disruptive player and the average age of one’s connections. In other words, the more
durable a given node’s connections are, the more likely they are to engage in disruptive behavior.
Interestingly, the models from the latter two sampling periods display a spike in probability for
short-term relationships before continuing the trajectory of a general curvilinear increase. This
may be due to temporal changes in the Steam enforcement structure which will be discussed
further in the subsequent section.
CRISIS AND STASIS 95
While the weighted rare event logistic regression model did not perform as well as the
gradient boosting approach the results are still useful for triangulating the findings from the
statistical learning models.
Table 6: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Private-
Collaborative Model.
Dependent variable: Banned (y/n)

Period One Period Two Period Three Period Four

Network Constraint -1.978
***
-1.989
***
-1.378
***
-2.071***
(0.051) (0.051) (0.039) (0.048)

Effective Network Size 0.001
***
0.001
***
-0.003
***
-0.118
(0.0002) (0.0002) (0.0002) (0.001)

Average Edge Age 0.001
***
0.001
***
0.001
***
0.001***
(0.000) (0.000) (0.000) (0.00)

Number of Connections
to other Disruptive
Players
0.040
***
0.023
***
0.123
***

1.359***
(0.009) (0.007) (0.011) (0.007)

Constant -5.999
***
-6.127
***
-5.320
***
-3.780***
(0.021) (0.020) (0.013) (0.019)

Observations 1,873,808 2,268,117 3,336,994 3,771,831
Log Likelihood -64,883.890 -67,711.260
-
124,933.800
-85,457.300
Akaike Inf. Crit. 129,777.800 135,432.500 249,877.600 170,924.600
McFadden’s Pseudo R
2
0.288 0.295 0.162 0.457
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01

CRISIS AND STASIS 96
The results of the rare-event logistic regression help flesh out the picture established by
the statistical learning model. The average duration of a given node’s connections was a
consistent positive predictor of disruptive behavior. In other words, both models confirm that
disruptive players are more likely to have long-term, stable connections. The positive
coefficients for the number of disruptive players’ social network and its increasing importance
across the statistical learning models suggests that these connections are directed towards other
disruptive players as opposed to the general community.
The picture surrounding effective network size and constraint is more complex. Within
the statistical learning model they did not play an important role in the prediction model.
However, both variables were significant at p<0.01 for the logistic regression model, although
the large sample size means that these results may be distorted (M. Lin, Lucas, & Shmueli,
2013). With these disclaimers in mind it appears that network constraint has a slight negative
influence on the likelihood of disruption. The picture surrounding effective network size is more
complex as it is a positive predictor in regression models from the first two sampling periods,
before shifting to a negative or insignificant role for the latter two models. This suggests that the
changing tolerance for disruptive play from the developer may be influencing the structural
signatures of players.
Overall these results create a mixed picture. H3a was partially supported. It stated that
disruptive players will be more likely to connect with other cheaters. The number of connections
also engaged in disruptive behavior was not an important predictor except in the last sampling
window when it was extremely important. These results were collaborated in the regression
model as the coefficient jumps significantly in the last sampling window after the
aforementioned ban-wave. This suggests partial support for H3a as the number of disruptive
CRISIS AND STASIS 97
contacts is important under some conditions but also means that it is critical to explore the scope
and reasons behind the change to community management strategy in order to determine why
these coefficients shifted. Before launching into this exploration in the discussion section it is
important to assess the other hypotheses for this chapter so it is possible to create a cohesive
picture of how the shift in base conditions interacted with the structural signatures of disruptive
players.
H3b on the other hand is supported across both models. It stated that players with more
long-term connections to the surrounding network would be more likely to engage in disruptive
behavior The average duration of a given node’s edges was highly important in three of the four
statistical learning models and positively associated with the likelihood of engaging in disruptive
behavior.
H3c and H3d both did not receive robust support. H3c argued that disruptive players
would have higher than expected rates of local closure while H3d suggested that they would
avoid positions of indirect brokerage. Effective network size was not an important predictor in
any of the statistical learning models. Additionally, it was not significant in the fourth regression
model while the directionality of the effect shifted between period two and period three from
negative (less closure means greater likelihood of disruption) to positive (closure increases the
rate of disruption). Similarly, network constraint rarely performed better than random data in the
statistical learning models. Additionally, the results from the regression run against H3d as more
constrained indirect networks lead to a lower likelihood of disruptive behavior, contrary to the
assertion that disruptive players would avoid positions of brokerage.
The final composite picture which emerges is one where private-collaborative disruptive
players are embedded in relatively stable, long-term networks. These connections are likely to
CRISIS AND STASIS 98
include other disruptive players, especially in the later sampling periods. However, aside from
their relative stability and the increasing importance of homophily there does not appear to be a
distinctive local or indirect pattern of closure or brokerage inherent to this particular brand of
disruptive play.
Rates of Disruptive Play over Time
The statistical learning models demonstrate that private-collaborative disruptive players
exhibit a distinctive structural signature that can be used to statistically differentiate them from
their host network. The question now becomes, does the presence or absence of disruptive play
correlate with major shifts over time in a developer’s policy (H2)?
In order to explore this hypothesis, a seasonal extreme studentized deviation anomaly
detection test was applied to the rate of captured disruptive play over time. This test decomposes
the time series of reported disruptive play and establishes a boundary that contains expected
variations. When the relative rate of disruptive play exceeds or falls below this window the point
is flagged as an anomaly.
CRISIS AND STASIS 99

Figure 6: SHESD results on bans over time. Vertical lines represent the periods when data was
collected for the classification models.
Anomaly detection isolated a number of small spikes in 2006 as well as a large and
protracted increase in the number of caught disruptive players in late 2015.
Due to the antagonistic relationship between developers and many private-collaborative
disruptive players, patch notes detailing changes to the Valve Anti Cheat system are unavailable
as a precaution against reverse engineering. As a result, it is not possible to work forwards from
major patches and pair them with periods of anomalous activity. Instead, the periods with
abnormally high activity were used as the floor for a search of press releases and trade materials
to determine if there was a significant realignment in Valve’s corporate strategy as either a
leading or lagging indicator of the periods of high activity. The anomaly windows were used as a
floor for the search as a ban represents the tail end of a disruptive player’s career, taking place
after they have circumvented a game’s rules.
CRISIS AND STASIS 100
The qualitative search of patch notes and media coverage related to the V AC system
uncovered two periods of note. The 2005-2006 spikes appear to be associated with a major
update to the V AC software called VAC2. This expanded the software’s detection range and
allowed Valve to catch new forms of disruptive behavior. This appears to have led to the period
of high bans as disruptive developers adjusted to the new regulatory system and older cheating
techniques were caught and weeded out. There does not appear to be a particularly well
documented shift in corporate strategy behind the release of VAC2, rather it appears to be a
regular update in order to keep up with changes in technology.
The spike in late 2014 and throughout 2015 is much more interesting. It appears to center
around the popular game Counter Strike: Global Offensive (CSGO). CSGO is one of the most
popular games on Steam and features a large professional community. Pro players compete in
tournaments for prize money and use their skills to earn a living. The popularity of pro-gaming
has increased with the introduction of live streaming services such as Twitch which makes it
easier for professionals to broadcast their matches to fans. Games such as League of Legends by
Riot Games pioneered the creation of a professional circuit, where the most skilled players form
teams and tour playing for crowds around the world. In early 2014, Valve began to sink resources
into professional CSGO to increase the game’s profitability and stature. This commitment
included establishing a $250,000 prize pool for major tournaments and promoting professional
CSGO heavily (Chalk, 2014). Shortly after increasing its involvement in professional CSGO and
the distribution of two prize pools, a series of bans were handed out to professional players who
had been reported for cheating. The investigation into professional cheating lead to changes in
the V AC system that caught a number of other players, leading to the 2014-15 “ban wave” which
players referred to as “the great V ACation” (Congo, 2014).
CRISIS AND STASIS 101
In this case, the chain of events runs contrary to the process laid out in H2. Instead of
levels of disruptive play causing issues within a community which lead to a major reorientation
or adjustment by a developer, the process is reversed. External pressure from competitors
prompted the development of a professional CSGO scene. This increased the visibility of
cheating which prompted a response in the form of a ban-wave.
Discussion
Given the results of the classification models and the findings from the anomaly analysis
there may be an interplay between changes in the regulatory structures within Steam and player’s
structural signatures. The samples from Periods One and Two take place before the significant
start of the ban-wave and crackdown on cheating with the CSGO section of the Steam
community. Meanwhile, the samples from Periods Three and Four were drawn during the height
of the banning and after the wave had passed, respectively. This temporal aspect may serve to
explain the increasing importance of being connected to a banned player in the later analysis
windows. Specifically, the ban-wave represents an attack on social networks in and around
disruptive players. Being banned removes players form the CSGO community by limiting their
ability to join the vast majority of servers that host Counter Strike content. This has two effects;
first it restricts the size of the remaining disruptive network due to the removal of nodes.
Secondly, those nodes that do remain have to rebuild shattered sub-networks and work to reform
the disruptive community. As a result, the community may have pulled inwards, leading to a
greater degree of homophily as disruptive players attempt to keep a low profile and avoid the
ban-wave. Additionally, the disruptive network has the potential to bounce back. The variable
prediction plots for the latter two sampling periods show a spike in the likelihood of disruption
among players with a low average connection age. This suggests that a portion of the disruptive
CRISIS AND STASIS 102
network is forming new, younger ties, potentially either restoring or reconnecting to other
elements of the sub-community. This is supported by the conceptual drift demonstrated in the
statistical learning models. After the ban wave the structural signature associated with disruptive
play shifts dramatically. Players with a low average edge age and more connections to other
disruptive members of the community become much more likely to be engaged in disruptive
play. This shift in variable importance scores and the variable prediction plots represents the
model picking up on a shift in the patterns associated with disruptive paly. Working backwards
from these shifts and the spike in bans seen in the time series a change in fundamental network
pattern associated with disruptive play appears to be correlated with the introduction of the ban-
wave, suggesting that the intervention shifted the network structure of the disruptive sub-
network. Further exploration of the networks before and after the ban-wave is needed, but this
provides a useful examination of how the presence of disruptive behavior can damage host
networks, and what the recovering process looks like.
Specifically, private-collaborative disruptive players may damage their host networks due
to their conservative tie formation. Networks rich in bridging capital are predicated on the
constant establishment of new, weak, connections that serve as opportunities and bridges for
information flow (R. D. Putnam, 2001). The classification models demonstrate that disruptive
players have a tendency towards older, more established connections, most likely to trusted allies
or confidants. This makes it more difficult for new friendships to break into sections of the
network rich in private-collaborative disruptive play. New connections pose a risk due to the
dangers inherent in this variety of disruptive behavior so establishing a new connection may be a
less attractive option, which in turn reduces the potential for brokerage. The spike in disruptive
players establishing new, fresh connections after the ban wave suggests that the changes made
CRISIS AND STASIS 103
this strategy untenable. This may help force disruptive players to reengage with their surrounding
networks. Whether the disruptive sub-community returns to its previous patterns cannot yet be
established, but it may be an interesting topic for study moving forwards.
To summarize, this chapter explored private-collaborative disruptive behavior, which was
defined as cheating or hacking. Using four samples from the Steam community and a limited
slate of network data a series of statistical learning models were able to achieve acceptable
performance. The results from these models indicate that the best predictor of disruptive
behavior is the age of a given player’s connections, with an older more stable local network
serving as the best predictor of disruptive behavior in the classification model. . Conversely local
and indirect brokerage and closure were not strong predictors. Homophily was also unimportant
until the last sampling window. This period occurred after the “great V ACaction”, a wave of bans
that was prompted by Valve’s move towards the competitive gaming scene and the resulting
scrutiny of professional players. The drift in predictive power between the various models may
be attributable to these shifting conditions, with the increasing importance of homophily and
younger connections suggesting a rebuilding process. However further observation of the Steam
disruptive community since the end of the ban-wave is needed to validate these effects.

CRISIS AND STASIS 104
Chapter Six- Private-Isolated Disruptive Behavior
This chapter continues the analysis section of the dissertation by exploring private-
isolated forms of disruptive behavior. This variety of disruptive behavior is in some ways the
simplest to understand. It centers around individuals acting on their own to break through various
regulatory structures within a given community. If successful, the benefits from these
breakthroughs only accrue to the person engaged in the disruptive behavior, hence the label
private-isolated (Dilla et al., 2013). Using data drawn from the game Team Fortress Two and its
associated economy, this chapter explores the influence of this disruptive behavior on other
players as well as the developers who have created and maintain the host community and its
regulatory structures. The remainder of this chapter presents the specific networking patterns that
previous research has associated with this brand of disruptive behavior as well as exploring how
these structural signatures displace other forms of social capital within the community. This
analysis sets the stage for a secondary examination to look at the relationship over time between
rates of disruptive play and significant changes in a developer’s corporate strategy. The results of
this secondary study serve to connect the rise and fall in levels of disruptive play to the process
of punctuated equilibrium in an organizational setting.
Private-Isolated Disruption
Private-isolated disruptive behavior generally takes the form of individuals targeting
weaker or less experienced members of a community through means explicitly prohibited by the
rules. By exploiting the relative difference in experience or skill between themselves and their
mark, the disruptive player can harvest resources or other benefits above and beyond what would
be available if they conformed to a community’s regulatory structures. Generally this process is
perpetrated by individuals as opposed to groups of players as the results of the disruptive
CRISIS AND STASIS 105
behavior are exclusive, so the more participants, the more split any reward (Dilla et al., 2013).
The signature example of this form of behavior is in-game scamming or fraud.
Scamming can take a number of different forms depending on the community in
question, and there is no exhaustive typology of scams in the current literature. However,
exploring two popular techniques should be sufficient to demonstrate why scamming falls into
the private-isolated category of behavior.
The first popular technique is the “too good to be true” deal. The scammer will approach
a player offering a seemingly valuable item for cheap. The mark will readily agree to this
excellent deal only to be fooled at the last minute when the scammer instead substitutes a similar,
but less valuable item, and comes away from the deal with a net profit (Rubin & Camm, 2013).
The second technique is to impersonate a trusted figure within the community by copying their
appearance, name and style (Bardzell et al., 2007; Jagatic, Johnson, Jakobsson, & Menczer,
2007). By piggybacking on the trust that this established figure has built, the scammer can offer
to hold items in escrow or act as a middleman before making off with the good held in trust and
changing their identity.
These two techniques are a sample of the ever expanding range of tools and approaches
open to this variety of disruptive player. The common factor among all of these techniques is that
they involve a small number of disruptive players (one, occasionally two) and the benefits from
the scam go directly to only those involved with the disruption (Dilla et al., 2013). This makes
scamming a textbook form of private-isolated disruptive behavior.
Existing literature on these behaviors suggests that engaging in private-isolated disruptive
behavior distorts the incentives that help shape a given person’s structural signature within a
network. Specifically, the focus on short-term gains through deception means the scammers often
CRISIS AND STASIS 106
attempt to avoid network closure. Criminologists examining fraud in various settings have
coined the term bogus brokerage to describe the alternative network dynamic fostered by
scamming and other behaviors associated with private isolated disruption (Burt, 2001; Tillman &
Indergaard, 1999).
Bogus brokerage refers to an actor leveraging the powers granted by occupying a
brokerage position within a given network by acting in bad faith (Burt, 2001; Tillman, 2003;
Tillman & Indergaard, 1999). By brokering connections between several different actors, the
disruptive player in question has access to a wide array of information from the segments of their
direct and indirect network neighborhood. Instead of using their privileged position to bring
together members of the community and close structural holes in their local network, a bogus
broker works to maintain and increase the number of gaps in the network (Tillman, 2003).
Provided that the bogus broker does not act in good faith, each gap in their immediate or indirect
network is effectively a structural hole, as no information can flow between a node’s connections
without being exposed to disruptive player at the heart of the network. By mediating the flow of
information between these contacts, the bogus broker can target any member of their local
network with a scam (Tillman, 2003). Because the mark has no reference from another person
exposed to the scammer, they are less likely to receive evidence or advice against further
association with a disruptive player, increasing the likelihood of success for any subsequent
scams.
H4a –Private-isolated disruptive actors will be more likely to broker connections between
members of their host network.
Given the importance placed on bogus brokerage within the literature it stands to reason
that scammers will avoid network closure. This closure defeats all of the advantages that spring
CRISIS AND STASIS 107
from occupying a bogus brokerage position. If a mark is connected to another target of the
scammer, particularly one who has already been victimized, they can share information regarding
their disruptive neighbor and damage the likelihood that any scam will be successful (Baker &
Faulkner, 2003). This sharing effectively collapses the structural holes around the disruptive
player and removes them from a brokerage position, reducing their ability to manipulate
information flows to their advantage.
H4b- Private-isolated disruptive actors will have lower levels of closure than equivalent
participants in their host networks.
Finally, there is little incentive for disruptive players to maintain relationships after they
have achieved their objectives. Once taken advantage of the likelihood increases that a mark will
become aware of the fraud and push back against the disruptive player. Instead of pushing their
luck by repeatedly exploiting the same target the literature within criminology suggests that the
frauds will cut ties and move on, exploiting new members of their host network who are unaware
of their aims and goals. As a result the average duration of a private-isolated disruptive player’s
edges within a network will be significantly lower than the rest of the community, as they have
little need to stick with a friendship and acquire benefits through reciprocity or mutual aid.
H4c – Private-isolated disruptive players will have a lower average edge age when
compared to other players in their host network.
These network structures displace the standard processes of bridging and bonding within
a social network. Private-isolated disruptive players actively work to cultivate structural holes
above and beyond what would be normally expected in a given social network. This decreases
the ability of players to bridge gaps and find new contacts, which in turn damages the
community’s bridging social capital. Additionally, the danger posed by network closure means
CRISIS AND STASIS 108
that players engaged in private-isolated forms of disruptive activity are less likely to be
embedded in tight overlapping networks of contacts. This negatively influences the overall level
of bonding capital within the community and prevents sub-groups from cohering within the
broader network.
Aside from their structural position, private-isolated disruptive players also damage the
overall health of the community by breaking down the cycle of reciprocity that is necessary to
build trust within a network. When exploring the concept of social capital Robert Putnam argued
that trust represents a form of virtuous circle (1994). The more trust that exists within a
community the more trust it creates, leading to ever higher levels of social capital (R. D. Putnam
et al., 1994). By breaking through regulatory structures designed to create and propagate trust for
their own benefit, private-isolated disruptive players short circuit this process. Their marks may
no longer trust future contacts after experiencing a scam or theft, and require more elaborate or
burdensome assurances from their partners in future dealings (Gregg & Scott, 2006). These three
factors, the propagation of structural holes, avoiding network closure and the creation of an
atmosphere where it is more difficult to engage in reciprocal exchange, all directly damage
stocks of bridging and bonding social capital. This damage creates an incentive for developers
responsible for a particular community to step in and make changes in order to foster a healthy
player-base. This tendency is captured in hypothesis two:
H2- Changes in regulatory structures by administrators of online communities are
correlated with increases in disruptive behavior.
Team Fortress Two
As described in the methods section, the core dataset for this chapter draws from Team
Fortress Two (TF2). TF2 is a multiplayer shooter game. Players join small teams generally no
CRISIS AND STASIS 109
bigger than 18 players per side and fight through a series of maps with different objectives and
goals. Victory conditions range from attempting to capture a flag, to securing certain points on
the map or scoring the most points within a predefined time period. The core gameplay of TF2 is
not entirely persistent, that is to say a player’s status does not carry over from match to match.
However, the game’s developer Valve Corporation has implemented a persistent layer that sits on
top of the primary game in the form of items and other improvements that players can earn
through playing. As an example, when players first join the game they have access to a relatively
limited roster of tools and weapons. By playing or spending currency within the game they can
unlock additional items ranging from cosmetic accessories to new tools that open up different
play styles within the game. These items can be traded to other players through a complex barter
system or sold within a community market for in-game credit that can be put towards games or
other items.
The TF2 dataset plays host to private-isolated disruptive behavior in the form of hat
scammers. Some items within the TF2 economy are more or less rare than others. Certain items
were only introduced for a limited time or handed out to a select bunch of players. This scarcity
and the demand for powerful or cosmetically appealing items within the game means value of
certain items is inflated in relation to others, with some rare items selling for more than $6000
(US) (Peterson, 2013). Since items are distributed through a pseudo-random process, wealth
within the game is not explicitly correlated with time spent playing. A new player may be
extremely lucky and obtain a valuable item, while a veteran can play for dozens of hours without
the same level of success. Additionally, the distinctions between valuable and commonplace
items are often subtle, such as a slight in-game visual effect, which can make differentiating
between different classes of goods difficult to an untrained eye. Due to these factors, a
CRISIS AND STASIS 110
community of “scammers” has grown up within the game (“Steam Reputation FAQ,” 2015).
Using a variety of social engineering tricks, these scammers exploit new or inexperienced users
by making unfair trades or engaging in social engineering. These behaviors are explicitly against
the terms of use for TF2 and are frowned on by the developer (Valve, 2015a). This illicit nature
combined with the fact that scammers generally keep their gains for their own profit and work
individually make this practice an example of private-isolated disruptive behavior.
While Valve has taken steps to ban scammers from the in-game economy, the most
complete listing of disruptive players can be generated from third-party data sources. The site
SteamRep (Steam is the gaming service that hosts TF2) was founded in 2011 and is run by a
committee of experienced players. Steamrep maintains an extensive historical database of
scammers who have been accused and evaluated by the site’s moderators over the past five
years. To be included in the database, a victimized individual has to submit evidence that they
traded with the alleged scammer and that they were deceived (“Steam Reputation FAQ,” 2015).
This evidence generally takes the form of screenshots of chat logs. The allegations are then
reviewed by the moderating staff of the website and if accepted the scammer is flagged so that if
any of their future marks search for them online they will be informed of their bad behavior.
By cross-referencing the SteamRep database with the various friendship networks drawn
from the TF2 community, it is possible to both isolate disruptive players within the game’s social
fabric and correlate rises and falls in disruption with policy shifts from the developer. This
provides everything needed for the first study to address the research questions put forward at the
start of this examination. Namely, do disruptive players have a distinctive structural signature
and are changes in the prevalence of this signature associated with shifts in developer strategy?
CRISIS AND STASIS 111
Method and Data
The TF2 data consists of two paired samples taken six months apart. The 2014 samples
were drawn in two periods from that calendar year, January and February (Period One) and July
and August (Period Two). The 2015 samples were drawn from June and July (Period Three) and
December, into early January of 2016 (Period Four). The rationale behind this sampling strategy
is two-fold. Changes among players may have a seasonal effect. Team Fortress Two and other
games often attract a school-age audience, so taking samples during the school year and over
summer break will help control for the shifting demographics of players based on their
availability. Having two sampling periods enables comparisons across the time during which
Valve has consistently patched and altered TF2 with several significant updates.
The data were collected using an iterative snowball sample. Starting with a random seed
of known TF2 players, a collection program iterated outwards traversing edges within the
network and filling in relationships between players. The sampling process continued until
results converged and the iterator stopped returning new players. However, it should be noted
that the size of the sample is lower than the reported number of Steam users who have played
TF2 (Baumer & Kephart, 2015). Therefore, this data should be seen as a network sample as
opposed to the population study.
To maintain speed, the data collection script only gathered basic social network data. So
in addition to constructing a social network, a random sample of players was further examined to
get their playtime and relative wealth within the Team Fortress Two dataset. Due to data
collection limitations, this in-depth sample is smaller than the overall network. In order to avoid
distortion to the network data, the sampled players were situated inside the larger, simple
network. In other words, all network metrics were generated using the larger, simple network and
CRISIS AND STASIS 112
joined onto the smaller, detailed subset of data. Therefore, any variables related to the TF2
network are drawn from the larger sample, which captures the complete local neighborhood of
nodes being examined in-depth, reducing distortion from network sampling. The composite of
these two datasets (per sampling period) forms the basis for bulk of this chapter.
Table 7: Sample Sizes. Private-Isolated Model.
Sample Players
Period One, Winter 2014 106,051
Period Two, Summer 2014 124,240
Period Three, Summer 2015 50,952
Period Four, Winter 2015 45,150

Each node in the networks represents a TF2 player, the edges represent mutually agreed
upon friendships. These friendships are one route of trade and exchange within the TF2 economy
and a common vector for disruptive players to engage with their marks.
The aforementioned data was cross referenced with a list of 11,249 scammers harvested
from SteamRep. To be included in the list of known scammers, an individual must have been
reported to SteamRep and judged to be disruptive by the in-house committee on that site. Cases
where there was insufficient evidence or the committee was split were discarded. This reference
dataset includes both the ID of the scammer as well as the date they were reported.
The various datasets at work create two broad categories of variables, those drawn from
the network and those inherent to the players who comprise the TF2 community. The network
variables include the standard measures of effective network size and brokerage described in the
methods chapter as well as the number of edges a given node has that link to a scammer. These
three variables cover the core hypothesis of this examination.
The network measures are also joined by a series of measures related to time played and
in-game wealth in order to capture player’s experience and status within the in-game
CRISIS AND STASIS 113
marketplace. The former is represented as hours of playtime. The number of items a player has
within TF2 is further broken down into several categories based on their relative rarity and value.
 Normal items are the core tools which every player receives. Each node in the network
has the same number of normal items and they cannot be traded in the economy, so this
variable does not figure into the predictive model.
 Vintage items are older pieces of equipment that were grandfathered into the game before
items were tradable. Like unique items there are a few rather rare vintage items but on the
whole they occupy the lower value tiers within the TF2 economy.
 Genuine items come from promotions, giveaways or other in-game events. They can vary
wildly with regards to value with some items being quite rare and expensive.
 Strange items are distributed through specific events as well as randomly given to players
within the game. They feature the ability to track player achievements and change in
response to their success make them more valuable and coveted.
 Haunted items can only be acquired during the game’s annual Halloween celebration,
making them relatively rare and valuable due to the constrained supply.
 Usual items are variants on any of the preceding tiers. They feature a distinctive visual
effect that makes the item stand out within game. These items are randomly distributed as
players earn rewards and are generally considered extremely rare or desirable.
Other categories exist but they are often restricted or un-tradable such as special
weapons reserved for employees of the developer who happen to be playing. While these
categories are not iron-clad (a particularly valuable or aesthetically pleasing, strange item may be
worth more than an ugly unusual item) they do provide a rough tool for assessing a player’s
wealth within the game.

CRISIS AND STASIS 114
Table 8: Dependent and Independent Variables. Private-Isolated Model.
Name Description Category Variable Role
Effective Network
Size
See methods chapter. Network Independent
Constraint See methods chapter. Network Independent
Disruptive ties Number of players in a
nodes immediate
network who have
been identified as
disruptive.
Network Independent
Average connection
age
Mean duration of a
player’s friendships
Network Independent
Time Played Minutes spent playing
TF2
Player data Independent
Items (Unique,
Vintage, Strange,
Haunted, Unusual,
Genuine)
Number of items
owned which fall into
a given class
Player data Independent
Disruptive If a player has been
flagged on SteamRep
for scamming
SteamRep data Dependent

Results
The first step in assessing the role of private-isolated disruptive play is to determine if
these players have a distinctive structural signature that varies from their host community. This
analysis was carried out using a combination of statistical learning models and rare-event
weighted logistic regression. The former provide a robust, but more difficult to interpret
mechanism for classifying players, while the latter may not capture some complex interactions in
the dataset.
For the statistical learning model a variety of ensemble approaches were applied in an
attempt to classify players into two bins, disruptive and non-disruptive based on their flag within
the SteamRep database. Random forests provided the best overall performance followed by
gradient boosting and rare-event logistic regression respectively.
CRISIS AND STASIS 115
All of the models performed well, with area under the ROC curve scores ranging between
0.72 (satisfactory) to 0.9150 (excellent) (Bradley, 1997).
Table 9: Absolute values of the Delong's D between each ROC curve. Significant scores mean
that the model with a higher ROC score performed significantly better than the alternative.
Period One
Gradient Boost
ROC = 0.915
Random Forest
ROC = 0.906
Rare Event
ROC = 0.747
Gradient Boost
ROC = 0.915

-- -- --
Random Forest
ROC = 0.906
2.032** -- --
Rare Event
ROC = 0.747
17.078*** 15.414*** --
Period Two
Gradient Boost
ROC = 0.832
Random Forest
ROC = 0.841
Rare Event
ROC = 0.774
Gradient Boost
ROC = 0.832
-- -- --
Random Forest
ROC = 0.841
1.268** -- --
Rare Event
ROC = 0.774
3.882*** 3.955*** --
Period Three
Gradient Boost
ROC = 0.735
Random Forest
ROC = 0.717
Rare Event
ROC = 0.728
Gradient Boost
ROC = 0.735
-- -- --
Random Forest
ROC = 0.717
0.656 -- --
Rare Event
ROC = 0.728
0.215 0.295 --
Period Four
Gradient Boost
ROC = 0.791
Random Forest
ROC = 0.874
Rare Event
ROC = 0.721
Gradient Boost
ROC = 0.7901
-- -- --
Random Forest
ROC = 0.874
2.464** -- --
Rare Event
ROC = 0.721
0.407 1.02 --
*
p<0.1;
**
p<0.05;
***
p<0.01

CRISIS AND STASIS 116
The gradient boosting model and random forest split on samples exhibiting the best
performance, however the random forest offered the highest average performance and performed
significantly better than the alternative approaches more often. In order to avoid an apples-to-
oranges comparison of variable importance, random forest was therefore selected as the
representative statistical learning model as it displayed the highest overall performance.
Using the variable importance metric described in the methods chapter, the following
table reports the relative importance of each variable in the random forest models (Breiman,
2001a, 2001b). The scores are scaled with the most important variable (i.e. the one which
provides the greatest gain in predictive power over randomness) is 100, and the least important is
0. Less important variables may still provide useful information, but their contribution is less
critical to model performance than their more important counterparts. These variable importance
scores represent a first step in determining the structural signature of disruptive players. While
each score states if a variable contributed to the model it does not provide the direction or
magnitude of their effect. However, these scores are useful for selecting a subset of variables for
deeper examination using variable prediction plots in the next section, a technique that does
provide data on the direction and shape of an effect but not necessarily its importance. Therefore,
the two approaches need to be taken together as a composite.

CRISIS AND STASIS 117
Table 10: Variable Importance: Private-Isolated Model.

Period One
2014
Period Two
2014
Period Three
2015
Period Four
2016
Average
Importance
Network Constraint 14.932 64.635 94.243 100 68.453
Effective Network Size 18.900 51.789 100 86.786 64.369
Average Age of
Connections (hours)
100 100 84.363 39.655 81.005
# of Genuine Items 4.754 12.222 0 10.359 6.834
# of Vintage Items 17.312 21.817 15.792 23.729 19.662
# of Unusual Items 18.732 31.956 66.799 17.011 33.624
# of Unique Items 18.531 43.487 53.563 44.600 40.045
# of Strange Items 22.689 47.048 48.056 50.253 42.011
# of Haunted Items 0 0 10.133 0 2.533
Total Time Played (hours) 37.056 63.918 77.665 84.511 65.788

Examining the variable importance of all of the models side by side and in chronological
order demonstrates some interesting trends. The factors of Network Constraint and Effective
Network Size increase in importance as time passes. In the Winter 2014 sample, they are
marginally important but their role grows throughout the remaining datasets. However, despite
these trends they all remain relatively important predictors and no variable importance score
shrinks to zero. Conversely, the average age of a connection drops in importance as time passes.
Player wealth plays a variable role, with the acquisition of more expensive unusual, unique and
CRISIS AND STASIS 118
strange goods out performing other tiers of item quality. Finally, total time played is a relatively
consistent predictor across all models.
While variable importance is useful for assessing the relative contributions of given predictor to
model performance, it does not provide any information regarding the directionality and shape of
an effect, this is where variable prediction plots come in. Each of the following figures lays out
the influence of a variable related to this chapter’s hypothesis on the predicted probability of a
given player engaging in private-isolated disruptive behavior.

Figure 7: Variable prediction plots, network constraint.

The picture that surrounds network constraint is extremely consistent. Disruptive players
actively avoid engaging in a constrained network at the local or indirect level. This means that
they assume brokerage roles in communities beyond their immediate social neighborhood, fitting
with the bogus brokerage hypothesis of offline fraud. In other words, private-isolated disruptive
CRISIS AND STASIS 119
players actively avoid embedding themselves within indirect networks where their connections
may know each other.

Figure 8: Variable prediction plots, effective network size.

Effective network size is a measure for the local level of closure-brokerage a node has.
Lower effective network size means that a node is embedded in small or more closely clustered
networks. The results from the variable prediction plots appear to confirm the existing literature.
The probability of a given player engaging in disruptive behavior increases steadily with
effective network size before flattening out. With the exception of the curvilinear effect seen in
the last sample, the less closure a player has and the more local brokerage they assume the more
likely they are to engage in disruptive behavior. The curvilinear effect in Winter 2015 may be
CRISIS AND STASIS 120
due to the larger sample size in that window providing more cases which extremely large
effective network sizes, dragging the fitted line downwards.

Figure 9: Variable prediction plot, average edge age.

Finally, the average duration of a connection also displayed a relatively clear trend across
samples. Private-disruptive players were more likely to have an edge for a shorter period of time
than their host community. Once a node’s edges passed, a low level of maturity the probability of
disruptive behaviour tailed off in all samples, with the exception of the summer 2015 case.

The rare event weighted logistic regression models used to triangulate these results
present a similar picture (King & Zeng, 2001). The coefficients, standard errors and significance
levels are presented below.
CRISIS AND STASIS 121

Table 11: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Private-
Isolated Model.

Dependent Variable: Identified Fraudulent Player (y/n)

Variable Period One Period Two Period Three Period Four
Network Constraint 0.286 -7.767
***
-20.868
***
-7.892
(0.200) (2.287) (7.642) (5.103)

Effective Network Size -0.001
**
0.002
***
0.001
***
0.002
***

(0.0003) (0.0003) (0.0002) (0.0004)

Average Edge Duration -0.001
***
-0.001
***
-0.0001 -0.0001
(0.0001) (0.0002) (0.0001) (0.0001)

Genuine Items -0.006 -0.006 0.004 0.005
(0.005) (0.005) (0.005) (0.006)

Vintage Items -0.015
***
0.002 0.002 0.007
(0.004) (0.002) (0.003) (0.004)

Unusual Items 0.032
***
0.008 -0.002 0.012
(0.011) (0.006) (0.008) (0.013)

Unique Items 0.002
***
0.001
***
0.001
**
0.002
***

(0.0002) (0.0002) (0.0003) (0.0004)

Strange Items 0.006
***
0.003 0.004
**
0.003
(0.001) (0.001) (0.002) (0.003)

Haunted Items -0.009 0.003 -0.005 -0.002
(0.006) (0.004) (0.007) (0.006)

Total Time Played 0.0001
**
0.00001
***
0.00001
**
0.00001
***

(0.0001) (0.00001) (0.00001) (0.00001)

CRISIS AND STASIS 122

Overall, the results of the rare-events logistic regression generally mirror the statistical
learning model, especially for Period Two and Period Three models where the area under the
curve performance metric was closest to the statistical learning models. For these two models,
network constraint demonstrated a strong negative effect, showing that the less a given node
assumed a brokerage role in its indirect network the less likely it was to engage in private-
isolated disruption.
Effective Network Size was positively associated with a greater likelihood of disruptive
behavior in every model except the one addressing Period One. This means that the less local
closure a given node has and the fewer people in their immediate network who know each other,
the more likely it is to be a disruptive player.
Finally, average edge duration was negatively associated with disruptive behavior in the
models associated with Periods One and Two. Therefore, the older a node’s connection is, the
less likely they are to be a scammer. This effect was reduced to statistical insignificance for the
latter two models.
There are two major points of contention between the random forest and rare-event
logistic regression models. The first is the significant negative influence of effective network size
on the likelihood of disruptive behavior in the Period One regression model. These findings were
contradicted by the random forest for that dataset, which stated that ENS was a positive.
Constant -3.903
***
-5.464
***
-5.783
***
-7.284
***

(0.064) (0.135) (0.194) (0.283)
Observations 106,051 124,240 50,952 45,150
Log Likelihood -5,006.183 -1,963.707 -1,109.394 -313.363
Akaike Inf. Crit. 10,034.370 3,949.414 2,240.789 648.726
McFadden R
2
0.078 0.086 0.041 0.136
Note:
**
p<0.05;
***
p<0.01
CRISIS AND STASIS 123
Similarly, the Period One and Period Four regression models’ lack of a significant result for
network constraint does not fit with the relatively high importance and strong negative influence
demonstrated in the statistical learning models.
In both these cases, the irregularities are concentrated in and around the Period One and
Period Four models. These are also the cases where the statistical learning approach dramatically
outperformed the rare-event logistic model (Period One AUCs 0.9065 v. 0.7474, Period Four
AUCs 0.8744 v. 0.7207). The protocol laid out in the preceding methods chapter therefore
suggests a heavier weighting for the results from the random forest when assessing this chapter’s
hypotheses.
To summarize, the classification models aimed to assess if private-isolated disruptive
players have a distractive structural signature with regard to their position within TF2’s
community. H4a argued that disruptive actors will assume a position of indirect brokerage. H4b
states that these players will also work to avoid closure in their local network. H4c suggested that
these players may have short, transitory ties with other members of the community.
Hypothesis 4a was generally supported. Network constraint was considered a good
predictor for every dataset except that drawn at Period One. The variable prediction plots for
each random forest demonstrated a steep drop off in the likelihood of disruptive behavior the
instant a player’s network was even slightly constrained. In other words, even the presence of
one closed triad (two connections of a node who are directly or indirectly tied to each other and a
disruptive player) dramatically reduced the likelihood of scamming. These results were also
borne out in the logistic regression, which demonstrated a strong negative effect when the model
fit approached that of the statistical learning approach.
CRISIS AND STASIS 124
Additionally, H4b was also generally supported. The larger a node’s effective network
size, the less closure it has in its immediate, local social environment and the more likely it is to
engage in disruptive play. ENS was generally an important predictor in the random forest
models. Additionally, increased network size was positively related to a higher likelihood of
disruption in all regression models except for Period One, which demonstrated significantly
poorer performance than the random forest model that presented opposite results.
Support for H4c was also strong in the random forest model. Variable importance metrics
consistently ranked average connection duration as one of the most important predictors. The
variable prediction plots demonstrated a spike in the probability of disruptive players for short
term relationships before levelling off for the models in Period One and Period Two. While
remaining an important predictor in the random forest, the variable prediction plots show a
reduction in the scale of the effect for the Period Three and Period Four models. This finding was
supported by the rare-event logistic regression where average connection age was significant
only in Period One and Period Two.
Taken together, these results paint a picture of private-isolated disruptive players
occupying a very distinctive position within the TF2 community network. Disruptive players are
more likely to assume brokerage positions and avoid closure as a byproduct of the process of
disruptive play. Due to their role as a bogus broker this fosters structural holes in their immediate
and indirect networks, bypassing the tendency towards bonding in most social networks.
Disruptive players are able to achieve this goal due to the relatively transparent social network
structures within TF2. Players can see who their friends are if they have any contacts in common,
making closure rather easy to detect. Given the obvious danger of a previous victim tipping off a
potential mark once they see a scammer re-enter their immediate social network avoiding closure
CRISIS AND STASIS 125
represents a sound and easy to execute strategy to increase the odds of success. Additionally,
their edges are more transitory, leading to greater churn in the development of relationships and
bypassing the temporal nature of bonding social capital. Once again this signature is afforded by
the technical nature of the Steam community. While friendships have to be mutually agreed upon
only one party is required to sever a connection. This helps protect against harassment or
unwanted connections but also allows scammers to cut and run after they have completed an
operation. These results point toward a structural signature that is associated with an excess of
bridging capital. When taken with the isolated nature of scamming (the benefits of a scam only
accrue to the perpetrator) even the development of bridging capital may be short circuited by the
tendency of disruptive players to engage in bogus brokerage instead of acting in the best interest
of the community
Having established that disruptive players occupy a distinctive position within the TF2
social network and that this position differs significantly from the majority of the population, the
next step is to see if increased levels of disruptive play are associated with structural course
adjustments in the TF2 community. Achieving this goal requires pairing two distinctive
timelines, the relative rate of disruption within the community and the introduction of significant
changes by the developer.
The first timeline was constructed using data pulled from SteamRep. Whenever a
disruptive player is reported to SteamRep, the report is adjudicated by the site’s administrators. If
accepted, the report is archived to provide quick reference in case the player is reported again.
This is in order to keep the site’s community from repeatedly judging the same event/player.
These archives include a unique identifier for each player recognized as disruptive as well as the
date of their report(s). Taking the first report for each player identifier, the date stamps of these
CRISIS AND STASIS 126
reports were used to generate a timeline stretching from early 2012 to early 2016 that includes
the number of reports for each week.
The timeline of changes to the TF2 economy is drawn from the game’s documentation.
Whenever a given game is updated, the vast majority of developers tell their players what the
updates were and why they were undertaken. These “patch notes” help players adjust to changes
in the rules or technical structure of the game and foster communication between the developer
and their constituent community. By moving backward through these texts and the associated
community para-texts that they generate, it is possible to find major changes to the technical or
regulatory infrastructure of the TF2 economy.
A qualitative review of the patch notes from 2011 to 2016 revealed six points of
significant change that introduced alterations to either the technical or rule systems that regulate
TF2 and its associated economy.

CRISIS AND STASIS 127
Table 12: Major events in the TF2 timeline.
ID Date Name Description Significance
1 2012-06-
14
Valve Hires
Economist
Valve hires economist
(and the future Greek
finance minister during
that country’s debt crisis)
Yanis Varoufakis to
analyze the Steam
Economy (Varoufakis,
2012).
Demonstrates pivot away
from seeing the Steam
marketplace as a game
add-on and towards
viewing it as an economy
2 2012-08-
15
Mann (sic) vs.
Machine Update
Introduced new
cooperative gaming mode
to TF2 (Valve, 2012).
Opened up the game to
players who do not enjoy
player versus player (PVP)
combat by enabling player
versus environment (PVE)
play.

3 2013-09-
04
New Trading
Infrastructure
Changed protocol for
trading allowing
prepackaged deals.
Additionally added new
user interface for trading
(Valve, 2013).
Altered the structure of
trades enabling new forms
of exchange. Also changes
how users view and
confirm a given deal.

4 2013-11-
01
Economist leaves
Valve
Dr. Varoufakis leaves
Valve for an academic
position (Cook, 2015).

No replacement academic
economist brought in as a
replacement

5 2015-12-
09
Additional trade
security
Added additional security
features and restrictions to
in-game trading. (Valve,
2015c).

Valve acknowledges that
restricting the flow of
goods within the market is
necessary to avoid fraud
and account theft.

6 2016-03-
03
Additional trade
security
Further upgrades to trade
security (Valve, 2016).

Valve extends previous
trade restrictions to further
combat the problem. Stops
restoring items after
successful trades if players
complain to the company’s
customer support.

CRISIS AND STASIS 128
Having these two timelines does not establish a relationship between them. Using the
approach laid out in the back half of the methods chapter, a seasonal hybrid extreme studentized
deviates time series approach (Kelly & Ahmad, 2015). To review, this approach decomposes any
potential seasonal trends within a dataset and establishes robust upper and lower boundaries
beyond which a score may be considered an anomaly. By looking for spikes or dips in the rate of
reported disruptive behavior within the TF community and associating this with the introduction
of patches or other significant changes, it is possible to see if the developer’s intervention
precedes or follows changes in the rate of disruptive play.

Figure 10: Rates of reported private-isolated disruption, statistically anomalous periods in red.
Vertical lines represent points from Table 11.
The anomaly detection procedure picked up one highly anomalous period of disruptive
behavior during the middle of September 2013. Interestingly, this spike did not come before an
intervention from Valve, but lagged behind it. Patch 3, the new trading system altered the user
interface for trades within TF2 and allowed for the creation of trade offers, and preset bundles of
CRISIS AND STASIS 129
goods that a prospective partner could accept, modify or reject. In reviewing the SteamRep
reports for this period, a common motif is that players were adapting to the new user interface.
This process created a window of opportunity where disruptive players could leverage the
community’s unfamiliarity to present duplicitous offers. The community eventually adjusted and
the new system appears to have lowered the overall rate of disruptive play, after the period of
adjustment.
These results do not support hypothesis two. Instead of functioning as a leading indicator,
creating a situation where the developer was forced to make changes to the regulatory structure
due to the displacement of social network structures by disruptive behavior, the results point
towards a lagging process. Disruptive players may have taken advantage of a shakeup in the
technical infrastructure of their community and exploited the new gray area to their own benefit.
Qualitative assessment of the SteamRep forums supports this assessment as player complained
about the confusing nature of the new system, but further research into the patching process is
needed to firmly establish this causal chain (SteamRep, 2013). This process differs from
emergent behavior in so far as the process of scamming is not new to the TF2 economy (Pearce,
Boellstorff, & Nardi, 2011). Rather the shift in the regulatory system provides more room for
existing patterns to manifest at an increased rate.
Discussion
Taken together the results of the classification and anomaly detection models present an
interesting composite picture. On one hand, disruptive players appear to occupy a distinctive
niche within a host community’s social network. The high variable importance scores for most
network-derived features indicate that the inclusion of social network data leads to an increase in
predictive power over straight demographic models. These network features demonstrate that
CRISIS AND STASIS 130
private-isolated disruptive players actively avoid network closure at the local and indirect level.
Additionally, what connections they do make tend to be shorter lived than their non-disruptive
counterparts.
Although the network signature of disruptive players differs from their host community,
spikes in disruptive play do not seem to generate a significant response from the developer. The
only anomalous jump in disruptive behavior detected by the SHESD model occurred after a
patch which changed the technical systems that govern trading, This suggests that private-
disruptive players look for emerging gaps in a community’s regulatory structure and exploit
these opportunities in the short term (as evidenced by their low tie duration). When a new set of
rules or regulations is introduced for an exogenous reason, such as improving user retention or in
response to a competitor’s pressure, the regulatory structures that govern a community are
temporarily perturbed. This creates room for further disruption. In other words, instead of
carving out a niche for themselves by breaking through existing regulatory structures, private-
isolated disruptive players appear to exploit exogenous gaps in the system.
Either way, the presence of these actors appears to be a significant problem given the
volume of security updates and content introduced to reduce the overall level of disruptive play.
However, after the 2013 spike, the rates of private-isolated disruption withinTF2 are remarkably
stable. This suggests two things, that both the community and the desire to break through
regulatory structures are relatively stable. If the TF2 community was not robust, it would not be
able to accommodate the introduction and creation of structural holes by private-isolated
disruptive players. But since the player base for the game has remained relatively stable, the
market continues to operate, providing a continued incentive for subsequent disruption.
CRISIS AND STASIS 131
The next chapter represents a significant change of focus towards public-disruptive
behavior. This change in pace allowed the results of the previous two chapters to be
contextualized within the context of a different community and a different mode of disruptive
behavior. After exploring public-isolated and public-collaborative disruptive play, each of the
various threads presented in the analysis chapters will be drawn together in a final discussion of
the relationship between disruptive play, and crisis and stasis within online communities.

CRISIS AND STASIS 132
Chapter Seven – Public-Isolated Disruptive Behavior
Public-isolated disruption captures glitches, bugs and other unintended consequences
embedded within a given regulatory system. These cases represent instances when elements of a
community’s regulatory system interact with each other in an unexpected or unintended fashion.
Players who discover one of these instances are occasionally able to exploit the gap to make a
given regulatory structure function in a way in which it was not intended by the developers
(Bainbridge & Bainbridge, 2007b). A well-known example of this type of behavior is the
“skiing” functionality in the game Starsiege: Tribes (usually referred to simply as Tribes). By
exploiting a bug in the game’s physics engine, players discovered that if they jumped at certain
times while moving down a hill they could start to accumulate momentum. With enough
distance, a player could build up enough momentum to go flying through the air at an incredible
speed, bypassing obstacles and making it difficult for other players to hit them. The bug was
labelled “skiing” and players began to exploit it, breaking the game’s rules, to dominate other
players who did not know the secret to this super-fast movement (“Skiing,” 2015). While these
disruptive players were frowned upon at first, skiing eventually spread throughout the
community representing a form of “emergent behavior” (Pearce et al., 2011). The developers did
not change the code to patch skiing out of the game and enterprising players wrote add-on
scripts, which in turn helped make skiing easier for other players (“Starsiege,” 2010). As a result,
the stigma against exploiting this particular bug fell away and the game adapted, becoming a
fast-paced airborne game with players flying in every direction.
Glitches, bugs and “cheese” strategies emerge from the complex interplay between the
various regulatory structures within a given gaming community (Moeller et al., 2009). These
structures are generally multifaceted and are usually generated by teams of designers, engineers
CRISIS AND STASIS 133
and other specialists. Fitting together these various structures to form a cohesive set of
regulations is not an easy task. Subtle interactions between various aspects of the code or rules
developed by different teams can lead to unexpected loopholes that violate either the spirit or
letter of various regulations (Consalvo, 2007; Kuo, 2001). As an example, the skiing bug came
about due to an interaction involving the sense of friction coded into the game. By jumping while
moving down a slope players could confuse the code to believe that they were not losing
momentum, allowing a character to reach incredible speeds (“Skiing,” 2015).
Skiing and other bugs represent public-isolated disruptive behavior because they are
generally discovered by individuals, but it is extremely difficult for a given person to keep a
monopoly on the benefits of this disruptive behavior (Meades & Canterbury, 2012). In the Tribes
example, one unknown player discovered the ability to ski and started to exploit the bug, but
because the game’s code universally applies to all the members of a community it was extremely
difficult to keep other people from making use of the same loophole. Other players could observe
the first skier within Tribes and notice that they gained speed when moving down a hill in a
certain way. By copying this behavior, they also gained the ability to ski and were able to join in
the process of disruptive behavior (“Skiing,” 2015). This differs significantly from the private
forms of disruption covered in the previous chapters as the benefits from these disruptive
practices (abilities earned through cheats or ill-gotten items) cannot be internalized and restricted
to only the disruptive player in question (Meades, 2015).
This difference in reward structure means that many of the structural signatures used in
the previous chapter may not apply. Unlike criminal or fraud networks, the rewards from public-
isolated disruptive behavior can flow freely, bypassing the need for network structures that
concentrate benefits or power on an individual or group. In other words, skiing within Tribes
CRISIS AND STASIS 134
does not detract from the ability of other players to exploit that bug, removing the need to keep
control over the benefits. In order to contextualize actions like skiing within existing literature, a
different approach is needed.
One promising approach is to conceptualize the knowledge of a strategy or glitch as a
form of information that diffuses through a given community. Players can discover these gaps in
a regulatory structure by accident or by testing the limits built into a community in an attempt to
discover an exploit (Bradenbug, 2006; Meades & Canterbury, 2012). Once the strategy or glitch
has been discovered and used, it begins to spread through several different channels. Players who
witness a disruptive player may be incentivized to copy them, learning how to bypass the game’s
regulations through mimicry Disruptive players may also publish guides or create other forms of
para-texts as a means of increasing their status within the community and sharing their discovery
(D. A. Fields & Kafai, 2010).
Regardless of the route, knowledge of an exploit or unfair strategy tends to spread
throughout the community, just like the practice of skiing within tribes. Not every member of the
community will be exposed to this information, and those who are need to have the motivation
and the knowledge to exploit it (Consalvo, 2007). This combination of factors closely mirrors a
diffusion problem, such as the classic diffusion of innovations paradigm (Everett, 2003b). This
sub-field of network analysis considers the spread of information (whether a new innovation,
meme, rumor or some other piece of data) through a network (Shifman, 2012; Valente, 1995). By
positioning the discovery and exploitation of a public-isolated disruptive practice as another
piece of diffusing information it is possible to draw from insights from this literature to establish
a likely structural signature for this class of disruptive player.
CRISIS AND STASIS 135
The first step down this path is establishing what type of diffusion best captures the
spread of public-isolated behavior. Speaking broadly, disruptive behavior can be broken up into
two general categories based on the processes that govern the spread of “contagious” information
throughout a network. Simple contagion mimics the spread of a disease (Centola & Macy, 2007).
Viruses and bacteria play little direct attention to the social circumstances of their host. Instead,
they exploit edges within human networks as they present themselves. Oftentimes a single
exposure is all that is needed to diffuse a disease within a given network. As an example, simple,
beneficial pieces of information often diffuse through a network like a disease (Weng, Menczer,
& Ahn, 2013). The news of a major world event spreads from person to person easily and
quickly, traversing single connections between disparate groups easily as word spreads.
Conversely, complex contagion includes socially sensitive or costly topics that diffuse
within a network (Brockmann & Helbing, 2013). The classic example of this is a new innovation
in policy or technology that is introduced into a community. Individuals may be reticent to adopt
the new innovation given the risk involved or costs sunk into the old way of doing things.
However, if enough members of the community adopt a behavior, this alters the social pressures
exhibited on hold-outs and late adopters (Kempe, Kleinberg, & Tardos, 2005; Valente, 1995).
Eventually pressure builds until a point of critical mass where diffusion accelerates rapidly until
the network is saturated.
The social stigma surrounding disruptive behavior and the competitive advantage that it
offers to those who do adopt it suggests that a simple, information contagion approach provides
the most useful framework. In many competitive games the use of bugs or “cheese” strategies
run contrary to common norms of fair play and equal competition (Moeller et al., 2009). As a
result, the social pressure within the community is sometimes oriented towards halting the spread
CRISIS AND STASIS 136
of disruptive information as opposed to encouraging it (Consalvo, 2007). Therefore, the social
pressure to spread a given innovation inherent in complex contagions is generally not at work for
disruptive strategies. Additionally, it is important to remember that in many games a player’s
social network includes both allies and competitors. As a result, the more connections who have
adopted a strategy the less useful it is as one’s competitors are already aware of the disruptive
behavior and can adjust accordingly.
Despite these points there is not a large-scale study of the diffusion of public-isolated
disruptive behavior over time within a gaming network. So while the aforementioned factors
suggest a simple diffusion process, it is important to leave room to disprove this assertion in the
subsequent hypothesis. In other words, if public-isolated disruptive play spreads through a
process similar to complex social contagion, the tests of the hypothesis later in this section
should be able to detect it.
Moving forward under the supposition that the spread of public-isolated disruptive
strategies within a gaming community mimics a simple contagion process, the next step is to
determine what structural signatures are associated with a greater likelihood of infection. While
the knowledge of a glitch or bug could theoretically originate anywhere in a community, certain
members of the associated network are more likely to be exposed due to their social position
(Meades & Canterbury, 2012).
Existing literature on the diffusion of memes and other forms of simple information
within online social networks suggests that individuals involved in tightly knit clusters rich in
closure will be less likely to adopt a contagious behavior. These tightly knit groups have two
factors working against exposure to contagious information. Since networks rich in closure are
often high in bonding social capital, participants generally have fewer connections, with their
CRISIS AND STASIS 137
time and energy being directed towards a smaller number of high intensity reciprocal
connections (R. D. Putnam, 2001). Additionally, the closed and inward looking structure of the
subgroup means that social attention is directed inside the local network. Therefore, unless the
glitch or bug is discovered within the closed subgroup or introduced through an alternative
channel, the lack of outward facing edges means that it is less likely to spread from the
surrounding community (Weng et al., 2013).
H5a- Public-isolated disruptive players will have lower levels of local network closure
then equivalent members of the host network.
Conversely, individuals in a brokerage position are significantly more likely to be
exposed to a simple contagion. By bridging the holes between various otherwise unconnected
parts of a network, brokers place themselves at the cross roads of numerous information flows
increasing the likelihood of exposure to new information (Burt, 1999). Additionally, nodes rich
in brokerage tend to have more, weaker ties as opposed to the few stronger reciprocal
connections favored by closure focused nodes (R. D. Putnam, 2001). As a result, brokers will
generally have more connections, each of which could be a potential channel for the spread of a
contagious piece of information (Weng et al., 2013).
H5b- Public-isolated disruptive players will be more likely to be in a position of indirect
brokerage in their host network.
Finally, it is important to recognize that the ability to recognize and exploit a glitch or
other loophole in a game’s regulatory system requires a degree of experience and knowledge
about these systems. A player needs to both recognize that the strategy in question provides
advantages above and beyond what would normally be available to them while having the
experience to implement the technique going forward (Meades & Canterbury, 2012). Both of
CRISIS AND STASIS 138
these abilities therefore require a degree of familiarity and mastery over the game and its
associated systems, two factors that are generally associated with time spent in game (Reeves,
Brown, & Laurier, 2009). Therefore, experienced players should be more likely to be disruptive
due to their ability to recognize and replicate flaws in the game’s code or rules.
H5c- Public-Isolated disruptive players will be more likely to be experienced members of
the community.
These tendencies have the potential to displace some predispositions towards closure
within a gaming community. Because disruptive players have access to abilities or roles that are
normally closed off by in-game regulations, they generally have an advantage over competitors
and challenges embedded into the game’s environment by its developers (Consalvo, 2007). This
decreases the need to form long-term connections with other players to overcome these
obstacles. As an example, before the advent of skiing, players in Tribes had to closely coordinate
their movements and help each other across the large in-game map (“Skiing,” 2015). Skiing
removes the structural need for this cooperation and lets an individual move quickly without
having to rely on teammates to get from one place to another. Studies of gaming communities
have repeated demonstrated that systems where no one player can achieve a high level of success
on their own lead to the formation of teams as members of the community are forced to work
together to overcome barriers (Ratan, Taylor, Hogan, Kennedy, & Williams, 2015; Shen, 2014;
Shen, Monge, & Williams, 2012). These initial contacts provide the seed of longer term
relationships which mature over time into smaller, stable networks based around co-play and a
shared in-game identity (D. Williams et al., 2006).
Disruptive players are less reliant on their teammates because they can draw advantages
from gaps in the game’s rules or code. As a result, the incentive to invest resources into long-
CRISIS AND STASIS 139
term reciprocal ties and form a stable team is reduced, a process compounded by the fact that
other players may look down on the disruptive player for violating norms of sportsmanship (Ang
& Zaphiris, 2010). This in turn degrades network closure and its corresponding stocks of
bonding social capital. This overall degradation to the health of the community poses a threat to
the health of the network. Players engaging with disruptive players are less likely to have a
positive experience or form durable social connections between themselves and their regulation
breaking counterparts (Moeller et al., 2009). This can lead to network fragmentation as non-
disruptive players become frustrated by their lack of in-game success and the social benefits that
spring from it. As a result, spikes in disruptive behavior should be associated with major changes
within the community by a given developer.
In order to assess these hypotheses this chapter draws from the popular multiplayer
online battle arena DOTA2.
DOTA.
The Defense of the Ancients Two (DOTA 2) is a multiplayer online battle arena game
(MOBA). MOBAs are a distinct genre where two teams of three to five players compete within a
limited map for resources (Drachen et al., 2014; Ferrari, 2013). MOBAs differ from shooters like
TF2 in so far as they use an overhead or isometric perspective, with players looking down on
their character as opposed to through their eyes. Additionally, MOBAs present complex resource
allocation problems to teams, with allies having to coordinate and split their time and effort
among a number of different tasks with varying degrees of specialization (Pobiedina, Neidhardt,
Calatrava Moreno, & Werthner, 2013). This provides a useful testbed for exploring public-
isolated disruption as the relative complexity of MOBAs provides a large room for potentially
disruptive players to find tactics which suit their needs.
CRISIS AND STASIS 140
The DOTA dataset consists of a sample of one month of matches in February and
September 2014. In the September sample, players were dealing with the presence of a disruptive
“cheese” strategy represented by the overuse of a character called “Tinker” within the game.
Tinker’s abilities enabled him to control several aspects of the game and made him extremely
difficult to beat, especially when used in conjunction with other characters who complemented
his abilities (“Is it really necessary to nerf Tinker?,” 2014). To counter this development, a patch
was released half way through the sampling period that “nerfed” Tinker, reducing his power
within the game and taking abilities away in order to bring his power in line with the rest of the
game’s roster (“6.82 Patch Notes,” 2014).
These data consist of logs of various matches taking place over the course of the month,
including information on who chose what character within the game, how they performed and
what team won. This enabled several networks to be constructed, one consisting of characters
joined by co-occurrence on a team and the other linking players who have identified each other
as friends within DOTA. The first network can be manipulated and permutated to create an
algorithm for detecting changes in the strategic landscape of the game, such as abusing Tinker or
other overpowered characters. These findings can be paired with the in-game social network to
determine the structural signature of disruptive players while controlling for their relative
experience and team performance.
Identifying Disruptive Players
Qualitative examination of DOTA related para-texts from 2014 reveals that there was a
series of complaints from that game’s community regarding the use and abuse of the character
Tinker. Specifically, one of Tinker’s abilities (March of the Machines) allowed him to exploit the
artificial intelligence that controls powerful non-player characters (NPCs) on DOTA’s map.
CRISIS AND STASIS 141
These NPCs provide bonuses to whichever character eliminates them, creating objectives and
points of contention within the map where the two teams have a chance to come together and
contest objectives (Eggert, Herrlich, Smeddinck, & Malaka, 2015). The NPCs in question had
been implemented as particularly tough enemies. Characters at the start of a DOTA match are
relatively weak and generally could not take on these foes and earn powerful bonuses. Instead,
eliminating them represented a mid to late game objective once the two teams had accumulated
enough power to overcome the strong computer controlled obstacles.
Tinker’s “March of the Machines” ability allowed him to damage these enemies from a
position where they cannot hurt the player back. This skill causes the character to summon a
wave of small, damaging robots that march across the game map hurting anything that they run
into across an extensive area. By camping out well away from the aforementioned NPCs and
using March of the Machines, a disruptive player could take advantage of a limitation in the
NPC’s artificial intelligence that keeps them from chasing a player too far. March of the
Machines outranged the hard limit coded into the game that determines how far an NPC could
travel, so a Tinker player can exploit the difference between their ability’s range and the limits
placed on NPCs to damage them without facing retaliation. As a result, a player using Tinker
could eliminate enemies that would crush other characters of an equivalent level
(AnEvilSnowman, 2013). The large rewards from defeating such a tough foe gives the disruptive
player a leg up and allows them to accumulate power more quickly and defeat the other team
before they are equipped to counter a high level threat.
Tinker’s abilities represent a glitch or “cheese” strategy. There is nothing stopping a
given player from adopting these tactics aside from the awareness that the March of the
Machines ability interacts with DOTA’s NPC intelligence. Therefore, the ability to engage in this
CRISIS AND STASIS 142
brand of disruptive behavior is public as the benefits cannot be actively restricted, anyone who
knows about the glitch could use it, even the other team. However, because only one character is
needed to use the Tinker strategy, the disruption falls into the isolated category of the two by two
typology established earlier.
Thus far the identification of the Tinker/NPC strategy as a form of disruptive behavior is
based on a qualitative assessment of player commentaries and para-texts. However, it is entirely
possible that these texts have been produced by a vocal minority, and other players have found
ways to counteract or avoid Tinker’s in-game advantages. While this does not change the fact
that the March of the Machines strategy is built around a loophole within the game’s artificial
intelligence, it does pose issues for detecting disruptive behavior quantitatively. If the disruptive
strategy was relatively obscure and Tinker did not become a primary character for disruptive
players, it may be difficult to disentangle disruptive players from their surrounding network.
By drawing two samples of DOTA games, one from February 2014 when the March of
the Machines strategy was not widespread versus one in September of that year then it reached
its peak, it is possible to compare the influence of the introduction of a disruptive strategy on a
heroes’ popularity. Specifically, the first sample is useful for establishing a baseline of Tinker’s
popularity within the game. Certain characters are more or less popular within DOTA and other
MOBAs due to a variety of factors. As an example, some characters are easier to play and
therefore attract a greater slice of the player-base. Others may have aesthetic qualities that make
them more appealing to certain demographics. If the Tinker disruption was widespread in the
DOTA community, it should incentivize potentially disruptive players to start using the character
at a higher rate in order to benefit from the aforementioned bug. While the community can push
back against the use of Tinker by using the “ban” feature which allows teams to prevent their
CRISIS AND STASIS 143
opponents from using specific characters this affordance requires that the other team know about
Tinker’s characteristics in order to make the informed choice and ban the use of the character.
Additionally, not all modes of DOTA feature the ban functionality, allowing Tinker use to
increase unchecked in these alternative modes (DOTA, 2016).
To assess this shift and demonstrate that the discovery of a disruptive strategy based
around Tinker increased that character’s popularity, two samples of 300,000 random matches
were drawn from February and September of 2014. Each match consisted of 10 players paired by
a matchmaking algorithm of two teams of equal skill. The selection rate of each hero was then
calculated, simply reflecting the number of times that hero was chosen divided by the number of
selections made in each sample. By comparing the difference in selection rates across the two
samples it is possible to determine which characters experienced increases or decreases in player
selection.

Figure 11: Change in hero popularity, February to September 2014.

CRISIS AND STASIS 144
Tinker experienced one of the larger positive swings out of all of the DOTA2 characters,
with only “the Faceless Void” demonstrating more growth. Faceless V oid’s improvement in
popularity can be attributed to a number of improvements or “buffs” that he received from the
developers over 2014 (FishingSloths, 2014). Tinker did not receive a major improvement in the
same time period, but experienced nearly as much growth. While this does not necessarily mean
that the emergence of the March of the Machines glitch drove all of this growth it does
demonstrate that Tinker became much more popular between the two sampling periods.
While measuring popularity points to an increase in Tinker’s usage, it is also important to
determine that character’s ability to engage in disruptive behavior shifted its location in the
strategic landscape of DOTA’s meta-game. The meta-game refers to the collected wisdom
accumulated through millions of DOTA matches regarding what strategies and combinations
work better together (Donaldson, 2015). Certain team compositions are more conventional
within the meta-game as previous experience has suggested that a specific selection of characters
or roles works well together. These conventions guide how players view their responsibilities
within the game and how different groups of five coordinate and assign roles to each other. If the
introduction of Tinker’s March of the Machine’s strategy led to widespread disruptive play, it
should also alter the meta-game as it adjusts to accommodate that character’s increase in power.
Specifically, Tinker should become a more conventional pick within the meta-game as
knowledge of the disruptive behavior spreads. In other words, if Tinker’s increase in popularity
was due to a factor other than the March of the Machines strategy, that character’s role in the
meta-game may not change. However, if players start adopting the new strategy, Tinker will be
used in different ways, shifting that character’s role in a team and altering DOTA’s meta-game.
CRISIS AND STASIS 145
Measuring change within a complex system such as DOTA’s meta-game is difficult.
Characters vary in their popularity over time and each team consists of five players interacting in
different ways. In order to say that a given team occurs more or less likely than would be
expected, it is essential to first establish a baseline of what those expectations are. One approach
for setting this bar which has been useful in other fields is to conceptualize the game as a
weighted network (Uzzi, Mukherjee, Stringer, & Jones, 2013). Each node in the network
represents one of the 108 different characters available within DOTA in September 2014. If any
two characters appear on the same team they are linked by edge, subsequent appearances
increase the weight of that edge by one. Following this protocol, the relative popularity of a
given character is expressed in its weighted degree score. Characters chosen more often by the
player-base for any reason (aesthetics, utility, promotions, etc.) will have a higher weighted
degree score. This network also captures the current meta-game. Popular combinations of
characters are expressed as sub-graphs within the network with more heavily weighted edges.
Conversely, less popular combinations will have connections with a lower total weighting.
Since the meta-game establishes certain combinations of characters that work well
together, the edges between the heroes in this combination will have a heavier weight within the
network, representing players adhering to conventional knowledge. Conversely, combinations
not popular within the meta-game will have a lower weighting. In other words, the aggregate
total of a node’s weight represents its relative popularity, but these connections are unevenly
distributed throughout the co-play network based on what team compositions work best with a
given character. By shuffling the distribution of these weights, it is possible to generate an
alternative concept of the meta-game that preserves the relative popularity of each hero while
shuffling the team compositions. A simple example helps illustrate this point. A random node in a
CRISIS AND STASIS 146
subset of the hero network might have four edges representing 100 matches with each edge
weighted 75, 10, 10 and 5. The heavily weighted edge represents a popular combination within
the current meta-game, while the remaining edges capture fewer conventional compositions. By
randomly permuting the weights, a new graph might have the same node with a distribution of 5,
20, 55 and 20. The total number of appearances is still 100, but the distribution of these uses
across different teams has been randomly altered. By comparing the known frequencies of a
given paring versus several thousand random permutations, it is possible to say that, controlling
for the relative popularity of each hero, a given pair occurs more or less often than random
chance. Each team consists of 10 pairs (hero 1-2, 1-3, 1-4, 1-5, 2-3, 2-4, 2-5, 3-4, 3-5 and 4-5)
each of which is more or less likely than random chance. This likelihood can be expressed as a z-
score, with a positive z-score representing a greater than random occurrence (i.e. adhering to the
meta-game, or being conventional), a negative z-score showing a selection rate below random
and zero representing the same selection rate as would be expected if heroes were chosen at
random (Uzzi et al., 2013).
Using this technique on the February and September samples, it is possible to evaluate
how Tinker’s role in the meta-game changed over the intervening months.
CRISIS AND STASIS 147

Figure 12: Z-score changes with 95% CI from February to September, 2014.
Table 13: Shifts in conventionality between February to September, 2014.

Hero Feb. Z-Score Sept Z-Score Difference
Legion Commander 1.727 0.563 -1.165
Alchemist 0.504 -0.233 -0.736
Bounty Hunter 1.260 0.617 -0.643
Crystal Maiden 0.641 0.056 -0.585
Witch Doctor -0.094 0.477 0.571
Phantom Assassin 0.824 1.497 0.673
Skywrath Mage -0.316 0.423 0.739
Sniper 1.545 2.287 0.742
Tinker -0.561 0.685 1.246
Faceless Void 0.071 1.478 1.407

Aside from Faceless V oid who received the aforementioned buffs, Tinker experiences a
dramatic increase in overall conventionality. This suggests that the increase in popularity shown
CRISIS AND STASIS 148
in table 11 reflects a repositioning of that character in DOTA’s strategic landscape. The
combination of these findings and the qualitative assessment of Tinker’s positon within DOTA
during the summer of 2014 suggests that the character was associated with widespread disruptive
play based on the March of the Machine ability. The corresponding increase in the Faceless
V oid’s popularity also offers an interesting opportunity to compare and contrast the structural
signatures of players who play a characters who’s increase in popularity is associated with
disruptive play (Tinker) with one whose popularity comes from within the boundaries of the
game’ regulatory structures (Faceless V oid).
Method and Data
Having established the prevalence of a disruptive strategy based around the use of the
DOTA character Tinker and a bug in that game’s artificial intelligence, the next step is to model
whether Tinker players have a distinctive structural signature within the friendship network
associated with DOTA players.
The data behind these models comes from February and September of 2014. The former
represents a snapshot of DOTA before the Tinker disruption became widely known, while the
later captures the game after the new strategy emerged.
The data consist of two parts for each sampling period. The game logs capture a record of
who participated in DOTA matches during the sampling window, what character they chose, and
general performance metrics. These data were cross referenced with social network data
capturing the number of friends a player has and the various metrics described in the methods
section such as constraint and network size. It is important to note that a significant portion of the
player-base have their DOTA profiles set to private. While a private profile restricts third parties
from viewing a given player’s data it does not change their visibility to other members of the
CRISIS AND STASIS 149
community in the game with the private player. There should be no significant difference
between the two groups, although further investigation is needed to confirm this supposition.
This means that it is not possible to situate these players in the network data as their unique
identifier is replaced with a generic string. After filtering these private players, the available
sample sizes are shown below
Table 14: Sample Sizes: Public-Isolated Disruption Model
Sample Players
Feb. 2014 153,028
Sept. 2014 103,643

Both datasets feature variables derived from the game log and network data.
Table 15: Dependent and Independent Variables: Public-Isolated Disruption Model.
Name Description Category Variable Role
Effective Network
Size
See methods chapter. Network Independent
Constraint See methods chapter. Network Independent
Friends who play
Tinker
Number of players in a
nodes immediate
network who have
played as Tinker during
the sampling window
Network Independent
Average connection
age
Mean duration of a
player’s friendships
Network Independent
Skill Level Players accumulated
ranking in DOTA two.
Proxy for skill and time
played
Player data Independent
Games Played Games played during
sampling window.
Binned into four
categories.
Player data Independent
Tinker If a player has played as
Tinker during the
sampling period.
Player data Dependent
Faceless V oid If a player has played as
the Faceless V oid during
the sampling period.
Player data Contrast
CRISIS AND STASIS 150

Results
Following the protocol laid out in the methods section a series of statistical learning
models and weighted regressions were fit to explore if it is possible to differentiate disruptive
players from their host networks while controlling for other factors.
The performance of each model was assessed using the area under the ROC curve metric.
Table 16: Absolute values of the Delong's D between each ROC curve. Significant scores mean
that the model with a higher ROC score performed significantly better than the alternative.
Gradient Boost
ROC = 0.723
Random Forest
ROC = 0.711
Rare Event
ROC = 0.723
Gradient Boost
ROC = 0.723
-- -- --
Random Forest
ROC = 0.711
0.915 -- --
Rare Event
ROC = 0.723
0.049 0.779 --
*
p<0.1;
**
p<0.05;
***
p<0.01

The performance of the various models in both sampling periods was tightly clustered. In
both cases the weighted rare event regression models performed better than, or close to, the
selected statistical learning approaches. Delong’s test of two correlated ROC curves was
insignificant between all three tests indicating equal performance across all three techniques
(DeLong, DeLong, & Clarke-Pearson, 1988). Given the easier interpretability of the regression
and its relatively comparable performance, it functions as a means to assess the hypothesis laid
out at the start of this chapter without having to rely on variable prediction plots or variable
importance. The coefficients for both models are presented below.

CRISIS AND STASIS 151
Table 17: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Public –
Isolated Model.
Dependent variable: Tinker Use

Feb. - Tinker Sept. - Tinker Sept.- Faceless
Network Constraint -0.084 0.402 0.060
(0.103) (0.241) (0.053)

Effective Network Size -0.001 0.002
***
-0.0002
(0.0003) (0.001) (0.0003)

Average Connection Duration -0.001 -0.001 -0.001
***

(0.000) (0.000) (0.00001)

Skill Level -0.122
***
0.099
***
0.131
***

(0.011) (0.012) (0.005)

5-11 Games -0.801
***
-2.651
***
0.561
***

(0.079) (0.193) (0.041)

12 – 19 Games -1.318
***
-3.532
***
1.035
***

(0.078) (0.257) (0.038)

20+ Games -1.910
***
-5.308
***
1.580
***

(0.077) (0.709) (0.039)

Number of Connections
with Other Tinker Players
0.009 0.010 --
(0.011) (0.008) --

Constant 1.259
***
-3.877
***
-6.043
***

(0.239) (0.252) (0.103)

Observations 82,915 122,423 122,424
Log Likelihood -15,346.510 -2,808.528 -34,547.960
Akaike Inf. Crit. 30,709.010 5,635.056 69,111.920
McFadden Pseudo R
2
0.0761 0.1429 0.0862
CRISIS AND STASIS 152

Note:
**
p<0.05;
***
p<0.01

Discussion
The results indicate a strong discrepancy between the two time periods. The different area
under the ROC curves and R
2
s indicate that the September model was able to use the provided
features to differentiate between Tinker players and the surrounding population with a higher
degree of accuracy. With regards to the hypotheses, support was mixed. During the September
sampling window when Tinker was being used as a disruptive character, effective network size is
positively associated with usage ( Est= 0.002, p < 0.01). This suggests that players with more
contacts and less closure in their immediate local networks are more likely to engage in
disruptive behavior. Indirect brokerage as operationalized through network constraint did not
have a significant influence on the likelihood to disrupt. However, skill level was positively
associated with Tinker use.
While this structural signature is also associated with the diffusion of new, legitimate
strategies within the DOTA network it is important to note that the Tinker strategy occupied a
different social space than its non-disruptive counterparts. Because the crux of the strategy
depends not on the skill of a given player or the interaction between members of a team but a
bug in the game’s code the player-base tended to view the March of the Machines approach as a
form of “cheese” strategy. Unlike other players or teammates NPCs within DOTA are static and
unchanging from match to match. So while a new strategy can be countered players cannot
suddenly make the NPCs immune to Tinker’s abilities
The shifts in the coefficients between the two periods demonstrates how disruptive play
interacts with a player’s network structure. Tinker players from the February time slice are not
CRISIS AND STASIS 153
significantly different in their network structure from the surrounding network. In other words,
there is no distinctive network signature associated with playing Tinker before the disruptive
glitch was identified. Additionally, as players increased in skill level, their likelihood of playing
Tinker decreased, suggesting that the character was avoided by high level players. This finding is
further supported by the exploration of the DOTA meta-game that shows Tinker being used less
often than would be expected by random in the February sampling window. However, once the
exploit combining March of the Machines and a bug in DOTA’s artificial intelligence was
discovered, the direction of these effects reverse. Tinker players become more likely to have a
large number of local ties with little closure. Additionally, skill level becomes a positive
predictor of Tinker use, with more experienced players adopting the character. This suggests a
shift among Tinker players towards a new section of the player-base, one which is experienced
and has a lot of connections to different parts of the attached social network.
Interestingly the networks patterns associated with Faceless V oid players differed
significantly from Tinker players. Players who chose Faceless V oid did not occupy a distinct
position in their surrounding social network except their connections tended to be newer than
other players. In other words, while both characters experienced comparable shifts in popularity
and an increased role in the meta-game the network signature associated with Faceless V oid do
not mimic that linked to Tinker.
The alignment between the structural signature associated with Tinker and the process of
simple diffusion highlights the similarity between these two processes. This is telling given the
contrasting case provided by Faceless V oid. While the changes in the regulatory structures that
benefitted Faceless V oid were publically communicated by the developer Tinker’s advantages
appear to come from an understanding of the game’s para-texts or knowledge of the particular
CRISIS AND STASIS 154
bug which facilitated the March of the Machines strategy. In other words, disruptive behavior
may be forced to diffuse throughout the network due to its illicit nature, while buffs and other
changes rooted in intentional shifts in the game’s regulatory structure are well documented and
evenly available to all members of the community. Further research is needed to disentangle
whether these patterns simply resemble diffusion or capture the actual process the difference
between the structural signatures of Tinker and Faceless V oid point towards the different routes
that a particular strategy can take towards popularity, and how these paths are reflected in their
adherent’s social networks. The growth and diffusion of the Tinker disruptive strategy led to an
intervention to counter the glitch. In a patch, which is cited as one of the most wide ranging
changes in the history of the game, Tinker’s abilities were altered so that power NPCs were
immune to March of the Machines (“biggest nerf in dota history?,” n.d., “Is it really necessary to
nerf Tinker?,” 2014). As a result, the entire basis of the disruptive strategy was removed. These
abilities “nerfed” Tinker to such a degree that the character fell out of favor within the
community, relegating the character to second-class status.
The fact that players with a high degree of experience and many non-redundant
connections were more likely to engage in disruptive behavior may explain the scope of this
shift. The experienced members of the community had a larger than expected number of
connections to non-redundant groups, so simply banning the players who made use of the Tinker
bug would remove a significant and well connected segment of the community. While this would
send a message that exploits and other forms of public-isolated disruptive behavior are not
welcome within DOTA, removing experienced local brokers with relatively durable connections
is also a great way to fragment a social network. Therefore, if the player side of disruptive play
cannot be changed, the only remaining mutable factor is the regulatory structure itself. As a
CRISIS AND STASIS 155
result, patch 6.82 in late September introduced the aforementioned changes and dramatically
altered the balance of DOTA, forcing the community to reorient and adapt to a new set of
regulations embedded within the game’s code.
CRISIS AND STASIS 156

Chapter Eight - Induced Disruption
This chapter tackles a special case of disruptive play, which takes place when players
within a gaming community find themselves breaking through regulatory structures in response
to those structures shifting around them. To explore this phenomenon and understand how it
feeds back into the process of change over time within a gaming community, this chapter will
first summarize the research on the social capital structures generated through non-disruptive
gameplay. This discussion is followed by a brief exploration of online protest, a common form of
induced disruption that links the structures necessary to engage in protest with the cooperative
structures created by players through day-to-day play. This conceptual model provides the
framework for the quantitative aspects of this study, an examination of the 2011 EVE Online
protests and their subsequent fallout.
Play and Protest
As stated in the opening chapters of this examination, the construction of network
structures associated with social capital is often an implicit design goal for developers creating
an online gaming community (Koster, 2013). The two varieties of social capital, bridging and
bonding, offer a number of advantages to both players and developers. To recap, bonding capital
refers to networks of trust and reciprocity formed within groups over time, bridging captures
connections that span these groups (Adler & Kwon, 2002). These ties are often weaker than
bonding links, but offer a route for information and coordination across gaps within the host
social network and facilitate the flow of information between groups (Burt, 2009; Granovetter,
1983).
CRISIS AND STASIS 157
The existing literature on gaming communities suggests that both varieties of social
capital are present within these venues but bridging capital is more common (Steinkuehler &
Williams, 2006). Bridging offers a number of advantages to players. It allows players to draw
upon information and resources from a wide variety of groups within the game. For new players,
this promotes learning and development as they are introduced to different strategies and tactics
within the game-world (Shen et al., 2011). Numerous studies have demonstrated that players
embedded in networks rich in bridging capital display higher performance within the game and
have access to groups or information not available to their less well positioned counterparts
(Shen, 2014; Shen et al., 2012).
Bonding social capital is generally less common within gaming communities but is by no
means unheard of. Bonding represents the overlapping, reciprocal relationships of trust that
mutual obligation that form within groups over time. At its most intense level, bonding captures
the long-term obligations between members of the same family or close friends. Therefore, this
form of bonding capital reflects a feeling of trust and reliability that emerges from being
embedded in densely clustered networks of social engagement (J. S. Coleman, 1988). As an
example, a gaming group composed of members who are all friends with each other creates a
tightly knit network within the game’s social graph. These overlapping bonds to the various
members of the community mean that if a member was to drop out or slack off with regards to
their responsibilities within the group, they face the overlapping social sanction of their
colleagues (J. S. Coleman, 1988). The threat of this sanction helps maintain group coherence
while allowing for better coordination between the bonded individuals (Burt, 2001; R. D.
Putnam, 2001). Additionally, the repeated and highly redundant network structures promoted by
bonding social capital are more robust against a single member of the group leaving (due to
CRISIS AND STASIS 158
unforeseen circumstances or simply choosing to quit the game) as the other ties between
members can shift to accommodate any changes.
Developers encourage the formation of bridging and bonding capital in a variety of ways
(Ducheneaut, Moore, & Nickell, 2007). Within many games bridging is fostered through the
development of an extensive market system. Many multiplayer games feature complex and
responsive in-game economy where players trade and exchange virtual goods (Castronova,
2001). Within these markets certain goods are often only available to certain sections of the
population, or are rare and therefore experience relatively high demand. Many players earn in-
game wealth by transporting or trading items or virtual wealth within their game’s economy. This
flexibility promotes the development of ties across and between groups as players look for new
opportunities to trade or find profit (Lehdonvirta & Castronova, 2014).
Further reinforcing this trend is the need for players to specialize on certain tasks within
the game. Many games feature a role system that encourages cooperation between players. A
single person can specialize their in-game character(s) on a small number of skills or abilities.
While a given player can become incredibly proficient in these skills, time or in-game
restrictions on resources mean that they cannot master every ability or role at the same time
(Ducheneaut et al., 2006; Shen et al., 2011). In order to cover their weaknesses or gain
advantages that are otherwise closed off due to their choice of specialization, players need to
search for compatriots who have specialized complementary abilities (Shen, 2010). As certain
skills are more or less common, based on the time and effort needed to develop them, it results in
players casting a wider and wider social net to find allies with the proper abilities to complement
their own (Shen et al., 2011).
CRISIS AND STASIS 159
While bridging social capital is often rooted within a game’s market and skill systems
bonding tends to emerge from the grouping functionality included in many popular games. These
organizations, often called guilds or corporations, are persistent in-game groups with a delineated
membership and identity (D. Williams et al., 2006). Players can apply to join a guild and get
access to a pool of shared resources. Additionally, many guilds feature the ability to call upon the
group’s reputation in dealings with third parties as well as a common set of enemies and allies
shared among all members of the group. Players involved in the same guild have assurances built
into the social fabric of the game and its code, which makes it difficult to attack each other. This
combination of common enemies and a baseline level of safety when dealing with members of
the same group means that members of the same guild are likely to form friendships or alliances
with each other (Ratan et al., 2010). Overtime, this proclivity leads to the development of dense,
reciprocal networks within some guilds (D. Williams et al., 2006). This facilitates trust and the
creation of bonding social capital as each member of the group works to support their
compatriots due to their shared identity and the threat of social sanction from mutual friends.
Network Flexibility. When a community is functioning normally, these structures and
the activities they coordinate are channeled towards objectives constrained within the regulatory
boundaries of the game. However, both the network structures and the resources they create are
agnostic in relation to how they are used (McAdam, 1999). In other words, the resources and
network structures nurtured by developers and used by players are not required to be spent on in-
game objectives. Trust, contacts, information and other characteristics of networks can be
repurposed and remobilized quickly in light of changes in the network’s setting.
Sometimes these changes to a network’s setting alter the regulatory structures that
constrain or incentivize certain patterns of behavior within a community. Induced disruptive
CRISIS AND STASIS 160
behavior represents a situation where the network’s structures fostered by developers are
repurposed and directed against regulatory structures. Often this remobilization is the result of
shifts in structures due to exogenous market or technical forces acting on the developer.
An example of this shift is the well-studied case of Star Wars Galaxies (SWG).
Developed by Sony Online Entertainment (SOE), Star Wars Galaxies was a massively
multiplayer gaming community based around the Star Wars intellectual property. Players took on
different roles from the Star Wars universe, acting as bounty hunters, smugglers and other space
faring adventure seekers (Ducheneaut, Moore, et al., 2007). The game was successful and
fostered a medium sized, dedicated community of players. However, facing pressure to expand
the game’s popularity, SOE released a series of changes to the game called the “New Game
Enhancement” which substantially altered the rules and gameplay of SWG (Schiesel, 2005).
These changes shifted the game’s regulatory structures, opening up previously restricted abilities
for new players and simplifying the game as a whole. For existing players, the New Game
Enhancement represented the regulatory structures which helped structure their participation in
the community shifting around them. Previously available goals or forms of social interaction
were simplified or removed altogether, placed behind the formidable barrier of the game’s
modified code. Existing social structures within the community remobilized and repurposed as a
result, demonstrating network flexibility. Some guilds simply pulled up stakes and left,
transferring their money and stocks of social capital en masse to competing games (Schiesel,
2005). Others protested and campaigned to have the changes reverted so SWG’s regulatory
structures would return to their original state (Jenkins, 2006c). Under the typology advanced
earlier in this dissertation, these actions would be considered a public-collaborative form of
disruptive play. Those players angry about the shift in SWG’s rules were working towards a goal
CRISIS AND STASIS 161
that would be shared among all members of the community, not just the activists, making it a
public form of disruptive play. However, a single player cannot bring enough pressure to
influence a developer, so collaboration is needed, filling in the second half of the classification
schema.
The key take-away at this point is the relationship between regulatory change and the
repurposing of in-game stocks of bridging and bonding social capital towards protest or other
forms of mobilization. This common thread provides a route for importing other concepts from
social movement theory to describe the process of induced disruption. Obviously a protest within
an online gaming community does not require the same organization as an offline social
movement studied by political scientists, and it usually does not carry the same risks (Bennett &
Segerberg, 2012). However, the basic principle that advocacy groups often exploit existing social
networks and stocks of social capital is a key point for understanding induced disruptive play.
The structures created through the day to day process of play can be re-appropriated and re-
purposed for the new movement and the stocks of bridging and bonding social capital generated
by a community can be redirected to overcome one of the key barriers any form of mobilization,
a collective action problem.
Collective Action Problems. Collective action problems address the management and
distribution of public goods, a non-excludable and non-rivalrous benefit from some form of
action (Olson, 1965b). As an example, imagine a generic lobbying organization that is acting on
behalf of some group within society. The members of this group expend time, effort and
resources to achieve a specific goal that benefits both themselves and their community. Often
this benefit is non-rivalrous as one person taking advantage of it does not diminish its utility for
other people (Olson, 1965b). Additionally, the outcomes of this activism may also be non-
CRISIS AND STASIS 162
exclusionary. That is to say that it is impossible (or at the very least rather difficult) to prevent
other people from taking advantage of a benefit. This situation creates a dynamic where members
of the community in question have an incentive structure that promotes non-participation,
otherwise known as free riding (Olson, 1965b). If the benefits fall evenly to all members of the
community, regardless of whether they agitated to obtain them, the most effective solution is to
leave the hard work to others while reaping the rewards of their effort (Olson, 1965b).
The dynamic between the public goods dilemma and consumer or social movements has
been extensively studied. Indeed, some authors have asserted that the fundamental goal of social
movements is to provide incentives and reasons to participate that go above and beyond the
latent impulse to free ride (Moore, 1995). These incentives can take many forms. The most
common of which is to co-opt existing traditions of reciprocity and mutual obligation and
redirect these obligations towards a new goal. As an example, in a dense, closed network rich in
bonding social capital, there may be longstanding debts and obligations between members that
are tolerated due to a high level of interior trust. If a member of the community decides to
contribute to a movement working towards a collective action problem they can “call in” these
social obligations to promote participation (Tarrow, 1998). This process is further compounded
by the high level of closure within networks rich in bonding social capital. If multiple people
within a group end up participating, it can create a tipping point scenario whereby the remaining
members find that the majority of their local network has chosen to contribute (González-Bailón,
Borge-Holthoefer, Rivero, & Moreno, 2011). This leads to social pressure on stragglers not to
free ride and helps spread participation even further.
In addition to the use of network structures associated with bonding social capital,
organizations that address the public goods problems also draw upon bridging capital. The
CRISIS AND STASIS 163
existing literature has demonstrated the importance of leadership figures in these efforts (Morris
& Staggenborg, 2012). Leaders are often well respected by their constituent communities and
have a great deal of influence over those beneath them. When they throw their weight behind a
collective action issue they tend to bring their constituent networks with them.
This research also argues that leaders within networks are often rich in bridging social
capital. Elite members of a network often have to manage multiple flows of information and
coordinate their subordinates (Balkundi & Kilduff, 2006). These demands promote the adoption
of a brokerage position by leaders, connecting groups of subordinates and controlling the flow of
information between them (Diani, 2003). As a result, individuals rich in bridging social capital
and occupying a brokerage position within a network are often critical actors who can incentivize
participation amongst their subordinates. Leadership can take many forms within a gaming
community. Players can be leaders in their guild, their social network or particular section of the
in-game economy, Leadership of any form does not necessarily entail participation in the EVE
protests, but leaders are in a position where if motivated to participate they have the resources
and skills to operate effectively. In other words, leadership within a gaming community provides
a helpful, but not essential framework for mobilization.
To summarize, protest or consumer activist movements face a collective action problem
due to the public nature of their goals. Overcoming this barrier requires the application of liberal
quantities of bridging and bonding social capital to provide leadership and a social sanction of
the nascent movement. The source of these resources is of secondary importance as network
structures can be repurposed and remobilized in light of a changing situation, as demonstrated by
the rapid repurposing of ties formed through social networks such as Twitter or Facebook to form
protest networks (Theocharis, 2013; Tremayne, 2014). Therefore, the structures fostered by
CRISIS AND STASIS 164
developers to improve the health of their constituent communities also provide a route for these
communities to push back against the developer in the event of unpopular shifts in one or more
regulatory structures. Within the typology laid out in Chapter Two this makes these players a
distinct and special brand of disruptive players. While the protests within EVE broke through
existing regulatory systems by attacking the in-game economy and servers the benefits of this
disruption relied on the cooperation of a large number of protesters. Additionally, these benefits
could not be effectively restricted to those who chose to participate in the protest. Players who
were not active or did not care about the Incarna patch were still effected by the results of the
mobilization. As a result, this form of induced disruption occupies the unique public-
collaborative disruptive quadrant within the framework established in Chapter Two. While the
flaunting of in-game regulatory systems places the EVE protesters in the same broad family as
glitch users or other parties which exploit the technical infrastructure of a game the organization
and objectives of the movement set them apart, hence the special label, “induced disruption”
which captures the most common case where public-collaborative disruptive behavior emerges.
Taken together this framework suggests that players rich in bridging and bonding social
capital will be more likely to respond with mobilization and protest if they find themselves
disappointed by shifts in a game’s regulatory structure. Since these resources do not emerge
overnight, disruptive players must have experience and time spent in game to foster their stocks
of social capital. Therefore:
H5a- Public-collaborative disruptors will be more experienced than non-disruptive
players.
CRISIS AND STASIS 165
Even with time spent in-game there is still the incentive to free-ride due to the public
goods dilemma inherent in most collective action problems. Therefore, disruptive players should
have high stocks of bonding social capital to push them towards participation.
H5b -Public-collaborative disruptive players will have higher than expected levels of
network closure given the size of their local networks.
Finally, complex forms of mobilization like an in-game protest require coordination and
leadership. These roles tend to fall to people with higher stocks of bridging capital, making them
more likely to participate in induced disruptive behavior.
H5c- Additionally, public-collaborative disruptive players will be more likely to assume
brokerage positions.
Having explored the relationship between game design, social capital and mobilization, it
is now time to explore an applied case study of induced, public-collaborative disruptive play; the
2011 EVE Online Incarna protests.
EVE and Micro-transactions. EVE is a massively multiplayer online game set in space.
Players take the role of a "capsuleer" a powerful pilot who can control ships ranging from small
fighters to massive sky-scraper sized capital ships (CCP Games, 2011a). One of the defining
characteristics of the game (according to its players) is its relative difficulty (Bergstrom et al.,
2013). The EVE community prides itself in being distinct due to the fact that EVE is such a
complex game, with numerous moving parts and overlapping systems. The game's steep learning
curve means that those who do succeed within the community cast themselves as an elite class
due to their ability to survive and thrive in a hostile environment (Bergstrom, 2013).
While EVE's difficulty leads to an extremely dedicated player base, it also limits the
relative size of the game as new players will often quickly leave the community (Bergstrom et
CRISIS AND STASIS 166
al., 2013). CCP Games' revenue model is based around subscriptions, which means that a smaller
player base directly damages their profits. Having recently acquired the intellectual property to
create a new game, the studio began to explore ways to increase the total revenue being
generated through EVE by both increasing the value of existing members of the community and
providing a mechanism that makes it easier for new players to join (Grayson, 2011).
This strategy eventually coalesced into the "Incarna" patch, an update to the game that
introduced the sale of in-game goods for hard currency (Smith, 2011). In its first iteration,
players could only purchase cosmetic goods, items which customized an in-game avatar without
influencing performance in any way (Smith, 2011). However, a series of leaked memos and
strategy documents indicated that CCP was planning to expand the program to include the
purchase of other goods such as in-game ships (Edwards, 2011).
These leaks prompted a number of responses, ranging from classic consumer advocacy
techniques like boycotts and social media pressure to in-game protests. The latter of these two
actions is particularly interesting from the perspective of disruptive play. Disgruntled EVE
players began to actively disrupt the in-game economy which supported them (Groen, 2011).
Congregating in the “Jita System” (a central trade and exchange hub within EVE) players
abandoned competition and began to spend their resources firing on a large central monument
erected by CCP (Edwards, 2011).
This maneuver represents an active challenge to the norms, rules and code of EVE. From
the perspective of in-game norms, the cooperation across rival factions for purposes other than
profit is a striking divergence from the “winner-take-all” ethos popular among EVE players
(Bergstrom et al., 2013). The protests locked down major trading hubs within the game, making
them all but impassable and fragmenting the transportation network that many players relied on
CRISIS AND STASIS 167
to earn a living within the game. Additionally, loading so many players into one area within the
game created a high level of stress on the game’s servers forcing the developers to modify the
game and slow time within the affected areas so that the servers could keep up with the load.
The Incarna protests are an excellent example of an induced, public-collaborative form of
disruptive play. The protesters were aiming for the rollback of the patch and a refocusing of
CCP’s resources onto their community, a public good that would affect all EVE players, not just
those protesting. The scale and scope of the protests also required collaboration among players; if
everyone worked independently there would not be enough disruptive players in one place to
successfully attack EVE’s economy and technical infrastructure. These two factors demonstrate
the public-collaborative nature of the Incarna disruptions.
Having established the setting for this chapter’s exploration of disruptive play it is time to
begin exploring the data and the various features at work within it.
Method and Data
In order to assess these hypotheses, this section draws from a population level dataset of
EVE online players. The dataset runs from May 2003 when the game started to the last week of
June 2011, stopping immediately before Incarna protests. The data originates from CCP Game’s
logs of EVE Online so while it is remarkably complete the alignment with the protest period is
coincidental and the dataset cannot be extended to cover the actual protests.
For each player within the EVE community, the data include basic metrics capturing what
corporations they joined within EVE, how many characters within the game they created, and
basic user demographics. This data is supplemented with static information on EVE’s directed
friendship network, which was drawn seven days before the eruption of protest over the Incarna
patch. Before exploring how this data was synthesized and constructed into a social network, it is
CRISIS AND STASIS 168
important to summarize what each distinct table within the dataset contains, and what is included
within the subsequent analysis.
Demographics. Information about the members within the EVE community can be
divided up into two broader categories. Character-level metrics capture information associated
with particular avatars within the game; user level data is tied to specific people playing the
game. Each user can have up to three characters tied to a given account at one time.
At the character level the data include the virtual race of the character (players can select
from one of four in-game races, each with different advantages and disadvantages), the age of an
avatar, its gender and how powerful it is within the game (i.e. how many skill points have been
awarded to that character). In addition to metrics associated with a given avatar, there is also
demographic data tied to the individual controlling that character, namely their gender and date
of birth. For a small subset of the population, no date of birth was provided, leading the database
to assign them a default birthdate in the late 1970s. To control for this discrepancy, an additional
variable was created to control for users who lack this information.
Constructing a social network out of the EVE data relies on static snapshot of all of the
social relationships within the game. Social connections within EVE work differently than most
other online communities. Connections within EVE represent a form of directed relationship, so
if one person adds another to their contact list there is no need for reciprocation for an edge to be
formed. Additionally, once someone has been added a user can flag them as a friend or an enemy
within the game. The valence of a relationship influences how the contact appears within the
game. When a contact is first added they appear in white, the same color as other random players
within the game. Marking a contact as a friend shifts their color to green, while enemies are
drawn out in red.
CRISIS AND STASIS 169
With this combination of network and contact data it is possible to construct a static
social graph of the entire EVE network several days before the Jita protests. Each node in this
network represents a user, with the descriptions of all of their characters aggregated upward to
one account. Collapsing multiple avatars onto one account ensures that the social graph is not
distorted by the fact that users tend to add their own avatars as contacts within the game, altering
their local social network (Leavitt et al., 2016). Each of these nodes is linked by any contact list
relationship where the user has flagged their connection as being neutral or a friend. This ensures
that negative relationships, which are generally used within EVE for surveillance as opposed to
social exchange do not influence any subsequent network metrics. Once constructed this network
serves as the basis for generating the network metrics laid out in the preceding methods section,
namely Burt’s Constraint and Effective Network Size.
Table 18: Independent Variables: Private-Collaborative Model
Table 8
Name Source Description Role
Time Played Character Data Total minutes played
aggregated across all
avatars associated
with an account.
Assesses H4a
Skill points Character Data Total skill points
acquired aggregated
across all characters
associated with an
account.
Assesses H4a
(alternative
conception of time in
game)
User Age User Data User’s age in years Control
Age provided User Data Binary control for if
an age was provided
for a user.
Control
User Gender User Data Years gender (M/F) Control
User Number of
Characters
User Data Number of Characters
associated with an
account
Control
Effective Network
Size
Contact Network See “Methods”
chapter
Assesses H4b
CRISIS AND STASIS 170
Constraint Contact Network See “Methods”
chapter
Assesses H4c

Identifying Disruptive Players. The previous variables are all explanatory features that
attempt to predict disruptive behavior; the next step is to identify players who engaged in
disruptive behavior within the game itself. Because the dataset containing the independent
variables ends a few days before the Jita protests it cannot be used to isolate protesting players.
However, EVE exists within a constellation of associated forums, message boards and associated
communities that are broadly labelled as para-texts (Consalvo, 2007). The official CCP Games
forums were a critical space for the Jita protests where users could interact with the developers
and make their case. Because both the forum and the game itself are run by CCP they share a
common identity system. Players use the same account to the login to both venues. Therefore, it
is possible to cross-reference the forum with game data and connect opinions within the para-text
to in-game actions.
The first step in cross referencing these data is to collect supplemental data from the
forums. Using a Python scraper, every post on the EVE forums that was created during the Jita
protests was downloaded and collated into a database. Each of these posts is associated with a
unique numeric string that represents the avatar who wrote it, an identifier that can be cross-
referenced with the other dataset laid out earlier in this chapter. However, just because a user
posted in the forums during the Jita protest doesn’t mean that they were engaged in disruptive
behavior. Some players chose to carry on with business as usual while others actively opposed
the protests.
Separating these various groups requires parsing their contributions to the forums. Given
the volume of posts (12,253 posts made reference to the Jita protests during the week they were
CRISIS AND STASIS 171
active) manual coding is extremely difficult. Fortunately, advances in text analysis have made
identifying topics from large corpuses of text significantly easier.
Z-Label Latent Dirchlet Analysis (Z-LDA) is a semi-supervised text analysis technique
that allows for a directed exploration of large bodies of text (Andrzejewski & Zhu, 2009). Z-
LDA starts with a series of “seeds,” keywords that have been identified as important using
qualitative analysis of a corpus of text. From these core terms, the model iterates through the text
looking for other words that have a high co-occurrence with the seed words and low co-
occurrence in other settings. This leads the clustering of specific phrases into semantic topics,
words which have been found to frequently co-occur within the corpus. To take an example from
outside EVE, the word bird would be clustered with terms like fly, peck or sing while the term
dog would find itself in a topic with words like bone, bark or wag. Words more likely to occur in
a given topic are assigned a greater weight, and documents within the corpus are sorted based on
the relative weight of the words the author used.
One of the difficult aspects of Z-LDA models is deciding how many topics the model
should create out of a corpus. Too few topics will create overlap and confusion between
unrelated concepts, too many will lead to fragmentation and make any results more difficult to
interpret. Instead of picking the number of topics in an ad hoc way, a series of statistical
measures have been developed that gauge what topics maximize the amount of explained
variance within the corpus while minimizing intra-topic error. This examination uses the most
recent of these metrics by Deveaud, Romain and SanJuan (2014). This approach works by
searching iteratively over a number of different topic modelling solutions looking for the number
of distinct topics that maximizes divergence between the various groups. Choosing the number
of topics is therefore similar to examining a scree plot from a principle components analysis. The
CRISIS AND STASIS 172
optimal number of topics represents the point where the between topic divergence stops
increasing at a high rate.

Figure 13: Z-LDA topic selection results.
By these metrics the Z-LDA model achieved optimal convergence at 5 topics. Each topic
features distinct words which are heavily weighted in that category. Exploring these top words
therefore helps determine what each topic represents.

CRISIS AND STASIS 173
Table 19: LDA keywords.
Topic Selected Top Loading Keywords

0 EVE, CCP, develop item, player, vanity, option, ship, time, don’t, pay, display,
run, newsletter

1 Game, item, play, money, ship, new, more, buy, player, make, pay, plex, isk,
price, thing use, real

2 CCP, game, people, EVE, want, up, think, player, make, time, now, see, know,
one, way

3 Account, cancel, origin, month, unsub, sub, CCP, out, one, day, EVE, run,
protest, alt. subscription, jita

4 Origin, vanity, EVE, item, people, CCP, beyond, thing, player, move, stand,
game, AUR, good, post, make, part, isn’t

To further contextualize how these topics differ a post with a high loading (>0.9) was
randomly selected from each topic. The posts are presented below.
Table 20: Selected posts.
Topic Sample Text
0 “CCP has already stolen Subscription fees for something that can not be
described as Content in any way.”

1 I'm not sure I fully understand what all the outrage is if "microtransactions"
have been with us for a long time now. Bear with me for a bit. The way I
understand microtrnsactions is "dollars for game items". According to CCP's
ideas, we would be able to buy a faction carrier using RL money... or trained
skills... or standings. Well, we can do that already through PLEX. I buy PLEX
using RL money, sell it for isk and then with that isk I buy a fully specced
carrier V toon and with the spare change I go and buy a faction fitted Archon
for him. We've been able to do this for a while. I bet more than one alliance buy
their officer fittings for the Alliance Tournament through PLEX. So maybe the
issue is just the prices? If they chage $80 for a vanity item, how much will they
charge me for an Archon? $500?I think the "microtransactions for vanity items
only" battle was lost when they introduced PLEX.

2 I was on massively and they had a link to the now infamous newsletter, which I
thought I had already read on evenews24.But it said something about "missing
pages". This peaked my interest so I downloaded it to compare to the first one.
There are actually 20 pages in total in the original. It's pretty much the same
thing, it appears the pages themselves were not edited, but pages were removed
which dropped the context of the remaining pages. So while I'm still upset at
those individuals my anger towards CCP as a whole has diminished. The most
notable line is: DISCLAIMER: The views put forward in this magazine do not
reflect general CCP company policies or decisions and are strictly individual
CRISIS AND STASIS 174
opinions, written by CCPers or about CCPers who feel strongly about these
issues. This is confidential internal information. Please respect that every
company has its trade secrets and that you are privy to those at CCP. Though it
still upsets me that certain individuals at CCP think as outlined in the PDF, I do
feel a little better with that simple paragraph. It definitely opens the door for
them to fix it, where before I didn't see a chance in hell of that happening.
Compare them for yourself:EVENews24.com tldr difference is basically this is
just asking people within CCP their opinions on the matter and doesn't actually
represent any actual plans. I still want confirmation in writing and signed by the
CEO himself before I believe anything that CCP says, but I'm a bit less
ragequit-y now. While this doesn't restore my trust/faith in CCP, it damages
what little I had to start with in evenews24. Ok, well, I never actually trusted
evenews24 at all, but now I definitely won't ever trust them :P Of course they
left things out, they always do. Easier to take things way out of context that way
when you only present the parts you want people to see. Misinformation is a
powerful tool, especially when you're trying to get more people involved in
your asshat campaign. I'm glad some else actually bothered to read it with an
objectionable attitude and reasoning. Oh and common sense, that helps too.
Then again, lynch mobs are fun I suppose.

3 EVE Online Subscription Cancelled You have cancelled your EVE Online
subscription. Your account will be suspended at the end of the current
subscription period. Firing my missiles at Jita than I’m out!

4 These are the people who have allowed you to get to where you are today.
These are the people who stuck by you over the past 8 years - even if they do
call themselves "bittervets". To turn your back on those people is at best poor
form, and at worst commercial suicide. Because Star Wars Galaxies has shown
you can't just trade in one group of customers for another. Why golf is not the
same as eve Specifically, buying golfclubs isn't the same as buying items in
Eve. The equivalent would be if you could pay $100 to get one
swing/strike/whatever it's called removed from your record. Instead of making a
hole in 5, you could pay $400 and get it listed as a hole in one! Suddenly, your
golf stats won't depend on how good you are, but simply on how much money
you're willing to spend. Try to suggest that to golfers and see how they like it. In
Eve, just flying a spaceship isn't all there is to the gameplay. The effort to GET
that spaceship is part of the game. The fact that it isn't easy to obtain certain
things, achieve certain things or do certain things is exactly what makes these
things meaningful. Meaningful to me AND to others, who can see what I have
achieved. And now some game designer would like to take that away? Bad
move CCP.

Topic Zero is primarily concerned with the content of the Incarna patch, which many
players felt was not up to par with previous updates by CCP Games. Topic One Is generally
CRISIS AND STASIS 175
neutral to supportive of the introduction of micro-transactions and wants to see any changes
managed properly. Topic Two is actively critical of the protesters viewing them as a toxic
component of the game’s community. Topic Three consists of people unsubscribing from the
game, many of whom state that they will stop by Jita and protest before leaving. Topic Four is
more concerned about the concept of fair play and how changes to the game may ruin it moving
forward.
Each of the topics captures a different element of the Jita event. However, Topics Three
and Four appear to be the best fit for capturing players engaged in disruptive behavior. Topic
Three captures people who were fed up with changes to the game and choosing to make their
voice heard while threatening to quit. Topic Four is more concerned with the norms within EVE
and how players feel left out due to the shift around them. While Topic Zero does also capture
some anger it seems more concerned with value as opposed to the nature of gameplay,
additionally post hoc analysis demonstrated that there were no instances of players using the
words “jita” or “protest” in Topic Zero.
As with the other examples in this dissertation, players identified as disruptive represent a
small minority. Specifically, 1,452 people (0.5 percent of the active player base) were flagged as
posting in the CCP forums with the intention of organizing, supporting or recruiting for the Jita
protests. Specifically, players whose forum posts were loaded with a confidence of 80% or
higher onto topics three or four were considered disruptive. Since the forums use the same
unique IDs as the in-game dataset these disruptive players can be situated in the data pulled from
the server logs, providing a complete image of their position within the entire EVE social
network a week before the start of the protests. This distinction between these players and the
CRISIS AND STASIS 176
rest of the EVE community therefore represents the dependent variable for all subsequent
classification models.
Results
Having established a variable that captures which players are engaged in disruptive
behavior and constructed a social network out of EVE's contact list/friendship system, the next
step is to fit a classification model based on the protocol laid out in the methods section. This
model should capture any structural differences between disruptive and non-disruptive players
and help inform a discussion of how EVE changed in response to the Jita Protests
Statistical Learning Model. Using all of the variables laid out in the previous section, a
series of statistical learning models were fit attempting to parse the differences between
disruptive and non-disruptive players. Stochastic gradient boosting, an ensemble model that uses
decision trees that are sequentially trained to improve fit on hard to classify cases, offered the
best performance with an area under the ROC of 0.74, which is generally considered to be
satisfactory performance. Random forests and bagged trees presented less optimal solutions with
area under the ROC of 0.71 and 0.68 respectively.
Using Delong’s test for the comparison of two ROC curves the gradient boosting model
demonstrated significantly better performance than the rare-event logistic regression (DeLong et
al., 1988). In addition, it performed marginally better than the random forest, while the forest did
not outperform the regression by a significant margin.
Table 21: Absolute values of the Delong's D between each ROC curve. Significant scores mean
that the model with a higher ROC score performed significantly better than the alternative.
Gradient Boost
ROC = 0.74
Random Forest
ROC = 0.71
Rare Event
ROC = 0.68
Gradient Boost
ROC = 0.74
-- -- --
Random Forest
ROC = 0.71
1.7915* -- --
CRISIS AND STASIS 177
Rare Event
ROC = 0.68
2.0495** 0.397 --
*
p<0.1;
**
p<0.05;
***
p<0.01

Using the standard variable importance described in the methods section a post-hoc
analysis of the model demonstrated that network constraint was by far and away the most
important predictor of participation in the Jita protests. Effective network size was a substantially
less powerful predictor.
Table 22: Variable importance: Private-Collaborative Model.
Variable
Scaled Relative Importance
(Higher is more important)
Network Constraint 100
Minutes Played 22.450
Age 18.014
Skill points Earned 16.188
Number of Active Characters 10.173
Effective Network Size 3.532
Number of Mutual Connections 0.273
Gender (Reference = Female) 0
Age Provided (Y/N) 0

While variable importance tells us which metrics are the best predictors of disruptive
behavior, it does not provide any data on the direction or shape of this effect. Following the
protocol laid out in the methods chapter, the next step is to generate a series of variable
prediction plots that capture the shift in the predicted probability of disruptive behavior as a
given variable shifts. Given that network constraint, effective network size and dedication to the
game represent the core hypotheses of this study, variable prediction plots were generated for
each variable.
CRISIS AND STASIS 178

Figure 14: Partial dependency plots for key variables.
Results indicate that network constraint has a strong negative influence on the likelihood
of protesting. In other words, the more closed a user’s indirect network is, the less likely they are
to protest. Conversely, unconstrained actors, or those occupying a brokerage role in their wider
network, are significantly more likely to engage in disruptive behavior.
Effective network size was not a particularly important predictor. Its influence remains
relatively flat after a sharper increase as soon as a player has a size above zero. This reflects the
need to have some form of integration into the EVE social network to participate. which captures
the minimum threshold of social involvement needed to participate in the EVE network in a
meaningful way.
Both skill points and minutes played demonstrate a curvilinear relationship. The more
time and energy a player has invested into EVE the more likely they are to protest. However
CRISIS AND STASIS 179
eventually this effect tails off and decreases suggesting that the most grizzled veterans of the
game were actually less likely to protest. Before delving into the possible reasons behind these
findings it is important to triangulate the results with a logistic regression model.
Rare Events Logistic Regression. The results from the statistical learning model were
validated using a rare events logistic regression on the same dataset. This model achieved an area
under the ROC curve score of 0.70, which is satisfactory but not as strong as the statistical
learning model.
Table 23: Rare Event Logistic Results. Logged Odd Coefficients and Standard Errors. Public-
Collaborative Model.

Variable Log Odds Coefficient
Effective Network Size 0.0003
(0.001)

Network Constraint -2.351
***

(0.189)

Minutes Played 0.00001
***

(0.0001)

Skill Points Earned 0.0001
**

(0.0001)

Gender 0.217
(0.155)

Number of Active Characters 0.309
***

(0.038)

Age -0.022
***

(0.003)

No Age Given -0.081
CRISIS AND STASIS 180
(0.182)

Mutual -0.039
(0.056)

Constant -5.254
***

(0.206)

Log Likelihood -8,726.402
Akaike Inf. Crit. 17,472.800
McFadden R2 0.05
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01

The results generally align with the findings from the gradient boosting approach with
regards to relative importance and direction of effects.
Taken together, these results suggest that disruptive players during the Jita protests were
relatively experienced members of the community who occupied brokerage roles. As such, H5a
(disruptive players will be more experienced) was supported (Variable Importance Rank = 2,
Regression Est = 0.001, SE = 0.001, p < 0.001) as was H5c (disruptive players will occupy
brokerage positions) (Variable Importance Rank = 1, Regression Est = -2.351, SE = 0.189, p <
0.001). However, H5b, the argument that disruptive players will have higher amounts of local
closure was not supported (Variable Importance Rank = 6, Regression Est = 0.0003, SE = 0.001,
p > 0.05).
Discussion
These results partially support the collective action hypothesis. Players with experience
and brokerage are well positioned to be leaders in any form of collective action mobilization.
They can draw from large and diverse networks of contact to spread information or be exposed
CRISIS AND STASIS 181
to new messages. However, the lack of localized network closure runs contrary to expectations.
In most other settings, network closure and its associated bridging capital are essential for
mobilizing individuals to engage in collective action. Without the overlapping obligations
embedded in closed networks, individuals who hear about a movement may lack the incentive or
social sanctioning necessary to move them away from free riding. As such, the lack of a
significant effect for localized closure (as operationalized through effective network size) means
that the relative composition of a protester’s local network is not as significant as the indirect
structures which they are embedded within.
One possible reason for these results can be found in the literature surrounding the
“connective action” view of social action (Bennett & Segerberg, 2012). This theory emerged
after the Occupy Wall Street Protests and other forms of network activism in the early 2010s. Its
proponents argue that public goods problems pose less of an organizational issue for groups
whose primary form of participation is mediated by technology. Because online communication
increases the reach and scope of a given message it is easier for information regarding a
movement to spread (Bennett & Segerberg, 2013). Additionally, forms of collective pressure
hosted by online communities often feature lower thresholds for participating (Bennett &
Segerberg, 2012). The relative amount of energy required to go to the Jita system and target CCP
Games’ associated structures while writing forum posts can be seen as relatively low compared
to other forms of mobilization. The connective action school argues that this combination of
easier participation and freer flowing information rebalances classical collective action
approaches. Free riding is less of an issue because each contribution is relatively small, so the
cost of participating is low. The small weight of these contributions adds up when communicated
and projected over a larger network and their aggregate power can still effect significant changes.
CRISIS AND STASIS 182
This alternative conceptualization may provide a better fit for the results gleaned from the
previous analysis. Players with relatively unconstrained networks are in a position to gather
information from a variety of sources. Given this network position, they have a greater likelihood
of hearing about CCP’s actions and the shift within the game’s regulatory structures. Driven by
the fact that they have invested a significant portion of time and energy into EVE, these players
are therefore more likely to mobilize despite not having a closed network full of overlapping
obligations around them. Ultimately, this lack of structure associated with bonding social capital
was not too detrimental to their performance due to the relatively low threshold for participation
in the Jita protests.
While closure was not a significant predictor of protest this does not mean that network
structures played no role in mobilizing players. Groups formed through “connective actions” are
not simply mobs moving in the same direction because a course seems fun or appealing (Bennett
& Segerberg, 2013). While networked coordination and organization make forming a movement
easier they do not remove all costs from participation. In the case of the Jita protests the
congregation of so many players in one spot put a large load on the game’s severs. The resulting
lag (or “time dilation” in EVE’s in-game slang) slowed the space around Jita, making EVE a
rather expensive screen saver for those in the system and reducing the normally fluid process of
flying and shooting to a glacial crawl (Edwards, 2011). Time spent in Jita therefore represented a
cost to the protesting players, as the rest of the game continued to move along at a normal pace
outside the protest zone, allowing non-disruptive players to get ahead while their colleagues were
stuck due to time dilation. Additionally, in-game pirates and other third parties took advantage of
the tightly packed protesters to fire area of effect weapons into the crowd and loot the resulting
CRISIS AND STASIS 183
wrecks (Edwards, 2011). So, while protesting may not require the strong, closed networks
traditionally associated with social movements it was also not without costs.
The idea of a personal action frame provides a way to reconcile the logic of connective
action with the costs associated with protesting. Personal action frames are unique and distinctive
perspectives on a given issue that are constantly shared and modified by members of a social
movement (Bennett & Segerberg, 2013). The archetypal example is the Occupy Wall Street
Slogan “We are the 99%”. The exact meaning and significance of the slogan changes from
person to person. It can be framed as a democratic statement of majority rule, a complaint about
wealth distribution or an argument for class warfare (Bennett & Segerberg, 2012). The important
thing is that the personal action frame provides a flexible way for people of varying backgrounds
and ideas to translate their own motivations into a common cause and slogan.
Within EVE the personal action frame seems to revolve around the norm of fairness. The
topic modelling of the forum posts demonstrates the concerns surrounding new players entering
the community and bypassing the hard work needed to be successful with a swipe of their credit
card. Players likened it to cheating at golf or engaging in “pay to win” behavior (Edwards,
2011). References to Origin, Electronic Art’s digital distribution service in Topic Three are also
telling as this studio pioneered the use of micro-transactions. The fairness personal action frame
also fits with the curvilinear relationship between time spent in game and the likelihood of
protesting. Purchasing in-game items is strongly correlated with a decrease in status in the eyes
of the other players in the community (Evers, van de Ven, & Weeda, 2015). For new players the
social cost of purchasing is less of an issue as they are already towards the bottom of the social
hierarchy within the game. Old players do not have as much of an incentive to purchase
functional items as they have already invested time and effort into the game to establish their
CRISIS AND STASIS 184
social networks and gain status (Shen, 2014). Players in the middle tiers of experience therefore
find themselves stuck. While they can benefit from purchasing items they will face the scorn of
older members of the community who they have established ties with. But if they do not
purchase functional goods they may be overtaken by new players who do not have these social
ties. Their concerns, coupled with the broad personal action frame that EVE should be fair
provided the motivation to overcome the dangers surrounding protesting in Jita. As this action
frame spread through para-texts and the in-game social network (especially through brokers as
the classification model demonstrates) the scale of disruptive behavior grew. Eventually the scale
of the Incarna protests presents a challenge to the developer, CCP Games. Leading to the next
topic of discussion.
Change Over Time. The EVE dataset is different from the other studies in this
dissertation in so far as it features static network data. This makes correlating changes in the
network structure with shifts in CCP Games policy difficult. However, the public record
surrounding the Jita protests demonstrates both how these events were prompted by major
changes at CCP and their influence on the trajectory of EVE.
First, it is clear that the Jita protests exist in response to a major change in strategy by
CCP. The need to develop new properties created fiscal demands on the studio, which forced a
shift from the status quo within EVE and prompted the introduction of the Incarna patch (Zacny,
2011).
Within the language of social movement theory the patch created space for mobilization
and a political opportunity. Broadly speaking, these moments come into being when previously
accepted assumptions and rules are suddenly open to questioning (Collier & Collier, 2002, p.
30). This provides opportunities for any radicalized or previously discontented individuals to
CRISIS AND STASIS 185
present opposition as both plausible and desirable, enabling the growth of a movement (Meyer,
2004)
Players are able to exploit these moments of disjuncture due to the fact that the game's
developer can neither totally repress nor appease them. While developers have powers within the
game that would make most dictators insanely jealous (such as the ability to create and destroy
in-game currency, ban players from the game world, and generally do whatever they want) the
use of these abilities is limited by the fact that they are ultimately dependent on the player base to
pay their salary. Simply put, antagonizing customers by not listening to them and keeping them
from the game, leads to lost profits, which in turn damages the development team (Groen, 2011).
However, due to exogenous market pressures simply appeasing the player-base is also not an
option. While CCP games has made attempts to listen to players when making changes within
EVE it is very careful to note that the player-base only serves in an advisory role (Woodford,
2013). Strategic choices which help both the player base and the developer are obvious win-wins
that can make both parties feel engaged, but as the Incarna patch indicates CCP is more than
willing to go against the will of the player-base when needed.
Therefore the Jita protests are a reaction to shifts in developer strategy, but the
repercussions of widespread induced disruption also served to jar CCP out of its new strategy
and force them to find a different path forward.
Following a week of protesting CCP Games made a firm commitment that there would be
no micro-transactions for performance related products within EVE (Groen, 2011). This
disavowal of earlier strategy memos helped calm the community. Additionally, CCP redirected
resources away from the existing micro-transaction framework to address issues that were of
greater concern to the community, such as graphical performance and network efficiency (CCP
CRISIS AND STASIS 186
Games, 2011b). As a result, the item store introduced within the Incarna patch has languished,
with few updates or new content to drive players to participate.
In addition to the technical infrastructure of the game, the Jita Protests directly influenced
CCP Games’ bottom line. In an attempt to serve their existing customer base better, CCP Games
restructured in response to player protests, laying off 20 percent of their workforce and
redirecting efforts away from developing new games and towards serving the EVE community
(Zacny, 2011). A long-term result of this reprioritization has been the failure of CCP Games to
develop new intellectual property outside of the EVE brand. Despite purchasing the rights to
White Wolf Gaming’s stable of tabletop games, CCP could not bring a new massively
multiplayer online game based on this property to market and ultimately sold the rights to
another developer (Meer, 2015). While CCP Games still exist, their entire portfolio of games is
now invested in the extended EVE Online universe, a move that began as a response to player
protests following the Incarna patch.
Numerous CCP Games spokespeople have repeatedly argued that the realignment and
refocusing of the studio in late 2011 was a direct response to player mobilization and the Jita
crisis (Zacny, 2011). From this perspective, the disruptive behavior of the Jita protesters had a
significant influence on the organizational trajectory of the studio. The exact causal flow of this
process may follow two trajectories. Either the presence of protests served as a warning sign to
CCP that they were overextended and prompted a reevaluation, or the protests were a symptom
of latent problems within the studio and player’s feeling of neglect. Either way the presence of
the Incarna protests has been directly linked through interviews and media coverage with a re-
evaluation of CCP’s long term strategy (Zacny, 2011).
CRISIS AND STASIS 187
Part of the distinction may lie in the alignment between the network structures (partially)
associated with protesting those linked to successful gameplay. Because the disruptive players in
this situation were both relatively experienced and located in a brokerage position within their
broader network, they represent critical actors in the EVE social network. If these brokers are
removed, either temporarily (by going to Jita and attacking EVE’s economy) or permanently (by
eventually quitting in disgust), they can fragment the overall network of the game. This in turn
degrades stocks of bridging social capital within EVE and negatively influences the overall
health of the community. Therefore, dealing with the protesters represents more than a public
relations headache for CCP Games. Their attempt to shift the code and norms of EVE around the
player base moved existing players into a space where they felt that their conception of the game
was no long valid, creating disruptive players who actively contradicted the new vision for EVE
Online.
This also created a secondary pressure on CCP, aside from developing a new revenue
stream they also had to preserve the community that they had already created. Ignoring the fact
that experienced players embedded in brokerage positions within their constituent network were
actively challenging the regulatory structures and norms that preserved the network could lead to
existing revenue streams disintegrating. Conversely, not pressing forward with micro-
transactions limits the revenue potential of the EVE community moving forward. CCP’s actions
in the summer and fall of 2011 demonstrate that the threat posed by experienced brokers actively
disrupting their title was significant enough to warrant forgoing micro-transactions and the
revenue associated with them.
With these results, along with the findings from the previous chapters in hand, the closing
chapter of this dissertation will discuss all of the findings in context and attempt to synthesize a
CRISIS AND STASIS 188
unified framework for understanding the relationship between disruptive play and change within
online communities.
CRISIS AND STASIS 189
Chapter Nine – Conclusion and Limitations
The four studies at the heart of this dissertation attempted to determine if players who
chose to break through one or more regulatory systems within an online community occupy a
distinctive place within their local social networks. These positions often differ from structural
signatures associated with the development of stocks of bridging and bonding social capital. As a
result, it was hypothesized that sizable increases in the rate of disruptive play are associated with
changes in the regulatory structures that govern a given community.
Structural Signatures
Using trace data from four gaming communities, an exploration of four distinctive types
of disruptive behavior generated mixed results. At the highest level of analysis, these models
were successful. With a degree of accuracy that ranged from satisfactory to excellent, various
statistical learning and data models were able to parse between disruptive players and their host
networks. The overall distribution of AUC scores across models ranged from 0.70 to 0.95, or
from satisfactory to excellent (Bradley, 1997). Additionally, variables derived from in-game
social networks were the most important predictors in three of the four analysis chapters. Taken
together, these findings suggest that situating disruptive players in a social network provides
additional information to the classification process, leading to higher performance. From a purely
technical standpoint, this finding would interest academics, gaming scholars and industry
researchers who are looking to understanding disruptive play within a community.
The good performance of the classification models and the importance of network
variables suggests that H1 (Disruptive players will occupy distinct positions within their host
network) is supported. The shape and form of these signatures varied from community to
community based on local regulations and the type of disruptive behavior in question. To
CRISIS AND STASIS 190
summarize, cheaters in the Steam Community were much less likely to form new edges and
displayed an increasing tendency towards homophily after a crackdown by the developer. Team
Fortress Two (TF2) scammers engaged in private-isolated disruptive behavior, actively avoided
network closure and had significantly more transitory connections than the surrounding network.
Tinker players in the Defense of the Ancients (DOTA) tended to be experienced players in a
brokerage position, but there was no correlation between using Tinker and network closure.
Finally, protesters operating in a situation of induced disruption tended to fall into a brokerage
role as well, especially among players in the middle tiers of in-game experience.
Change Over Time
Hypothesis two posited that increases in disruptive play would prompt changes from
developers. This chain of events was rooted in the evolutionary dynamics of developers and their
constituent communities. Disruptive play represents an alternative approach, which bypasses
many of the technical and social structures put in place that help generate bridging and bonding
social capital within a game. The increasing pressure created by disruptive actors on the health of
their host communities was thought to eventually build to a point where the developer was
forced out of a period of stasis to react and shift strategies to deal with the threat.
Support for this model was highly contingent on the type of disruptive play being
discussed. At a high level, the form of change over time can be correlated with whether the
disruptive behavior in question was public (i.e. the benefits of disruptive action cannot be
internalized) or private.
Private-isolated (scamming) and private-collaborative (cheating) disruptive play was
generally a lagging indicator of major policy changes. Among TF2 scammers, peaks in disruptive
behavior came about after the introduction of technical changes as players exploited the
CRISIS AND STASIS 191
confusion and uncertainty around the new systems. Cheaters in the Steam Community went
relatively unchallenged from 2010 to 2014 until the shift toward professional e-sports brought a
number of high profile cases to the spotlight. This increased the salience of tackling disruptive
behavior and lead to a widespread crack down.
In both cases, changes appeared to come from exogenous sources within the gaming
industry or were due to long needed upgrades. As a result, the hypothesis that disruptive behavior
can predict changes in developer strategy is not supported. The alternative picture that surfaces is
one where disruptive play is a constant low level concern. Development teams in both cases were
not inactive, releasing updates and patches to software regularly and trying to help users avoid
disruptive behavior. However, rates of behavior never became problematic unless the technical
architecture or relative visibility of disruptive players changed due to other events. The overall
image emerging from these findings is one of management. Disruptive players were never fully
eliminated from either community but they were obstructed through the introduction of patches
and other interventions. While these changes often did not significantly decrease the reported
rates of disruptive play, neither did they increase. As a result, the picture is one of maintenance
and stasis. Disruptive play is not ignored but neither is it allowed to build to crisis levels.
Nevertheless, shifts in the technical structures that regulate the community or a developer’s
tolerance for certain levels of disruptive behavior, may upset this stasis and lead to a period of
readjustment. However, these factors are generally driven by exogenous influences, as opposed
to prompts from within the community.
The results are a little bit different for the two public forms of disruption. Public-isolated
and public-collaborative disruption both featured experienced players occupying positions of
brokerage within their communities. Spikes in these disruptive behaviors also promoted
CRISIS AND STASIS 192
developer reactions: the increase in Tinker’s raw popularity and dominance in DOTA’s meta-
game, led to a widespread rebalancing of both that character and the entire DOTA strategic meta-
game. The protests and public-collaborative disruption within EVE in response to the Incarna
patch contributed to the refocusing of the developer’s long-term strategy to favor that
community. In both cases, the changes came about as a lagging response to increases in
disruptive behavior. The difference between the influence of public and private forms of
disruption on the likelihood of crisis and stasis may be due to the structural signatures associated
with either form. Because the benefits of private disruption can be internalized and kept among
disruptive players, their structural signatures tended towards being either transitory (private-
isolated disruption) or unchanging (private-collaborative). This effectively marginalizes these
players within the broader network as their edges are either constantly in flux without time to
develop or rarely changing and not keeping up with the times. Conversely, both forms of public
disruption are much better connected through brokerage positions, even if they do eschew
network closure.
This correlation between public disruption and network brokerage means that simply
removing disruptive players as they crop up and managing player’s behavior through sanctioning
is more difficult. Banning all of the players who used Tinker or everyone who attacked EVE’s
economy and servers is possible given the power afforded to developers, but it would remove a
significant number of brokers from their constituent networks. As removing brokers is an
excellent way to fragment and destroy a social network, the application of developer power to
sanction public disruptive players becomes less appealing.
If the people breaking the regulations cannot be removed, the remaining route is to
change the regulations to make disruption impossible or unnecessary. Nerfing Tinker in DOTA
CRISIS AND STASIS 193
removed the technical loophole which enabled that character to be played disruptively. Stepping
back from introducing micro-transactions diffused the rationale behind the EVE protests. In both
cases, regulatory shifts following increases in disruptive behavior helped managed the issue.
Nevertheless, changing regulations can have a cascading effect on the other interlocking systems
that govern a gaming community. Changing Tinker led to the rebalancing of a large number of
other characters in DOTA, which profoundly changed how the game was played going forward,
while deprioritizing micro-transactions reinforced EVE’s current pattern of having a relatively
small but dedicated player-base (Bergstrom, 2013). Either way, the inability to simply manage
the problem by managing players (potentially due to their network position) means that
regulation and strategy need to be altered, partially supporting hypothesis one.
Implications of Theory
These mixed results have numerous implications for the various theories introduced in
the first two chapters of this dissertation. For evolutionary theory and concepts of punctuated
equilibrium, the summarized results summarized demonstrate the overall complexity of
conceptualizing change over time. At the macro-level, changes in developer strategy through
interventions into the constituent community such as patches or code changes appear to be
correlated with both exogenous events in the marketplace as well as endogenous shifts in the
community. This highlights the closely interlinked nature of these two realms. As an example, in
the discussion of EVE Online and the player protests in response to the Incana patch, disruptive
players frequently mentioned Electronic Arts’ new digital distribution platform “Origin” which
had a reputation for being full of micro-transactions similar to the ones CCP was aiming to
design (Savage, 2013). On one hand, the profitability of this strategy for Electronic Arts provided
competitive pressure as well as a template that may have played a role in CCP Game’s own shift
CRISIS AND STASIS 194
in strategy. However, the example of Origin also influenced player tolerances for micro-
transactions within EVE and gave them a concrete of example of what they believed their
community would become.
The awareness and responsiveness of members of the EVE community to the
marketplace and developer strategy point toward a possible extension of punctuated equilibrium
theory within the context of online communities. In its general formulation, punctuated
equilibrium theory posits that changes in performance, environment or competition will
eventually prompt the revolutionary re-orientation of the company after a period of prolonged
stasis (Romanelli & Tushman, 1994; Tushman & Romanelli, 2008). However, earlier
formulations have generally neglected the role of consumer pressures within the framework.
These pressures are especially important in online communities such as (but not limited to)
games as they represent cooperative productions between a developer and the associated
community. The developer provides infrastructure, content, affordances and regulations that
make a given community possible (Lessig, 2006b). Users adopt and repurpose this framework to
produce their own experiences and develop social networks and their associated forms of capital
that maintain the community (and the revenue it generates) in the long term (Koster, 2013). This
co-production gives the members of the community a role more akin to a minority shareholder
than a simple consumer. In other words, some members of the community are not just consumers
but have invested emotional, temporal or actual capital in the community and developer.
This places the developer in an interesting position. On one hand, they have the power to
implement a wide array of changes through their control of the community’s regulatory
structures. However, because the members of the community are necessary for it to be
successful, shifts that antagonize these stakeholders could pose a risk to the developer. Players
CRISIS AND STASIS 195
have the traditional weapons of consumers, exit (leaving the community) and voice (expressing
displeasure) but their reach is enhanced by the fact that they are also the commodity being
monetized by the developer (Hirschman, 1970). A manufacturer can still sell a product that the
marketplace does not like by discounting it or finding new marketplaces, but a developer
requires the community as both consumers and a necessary component of the product, adding
heft to their role as stakeholders in the game.
By including the player as a stakeholder, they can be brought into process of punctuated
equilibrium as a party weighing in on the timing of strategic change. This shifts the mechanism
through which a community feeds into the need for realignment from the degradation of
community health due to disruptive network signatures, to players exercising exit and voice to
leverage their stakeholder position to incentivize a change. This configuration also provides an
excellent framework for understanding why private disruptive behavior was not associated with
shifts from the developer but public varieties were. Both private-isolate and private-collaborative
disruptive behavior have structural characteristics that promote silence among the host
community. For private-collaborative disruption (cheating) this originates from the performative
aspect of disruptive behavior, in other words, cheating works best when other members of the
community do not know that you are cheating (Consalvo, 2007). As an example, members of the
community were surprised at the number of professional players caught cheating, never
expecting the best of the best to engage in such behavior (Grayson, 2014). Private-isolated
behavior also benefits from a silencing impulse. If executed properly, the victim of a scam may
not even know that they were victimized. Even if the victim does discover their misfortune, a
significant body of literature within the field of criminology suggests that victims of fraud are
less likely to report the event due to social stigma (Mason & Benson, 1996; Wyk & Benson,
CRISIS AND STASIS 196
1997). In either case, the players who lose the most from disruptive play are not in a position to
organize and work together systematically to apply pressure as a stakeholder group. Conversely,
public facing disruptive behavior is predicated on the fact that the rewards from the disruption
are open to anyone who chooses to participate. Public-isolated behavior such as glitching, has the
potential to become more widespread and is more difficult to conceal than private forms of
disruptive play. Finally, public-collaborative disruptive behavior is often directly formulated as a
mechanism for enacting pressure on the developer, making it arguably the purest manifestation
of player as stakeholder in the punctuated equilibrium process. In other words, the varieties of
disruptive behavior that are the most visible within the community are also the ones which have
the greatest likelihood of prompting a response, whether that pressure comes from the disruptive
players themselves who organize and coordinate through protest, or disgruntled players fed-up
with widespread glitching.
This extension has several implications for the theory of punctuated equilibrium. First it
suggests that the pool of potential influencers who directly contribute to a strategic shift using
persuasion or voice may have to be expanded. The original conception of punctuated equilibrium
conceptualizes the decision-making process as being largely internal to a given firm (Romanelli
& Tushman, 1994). Consumers indirectly play a role by moving between different products or
services creating competitive crises that may prompt realignment in the game. However, the
player-as-stakeholder view gives the members of the community a direct line into the decision-
making process due to their unique role as both content consumers and creators, expanding the
overall scope of the theory.
In addition to the impact on punctuated equilibrium theory, this perspective also has
implications for theories surrounding regulation and the structures that govern gameplay.
CRISIS AND STASIS 197
Specifically, the formulation of code as law put forward by Lessig begins to operate in a space
with a bit more ambiguity. Code being law means that direct control over the code is power.
Lessig conceptualizes corporate code as being “merchant-sovereigns” who are governed
primarily by the ability of competitors to offer alternative rule sets if their own regulations are no
longer agreeable to their customers (Lessig, 2006b, p. 287). He correctly points out that the
problem of sunk costs in a given community and the presence of un-transferable stocks of social
capital make this free flow of consumers difficult. As a result, he speculates that there may be a
gradual shift toward “citizen-sovereign” arrangements where users have a direct feedback and a
say in the development of code (Lessig, 2006b, p. 291). This citizen-sovereign model does not
appear to have developed within the field of gaming. While developers can rarely totally ignore
their users, neither are they beholden to the whims of the player-base. This places these online
communities in a nether region between the merchant-sovereign ideal, where the only way to
influence a regulatory system is to leave, and the citizen-sovereign system, where voices are
directly and consistently represented within the regulatory system. This middle ground most
strongly resembles a form of corporatism. Corporatism represents a series of attempts by various
powerful institutions throughout history to incorporate important members of the political sphere
into the political process via consultation (Gourevitch & Shinn, 2007; Martin, 1983). Various
groups within the system can organize and provide consultation to the sovereign (i.e. the body
that controls the code) but this actor is not under any direct obligation to heed the will of these
consultant bodies aside from their inherent clout due to their stake in society or the economy
(Martin, 1983). The corporatist framework provides a way to conceptualize players as one of
many stakeholders who have a direct or indirect seat at the table, but no formalized and direct
control over the regulatory system beyond the exertion of varying levels of influence. This
CRISIS AND STASIS 198
represents a greater role than the limited “leave if you do not like it” consumer within a free
marketplace, but in a lesser position than someone directly involved in the creation and
deployment of code.
There are numerous examples of this corporatist approach to the relationship between
players and code within the industry. CCP Games, the developer of EVE has pioneered this
strategy though the Council of Stellar Management (CSM). The CSM is an advisory body made
up of elected players who provide feedback to CCP without having direct control over changes
or the game’s code (Ireland, 2013). This incorporates the opinions of players without directly
seceding control, a hallmark of the corporatist philosophy extended into the game’s regulatory
structure.
The CSM is the most well developed corporatist structure currently available but there
are other examples. Valve Corporation recently introduced Overwatch, a system where players
can view other player’s game footage deemed suspicious by the developer and report disruptive
players (Stanton, 2016). The verdict from the collected Overwatch decisions is used in
consultation with other evidence to ban troublemakers missed by technical systems such as V AC.
The Overwatch system represents another corporatist philosophy manifestation into to an
element of the regulatory system. The developer is not relinquishing power to the player base,
and the terms of service still enables them to ban or punish players freely (Stanton, 2016).
However, by incorporating players into the system as stakeholders increases the efficacy of the
regulatory system while providing Overwatch participants with a way for their opinion to have a
chance of being used. This middle ground form of code control represents an interesting process
emerging out of the interplay between the economics of the games industry, punctuated
CRISIS AND STASIS 199
equilibrium within developers, and the need to manage complex and highly active communities
successfully.
This framework of corporatist regulation may be useful for future research focused on the
interaction of consumer and corporate networks and other sectors where users are both producers
and consumers of content, such as social networks. The existing literature or corporatism within
the fields of sociology offers a complex but useful series of insights on the interaction between
bodies with regulatory power and influential groups under their purview (Martin, 1983).
Drawing from these insights as metaphors for future friendly or contentious engagements
between developers and their constituent community’s may offer a route for a more nuanced
view of the interaction between these two parties and how it feeds back both into developments
within organizational strategy and stocks of online social capital.
Limitations and Future Research
This study has a number of limitations that should be taken into account to contextualize
these conclusions. First is the fact that many of the classification models were trained in
situations where the ground truth (i.e. the number of disruptive players) is not totally clear. This
lack of information can come from two sources. False positives represent cases where the
definition of disruption is too broad. As an example, it is likely that not every Tinker player
exploited glitches with that character. Alternatively, false negatives can occur when the definition
is too narrow such as scammers in the TF2 dataset are only included in the sample if they have
been reported to SteamRep. In either case, the lack of ground truth means that the accuracy and
findings from the classification models may shift as the variables they were trained on are
altered. This is not to say that the findings are not useful, but it does mean that there may be sub-
CRISIS AND STASIS 200
categories of disruptive players who are less likely to be reported or monitored who are not
represented in the classification model.
Additionally, the sampling periods sometimes did not cover moments of increased
disruptive behavior. This was most prevalent in the TF2 study where the major spike in
scamming took place in 2013, before data collection started. As a result, it is not possible to
gather data from before and after the patch to see how disruptive play shifted as the technical
regulations it runs up against also changed. While the findings in the chapter on private-isolated
disruptive behavior were buttressed by qualitative assessment and forum logs, this does detract
from the overall validity of that chapter’s conclusions.
Finally, the types of data included in the various models present some liabilities. The use
of log data means that while the sample sizes in all chapters are relatively large for studies within
the field of social science, the data is also shallow (Wood, Griffiths, & Eatough, 2004). The
various networks are generally unimodal and undirected, while player demographics such as
gender and age are largely missing from publically exposed API data. In some ways this
limitation is a testament to the efficacy of network variables in the classification models. The fact
that the classifiers were able to achieve fair to excellent performance with such a shallow pool of
predictors demonstrates the contribution that these features can make to predictive models going
forward. Conversely, deeper data on gender and age may help unpack the issue of change and
disruptive play even further while providing insights into the interaction between power, culture,
and rule breaking.
These limitations also open up routes for future research based on the findings presented
in this dissertation. Two potential avenues for subsequent work seem particularly appealing
based on these findings. The first is to interrogate the assumptions and processes that govern the
CRISIS AND STASIS 201
relationships between developers and their constituent communities. As evidenced by the EVE
Online case, these relationships can become quite complex, with discussion and debate between
both parties helping set the long-term trajectory of the company and community (Ireland, 2013).
Unpacking the behind-the-scenes dynamics with a qualitative assessment or focused survey
research would be extremely beneficial in situating and contextualizing the findings from this
examination.
The need for more focused work also extends to the exploration of player networks. The
simple, unimodal networks presented in this examination have a degree of extraction that does
not always mimic the complex interactions taking place within and across gaming communities
(Contractor, Monge, & Leonardi, 2011). A more focused exploration of a disruptive community
could benefit from constructing a multi-modal network that forms edges from in-game and out-
of-game links. As an example, the use of para-texts, guides and other third-party resources plays
an important role in both forms of public disruptive behavior (Consalvo, 2007). While there was
some attempt to capture these links in their respective chapters, a more focused study dealing
with a smaller, more manageable dataset could capture a wider variety of possible connections
between players.
Conclusion
This dissertation explored the interactions between players who chose to break through
one or more regulatory system within a game and their community and its developers. Drawing
data from a variety of different games and communities, it situated and examined disruptive
players within their host networks. Based on their position in a two by two typology of possible
disruptive behaviors, a distinctive structural signature was generated for each possible variety of
disruptive play based on the relevant findings from criminology, sociology, network science and
CRISIS AND STASIS 202
game studies. Using a combination of statistical learning and classic data modelling, each variety
of disruption was assessed based in a combination of network and player data. These
classification models revealed that disruptive players often occupy a distinct position within their
host networks, and it is possible to extract this data and make predictions based on it. The overall
rate of disruptive play in each community was then assessed to determine if increases in
disruptive behavior were correlated with major shifts in developer policy. Originally it was
assumed that these shifts would occur because disruptive players damage their surrounding
networks and force an intervention. However, when taken together, the results indicate a strong
difference between “public” and “private” disruptive behavior. Private disruption generally
eschews strong positions of brokerage or closure within the host network. As a result, these
players can be removed from the network without causing extensive fragmentation, freeing up
the developer to turn the coercive power granted to them in the terms of use against
troublemakers. Conversely, public disruptive behavior is associated with experienced players
occupying a brokerage role. As a result, simply removing these disruptors poses a threat to the
in-game social network. Therefore, the regulatory system has to change instead, leading to drift
in corporate strategy at both the short and long term level. Taken together these findings present
new opportunities for research within a number of fields, as well as providing valuable insights
for industry researchers. Therefore, this dissertation is best seen as a launching point for both the
concept of disruptive play, which brings together protest, scamming, glitching and cheating into
a cohesive framework and provides methodological suggestions to explore this framework
moving forward.

CRISIS AND STASIS 203
References
6.82 Patch Notes. (2014, September 25). Retrieved March 31, 2016, from
http://www.dotafire.com/dota-2/forum/general/list-of-the-most-important-changes-in-6-
82-9880
Adler, P. S., & Kwon, S.-W. (2002). Social capital: Prospects for a new concept. Academy of
Management Review, 27(1), 17–40.
Ahmad, M. A., Borbora, Z., Shen, C., Srivastava, J., & Williams, D. (2011). Guild play in
MMOGs: rethinking common group dynamics models. In Proceedings of the Third
international conference on Social informatics (pp. 145–152). Berlin, Heidelberg:
Springer-Verlag. Retrieved from http://dl.acm.org/citation.cfm?id=2050728.2050753
Ahmad, M. A., Keegan, B., Srivastava, J., Williams, D., & Contractor, N. (2009). Mining for
Gold Farmers: Automatic Detection of Deviant Players in MMOGs. In International
Conference on Computational Science and Engineering, 2009. CSE ’09 (V ol. 4, pp. 340 –
345). http://doi.org/10.1109/CSE.2009.307
Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study.
Administrative Science Quarterly, 45(3), 425–455.
Aldrich, H. (1999). Organizations evolving. Sage.
Aldrich, H. (2008). Organizations and environments. Stanford University Press. Retrieved from
https://books.google.com/books?hl=en&lr=&id=CkKsAAAAIAAJ&oi=fnd&pg=PR15&
dq=aldritch+organizations+evolving&ots=FufptiR3mN&sig=b61QeDa-
f09pVpOIjdbZIAqWE0o
Aldrich, H. E., & Pfeffer, J. (1976). Environments of organizations. Annual Review of Sociology,
79–105.
CRISIS AND STASIS 204
Anderson, M. (2015, October 29). Technology Device Ownership: 2015. Retrieved from
http://www.pewinternet.org/2015/10/29/technology-device-ownership-2015/
Andrews, S. F. (2010). The Guild Leader’s Handbook: Strategies and Guidance from a Battle-
Scarred MMO Veteran. No Starch Press.
Andrzejewski, D., & Zhu, X. (2009). Latent Dirichlet Allocation with topic-in-set knowledge. In
Proceedings of the NAACL HLT 2009 Workshop on Semi-Supervised Learning for
Natural Language Processing (pp. 43–48). Association for Computational Linguistics.
AnEvilSnowman. (2013). Tinker 9 min Boots of Travel - [DotA 2 Guide]. Retrieved from
https://www.youtube.com/watch?v=N1_TywuKdw4
Ang, C. S., & Zaphiris, P. (2010). Social Roles of Players in Mmorpg Guilds - A social network
analytic perspective. Information Communication Society, 13(4), 592–614.
Aniszczyk, C. (2015, August 31). Anomaly Detection. Retrieved May 26, 2016, from
https://github.com/twitter/AnomalyDetection
Asimov, I. (2010). Foundation: Foundation and Empire; Second Foundation. Everyman’s
Library. Retrieved from
https://books.google.com/books?hl=en&lr=&id=j2f6L1vuR3kC&oi=fnd&pg=PR50&dq=
foundation+asimov&ots=Q-2rF8TS67&sig=y20-HXwJr6NCfAx3saDvNI3C8MY
Astley, W. G. (1985). The Two Ecologies: Population and Community Perspectives on
Organizational Evolution. Administrative Science Quarterly, 30(2), 224–241.
http://doi.org/10.2307/2393106
Bainbridge, W. A., & Bainbridge, W. S. (2007a). Creative Uses of Software Errors: Glitches and
Cheats. Social Science Computer Review, 25(1), 61–77.
http://doi.org/10.1177/0894439306289510
CRISIS AND STASIS 205
Bainbridge, W. A., & Bainbridge, W. S. (2007b). Creative uses of software errors: Glitches and
cheats. Social Science Computer Review, 25(1), 61–77.
Baker, W. E., & Faulkner, R. R. (2003). Diffusion of fraud: Intermediate economic crime and
investor dynamics. Criminology, 41, 1173.
Bakioglu, B. S. (2009). Spectacular interventions of second life: Goon culture, griefing, and
disruption in virtual spaces. Journal For Virtual Worlds Research, 1(3).
Bakioğlu, B. S. (2012). Negotiating governance in virtual worlds: grief play, hacktivism, and
LeakOps in Second Life®. New Review of Hypermedia and Multimedia, 18(4), 237–259.
http://doi.org/10.1080/13614568.2012.746742
Bakker, R. M., Raab, J., & Milward, H. B. (2012). A preliminary theory of dark network
resilience. Journal of Policy Analysis and Management, 31(1), 33–62.
http://doi.org/10.1002/pam.20619
Bakos, Y ., Marotta-Wurgler, F., & Trossen, D. R. (2014). Does anyone read the fine print?
Consumer attention to standard form contracts. Journal of Legal Studies, 43(1), 09–40.
Bak, P., & Sneppen, K. (1993). Punctuated equilibrium and criticality in a simple model of
evolution. Physical Review Letters, 71(24), 4083.
Balkin, J. M. (2004). Law and liberty in virtual worlds. NYL Sch. L. Rev., 49, 63.
Balkundi, P., & Kilduff, M. (2006). The ties that lead: A social network approach to leadership.
The Leadership Quarterly, 17(4), 419–439.
Bardzell, J., Jakobsson, M., Bardzell, S., Pace, T., Odom, W., & Houssian, A. (2007). Virtual
worlds and fraud: Approaching cybersecurity in massively multiplayer online games. In
Proceedings of DiGRA 2007 Conference (pp. 451–742).
CRISIS AND STASIS 206
Bartle, R. (1996). Hearts, clubs, diamonds, spades: Players who suit MUDs. Journal of MUD
Research, 1(1), 19.
Bateman, C., Lowenhaupt, R., & Nacke, L. E. (2011). Player typology in theory and practice. In
Proceedings of DiGRA.
Baumer, M., & Kephart, C. (2015). Aggregate Dynamics in a Large Virtual Economy: Prices and
Real Activity in Team Fortress 2*. eScholarship. Retrieved from
http://escholarship.org/uc/item/9vm4h78s
Baumgartner, F. R., Jones, B. D., & Mortensen, P. B. (2014). Punctuated equilibrium theory:
Explaining stability and change in public policymaking. Theories of the Policy Process,
59–103.
Baum, J. A. (1999). Whole-part coevolutionary competition in organizations. Variations in
Organization Science, 113–135.
Baum, J. A., & Singh, J. V . (1994). Evolutionary dynamics of organizations. Evolutionary
Dynamics of Organizations. Retrieved from http://repository.ust.hk/ir/Record/1783.1-
62749
Bennett, W. L., & Segerberg, A. (2012). The Logic of Connective Action. Information,
Communication & Society, 15(5), 739–768.
http://doi.org/10.1080/1369118X.2012.670661
Bennett, W. L., & Segerberg, A. (2013). The Logic of Connective Action: Digital Media and the
Personalization of Contentious Politics. Cambridge University Press.
Bergstrom, K. (2013). EVE Online Newbie Guides: Helpful information or gatekeeping
mechanisms at work? Selected Papers of Internet Research, 3(0). Retrieved from
http://spir.aoir.org/index.php/spir/article/view/692
CRISIS AND STASIS 207
Bergstrom, K., Carter, M., Woodford, D., & Paul, C. A. (2013). Constructing the Ideal EVE
Online Player. Presented at the Proceedings of DiGRA 2013 Conference: DeFragging
Game Studies.
biggest nerf in dota history? • /r/DotA2. (n.d.). Retrieved May 9, 2016, from
https://www.reddit.com/r/DotA2/comments/2szz6j/biggest_nerf_in_dota_history/
Blackburn, J., Kourtellis, N., Skvoretz, J., Ripeanu, M., & Iamnitchi, A. (2014). Cheating in
online games: a social network perspective. ACM Transactions on Internet Technology
(TOIT), 13(3), 9.
Blackburn, J., Simha, R., Kourtellis, N., Zuo, X., Ripeanu, M., Skvoretz, J., & Iamnitchi, A.
(2012). Branded with a Scarlet “C”: Cheaters in a Gaming Social Network. In
Proceedings of the 21st International Conference on World Wide Web (pp. 81–90). New
York, NY , USA: ACM. http://doi.org/10.1145/2187836.2187848
Blackburn, J., Simha, R., Long, C., Zuo, X., Kourtellis, N., Skvoretz, J., & Iamnitchi, A. (2011).
Cheaters in a gaming social network. SIGMETRICS Perform. Eval. Rev., 39(3), 101–103.
http://doi.org/10.1145/2160803.2160871
Blaszczyński, J., Stefanowski, J., & Idkowiak, \Lukasz. (2013). Extending bagging for
imbalanced data. In Proceedings of the 8th International Conference on Computer
Recognition Systems CORES 2013 (pp. 269–278). Springer.
Blodgett, B. M. (2009). And the ringleaders were banned: an examination of protest in virtual
worlds. In Proceedings of the fourth international conference on Communities and
technologies (pp. 135–144). New York, NY , USA: ACM.
http://doi.org/10.1145/1556460.1556481
Blow, J. (2004). Game development: Harder than you think. Queue, 1(10), 28.
CRISIS AND STASIS 208
Borbora, Z. H., & Srivastava, J. (2012). User Behavior Modelling Approach for Churn Prediction
in Online Games. In Privacy, Security, Risk and Trust (P ASSAT), 2012 International
Conference on and 2012 International Confernece on Social Computing (SocialCom)
(pp. 51–60). http://doi.org/10.1109/SocialCom-PASSAT.2012.84
Borbora, Z., Srivastava, J., Hsu, K.-W., & Williams, D. (2011). Churn Prediction in MMORPGs
Using Player Motivation Theories and an Ensemble Approach. In Privacy, security, risk
and trust (passat), 2011 ieee third international conference on and 2011 ieee third
international conference on social computing (socialcom) (pp. 157–164).
http://doi.org/10.1109/PASSAT/SocialCom.2011.122
Bradenbug, H. (2006). Pathologies of the Virtual Public Sphere. In S. Oates (Ed.). New York,
NY: Routledge.
Bradley, A. P. (1997). The use of the area under the ROC curve in the evaluation of machine
learning algorithms. Pattern Recognition, 30(7), 1145–1159.
http://doi.org/10.1016/S0031-3203(96)00142-2
Breiman, L. (2001a). Random Forests. Machine Learning, 45(1), 5–32.
http://doi.org/10.1023/A:1010933404324
Breiman, L. (2001b). Statistical Modeling: The Two Cultures (with comments and a rejoinder by
the author). Statistical Science, 16(3), 199–231. http://doi.org/10.1214/ss/1009213726
Bright, D. A., Hughes, C. E., & Chalmers, J. (2012). Illuminating dark networks: a social
network analysis of an Australian drug trafficking syndicate. Crime, Law and Social
Change, 57(2), 151–176. http://doi.org/10.1007/s10611-011-9336-z
Brockmann, D., & Helbing, D. (2013). The hidden geometry of complex, network-driven
contagion phenomena. Science, 342(6164), 1337–1342.
CRISIS AND STASIS 209
Bryant, J. A., & Monge, P. R. (2008). The evolution of the children’s television community,
1953-2003. International Journal of Communication, 2, 33.
Burley, D. (2010). Penguin life: A case study of one tween’s experiences inside Club Penguin.
Journal For Virtual Worlds Research, 3(2). Retrieved from https://jvwr-ojs-
utexas.tdl.org/jvwr/index.php/jvwr/article/view/1894
Burn, A., & Carr, D. (2003). Signs from a strange planet: Role-play and social performance in
Anarchy Online. In Actes de conférence, COSIGN 2003: 3rd Conference on
Computational Semiotics for Games and New Media, University of Teesside (pp. 14–21).
Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge: Harvard.
Burt, R. S. (1995). Structural holes: the social structure of competition. Harvard University
Press.
Burt, R. S. (1999). The social capital of opinion leaders. The Annals of the American Academy of
Political and Social Science, 566(1), 37–54.
Burt, R. S. (2000). The network structure of social capital. Research in Organizational Behavior,
22, 345–423.
Burt, R. S. (2001). Structural Holes versus Network Closure as Social Capital. In R. S. Burt, K.
Cook, & N. Lin (Eds.), Social Capital: Theory and Research (Sociology and Economics.
Aldine Transaction.
Burt, R. S. (2004). Structural holes and good ideas. American Journal of Sociology, 110(2), 349–
399.
Burt, R. S. (2005). Brokerage and Closure: An Introduction to Social Capital. Oxford University
Press.
CRISIS AND STASIS 210
Burt, R. S. (2009). Structural holes: The social structure of competition. Harvard university
press. Retrieved from
https://books.google.com/books?hl=en&lr=&id=FAhiz9FWDzMC&oi=fnd&pg=PR5&d
q=burt+structural+holes&ots=vIO0bv7HFO&sig=nuQz8QouqMEnW112NmUX_1-
7aQM
Callele, D., Neufeld, E., & Schneider, K. (2005). Requirements engineering and the creative
process in the video game industry. In Requirements Engineering, 2005. Proceedings.
13th IEEE International Conference on (pp. 240–250). IEEE.
Campbell, D. T. (1965). Variation and selective retention in socio-cultural evolution. Social
Change in Developing Areas: A Reinterpretation of Evolutionary Theory, 19, 26–27.
Campbell, D. T. (1982). The “blind-variation-and-selective-retention” theme. The Cognitive-
Developmental Psychology of James Mark Baldwin: Current Theory and Research in
Genetic Epistemology. Norwood, NJ: Ablex Publishing.
Carr, M. (2000). Technological Affordance, Social Practice and Learning Narratives in an Early
Childhood Setting. International Journal of Technology and Design Education, 10(1),
61–80. http://doi.org/10.1023/A:1008986002620
Carter, M. (2013). Ruthless Play. Presented at the Foundations of Digital Games.
Carter, M., Woodford, D., & Bergstrom, K. (2013). Internet Spaceships Are Serious Business:
EVE Online Workshop Introduction. In Proceedings of Foundations of Digital Games
Conference 2013.
Castronova, E. (2001). Virtual worlds: A first-hand account of market and society on the cyberian
frontier. Retrieved from http://papers.ssrn.com/sol3/Papers.cfm?abstract_id=294828
CRISIS AND STASIS 211
Castronova, E. (2007). Effects of Botting on World of Warcraft «. Indiana University, 13.
Retrieved from http://virtuallyblind.com/files/mdy/blizzard_msj_exhibit_7.pdf
Castronova, E. (2008a). Exodus to the virtual world: How online fun is changing reality.
Palgrave Macmillan. Retrieved from
https://books.google.com/books?hl=en&lr=&id=BCBw_ByOsqIC&oi=fnd&pg=PP1&dq
=castronova&ots=Pv-PKAZGyW&sig=vVjV_XSusEhbj_cJJoOWOig-tGQ
Castronova, E. (2008b). Synthetic worlds: The business and culture of online games. University
of Chicago press. Retrieved from
https://books.google.com/books?hl=en&lr=&id=0OHVdwE5Kb0C&oi=fnd&pg=PR7&d
q=castronova&ots=i-rZvmrP11&sig=l8k5zX51wPe0_k4_iSHCSJ_tt7U
CCP Games. (2011a, July 11). EVE IS REAL. Retrieved December 9, 2011, from
http://www.eveisreal.net/#!/visualizer/
CCP Games. (2011b, October 5). Dev Blogs. Retrieved April 25, 2016, from
http://community.eveonline.com/news/dev-blogs
Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties1. American
Journal of Sociology, 113(3), 702–734.
Chalk, A. (2014, October 3). Valve puts $250,000 on the line at Dreamhack 2014 Counter-Strike:
Global Offensive tournament. Retrieved May 8, 2016, from
http://www.pcgamer.com/valve-puts-250000-on-the-line-at-dreamhack-2014-counter-
strike-global-offensive-tournament/
Chang, K. T.-T., Koh, A. T.-T., Low, B. Y .-Y ., Onghanseng, D. J. S., Tanoto, K., & Thuong, T. S.
T. (2008). Why I love this online game: The MMORPG stickiness factor. ICIS 2008
Proceedings, 88.
CRISIS AND STASIS 212
Chawla, N. V ., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: synthetic
minority over-sampling technique. Journal of Artificial Intelligence Research, 321–357.
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.
Chesney, T., Coyne, I., Logan, B., & Madden, N. (2009). Griefing in virtual worlds: causes,
casualties and coping strategies. Information Systems Journal, 19(6), 525–548.
http://doi.org/10.1111/j.1365-2575.2009.00330.x
Cheung, G., & Huang, J. (2011). Starcraft from the stands: understanding the game spectator. In
Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp.
763–772). ACM.
Chod’s Cheats. (n.d.). Retrieved May 8, 2016, from https://chods-cheats.com/
Cole, H., & Griffiths, M. D. (2007). Social interactions in massively multiplayer online role-
playing gamers. CyberPsychology & Behavior, 10(4), 575–583.
Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of
Sociology, S95–S120.
Coleman, S., & Dyer-Witheford, N. (2007). Playing on the digital commons: collectivities,
capital and contestation in videogame culture. Media, Culture & Society, 29(6), 934–953.
Collier, R. B., & Collier, D. (2002). Shaping The Political Arena (1st ed.). University of Notre
Dame Press.
Congo. (2014, November 21). Counter-Strike world rocked by professional players cheating
scandal. Retrieved May 8, 2016, from
http://www.mweb.co.za/games/view/tabid/4210/article/16798/counter-strike-world-
rocked-by-professional-players-cheating-scandal.aspx
CRISIS AND STASIS 213
Consalvo, M. (2005). Gaining Advantage: How Videogame Players Define and Negotiate
Cheating. In DIGRA Conf. Citeseer.
Consalvo, M. (2007). Cheating: Gaining Advantage in Videogames. MIT Press.
Consalvo, M. (2009). There Is No Magic Circle. Games and Culture, 4(4), 408–417.
http://doi.org/10.1177/1555412009343575
Consalvo, M., & Vazquez, I. S. (2015). Game platforms and the evolution of cheating practices:
An exploratory study. Journal of Gaming & Virtual Worlds, 7(1), 3–19.
Contractor, N., Monge, P., & Leonardi, P. M. (2011). Network Theory | Multidimensional
Networks and the Dynamics of Sociomateriality: Bringing Technology Inside the
Network. International Journal of Communication, 5(0), 39.
Cook, J. (2015, February 4). Greece’s new finance minister used to manage virtual economies in
video games. Retrieved April 25, 2016, from http://www.businessinsider.com/yanis-
varoufakis-valve-game-economy-greek-finance-2015-2
Corneliussen, H. G., Rettberg, J. W., MacCallum-Stewart, E., & Parsler, J. (Eds.). (2008). Role-
play vs. Gameplay: The Difficulties of Playing a Role in World of Warcraft. In Digital
Culture, Play, and Identity: A World of Warcraft® Reader. The MIT Press.
Cornes, R., & Sandler, T. (1996). The theory of externalities, public goods, and club goods.
Cambridge University Press.
Darwin, C., & Bynum, W. F. (1859). The origin of species by means of natural selection: or, the
preservation of favored races in the struggle for life. Retrieved from
http://www.ias.ac.in/resonance/V olumes/14/02/0204-0208.pdf
Davidovici-Nora, M. (2009). The dynamics of co-creation in the video game industry: the case of
World of Warcraft. Communications & Strategies, (73), 43.
CRISIS AND STASIS 214
Davies, J., & Davies, D. (2010). Origins and evolution of antibiotic resistance. Microbiology and
Molecular Biology Reviews, 74(3), 417–433.
DeLong, E. R., DeLong, D. M., & Clarke-Pearson, D. L. (1988). Comparing the areas under two
or more correlated receiver operating characteristic curves: a nonparametric approach.
Biometrics, 44(3), 837–845.
De Paoli, S., & Kerr, A. (2009). “ We Will Always Be One Step Ahead of Them” A Case Study
on the Economy of Cheating in MMORPGs. Journal For Virtual Worlds Research, 2(4).
Retrieved from https://journals.tdl.org/jvwr/index.php/jvwr/article/viewArticle/865
Deterding, S., Dixon, D., Khaled, R., & Nacke, L. (2011). From game design elements to
gamefulness: defining gamification. In Proceedings of the 15th international academic
MindTrek conference: Envisioning future media environments (pp. 9–15). ACM.
Retrieved from http://dl.acm.org/citation.cfm?id=2181040
Deterding, S., Sicart, M., Nacke, L., O’Hara, K., & Dixon, D. (2011). Gamification. using game-
design elements in non-gaming contexts. In CHI’11 Extended Abstracts on Human
Factors in Computing Systems (pp. 2425–2428). ACM. Retrieved from
http://dl.acm.org/citation.cfm?id=1979575
Deveaud, R., SanJuan, E., & Bellot, P. (2014). Accurate and effective latent concept modeling for
ad hoc information retrieval. Document Numérique, 17(1), 61–84.
Diani, M. (2003). Leaders or brokers? Positions and influence in Social Movement Networks.
Social Movements and Networks: Relational Approaches to Collective Action, 105–122.
Diani, M., & McAdam, D. (2003). Social Movements and Networks : Relational Approaches to
Collective Action: Relational Approaches to Collective Action. Oxford University Press.
CRISIS AND STASIS 215
Dibbell, J. (1994). Rape in Cyberspace or How an Evil Clown, a Haitian Trickster Spirit, Two
Wizards, and a Cast of Dozens Turned a Database into a Society, A. Annual Survey of
American Law, 1994, 471.
Dilla, W. N., Harrison, A. J., Mennecke, B. E., & Janvrin, D. J. (2013). The assets are virtual but
the behavior is real: An analysis of fraud in virtual worlds and its implications for the real
world. Journal of Information Systems, 27(2), 131–158.
Doctorow, C. (2003). Down and out in the magic kingdom. Macmillan. Retrieved from
https://books.google.com/books?hl=en&lr=&id=gfg13CM_kU8C&oi=fnd&pg=PA7&dq
=down+and+out+magic+kingdom&ots=qNCDlvuwDC&sig=4qKVctqVd_N1LwAgkdg7
ZZVWyLk
Doherty, S. M., Liskey, D., Via, C. M., Frederick, C., Kring, J. P., & Liu, D. (2014). An Analysis
of Expressed Cheating Behaviors in Video Games. Proceedings of the Human Factors
and Ergonomics Society Annual Meeting, 58(1), 2393–2396.
http://doi.org/10.1177/1541931214581498
Donaldson, S. (2015). Mechanics and Metagame Exploring Binary Expertise in League of
Legends. Games and Culture, 1555412015590063.
http://doi.org/10.1177/1555412015590063
Drachen, A., Yancey, M., Maguire, J., Chu, D., Wang, I. Y ., Mahlmann, T., … Klabajan, D.
(2014). Skill-based differences in spatio-temporal team behaviour in defence of the
Ancients 2 (DotA 2). In Games Media Entertainment (GEM), 2014 IEEE (pp. 1–8).
IEEE. Retrieved from http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7048109
CRISIS AND STASIS 216
Drain, B. (2011, June 21). Controversy brewing over EVE Online microtransactions. Retrieved
December 9, 2011, from http://massively.joystiq.com/2011/06/21/controversy-brewing-
over-eve-online-microtransactions/
Ducheneaut, N., & Moore, R. J. (2004). Let me get my alt: digital identiti (es) in multiplayer
games. In Proceedings of The CSCW2004 Workshop on Representation of Digital
Identities, Chicago, IL.
Ducheneaut, N., Moore, R., & Nickell, E. (2007). Virtual “Third Places”: A Case Study of
Sociability in Massively Multiplayer Games. Computer Supported Cooperative Work
(CSCW), 16(1), 129–166. http://doi.org/10.1007/s10606-007-9041-8
Ducheneaut, N., Yee, N., Nickell, E., & Moore, R. J. (2006). Building an MMO With Mass
Appeal. Games and Culture, 1(4), 281 –317. http://doi.org/10.1177/1555412006292613
Ducheneaut, N., Yee, N., Nickell, E., & Moore, R. J. (2007). The life and death of online gaming
communities: a look at guilds in world of warcraft. In CHI ’07: Proceedings of the
SIGCHI conference on Human factors in computing systems (pp. 839–848). New York,
NY , USA: ACM. Retrieved from http://doi.acm.org/10.1145/1240624.1240750
Duh, H., & Chen, V . (2009). Cheating behaviors in online gaming. Online Communities and
Social Computing, 567–573.
Duijn, P. A. C., Kashirin, V ., & Sloot, P. M. A. (2014). The Relative Ineffectiveness of Criminal
Network Disruption. Scientific Reports, 4. http://doi.org/10.1038/srep04238
Dumitrica, D. D. (2011). An exploration of cheating in a virtual gaming world. Journal of
Gaming & Virtual Worlds, 3(1), 21–36. http://doi.org/10.1386/jgvw.3.1.21_1
Duncan, S. C. (2011). Minecraft, beyond construction and survival. Well Played: A Journal on
Video Games, Value and Meaning, 1(1), 1–22.
CRISIS AND STASIS 217
Edwards, T. (2011, June 25). Eve Online players protest against monocle
prices/microtransactions. Lasers involved. Retrieved December 14, 2013, from
http://www.pcgamer.com/2011/06/25/eve-online-players-protest-against-monocle-
pricesmicrotransactions-lasers-involved/
Eggert, C., Herrlich, M., Smeddinck, J., & Malaka, R. (2015). Classification of Player Roles in
the Team-Based Multi-player Game Dota 2. In K. Chorianopoulos, M. Divitini, J. B.
Hauge, L. Jaccheri, & R. Malaka (Eds.), Entertainment Computing - ICEC 2015 (pp.
112–125). Springer International Publishing. Retrieved from
http://link.springer.com/chapter/10.1007/978-3-319-24589-8_9
Eklund, L. (2010). Social Play? A study of social interaction in temporary group formation
(PUG) in World of Warcraft. Proceedings of DiGRA Nordic 2010: Experiencing Games:
Games, Play, and Players (2010).
Ellison, N. B., Steinfield, C., & Lampe, C. (2007). The Benefits of Facebook “Friends:” Social
Capital and College Students’ Use of Online Social Network Sites. Journal of Computer-
Mediated Communication, 12(4), 1143–1168. http://doi.org/10.1111/j.1083-
6101.2007.00367.x
El-Nasr, M. S., Drachen, A., & Canossa, A. (2013). Game analytics maximizing the value of
player data. London; New York: Springer.
Ertekin, S., Huang, J., Bottou, L., & Giles, L. (2007). Learning on the border: active learning in
imbalanced data classification. In Proceedings of the sixteenth ACM conference on
Conference on information and knowledge management (pp. 127–136). ACM.
CRISIS AND STASIS 218
Esposito, F., Malerba, D., Semeraro, G., & Kay, J. A. (1997). A comparative analysis of methods
for pruning decision trees. Pattern Analysis and Machine Intelligence, IEEE Transactions
on, 19(5), 476–491.
Everett, R. M. (2003a). Diffusion of Innovations, 5th Edition (5th edition). New York: Free Press.
Everett, R. M. (2003b). Diffusion of Innovations, 5th Edition (5th edition). New York: Free Press.
Evers, E., van de Ven, N., & Weeda, D. (2015). The Hidden Cost of Microtransactions: Buying
In-Game Advantages in Online Games Decreases a Player’s Status. International Journal
of Internet Science, 10(1), 20–36.
Faraj, S., Jarvenpaa, S. L., & Majchrzak, A. (2011). Knowledge collaboration in online
communities. Organization Science, 22(5), 1224–1239.
Fellman, P. V ., & Strathern, M. (2007). The Symmetries and Redundancies of Terror: Patterns in
the Dark, A study of Terrorist Network Strategy and Structure. arXiv:0707.4028 [nlin].
Retrieved from http://arxiv.org/abs/0707.4028
Ferrari, S. (2013). From generative to conventional play: Moba and league of legends.
Proceedings of DiGRA 2013: DeFragging Game Studies, 1(1), 1–17.
Fields, D. A., & Kafai, Y . B. (2010). “‘Stealing From Grandma’”or Generating Cultural
Knowledge? Contestations and Effects of Cheating in a Tween Virtual World. Games and
Culture, 5(1), 64–87.
Fields, T. V . (2013). Game industry metrics terminology and analytics case study. In Game
Analytics (pp. 53–71). Springer.
Fine, G. A. (1983). Shared fantasy: Role playing games as social worlds. University of Chicago
Press.
CRISIS AND STASIS 219
FishingSloths. (2014). Dota 2 - Void 6.80 Chronosphere buff. Retrieved from
https://www.youtube.com/watch?v=cujV4He3j5Y
Foo, C. Y . (2008). Grief Play Management: A Qualitative Study of Grief Play Management in
MMORPGs. VDM Publishing.
Freeman, L. C. (1977). A set of measures of centrality based on betweenness. Sociometry, 35–41.
Friedman, E. J., & Resnick, P. (2001). The Social Cost of Cheap Pseudonyms. Journal of
Economics & Management Strategy, 10(2), 173–199. http://doi.org/10.1111/j.1430-
9134.2001.00173.x
Friedman, J. H. (2002). Stochastic gradient boosting. Computational Statistics & Data Analysis,
38(4), 367–378.
Galar, M., Fernandez, A., Barrenechea, E., Bustince, H., & Herrera, F. (2012). A review on
ensembles for the class imbalance problem: bagging-, boosting-, and hybrid-based
approaches. Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE
Transactions on, 42(4), 463–484.
Ganley, D., & Lampe, C. (2009). The ties that bind: Social network principles in online
communities. Decision Support Systems, 47(3), 266–274.
Gersick, C. J. (1991). Revolutionary change theories: A multilevel exploration of the punctuated
equilibrium paradigm. Academy of Management Review, 16(1), 10–36.
Goffman, E. (1974). Frame analysis: An essay on the organization of experience. Harvard
University Press.
González-Bailón, S., Borge-Holthoefer, J., Rivero, A., & Moreno, Y . (2011). The Dynamics of
Protest Recruitment through an Online Network. Scientific Reports, 1.
http://doi.org/10.1038/srep00197
CRISIS AND STASIS 220
Goodreau, S. M., Kitts, J. A., & Morris, M. (2009). Birds of a feather, or friend of a friend? using
exponential random graph models to investigate adolescent social networks*.
Demography, 46(1), 103–125.
Gould, N. E.-S. J. (1972). Punctuated equilibria: an alternative to phyletic gradualism. Retrieved
from
https://books.google.com/books?hl=en&lr=&id=3ULyAgAAQBAJ&oi=fnd&pg=PA239
&dq=punctuated+equilibrium+gould&ots=jZe3BVeAqk&sig=lXe0JM7B7Zz1PC0K8h4
w-3VeZgc
Gourevitch, P. A., & Shinn, J. (2007). Political power and corporate control: the new global
politics of corporate governance. Princeton University Press.
Granovetter, M. (1983). The Strength of Weak Ties: A Network Theory Revisited. Sociological
Theory, 1, 201–233.
Grayson, N. (2011, October 20). Interview: EVE Online’s Kristoffer Touborg - PC Gamer.
Retrieved April 1, 2016, from http://www.pcgamer.com/interview-eve-onlines-kristoffer-
touborg/?ns_campaign=article-
feed&ns_mchannel=ref&ns_source=steam&ns_linkname=0&ns_fee=0
Grayson, N. (2014, November 24). Top Counter-Strike Players Caught In Big Cheating Scandal.
Retrieved May 26, 2016, from http://kotaku.com/top-counter-strike-players-caught-in-
big-cheating-scand-1662810816
Greene, H. (1997). Undead Laws: The Use of Historically Unenforced Criminal Statutes in Non-
Criminal Litigation. Yale Law & Policy Review, 16(1), 169–194.
Gregg, D. G., & Scott, J. E. (2006). The role of reputation systems in reducing on-line auction
fraud. International Journal of Electronic Commerce, 10(3), 95–120.
CRISIS AND STASIS 221
Greiner, L. E. (1972). Evolution and revolution as organizations grow. Retrieved from
http://in8dna.com/Portals/0/documents/partners/HBR_EvolutionRevolution.pdf
Groen, A. (2011, June 28). EVE Online protests could cost CCP $1 million. Retrieved December
9, 2011, from http://www.vg247.com/2011/06/28/eve-online-protests-could-cost-ccp-1-
million/
Gross, R., & Acquisti, A. (2005). Information revelation and privacy in online social networks.
In Proceedings of the 2005 ACM workshop on Privacy in the electronic society (pp. 71–
80). ACM.
Haldane, J. B. S. (1949). Suggestions as to Quantitative Measurement of Rates of Evolution.
Evolution, 3(1), 51–56. http://doi.org/10.2307/2405451
Hamari, J., & Lehdonvirta, V . (2010). Game design as marketing: How game mechanics create
demand for virtual goods. International Journal of Business Science & Applied
Management, 5(1), 14–29.
Hau, Y . S., & Kim, Y .-G. (2011). Why would online gamers share their innovation-conducive
knowledge in the online game user community? Integrating individual motivations and
social capital perspectives. Computers in Human Behavior, 27(2), 956–970.
http://doi.org/10.1016/j.chb.2010.11.022
Have you been accused for hacking before? (2011, June 3). Retrieved May 8, 2016, from
http://forums.steampowered.com/forums/archive/index.php/t-1915641.html
Hirschman, A. O. (1970). Exit, voice, and loyalty. Harvard University Press.
Holin Lin, & Sun, C.-T. (2010). Cash Trade in Free-to-Play Online Games. Games and Culture,
6(3), 270–287. http://doi.org/10.1177/1555412010364981
CRISIS AND STASIS 222
Holme, P., Kim, B. J., Yoon, C. N., & Han, S. K. (2002). Attack vulnerability of complex
networks. Physical Review E, 65(5), 056109.
Huang, H., Tang, J., Wu, S., & Liu, L. (2014). Mining triadic closure patterns in social networks.
In Proceedings of the companion publication of the 23rd international conference on
World wide web companion (pp. 499–504). International World Wide Web Conferences
Steering Committee.
Hu, D., Kaza, S., & Chen, H. (2009). Identifying significant facilitators of dark network
evolution. Journal of the American Society for Information Science and Technology,
60(4), 655–665. http://doi.org/10.1002/asi.21008
Huizinga, J. (1971). Homo Ludens: A Study of the Play-Element in Culture. Beacon Press.
Hutchby, I. (2001). Technologies, texts and affordances. Sociology, 35(2), 441–456.
Huynh, K.-P., Lim, S.-W., & Skoric, M. M. (2013). Stepping out of the magic circle: Regulation
of play/life boundary in MMO-mediated romantic relationship. Journal of Computer-
Mediated Communication, 18(3), 251–264.
Ireland, J. (2013, January 5). “Democracy ” in a Virtual World: EVE Online’s Council of Stellar
Management and the Power of Influence. University of Wisconsin Milwaukee. Retrieved
from http://gradworks.umi.com/15/38/1538532.html
Is it really necessary to nerf Tinker? • /r/DotA2. (2014, September 25). Retrieved March 29,
2016, from
https://www.reddit.com/r/DotA2/comments/2h4ogo/is_it_really_necessary_to_nerf_tinke
r/
Jagatic, T. N., Johnson, N. A., Jakobsson, M., & Menczer, F. (2007). Social phishing. Commun.
ACM, 50(10), 94–100. http://doi.org/10.1145/1290958.1290968
CRISIS AND STASIS 223
James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An Introduction to Statistical
Learning: with Applications in R (1st ed. 2013, Corr. 5th printing 2015 edition). New
York: Springer.
Jeni, L. A., Cohn, J. F., & De La Torre, F. (2013). Facing Imbalanced Data–Recommendations
for the Use of Performance Metrics. In Affective Computing and Intelligent Interaction
(ACII), 2013 Humaine Association Conference on (pp. 245–251). IEEE.
Jenkins, H. (2006a). Convergence Culture: Where Old and New Media Collide. NYU Press.
Jenkins, H. (2006b, July 21). So What Happened to Star Wars Galaxies? Retrieved from
http://henryjenkins.org/2006/07/so_what_happened_to_star_wars.html
Jenkins, H. (2006c, July 21). So What Happened to Star Wars Galaxies? Retrieved from
http://henryjenkins.org/2006/07/so_what_happened_to_star_wars.html
Jenkins, H. (2006d, August 2). National Politics within Virtual Game Worlds: The Case of China.
Aca-Fan, MIT. Retrieved from
http://henryjenkins.org/2006/08/national_politics_within_virtu_1.html
Kabus, P., Terpstra, W. W., Cilia, M., & Buchmann, A. P. (2005). Addressing cheating in
distributed MMOGs. In Proceedings of 4th ACM SIGCOMM workshop on Network and
system support for games (pp. 1–6). New York, NY , USA: ACM.
http://doi.org/10.1145/1103599.1103607
Keegan, B., Ahmed, M. A., Williams, D., Srivastava, J., & Contractor, N. (2010). Dark Gold:
Statistical Properties of Clandestine Networks in Massively Multiplayer Online Games.
In 2010 IEEE Second International Conference on Social Computing (SocialCom) (pp.
201 –208). http://doi.org/10.1109/SocialCom.2010.36
CRISIS AND STASIS 224
Kelly, S., & Ahmad, K. (2015). Propagating Disaster Warnings on Social and Digital Media. In
K. Jackowski, R. Burduk, K. Walkowiak, M. Woźniak, & H. Yin (Eds.), Intelligent Data
Engineering and Automated Learning – IDEAL 2015 (pp. 475–484). Springer
International Publishing. Retrieved from http://link.springer.com/chapter/10.1007/978-3-
319-24834-9_55
Kempe, D., Kleinberg, J., & Tardos, É. (2005). Influential nodes in a diffusion model for social
networks. In Automata, languages and programming (pp. 1127–1138). Springer.
Retrieved from http://link.springer.com/chapter/10.1007/11523468_91
Kennedy, R. (2008). Virtual rights? Property in online game objects and characters. Information
& Communications Technology Law, 17(2), 95–106.
Khong, D. W. (2007). Orphan Works, Abandonware and the Missing Market for Copyrighted
Goods. International Journal of Law and Information Technology, 15(1), 54–89.
King, G., & Zeng, L. (2001). Logistic regression in rare events data. Political Analysis, 9(2),
137–163.
Koster, R. (2013). Theory of Fun for Game Design. O’Reilly Media, Inc.
Krebs, V . (2002a). Mapping Networks of Terrorist Cells. CONNECTIONS, 24(3), 43–52.
Krebs, V . (2002b). Uncloaking Terrorist Networks. First Monday, 7(4). Retrieved from
http://pear.accc.uic.edu/ojs/index.php/fm/article/view/941
Kucuk, S. U. (2008). Consumer Exit, V oice, and’Power’on the Internet. Journal of Research for
Consumers, (15), 1.
Kuo, A. (2001). A (very) brief history of cheating. In How They Got Game Archive.
Lan, X., Zhang, Y ., & Xu, P. (2009). An overview on game cheating and its counter-measures. In
Proceedings of 2009 International Symposium on Computer Science and Computational
CRISIS AND STASIS 225
Technology (ISCSCT2009). Huang-Shan China (pp. 195–200). Citeseer. Retrieved from
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.403.6124&rep=rep1&type=pdf
#page=207
Lastowka, G. (2011). Virtual Justice: The New Laws of Online Worlds. Yale University Press.
Lastowka, G., & Hunter, D. (2004). Virtual Crimes. New York Law School Law Review, 49, 293.
Laurens, P., Paige, R. F., Brooke, P. J., & Chivers, H. (2007). A Novel Approach to the Detection
of Cheating in Multiplayer Online Games. In 12th IEEE International Conference on
Engineering Complex Computer Systems, 2007 (pp. 97–106).
http://doi.org/10.1109/ICECCS.2007.11
Leavitt, A. (2015). “This is a Throwaway Account”: Temporary Technical Identities and
Perceptions of Anonymity in a Massive Online Community. In Proceedings of the 18th
ACM Conference on Computer Supported Cooperative Work & Social Computing
(pp. 317–327). New York, NY , USA: ACM. http://doi.org/10.1145/2675133.2675175
Leavitt, A., Clark, J., & Wixon, D. (2016). Uses of Multiple Characters in Online Games and
Their Implications for Social Network Methods. In Proceedings of CSCW.
Lehdonvirta, V ., & Castronova, E. (2014). Virtual Economies: Design and Analysis. Cambridge,
MA: The MIT Press.
Lessig, L. (2006a). Code: And Other Laws of Cyberspace, Version 2.0. Basic Books.
Lessig, L. (2006b). Code: And Other Laws of Cyberspace, Version 2.0. Basic Books.
Liaw, A., & Wiener, M. (2002). Classification and regression by randomForest. R News, 2(3),
18–22.
CRISIS AND STASIS 226
Lin, H., & Sun, C.-T. (2007). Cash trade within the magic circle: Free-to-play game challenges
and massively multiplayer online game player responses. In Proceedings of DiGRA (pp.
335–343). Citeseer.
Lin, M., Lucas, H. C., & Shmueli, G. (2013). Research Commentary—Too Big to Fail: Large
Samples and the p-Value Problem. Information Systems Research, 24(4), 906–917.
http://doi.org/10.1287/isre.2013.0480
Lowood, H. (2006). Storyline, dance/music, or PvP? Game movies and community players in
World of Warcraft. Games and Culture, 1(4), 362–382.
Magaloni, B., Diaz-Cayeros, A., Matanock, A. M., & Romero, V . (2011). Living in Fear:
Mapping the Social Embeddedness of Drug Gangs and Violence in Mexico (SSRN
Scholarly Paper No. ID 1963836). Rochester, NY: Social Science Research Network.
Retrieved from http://papers.ssrn.com/abstract=1963836
Martin, R. M. (1983). Pluralism and the New Corporatism*. Political Studies, 31(1), 86–102.
http://doi.org/10.1111/j.1467-9248.1983.tb01336.x
Mason, K. A., & Benson, M. L. (1996). The effect of social support on fraud victims’ reporting
behavior: A research note. Justice Quarterly, 13(3), 511–524.
http://doi.org/10.1080/07418829600093071
McAdam, D. (1999). Political Process and the Development of Black Insurgency, 1930-1970 (1st
ed.). University Of Chicago Press.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a Feather: Homophily in Social
Networks. Annual Review of Sociology, 27(1), 415–444.
Meades, A. (2015). Understanding Counterplay in Video Games. Routledge.
CRISIS AND STASIS 227
Meades, A., & Canterbury, U. (2012). Why we Glitch: process, meaning and pleasure in the
discovery, documentation, sharing and use of videogame exploits. Well-Played: A Journal
on Videogames, Value and Meaning, 2(02), 79–98.
Meer, A. (2015, October 29). Paradox Buys White Wolf – Including World Of Darkness &
Vampire The Masquerade – From EVE Online Firm CCP. Retrieved from
https://www.rockpapershotgun.com/2015/10/29/paradox-white-wolf/
Meir, R., & Rätsch, G. (2003). An introduction to boosting and leveraging. In Advanced lectures
on machine learning (pp. 118–183). Springer. Retrieved from
http://link.springer.com/chapter/10.1007/3-540-36434-X_4
Meyer, D. S. (2004). Protest and Political Opportunities. Annual Review of Sociology, 30, 125–
145.
Miller, V . D., Poole, M. S., Seibold, D. R., Myers, K. K., Park, H. S., Monge, P., … others.
(2011). Advancing research in organizational communication through quantitative
methodology. Management Communication Quarterly, 25(1), 4–58.
Milo, R., Shen-Orr, S., Itzkovitz, S., Kashtan, N., Chklovskii, D., & Alon, U. (2002). Network
Motifs: Simple Building Blocks of Complex Networks. Science, 298(5594), 824–827.
http://doi.org/10.1126/science.298.5594.824
Mitchell, L. E. (2004). Structural Holes, CEOs, and Informational Monopolies - The Missing
Link in Corporate Governance. Brooklyn Law Review, 70, 1313.
Moeller, R. M., Esplin, B., & Conway, S. (2009). Cheesers, pullers, and glitchers: The rhetoric of
sportsmanship and the discourse of online sports gamers. Game Studies, 9(2).
Monge, P., & Contractor, N. (2003). Theories of Communication Networks. Oxford ; New York:
Oxford University Press.
CRISIS AND STASIS 228
Monge, P., Heiss, B. M., & Margolin, D. B. (2008). Communication network evolution in
organizational communities. Communication Theory, 18(4), 449–477.
Morris, A. D., & Staggenborg, S. (2012). Leadership in Social Movements. In D. A. Snow, S. A.
Soule, & H. Kriesi (Eds.), The Blackwell Companion to Social Movements (pp. 171–196).
Blackwell Publishing Ltd. Retrieved from
http://onlinelibrary.wiley.com/doi/10.1002/9780470999103.ch8/summary
Morselli, C. (2010). Assessing Vulnerable and Strategic Positions in a Criminal Network.
Journal of Contemporary Criminal Justice, 26(4), 382–392.
http://doi.org/10.1177/1043986210377105
Morselli, C., Giguère, C., & Petit, K. (2007). The efficiency/security trade-off in criminal
networks. Social Networks, 29(1), 143–153. http://doi.org/10.1016/j.socnet.2006.05.001
Neuvirth, H., Finkelstein, Y ., Hilbuch, A., Nahum, S., Alon, D., & Yom-Tov, E. (2015). Early
Detection of Fraud Storms in the Cloud. In Machine Learning and Knowledge Discovery
in Databases (pp. 53–67). Springer.
Olson, M. (1965a). The Logic of Collective Action: Public Goods and the Theory of Groups.
Harvard University Press.
Olson, M. (1965b). The Logic of Collective Action: Public Goods and the Theory of Groups.
Harvard University Press.
Paap, J., & Katz, R. (2004). Anticipating Disruptive Innovation. Research Technology
Management, 47(5), 13–22.
Pearce, C., Boellstorff, T., & Nardi, B. A. (2011). Communities of Play: Emergent Cultures in
Multiplayer Games and Virtual Worlds. MIT Press.
CRISIS AND STASIS 229
Peterson, C. (2013, May 1). The $28000 Team Fortress 2 backpack. Retrieved from
http://www.pcgamesn.com/tf2/28000-team-fortress-2-backpack
Pisan, Y . (2007). My guild, my people: role of guilds in massively multiplayer online games. In
Proceedings of the 4th Australasian conference on Interactive entertainment (pp. 20:1–
20:5). Melbourne, Australia, Australia: RMIT University. Retrieved from
http://dl.acm.org/citation.cfm?id=1367956.1367976
Pobiedina, N., Neidhardt, J., Calatrava Moreno, M. del C., & Werthner, H. (2013). Ranking
Factors of Team Success. In Proceedings of the 22Nd International Conference on World
Wide Web Companion (pp. 1185–1194). Republic and Canton of Geneva, Switzerland:
International World Wide Web Conferences Steering Committee. Retrieved from
http://dl.acm.org/citation.cfm?id=2487788.2488147
Poor, N., & Skoric, M. M. (2014). Death of a Guild, Birth of a Network Online Community Ties
Within and Beyond Code. Games and Culture, 9(3), 182–202.
http://doi.org/10.1177/1555412014537401
Preece, J., & Maloney-Krichmar, D. (2005). Online communities: Design, theory, and practice.
Journal of Computer-Mediated Communication, 10(4), 00–00.
Preece, J., Nonnecke, B., & Andrews, D. (2004). The top five reasons for lurking: improving
community experiences for everyone. Computers in Human Behavior, 20(2), 201–223.
http://doi.org/10.1016/j.chb.2003.10.015
Putnam, R. (2000). Bowling Alone: The Collapse and Revival of the American Community. New
York: NY: Simon & Schuster.
Putnam, R. D. (2001). Bowling Alone: The Collapse and Revival of American Community (1st
ed.). Touchstone Books by Simon & Schuster.
CRISIS AND STASIS 230
Putnam, R. D., Leonardi, R., & Nanetti, R. Y . (1994). Making Democracy Work: Civic Traditions
in Modern Italy. Princeton University Press.
Raab, J., & Milward, H. B. (2003). Dark Networks as Problems. Journal of Public
Administration Research and Theory, 13(4), 413–439.
http://doi.org/10.1093/jpart/mug029
Ratan, R. A., Chung, J. E., Shen, C., Williams, D., & Poole, M. S. (2010). Schmoozing and
Smiting: Trust, Social Institutions, and Communication Patterns in an MMOG. . Journal
of Computer Mediated Communication, 16(1), 93–114. http://doi.org/10.1111/j.1083-
6101.2010.01534.x
Ratan, R. A., Taylor, N., Hogan, J., Kennedy, T., & Williams, D. (2015). Stand by Your Man An
Examination of Gender Disparity in League of Legends. Games and Culture,
1555412014567228. http://doi.org/10.1177/1555412014567228
Reeves, S., Brown, B., & Laurier, E. (2009). Experts at Play Understanding Skilled Expertise.
Games and Culture, 4(3), 205–227. http://doi.org/10.1177/1555412009339730
Rheingold, H. (2000). The virtual community: homesteading on the electronic frontier. MIT
Press.
Ribbens, W., Poels, Y ., & Lamotte, G. (2011). Fail with honour or win by cheating? Research
into the perceptions and motivations of cheaters in online multiplayer games. Vice City
Virtue. Moral Issues in Digital Game Play, Acco, Leuven, 177–203.
Romanelli, E., & Tushman, M. L. (1994). Organizational transformation as punctuated
equilibrium: An empirical test. Academy of Management Journal, 37(5), 1141–1166.
Rubin, V . L., & Camm, S. C. (2013). Deception in video games: examining varieties of griefing.
Online Information Review, 37(3), 369–387. http://doi.org/10.1108/OIR-10-2011-0181
CRISIS AND STASIS 231
Saed, S. (2015, February 8). Steam has over 125 million active users, 8.9M concurrent peak.
Retrieved from https://www.vg247.com/2015/02/24/steam-has-over-125-million-active-
users-8-9m-concurrent-peak/
Salen, K., & Zimmerman, E. (2004). Rules of play: Game design fundamentals. MIT press.
Savage, P. (2013, February 27). EA planning microtransactions for “all of our games” | PC
Gamer. Retrieved May 26, 2016, from http://www.pcgamer.com/ea-planning-
microtransactions-for-all-of-our-games/
Scacchi, W. (2010). Computer game mods, modders, modding, and the mod scene. First
Monday, 15(5). Retrieved from
http://uncommonculture.org/ojs/index.php/fm/article/view/2965
Schiesel, S. (2005, December 10). For Online Star Wars Game, It’s Revenge of the Fans.
Retrieved from http://www.nytimes.com/2005/12/10/arts/10star.html
Schumpeter, J. (1942). Creative destruction. Capitalism, Socialism and Democracy, 82–5.
Shen, C. (2010). The patterns, effects and evolution of player social networks in online gaming
communities. University of Southern California.
Shen, C. (2014). Network patterns and social architecture in Massively Multiplayer Online
Games: Mapping the social world of EverQuest II. New Media & Society, 16(4), 672–
691. http://doi.org/10.1177/1461444813489507
Shen, C., Monge, P., & Williams, D. (2011). The Evolving Virtual Relationships: A Longitudinal
Analysis of Player Social Networks in a Large MMOG (SSRN Scholarly Paper No. ID
1929908). Rochester, NY: Social Science Research Network. Retrieved from
http://papers.ssrn.com/abstract=1929908
CRISIS AND STASIS 232
Shen, C., Monge, P., & Williams, D. (2012). Virtual Brokerage and Closure: Network Structure
and Social Capital in a Massively Multiplayer Online Game. Communication Research.
http://doi.org/10.1177/0093650212455197
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. http://doi.org/10.1002/asi.23129
Shifman, L. (2012). An anatomy of a YouTube meme. New Media & Society, 14(2), 187–203.
Skiing. (2015, September 29). Retrieved September 29, 2015, from
http://tribes.wikia.com/wiki/Skiing
Smith, Q. (2011, June 24). Eve Online Dev Responds To Incarna Anger. Retrieved from
http://www.rockpapershotgun.com/2011/06/24/eve-online-dev-responds-to-incarna-
anger/
Stanton, R. (2016, January 12). A negative-sum game: Policing Counter-Strike: GO cheaters with
Overwatch | Ars Technica. Retrieved May 26, 2016, from
http://arstechnica.com/gaming/2016/01/how-valve-crowdsources-the-policing-of-
cheaters-in-counter-strike-go/
Starsiege-Tribes. (2010). Retrieved from http://www.freemmogamer.com/2008/09/starsiege-
tribes.html
Steam Reputation FAQ. (2015). Retrieved from http://forums.steamrep.com/pages/faq/
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. http://doi.org/10.1111/j.1083-6101.2006.00300.x
CRISIS AND STASIS 233
Sun, C.-T. (2005). The “White-Eyed” Player Culture: Grief Play and Construction of Deviance in
MMORPGs.
Susarla, A., Oh, J.-H., & Tan, Y . (2012). Social networks and the diffusion of user-generated
content: Evidence from YouTube. Information Systems Research, 23(1), 23–41.
Tarrow, S. (1998). Power in Movement: Social Movements and Contentious Politics. Cambridge
University Press.
Taylor, T. L. (2002). “ Whose Game Is This Anyway?”: Negotiating Corporate Ownership in a
Virtual World. In CGDC Conf.
Taylor, T. L. (2012). Raising the Stakes: E-sports and the Professionalization of Computer
Gaming. Mit Press. Retrieved from
https://books.google.com/books?hl=en&lr=&id=CiL8aPrSeKcC&oi=fnd&pg=PR5&dq=
professional+gamomg&ots=dCFlTE8rbJ&sig=rkyFAAxLCFULQu6CLHK7sD_5lTw
Theocharis, Y . (2013). The wealth of (occupation) networks? Communication patterns and
information distribution in a Twitter protest network. Journal of Information Technology
& Politics, 10(1), 35–56.
Tillman, R. (2003). Abandoned Consumers: Deregulation and Fraud in the California Auto
Insurance Industry. Social Policy and Society, 2(01), 45–53.
http://doi.org/10.1017/S1474746403001064
Tillman, R., & Indergaard, M. (1999). Field of Schemes: Health Insurance Fraud in the Small
Business Sector. Social Problems, 46(4), 572–590. http://doi.org/10.2307/3097076
tomatocarrotjuice. (2015, February 17). Ex-cheaters of CS:GO, why do you cheat and what made
you stop? • /r/GlobalOffensive. Retrieved May 8, 2016, from
CRISIS AND STASIS 234
https://www.reddit.com/r/GlobalOffensive/comments/2w5q4e/excheaters_of_csgo_why_
do_you_cheat_and_what_made/
Tremayne, M. (2014). Anatomy of protest in the digital era: A network analysis of Twitter and
Occupy Wall Street. Social Movement Studies, 13(1), 110–126.
Tushman, M. L., & Romanelli, E. (2008). Organizational evolution’. Organization Change: A
Comprehensive Reader, 155, 2008174.
Uzzi, B., Mukherjee, S., Stringer, M., & Jones, B. (2013). Atypical Combinations and Scientific
Impact. Science, 342(6157), 468–472. http://doi.org/10.1126/science.1240474
Valente, T. W. (1995). Network models of the diffusion of innovations. Retrieved from
http://www.sidalc.net/cgi-
bin/wxis.exe/?IsisScript=sibe01.xis&method=post&formato=2&cantidad=1&expresion=
mfn=019132
Valente, T. W., & Davis, R. L. (1999). Accelerating the Diffusion of Innovations Using Opinion
Leaders. The ANNALS of the American Academy of Political and Social Science, 566(1),
55–67. http://doi.org/10.1177/000271629956600105
Valve. (2012, August 15). Team Fortress 2 - Mann vs. Machine. Retrieved April 25, 2016, from
http://www.teamfortress.com/mvm/
Valve. (2013, September 3). News - New Steam Trading Feature - Trade Offers. Retrieved April
25, 2016, from http://store.steampowered.com/news/11395/
Valve. (2015a). Steam Trading Policy. Retrieved from f
https://support.steampowered.com/kb_article.php?ref=9958-MJDG-3003
Valve. (2015b, September 1). Steam Community. Retrieved from http://steamcommunity.com
CRISIS AND STASIS 235
Valve. (2015c, December 9). News - Security and Trading. Retrieved April 25, 2016, from
http://store.steampowered.com/news/19618/
Valve. (2016, March 1). News - Security and Trading: Update. Retrieved April 25, 2016, from
http://store.steampowered.com/news/20631/
Varoufakis, Y . (2012, June 14). It All Began with a Strange Email. Retrieved from
http://blogs.valvesoftware.com/economics/it-all-began-with-a-strange-email/
V oulgari, I., & Komis, V . (2011). Bending the Rules... and Adding Some New Ones: Legal and
Illegal Behaviours of Players in Massively Multiplayer Online Games. In International
Workshop DEG.
Ward, M. (2006). Gay rights win in Warcraft world (V ol. 2009). Retrieved from
http://news.bbc.co.uk/2/hi/technology/4700754.stm
Wasserman, S., & Faust, K. (1994). Social Network Analysis: Methods and Applications (1st
ed.). Cambridge University Press.
Weiner, J. (1995). The Beak of the Finch: A Story of Evolution in Our Time (Reprint edition).
New York: Vintage.
Weng, L., Menczer, F., & Ahn, Y .-Y . (2013). Virality Prediction and Community Structure in
Social Networks. Scientific Reports, 3. http://doi.org/10.1038/srep02522
Wiesenfeld, S. L. (1967). Sickle-Cell Trait in Human Biological and Cultural Evolution. Science,
157(3793), 1134–1140. http://doi.org/10.1126/science.157.3793.1134
Williams, D. (2007). The impact of time online: Social capital and cyberbalkanization.
CyberPsychology & Behavior, 10(3), 398–406.
CRISIS AND STASIS 236
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.
Williams, D., Kennedy, T. L. M., & Moore, R. J. (2011). Behind the Avatar: The Patterns,
Practices, and Functions of Role Playing in MMOs. Games and Culture, 6(2), 171–200.
http://doi.org/10.1177/1555412010364983
Williams, D., & Xiong, L. (2009). Herding Cats Online: Real Studies of Virtual Communities. In
E. Hargittai (Ed.), Research Confidential: Solutions to Problems Most Social Scientists
Pretend They Never Have. University of Michigan Press.
Williams, J. P., & Kirschner, D. (2012). Coordinated Action in the Massively Multiplayer Online
Game World of Warcraft. Symbolic Interaction, 35(3), 340–367.
http://doi.org/10.1002/symb.22
Woodford, D. (2013). Regulating virtual worlds : considering participant-driven approaches. In
Proceedings of DiGRA 2013 : DeFragging Game Studies. (pp. 1–14). Atlanta, GA:
Digital Games Research Association. Retrieved from http://www.digra.org/wp-
content/uploads/digital-library/paper_174.pdf
Wood, R. T. A., Griffiths, M. D., & Eatough, V . (2004). Online data collection from video game
players: methodological issues. Cyberpsychology & Behavior: The Impact of the Internet,
Multimedia and Virtual Reality on Behavior and Society, 7(5), 511–518.
Wyk, J. V ., & Benson, M. L. (1997). Fraud victimization: Risky business or just bad luck?
American Journal of Criminal Justice, 21(2), 163–179.
http://doi.org/10.1007/BF02887448
CRISIS AND STASIS 237
Xu, J., & Chen, H. (2008). The topology of dark networks. Commun. ACM, 51(10), 58–65.
http://doi.org/10.1145/1400181.1400198
Yang, P., Harrison, B., & Roberts, D. L. (n.d.). Identifying patterns in combat that are predictive
of success in moba games.
Yan, J., & Randell, B. (2005). A systematic classification of cheating in online games. In
Proceedings of 4th ACM SIGCOMM workshop on Network and system support for games
(pp. 1–9). New York, NY, USA: ACM. http://doi.org/10.1145/1103599.1103606
Yee, N. (2006a). Motivations for Play in Online Games. CyberPsychology & Behavior, 9(6),
772–775. http://doi.org/10.1089/cpb.2006.9.772
Yee, N. (2006b). The Demographics, Motivations, and Derived Experiences of Users of
Massively Multi-User Online Graphical Environments. Presence: Teleoperators and
Virtual Environments, 15(3), 309–329. http://doi.org/10.1162/pres.15.3.309
Yee, N. (2008). Maps of digital desires: Exploring the topography of gender and play in online
games. Beyond Barbie and Mortal Kombat: New Perspectives on Gender and Gaming,
83–96.
Yu, D., & Hang, C. C. (2010). A Reflective Review of Disruptive Innovation Theory.
International Journal of Management Reviews, 12(4), 435–452.
http://doi.org/10.1111/j.1468-2370.2009.00272.x
Zackariasson, P., & Wilson, T. L. (2008). Game on: Competition and Competitiveness in the
Video Game Industry. Competition Forum, 6(1), 43–52.
Zacny, R. (2011, October 18). Layoffs at “overstretched” CCP, World of Darkness team worst
hit. Retrieved April 1, 2016, from http://www.pcgamer.com/layoffs-at-overstretched-ccp-
CRISIS AND STASIS 238
world-of-darkness-team-worst-hit/?ns_campaign=article-
feed&ns_mchannel=ref&ns_source=steam&ns_linkname=0&ns_fee=0
Zhu, X., & Davidson, I. (2007). Knowledge discovery and data mining: challenges and realities.
Information Science Reference.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The evolution of multilevel organizational networks in an online gaming community

PDF

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

PDF

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

PDF

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

PDF

Video game live streaming and the perception of female gamers

PDF

The evolution of multidimensional and multilevel networks in online crowdsourcing

PDF

Behavior-based approaches for detecting cheating in online games

PDF

Ecology and network evolution in online innovation contest crowdsourcing

PDF

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

PDF

Running with newbies: understanding online communities through the eyes of second-generation gamers

PDF

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

PDF

The evolution of scientific collaboration networks

PDF

Predicting and modeling human behavioral changes using digital traces

PDF

Upvoting the news: breaking news aggregation, crowd collaboration, and algorithm-driven attention on reddit.com

PDF

Virtual worlds as contact zones: development, localization, and intergroup communication in MMORPGs

PDF

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

PDF

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

PDF

Contagious: social norms about health in work group networks

PDF

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

PDF

The Distance: a cooperative communication game to long-distance players

Asset Metadata

Creator Clark, Joshua (author)

Core Title Crisis and stasis: understanding change in online communities

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Publication Date 08/03/2016

Defense Date 05/27/2016

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

Tag cheating,communication,EVE Online,game studies,games,OAI-PMH Harvest,social network analysis,Team Fortress 2

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Williams, Dmitri (committee chair), Monge, Peter (committee member), Wixon, Dennis (committee member)

Creator Email josh.aclark@gmail.com,joshuaac@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-293897

Unique identifier UC11280689

Identifier etd-ClarkJoshu-4719.pdf (filename),usctheses-c40-293897 (legacy record id)

Legacy Identifier etd-ClarkJoshu-4719.pdf

Dmrecord 293897

Document Type Dissertation

Format application/pdf (imt)

Rights Clark, Joshua

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

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