THE PATTERNS, EFFECTS AND EVOLUTION OF PLAYER SOCIAL NETWORKS IN ONLINE GAMING COMMUNITIES

by Cuihua Shen

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

December 2010

Copyright 2010

Cuihua Shen

ii Dedication To my parents. To Minxue.

iii Acknowledgements This dissertation would not have been possible without the help and support from many people. First and foremost, I thank my advisor, Peter Monge, for his consistent guidance and support throughout my five years at the Annenberg School. In many respects you are my academic father. I have learned so much from you. I am greatly indebted to my excellent dissertation committee members: Janet Fulk, Peter Robertson and Dmitri Williams. Your feedback and support during the process has been invaluable. I especially thank Dmitri Williams for his insistence that I should spend more time playing the game. Without those hours spent talking to and raiding with my guildmates, I would not have the contextual knowledge of EQII necessary to really understand players and their behaviors. All my colleagues in the Virtual Worlds Observatory research team have been instrumental in providing help and feedback on the various stages of this project. In particular, I would like to thank Ron Burt, who has generously provided network metrics of brokerage and closure for analysis. Yun Huang, Muhammad Ahmed and Dora Cai have been extremely helpful in answering my questions, extracting datasets and giving me feedback. I feel very fortunate for being part of an awesome team of social scientists and computer scientists, and I look forward to many future collaborations. I would like to thank my wonderful colleagues at the Annenberg School, especially Helen Hua Wang, Drew Margolin, Matt Weber, Robby Ratan and Seungyoon Lee, for being my co-authors, sounding boards, and travel companions. Conversations, meetings and parties with colleagues in my cohort, the Annenberg Networks Network

iv research team and the Game Studies team (for lack of an official acronym) are among the most enjoyable moments of my five years at Annenberg. This dissertation project is partially supported by grants from the National Science Foundation (IIS-0729505 & IIS-0838548), Army Research Institute (W91WAW08-C-0106), and the Annenberg Program on Online Communities. Ganesh Gaikwad has provided excellent research assistance in extracting and preparing datasets. I also want to thank Sony Online Entertainment for providing access to the server logs of EQII and many players for participating in the survey. Finally, I am grateful to my parents for their love and support along the way. My husband Minxue has put up with my exhaustion, frustration and self-doubt, and responded with unconditional love and encouragement. Thank you for standing by me.

v Table of Contents Dedication ........................................................................................................................... ii  Acknowledgements ............................................................................................................ iii  List of Tables ................................................................................................................... viii  List of Figures .................................................................................................................... ix  Abbreviations ...................................................................................................................... x  Abstract .............................................................................................................................. xi  Chapter 1: Introduction ....................................................................................................... 1  Chapter 2: Conceptual Framework ................................................................................... 11  Understanding MMOGs.................................................................................................... 11  MMOGs as Online Communities ............................................................................ 11  Bandwidth and Social Information Processing ............................................... 12  Relative Anonymity ........................................................................................ 15  From Neighborhoods to Networks.................................................................. 16  Entry and Exit Costs ....................................................................................... 17  Characteristics of EQII ............................................................................................ 19  Game Content and Mechanics ........................................................................ 20  Character Creation .......................................................................................... 27  Social Interactions........................................................................................... 29  Player Guilds ................................................................................................... 41  Network Patterns ............................................................................................................... 46  Virtual Third Places ................................................................................................. 47  Prior Research .......................................................................................................... 50  Mapping the Social World of EQII .......................................................................... 52  Network Effects ................................................................................................................ 57  Social Impacts of Internet Use ................................................................................. 57  What is Social Capital? ............................................................................................ 60  Social Capital: Source or Outcome? ........................................................................ 62  Bridging and Bonding, Brokerage and Closure ....................................................... 64  Network Evolution ............................................................................................................ 75  Variation, Selection, Retention ................................................................................ 76  The Evolution of Personal Ties in EQII .................................................................. 83  Aging and Inertia ..................................................................................................... 86  Social Architecture................................................................................................... 92  Homophily and Proxmity ......................................................................................... 95 

vi Chapter 3: Methods ......................................................................................................... 100  Data Sources ................................................................................................................... 100  Behavioral Logs ..................................................................................................... 102  Player Survey ......................................................................................................... 105  Data Transformation ....................................................................................................... 107  Network Construction ............................................................................................ 107  Time Interval ................................................................................................. 107  Trade Network .............................................................................................. 108  Grouping Network ........................................................................................ 110  Mentoring Network ....................................................................................... 111  Collaboration Network.................................................................................. 112  Chat Network ................................................................................................ 112  Active Players ............................................................................................... 114  Network Measures ................................................................................................. 114  Network Size (Degree) ................................................................................. 114  Brokerage ...................................................................................................... 115  Closure .......................................................................................................... 116  Other Measures ...................................................................................................... 119  Demographics ............................................................................................... 119  EQII Play Time ............................................................................................. 119  Character Level ............................................................................................. 119  Proximity....................................................................................................... 120  Character Class ............................................................................................. 120  Guild Membership ........................................................................................ 121  Task Performance ......................................................................................... 121  Trust .............................................................................................................. 121  Sense of Community Online ......................................................................... 121  Data Analysis .................................................................................................................. 122  Analysis Overview ................................................................................................. 122  Exponential Random Graph (ERG) Models ................................................. 125  Stochastic Actor-Based Models for Network Dynamics .............................. 127  Event History Analysis of Tie Persistence/Decay ........................................ 130  Hypotheses Testing ................................................................................................ 134  Network Patterns ........................................................................................... 134  Network Effects ............................................................................................ 134  Network Evolution ........................................................................................ 135  Chapter 4: Results ........................................................................................................... 138  Network Patterns ............................................................................................................. 138  Preliminary Analysis .............................................................................................. 138  Who is connected? ........................................................................................ 142  Guild Membership ........................................................................................ 145  Factors Associated with Social Engagement ......................................................... 147  Network Effects .............................................................................................................. 154 

vii Brokerage ............................................................................................................... 154  Closure ................................................................................................................... 158  Network Evolution .......................................................................................................... 159  Size of Ego Networks ............................................................................................ 159  Tie Decay ............................................................................................................... 161  Chapter 5: Discussion and Conclusion ........................................................................... 168  Discussion ....................................................................................................................... 168  Network Patterns .................................................................................................... 168  Network Effects ..................................................................................................... 173  Network Evolution ................................................................................................. 177  Limitations and Future Research .................................................................................... 182  Limitations ............................................................................................................. 182  Directions for Future Research .............................................................................. 186  Different Forms of Sociability ...................................................................... 187  Social Networks of Competition and Conflict .............................................. 187  Participation Life Cycles............................................................................... 188  Comparisons Across MMOGs and Other Communities .............................. 189  Virtual Worlds as Testbeds for Social Science Theory ................................ 190  Conclusion ...................................................................................................................... 191  Bibliography ................................................................................................................... 193 

viii List of Tables Table 1. Character Classes of EQII

28 

Table 2. Summary of Research Questions and Hypotheses

99 

Table 3. Time Range of Data Sources

106 

Table 4. Data Sources of Selected Variables

122 

Table 5. Types of Changes in Networks

124 

Table 6. Descriptives of Chat, Trade and Collaboration Networks

143 

Table 7. Density, SD and QAP Correlations of Chat, Trade and Collaboration Ties

145 

Table 8. Correlations Among Study Variables

149 

Table 9. Logistic Regression Models Predicting the Likelihood to Chat, Trade and Collaboration (N = 8074)

150 

Table 10. Regression Models Predicting Degree in Chat, Trade and Collaboration Networks

150 

Table 11. Correlations Among Variables in Regression Models Predicting Character Level (N = 27770) 155  Table 12. Regression Models Predicting Character Level (N = 27770)

156 

Table 13. Correlations Among Variables in Regression Models Predicting Trust and Sense of Community (N = 802) 157  Table 14. Regression Models Predicting Trust and Sense of Community (N = 802)

158 

Table 15. Post Hoc Comparisons Between Character Level Groups

160 

Table 16. Event History Models Predicting Tie Decay

163 

Table 17. Summary of Hypothesis-Testing Results

167 

ix List of Figures Figure 1. Character Creation and Customization

21 

Figure 2. “Looking for More (LFM)” Interface

34 

Figure 3. Raiding in the Protector’s Realm

38 

Figure 4. Guild Interface

42 

Figure 5. A Simple Structural Illustration of Brokerage and Closure

71 

Figure 6. Active Characters on Guk from June 1 to August 30, 2006

108 

Figure 7. Accumulated Play Time by Character Level

139 

Figure 8. Character Class by Gender

141 

Figure 9. Percentage of Connected Characters by Level

144 

Figure 10. Guild Membership by Character Class

146 

Figure 11. Guild Membership by Character Level

147 

Figure 12. Estimated Marginal Means of Network Size Over 13 Weeks

161 

Figure 13. Hazard Estimates of Tie Decay by Guild Membership.

164

x Abbreviations CMC

Computer-Mediated Communication

EQ

EverQuest

EQII

EverQuest II

ERG

Exponential Random Graph

EXE

Exchange

MMOG

Massively Multiplayer Online Game

NPC

Non-Player Character

PvE

Player-versus-Environment

PvP

Player-versus-Player

QAP

Quadratic Assignment Procedure

RP

Role-playing

SOE

Sony Online Entertainment

V-S-R

Variation-Selection-Retention

WoW

World of Warcraft

xi Abstract This dissertation presents a critical examination of the social interactions among MMOG participants by focusing on network patterns, effects and evolution. It is situated in a popular MMOG, EverQuest II (EQII), drawing on a combination of unobtrusively collected behavioral server logs and a comprehensive survey conducted with the players directly through the game engine. An exploratory analysis of network patterns revealed that the social architecture of the world was quite effective in shaping the structure of interaction, as the involvement in various social networks was influenced by class choice and character level. However, sociability among players was quite diffuse, with a sizable number of players opting to play solo despite the built-in mechanisms that encourage collaborative play. Second, drawing on the theory of social capital, this study tested the effects of different structural properties of player social networks. Players who bridged diverse, otherwise unconnected partners were rewarded with better task performance in EQII. But contrary to expectation, players located in dense and closed cliques did not show higher level of trust towards guildmates or sense of community. Lastly, a longitudinal analysis of tie persistence and decay demonstrated the transient nature of social relationships in EQII, but these ties became considerably more durable over time. Also, character level similarity, shared guild membership and geographic proximity were powerful mechanisms in preserving social relationships.

1 Chapter 1: Introduction Over the last three decades a variety of social, economic, and technological changes have rendered obsolete a significant stock of America’s social capital. … Our growing social-capital deficit threatens educational performance, safe neighborhoods, equitable tax collection, democratic responsiveness, everyday honesty, and even our health and happiness. (Putnam, 2000, p. 367) Practical virtual reality emerged unannounced from the dark imagineering labs of the video game industry, got powered by high-speed Internet connections, and exploded across the globe, catching us all by surprise. Already, practical virtual reality immerses 20 or 30 million people in worlds of perpetual fantasy. Over the next generation or two, hundreds of millions will join them. The exodus of these people from the real world, from our normal daily life of living rooms, cubicles, and shopping malls, will create a change in social climate that makes global warming look like a tempest in a teacup. (Castronova, 2007, pp. xiv-xv) The motivation for this dissertation is based on the juxtaposition of two important societal trends in contemporary America, as evidenced in the two quotes above. The first is demonstrated in a long-lasting tradition in social theory regarding the loss of community and the decline of social capital in modern societies (Fischer, 1982; Norris, 2002; Putnam, 2000; Simmel & Hughes, 1949; Tönnies, 1974). Social contact with other human beings generates important instrumental and socio-emotional benefits (Burt, 1992; Coleman, 1988). The networks and membership in voluntary associations help maintain democratic institutions and enhance personal well-being (Putnam, 2000). Yet, our interpersonal environment is changing, and such a change is epitomized in Robert Putnam’s “Bowling Alone” hypothesis (2000), in which he argued that collective activities, from family dinners to participation in bowling clubs, are in a marked decline in American society in the last few decades. Putnam estimated that a

2 significant portion of the change—25%, to be specific—could be attributed to the massive adoption and diffusion of media technologies, television in particular, that have weakened social capital as people are watching more TV (especially entertainment programs), watching it more habitually, pervasively, and alone. The ebbing of community life is further observed in more recent data, as a close comparison between the 2004 General Social Survey with that of 1985 revealed shrinking individual discussion networks and lost connections with local neighborhoods and associations (McPherson, Smith-Lovin, & Brashears, 2006). The second societal trend is the massive diffusion of new media technologies and our increasingly mediated social life made possible by these technologies. We communicate with friends and colleagues via email, keep track of the latest happenings of our social circle on social networking sites such as Facebook and Twitter, and interact with others in various informational and recreational online communities (boyd & Ellison, 2007; Preece & Maloney-Krichmar, 2005). Among the various forms of computermediated communities, one particular genre, Massively Multiplayer Online Games (MMOGs) has become an increasingly popular phenomenon of much social, cultural and economic significance (Bainbridge, 2007; Castronova, 2005; Ducheneaut, Yee, Nickell, & Moore, 2006). MMOGs1 are persistent (always on) and immersive graphical environments where millions of participants engage in goal-oriented, gaming activities as

1

In this dissertation, MMOGs do not include sandbox-type of virtual worlds, such as Second Life. This type of virtual world allows for a much greater level of freedom. There are no developer-created goals and users are given tools to create their own version of the world at will (Williams, in press). MMOGs also exclude social games, such as Farmville, which are based on pre-existing social platforms (e.g., Facebook) and are non-immersive.

3 well as social interactions by controlling characters known as “avatars” (Castronova, 2005; Williams, Yee, & Caplan, 2008; Yee, 2006a). MMOGs afford high-fidelity two- or three-dimensional immersive environments, which allow users to interact with the environment as well as other users via multiple communication channels. MMOGs surpass the typical computer-mediated applications such as the standard chatroom or textbased web forums in that they combine multimedia, comprehensive user control, and interactivity to create an immersive environment that virtually mimics complex physical spaces. In essence, MMOGs symbolize a comprehensive “world” where networked participants could lead virtual lives in growing and developing their characters, crafting weapons, slaying dragons, decorating virtual houses, and more importantly, engaging in short- and long-term social groups. The rise of MMOGs and virtual worlds is poignantly captured in Castronova’s (2007) powerful metaphor of “exodus,” that people are leaving the physical and offline world, albeit temporarily, to join thousands of others (represented by avatars) in exploring and inhabiting the burgeoning virtual universes. By conservative estimates, there are more than 45 million accounts within the virtual worlds in the West (White, 2009), with perhaps double that figure in Asia. A recent PEW report shows that 43% of all adult gamers play games online, while 21% of all teen gamers participate in MMOGs (Lenhart, Jones, & Macgill, 2008). Among all online gamers, MMOG players are especially active and engaged, as 89% play at least a few times a week and nearly half (49%) play every day or almost every day (Lenhart et al., 2008), with research showing typical amount of time played per week ranging between 23 and 27 hours (Griffiths,

4 Davies, & Chappell, 2004; Williams et al., 2008; Yee, 2006a). Considering that the national average of number of hours spent on Internet use at home is 10.1 hours per week, and for the heavy Internet users, 21.8 hours per week (Center for the Digital Future, 2009), the level of time commitment from MMOG players dwarfs that of other types of online activities. Combined, these two societal trends prompt a series of critical questions. As Williams (2006b) speculated, the steady decay of civic and community life in America provides a backdrop by which to understand the rise of networked social games. The demand for social connections has never changed, but increasingly atomized modern societies may have failed to provide adequate structure and outlets to fulfill this demand. As a result, an increasing number of people go online to virtual communities to create fresh and maintain existing social ties (Rheingold, 1993; Wellman, 2001). Among these communities, MMOGs represent one of the most popular and most sophisticated categories and thus, are important sites to study the social dynamics within. Besides asking why people are drawn to MMOGs, it is equally important to study the consequences. To what extent are MMOGs beneficial in facilitating and fulfilling the human demand for social contact? What implications does the massive exodus to virtual spaces bring to the social climate of the offline world? And finally, what roles and functions can virtual worlds serve in the reintegration of society and restoration of social capital? To date, the research on the social dimensions of MMOGs has remained relatively scant, but empirical research has accumulated on the social consequences of the two

5 media that MMOGs are built from—Internet and video games. Since the popular adoption of the Internet in the 1990s, debate has ensued about the psychological and social impact of time spent online. This question persists because a decade of research on this topic has produced conflicting findings (e.g., Kraut et al., 2002; Kraut et al., 1998; Nie, 2001; Shklovski, Kiesler, & Kraut, 2006). For example, some scholars are concerned that Internet use would encroach on time that was previously spent with family and close friends and leave many feeling lonelier (Nie, 2001; Nie & Erbring, 2002), an argument that received early support (e.g., Kraut et al., 1998). In contrast, others praise the potential of the Internet to supply an additional avenue of social interaction. Not only could the Internet enhance one’s everyday communication with family and friends locally and over a distance (e.g., Wellman, 2001), it could also enlarge one’s existing social network by bringing together people with shared interests and values in virtual communities (e.g., Horrigan, Rainie, & Fox, 2001). The social functions of the Internet are also supported by empirical data (see the section on network effects; Boase, Horrigan, Wellman, & Rainie, 2006; Cole, Suman, Schramm, Bel, & Aquino, 2000; Katz & Aspden, 1997; Katz & Rice, 2002; Kraut et al., 2002; Quan-Haase, Wellman, Witte, & Hampton, 2002). A similar case has developed over the same time period for the uses and effects of video games. A large body of research has focused on testing the possible negative impacts of playing video games on players’ physical and mental health (e.g., aggression; Anderson, 2004), mainly with children and adolescents (e.g., Griffiths, 1997). Stereotypes that gamers are isolated, socially awkward young males have been

6 perpetuated by mass media (Williams, 2003). It was not until recently that scholars started to question these stereotypes by looking closely at the demographic and health profiles of gamers. A handful of studies have shown that, contrary to stereotypes, the majority of gamers are adults, male, white, and middle-class (Griffiths et al., 2004; Williams et al., 2008). The research on Internet and video games has stimulated an ongoing debate in academic as well as popular discourse and much of the controversy has been passed on to MMOGs, a unique genre that combines elements of both. Most of the game activities offered in MMOGs are already available in standalone video games. Therefore, what attracts millions of players to the online gaming worlds is the fact that they are playing with, or against, other fellow players rather than artificial characters controlled by the computer. The social factor, therefore, is arguably what makes MMOGs successful. Some scholars have suggested that MMOGs are examples of virtual "third places" where individuals can go online to socialize and maintain a social network as they share experiences in crafting weapons and slaying dragons (Ducheneaut, Moore, & Nickell, 2007; Steinkuehler & Williams, 2006). In other words, “gamers don’t bowl alone.” (Williams, 2006b) But beyond rhetoric, how exactly do MMOGs function as social spaces? Empirical data are scarce and far from conclusive. Qualitative studies have depicted vivid anecdotes about the many forms of sociability that are possible in these worlds (e.g., Nardi, 2010; Taylor, 2006). Several survey studies also reveal that the “social factor” contributes to enjoyment and many players are motivated to enter these spaces for social

7 reasons (Griffiths et al., 2004; Yee, 2006c). Yet, drawing on behavioral data from server logs, an examination of the social interactions within World of Warcraft, one of the most popular MMOGs on the market, has revealed that a significant collection of gamers play alone and choose not to interact with others (Ducheneaut et al., 2006). Clearly, more systematic and generalizable research is needed on the patterns and dynamics of interactions in MMOGs in order to offer a reliable picture of how the “social factor” really functions. This dissertation represents such an effort to critically examine the social dimension of these online game worlds. It is situated in a popular MMOG, EverQuest II (EQII). With the cooperation of the company that owns the game, a combination of data sources were made available to this project, including unobtrusively collected behavioral server logs for 9 months in 2006 and a comprehensive survey conducted to the players directly through the game engine. This unique dataset provides an unprecedented opportunity in new media and game research, one that allows for a thorough and representative analysis of behaviors, attitudes and demographic attributes, and an exploration into how they are related to each other and change over time. As this study focuses on social relationships among players, a network approach provides the appropriate analytical and theoretical framework. Specifically, this study addresses three broad research questions: 1) What are the basic patterns of player social networks in MMOGs, and how are they related to player attributes and game mechanics? 2) What impact do these networks have on social and task outcomes? 3) How do player social networks evolve over time? This research attempts to address these questions on

8 network patterns, effects and evolution based on the theories of social capital and network evolution. The first component of this study systematically evaluates the social experiences of players. As documented in previous literature, MMOG communities in general and EQII in particular accommodate many possible ways in which social interactions may arise in these virtual worlds, but such descriptions do not provide any empirical basis with regard to the actual level of sociability. To what extent is EQII a social space and what are the factors associated with informal sociability? This exploratory overview aims to establish a general account of the social life in EQII. It contributes to existing research by providing a much needed benchmark to anchor the discussion of the “social factor” and to facilitate comparison across different virtual worlds (cf. Ducheneaut et al., 2006). It also sets the stage for further analyses of the effects and evolution of player social networks. Based on the exploratory overview of the social experiences in EQII, the second part of the study focuses on the effects resulting from social networks established within online gaming communities. Grounded in the theory of social capital, this study offers a critique of the existing conceptualizations of online social capital that most empirical studies tend to consider and measure social capital as an outcome. Instead, this study proposes a structural approach to conceptualize and measure online social capital generated in player social networks, following the work of Granovetter (1973), Coleman (1988) and Burt (2005). The two distinct types of social capital, bridging and bonding (Putnam, 2000; Williams, 2006a), correspond to two distinct structures, brokerage and

9 closure (Burt, 2005). Therefore, measuring the structural properties, namely the degree of network brokerage and closure, offers valuable insights about the social consequences generated within networks. This dissertation represents a validity test of the theory of brokerage and closure in another variety of online environment: MMOG communities. It also contributes to the online social capital literature by using a network-based approach rather than an outcome-based approach. The third part of the study examines the longitudinal evolution of social networks formed in MMOGs. Are online relationships merely random, short-lived encounters or lasting and substantive connections? What makes some relationships more durable than others? These questions are important because they highlight the dynamic processes of relationship formation, maintenance, and demise in online worlds, an issue rarely examined in the extant literature despite repeated calls for more research employing longitudinal analyses (Ellison, Steinfield, & Lampe, 2007; Harris, Bailenson, Nielsen, & Yee, 2009; Lewis, Kaufman, Gonzalez, Wimmer, & Christakis, 2008; Williams, 2007). Further, an understanding of the stability, or rather, fragility, of online relationships provides a crucial context to qualify the significance and generalizability of findings obtained in empirical studies relying on cross-sectional data. Evolutionary theories, as represented by the work on socio-cultural revolution (Campbell, 1965), organizational ecology (Carroll & Hannan, 2000), and the recent extension to network evolution (Monge, Heiss, & Margolin, 2008), are particularly well-suited to study change. An evolutionary framework is thus applied to personal networks among players of EQII to trace the variation, selection and retention of social ties in virtual worlds. Hypotheses about tie

10 persistence and decay are derived based on three sets of factors: aging and inertia in networks, social architecture imposed by game design, and the natural tendency to connect based on homophily and proximity. This dissertation is organized as follows. Chapter 2 introduces the conceptual framework of the study. It first provides a description of MMOGs in the context of online communities, focusing on several key features of MMOGs that are shared with most online communities. A detailed depiction of EQII is then offered, including game content and mechanics, character creation, social interaction and player associations. This rich description provides a necessary context to understand the social architecture and the relationship dynamics relevant to the study of the patterns, effects and evolution of player social networks. The next section reviews prior theorizing and empirical work on the social dynamics of MMOGs and proposes a preliminary analysis of the social experiences. The following section reviews the theory of social capital and the extant literature on online social capital, proposing a structural perspective to connect player social networks in EQII with their task performance and relational outcomes. The next section builds on the evolutionary framework and its recent extension to network evolution. It discusses the many forces at play in the initiation, selection, and retention of social ties. Chapter 3 discusses the sources of data, the collection and transformation processes, and the methods for data analysis. Chapter 4 presents the results on network patterns, effects and evolution. Chapter 5 discusses the implications of the results, and suggests limitations and directions for future research.

11 Chapter 2: Conceptual Framework Understanding MMOGs MMOGs as Online Communities Although MMOGs may appear as a novel form of online social interaction, their characteristics could be better understood in the context of online communities, of which MMOGs are a recent breed. Broadly defined, online communities refer to groups of people “who come together for a particular purpose, and who are guided by policies (including norms and rules) and supported by software” (De Souza & Preece, 2004; Preece & Maloney-Krichmar, 2005). Different generations of technologies have been used to create these communities, including the early Usenet, Bulletin Board System, and web-based forums, as well as more recent social networking technology, blogging, and wiki, among others. Such communities are playing an increasingly important part in various aspects of human life, including collaborative production (e.g., open source software communities; Bagozzi & Dholakia, 2006), information sharing (e.g., tripadvisor.com or discretionary databases in organizations; Kalman, Monge, Fulk, & Heino, 2002; Wenger, 1998), political deliberation (e.g., political blogs; Adamic & Glance, 2005) and social support (e.g., online cancer support groups; Klemm et al., 2003). These communities comprise a considerable portion of Internet use. For example, as shown in a national telephone survey conducted in 2008, one third (35%) of all American adult Internet users have a profile on at least one social networking site, and the percentage is 75% for adults between 18 and 24 and 57% of adults between 25 and 34 (Lenhart, 2009). A 2009

12 national report shows that 15% of Internet users in the United States are members of at least one exclusively-online group, specifically defined as “a group that shares thoughts or ideas, or works on common projects, through electronic communication only” (Center for the Digital Future, 2009). This broad definition accommodates a great variety of online worlds with regard to their technological affordances, domain interests, and rules of interaction. Yet, because of their reliance on networked communication technology, interactions in online communities still share some critical features, which have been systematically examined and summarized in the literature of Computer-Mediated Communication (CMC) and online sociability. Bandwidth and Social Information Processing Early CMC research is characterized by a focus on the bandwidth of communication media, comparing CMC against the benchmark of face-to-face communication. The main components of this argument include social presence theory (Culnan & Markus, 1987), media richness theory (Daft & Lengel, 1986), and the lack-ofsocial-context-cues hypothesis (Kiesler, Siegel, & McGuire, 1984). Social presence, defined as a quality of the communications medium itself, is the feeling that other actors are jointly involved in communication interaction (Short, Williams, & Christie, 1976). CMC, with fewer nonverbal elements, is extremely low in social presence, which is argued to account for task orientation and impersonality of CMC (Culnan & Markus, 1987). Similarly, media richness theory suggests that media differ in their bandwidth, or their capacity to deliver different cues (Daft & Lengel, 1986).

13 Face-to-face is touted as the richest medium, while CMC is a very lean channel, because nonverbal cues are absent. As communicators would match the richness of each available medium to the ambiguity of the intended message, CMC is considered as a suitable medium for communicating simple and unequivocal messages. Some researchers also argued that CMC suffers from a lack of social context information (Kiesler et al., 1984). In plain text through electronic channels, people’s social status, power and prestige are not conveyed contextually through things like physical surroundings and clothes, nor are they communicated dynamically through gaze, touch, or facial expressions. Communication is predicted to be more impersonal because the rapid exchange of text, the lack of social feedback, and the paucity of social context cues redirect attention toward the message itself (Kiesler et al., 1984). Social presence theory, media richness theory and the lack of social context cues approach can all be summarized as “cues-filtered-out” approach (Culnan & Markus, 1987). They all suggest that CMC is inherently a medium of limited bandwidth, and is good for giving and receiving information, opinions, and suggestions. CMC is less suited for communicating ambiguous messages or social-emotional tasks involving conflict and negotiation. There is also more equality of participation in CMC than in face-to-face group interaction (Hiltz & Wellman, 1997). According to Sproull & Kiesler (1991), by reducing cues on hierarchical dominance and power information, social influence among communicators might become more equal and democratized. However, field research has contested the above claims by showing evidence for substantial emotional support and relation formation online in both socially close and

14 distant groups. Specifically, Walther (1992, 1996) argues that text-based CMC differs from face-to-face interaction only on the rate of information transfer. It just takes CMC longer than face-to-face to accomplish the same level of social interaction. Given sufficient time, computer mediation should have very limited effect on relational communication, as social information accumulates and gets processed. This is because participants could develop and adapt specific interaction strategies in CMC settings to suit their relational needs. Relationships developed and maintained online are much like the relationships formed offline. Moreover, ties initially impersonal and instrumental can broaden out to be socially supportive, which is why participants often become increasingly attached to online communities (Hiltz & Wellman, 1997). Participants are able to exploit and adapt the means of communication to achieve their relational goals, mitigating the presumably “lean” nature of CMC. This more functionally-oriented perspective, known as Social Information Processing (SIP) theory, suggests that the “limited bandwidth” of online communication is not a determining factor of the formation and quality of social relationships, even less so as the technology itself advances, dramatically increasing the bandwidth of CMC and boosting our capacity to transmit socioemotional information (Herring, 2004; Walther, 2006). In particular, most MMOGs on the market offer a variety of media channels, such as text-based chat, voice chat, as well as special commands called “emotes” to choreograph a series of dance moves or to send virtual hugs (Nardi, 2010). Combined, these features enable a much richer communication experience in MMOGs than the early text-based CMC environment, and studies have found that MMOGs can be fertile ground for the creation

15 and development of meaningful social relationships (Nardi, 2010; Taylor, 2006; Yee, 2009). Relative Anonymity The CMC literature on bandwidth and social information processing resonates with several notable qualities of online communities that impinge on social relationships built within them. The first quality is relative anonymity, as compared to offline and faceto-face interactions (McKenna & Bargh, 2000). Marx (1999) has proposed seven levels of identity knowledge: (1) legal name, (2) locatability, (3) pseudonyms that can be linked to legal name and/or location, (4) pseudonyms that cannot be linked to name or location, (5) pattern knowledge—a person’s distinctive appearance or behavior patterns, (6) social categorization, such as gender, age, and sexual orientation, and (7) symbols of eligibility/noneligibility information, such as the possession of knowledge, artifacts or skills. Online communities may vary widely with regard to the level of anonymity they tolerate. For example, some communities require or encourage participants to reveal their true name, location, gender, and other identity information, especially for individuals in social networking sites such as Facebook and LinkedIn. Other online worlds offer more anonymity, typically revealing identity knowledge for the fourth element, i.e., pseudonyms that cannot be linked to name or location. Legal name and location are both invisible, so are social categorization such as gender, age, and sexual orientation. Most MMOGs fall into this category and a pseudonym, or “screen name,” is the only piece of identity information required for participation. Researchers have noted that the relative

16 anonymity afforded in such online communities may reduce the risks inherent in selfdisclosure, thus facilitating the development of intimate relationships in a non-threatening environment (Bargh & McKenna, 2004). It should be noted, however, that MMOG participants may voluntarily reveal other identity information in addition to their pseudonym, such as their gender, location, and occupation, usually after repeated interactions and sometimes in offline gatherings (Skoric, Tang, Liao, & Poor, 2010; Taylor, 2006). Also, some players simply use MMOGs as an additional venue where they can interact with pre-existing social ties, thus online pseudonyms are mere indications of other layers of identity. From Neighborhoods to Networks A related feature of online communities is that they have transformed the notion of “communities” from densely-knit neighborhood groups to social networks of kin, colleagues and friends that may span the globe (Wellman & Gulia, 1999). Traditionally, without modern transportation and communication technologies, our social life was very much organized around locality. Communities are formed among people residing in the same village or neighborhood where everybody knows everybody else. Therefore, the term “community” provides an idyllic image, connoting feelings of solidarity, empathy, and intimacy among neighbors (Katz, Rice, Acord, Dasgupta, & David, 2004; Preece & Maloney-Krichmar, 2005; Tönnies, 1957). The dictating role of location has gradually faded away as globalization and the advances of transportation and communication technologies, such as railroads, cars, planes, the telephone, the Internet and mobile phones, have facilitated the formation and

17 maintenance of social connections among spatially dispersed individuals (Mok & Wellman, 2007), making McLuhan’s (1964) notion of a “global village” a reality. Just like Licklider and Taylor (1968) said several decades ago, “…life will be happier for the on-line individual because the people with whom one interacts most strongly will be selected more by commonality of interests and goals than by accidents of proximity.” As a result, communities online are formed not around neighborhoods any more, but around social networks comprising kin, friends, and more importantly, people of similar interests (Wellman & Gulia, 1999). Just as human interactions stretch beyond the confinement of locality, the conceptualization of “community” also moves away from the idealized notion of solidary and place-oriented groups to include broad and diverse computer supported social networks (Wellman, 2001; Wellman & Gulia, 1999). Entry and Exit Costs With a broader and more diverse conceptualization of community discussed above, a related quality of community life—entry and exit costs—also undergoes dramatic transformation (Galston, 2000). In an archetypical place-based community, membership is rather limited and the barriers to leaving old groups and joining new ones are high. Being in the same place, or “accidents of proximity” (Licklider & Taylor, 1968), is the prerequisite to community membership. Exiting the old community and entering a new one thus requires significant cost of relocation. When dissatisfaction arises, individuals typically have two choices: to leave the community, or to voice their grievances (Hirschman, 1970). Galston (2000) argues that such a choice depends on the barriers to exiting old groups and joining alternatives. Exit would be the preferred option

18 if costs are low; as costs rise, individuals are more likely to stay in the existing group and voice their discontent. Therefore, people in archetypical place-based communities are more likely to stay and try to improve the existing institution despite their dissatisfaction, which may in turn strengthen mutual bonds among members. By contrast, online groups have much lower barriers to entry and exit (Galston, 2000). As these groups attract members based on commonality of interests instead of shared location, they operate under the same principle of voluntary associations. Technologically, these online spaces tend to exert weak control over the admission and maintenance of membership so border-crossing behaviors are cheap. When discontent arises, individuals tend to leave an old community and join new ones that fit their needs better (Galston, 2000; Hirschman, 1970). Not surprisingly, online groups often experience rapid influx of newer members, as well as a high attrition rate of existing members. Their relatively high level of “metabolism,” resulting from lower entry and exit costs, would contribute to our understanding of the nature of social relationships built within them. In summary, as a new variety of online communities that has a history over two decades, several features and characteristics noted by scholars of CMC and online sociability are also present in MMOG environments. Even with limited bandwidth, participants in online communities can still build meaningful social relationships via CMC. These online spaces provide a platform where social interactions can occur in the absence of rich identity information such as legal name and location, where individuals are linked by choice rather than accident of proximity, and where the notion of

19 “community” expands beyond the densely-knit group of neighbors to include diverse and globe-spanning computer supported social networks. Characteristics of EQII The above literature review provides a theoretical starting point to understand MMOGs as a special genre of online communities. Yet, online worlds are so heterogeneous that these general features mentioned above can have very different effects on relationship formation and development in different contexts. Therefore, the discussion of sociability in a gaming community has to be grounded in its specifics— the opportunities and constraints of the virtual environment, the purposes of individual participants and groups, and the types and rules of interactions. Lessig’s “code is law” observation nicely summarizes the importance of the social architecture of online systems: We will see that cyberspace does not guarantee its own freedom but instead carries an extraordinary potential for control… Architecture is a kind of law: It determines what people can and cannot do… As the world is now, code writers are increasingly lawmakers. They determine what the defaults of the Internet will be; whether privacy will be protected; the degree to which anonymity will be allowed; the extent to which access will be guaranteed. They are the ones who set its nature. Their decisions, now made in the interstices of how the Net is coded, define what the Net is. (Lessig, 2006, pp. 77-79) In the same way, code in the world of EQII determines what players can and cannot do. As a participant observer of a popular MMOG, EverQuest II (EQII), I present in the following a description of its content and the characteristics that are relevant to the study of the patterns, effects and evolution of player social networks.

20 Game Content and Mechanics EQII launched in November of 2004 as a much updated sequel to EverQuest, an early online game that many considered to have defined the genre of MMOGs as we know it (Bartle, 2003; Lastowka, 2009). Although no longer the North American market leader, the EverQuest franchise presents game rules and goals that are nearly identical to World of Warcraft, one of the most popular MMOGs, and several other fantasy titles on the market. Altogether, these games represent 85% of the total MMOGs currently being played in the west (Woodcock, 2008). Players typically purchase a copy of the game in the store, install the game software on a computer, and then connect to the game servers over the Internet. The game operates on a subscription-based model, with every player (account) paying a monthly fee ($14.99 as of May 2010) to access the game. In EQII, the conflict between two factions – good and evil – provides the underlying storyline. Each faction has its own home city for players to enter the game, and provides a wide range of races and classes for players to customize their characters. Upon entering the game, players are prompted to create their virtual representation, known as “avatar” (also referred to as “toon”) in the game, to name the character, and to choose their gender, race, character class, and facial features (see Figure 1).

21 Figure 1. Character Creation and Customization

22 Once the characters are created and personalized, players can begin their journey in EQII at a tutorial island, which is designed to familiarize players with the core concepts of the game. The island introduces players to quests, combat, crafting and grouping before players set foot on the vast and dangerous world of EQII. Players also take this opportunity to master the controls and physical movements. EQII is designed with realistic physics principles. Players can sit, crouch, walk, sprint, jump, and swim through the environment, suffering harm when they hold their breath under water for too long or fall hard from a cliff. Players can switch between two perspectives: the first person perspective and the third person perspective. Just like other MMOGs, EQII operates on numerous servers, which are parallel versions of a persistent virtual environment. EQII operates on four types of servers: Player-versus-Environment (PvE), Player-versus-Player (PvP), Role-Playing (RP) and Exchange (EXC). All the servers provide the same general features such as questing, exploring the in-game world and socializing with other players. The PvE servers are the most paradigmatic of EQII as well as the MMOG genre, while other server types include additional features. For example, in addition to fighting Non-Player Characters (NPCs), players can also directly confront other players in combat on PvP servers. On RP servers, players may act and talk like the fictitious characters in the game world and avoid references to the offline world (Williams, Kennedy, & Moore, 2010). EXE servers allow players to conduct real-money transactions (RMT), e.g., buying virtual game currency (instead of earning them through game play) using U.S. Dollars, a feature usually considered illegitimate for most MMOGs (Castronova, 2005). For all activities in EQII

23 except chat, players are only allowed to interact with those who are on the same server, thus making the server a self-contained unit to observe player interactions. Transferring characters from one server to another is cumbersome and requires a charge ($25 as of May 2010), with many strings attached (e.g., transfers between PvP and PvE servers are not allowed). Therefore, servers can be considered as stand-alone parallel virtual worlds with a stable population of characters, providing ample opportunities for replication. Currently EQII is operating on 25 servers, 19 of which are for players located in North America. In this study, all the research questions and hypotheses were addressed using data from the Guk server, as it represents the most standard Player-versus-Environment (PvE) social architecture in virtual worlds. The Guk server is designed to accommodate players located in North America. Combat and questing comprise the main activities of the game. On PvE servers, combat is between players and NPCs such as gnomes and zombies, generally known as “monsters” or “mobs.” PvP servers allow players to directly confront and kill other players in combat, in addition to NPCs. Combat in EQII is fairly typical of the MMOG genre. Players use weapons such as swords and bows, and can cast various spells to deal damage to the target. After successfully killing a monster, players are awarded experience points (also known as XP), which are required for level advancement, and in most cases, in-game currency or other valuable items (e.g., weapons and armors) dropped by the monster, known as “loot.” Questing is the activity to complete game-designed tasks, such as talking to special NPCs and collecting in-game items. As the name “EverQuest” reveals, EQII has a special focus on quests, and completing quests is also rewarded in the

24 form of experience points, in-game currency and/or valuable items. While XP gain is a key motivator for killing mobs and completing quests, especially for players who want to level up, obtaining the precious “loot” so as to equip one’s avatar with superior gear becomes an increasingly important objective for more advanced players. As the coveted “loot” often comes with particularly challenging mobs and quests, players tend to collaborate in groups and raids. The distribution of loot, as a result, can be a subject of contention in these collaborations. The game progresses as players earn experience points and reach higher levels (the maximum level is 70 for the study period), which improves players’ existing skills, powers, and abilities and opens up new ones. Character level provides a reliable assessment of the possibility of defeating the target. For example, a level-20 player can confidently defeat a level-19 monster, but such a battle would almost always lead to player death if the monster is at level 22. In addition, monster encounters are classified into three categories: Normal, Heroic and Epic. The Normal monsters can be defeated by a single player of the same or higher level, but Heroic and Epic monsters are much more difficult and better fought by a group of players. The level and rarity of loot dropped by these monsters also correspond to their difficulty. Therefore, the more difficult the task is, the better the reward the player receives in terms of XP and loot. The vital statistics of the target are available to the player so that the order and strategies of combat can be planned accordingly. In EQII, player death results in no experience loss or lost levels. The liability of death is temporary rather than permanent. Upon death, characters would either be

25 resurrected by a friend on the spot or revive at specific revival locations. In the latter case, the character revives from a “respawn” point in the same zone. Both methods of revival would result in a minor experience “debt” that has to be repaid out of experience points gained later. The character’s equipment such as armors and weapons remains functional after death, but each death would accrue a 10% damage to the equipment, until it is fully worn out after ten consecutive deaths and has to be repaired for a fee. In summary, death in EQII is a moderate inconvenience rather than a serious failure, but death disrupts the flow of combat activities in that players have to do a fairly long “corpse run,” repay the experience debt, and sometimes mend the equipment before they can resume combat in full competence. Players typically experience numerous deaths in their game play. Deaths are often unavoidable in players’ search for the best combat tactics and strategies to defeat the target. The massive world of EQII is divided into numerous “zones”. Originally, zones refer to different graphic sets which are required to load in order to display the game (Yee, 2001). When players leave a zone and enter another, there is a temporary pause as the game loads the graphic set. These virtual boundaries become natural divisions of game play. In EQII, each zone is different with regard to its visual appearance, environmental settings, and the monsters at large. Zones can be aligned according to their level of challenge to the players. For example, Antonica presents relatively low threats as most monsters meandering in the zone are between level 10 to 20, while Everfrost is a higher level zone where only players above level 40 are able to enter the zone without risking imminent death.

26 Perhaps more so than other popular MMOGs, the game play experience in EQII is highly structured by level—players usually follow the same general path of progression. As quests, monsters, and zones are all aligned by level, players would have to tackle them in a relatively linear order (for an example, see the zone by level diagram at: http://eq2.wikia.com/wiki/Zones_By_Level). Skipping the linear order to interact with higher level objects (e.g., a level-28 player attacking a level-30 enemy, or a level 28 character travelling in a 30-plus zone) is not encouraged and typically results in player death. By contrast, a higher level player could safely interact with lower level objects (a level 28 player attacking a level 20 enemy) but such encounters offer very little reward. For example, a level 28 character could very easily kill a level 20 monster, but winning the battle would bring almost nothing in return. In other words, there is very limited flexibility for players to create their game experiences that disrupt the “line” of quests, monsters and zones. By way of game design, players are motivated to only interact with objects—including fellow players, quests, zones and mobs—that are closely aligned with their own level in order to play the game efficiently. This highly structured game play in EQII is in stark contrast with “sandbox” virtual worlds such as Second Life, where there is no developer-created goals or story line and participants are free to pursue their own goals and activities (Williams, in press). Even among structured MMOG worlds, the game mechanics in EQII are quite rigid; players have to follow the “line” closely and enjoy relatively less flexibility than in other MMOGs.

27 Character Creation In EQII, each player needs to choose the alignment (good, evil or neutral), race and class of each character. As of May 2010, EQII provides 19 fantasy races for players to choose from, and these races are categorized according to their alignment. For example, High Elf is a good race, but Dark Elf belongs to the evil side. The alignment also restricts the character class (discussed below) one could choose from. Each race is endowed with a set of starting attributes, such as intelligence and strength. But race has little bearing on what the character is capable of, as these starting attributes only have very limited importance at the beginning and lose meaning when the character develops. Probably the most important lasting effect of race is the appearance of the character. High elves, for example, are tall and slender while Dwarves are stout and sturdy. Unlike race, character class selection represents an important decision that has far-reaching implications on players’ entire game play experience. Character class defines the abilities as well as weaknesses of an avatar. EQII provides 24 character classes altogether for players to choose from, 8 of which are available to players of the good alignment, 8 for the evil alignment, and the rest of which are available to all (neutral). These classes could be categorized into four archetypes—Fighter, Scout, Priest and Mage. Each archetype contains two good classes, two evil classes, and two neutral classes (see Table 1).

28 Table 1. Character Classes of EQII Archetype Fighter Priest Mage Scout

Neutral Berserker, Guardian Fury, Warden Warlock, Wizard Dirge, Troubador

Alignment Good only Monk, Paladin Mystic, Templar Conjurer, Illusionist Ranger, Swashbuckler

Evil only Bruiser, Shadowknight Defiler, Inquisitor Coercer, Necromancer Assassin, Brigand

Classes belonging to the same archetype tend to share most of the same abilities and assume the same roles in combat. In general, fighters have excellent defense as well as good damage output, thus they can serve as the main "tanks" which absorb damage from enemies. Priests’ are excellent healers who specialize in restoring health for group members, but they typically have mediocre damage output. Mages and Scouts each have distinct capabilities in casting damage using various spells. Scouts can deal damage in close proximity. Mages can deal damage from a distance, but they are typically more fragile than Scouts. Many nuanced differences exist among classes of the same archetype. For example, Conjuror and Necromancer have the unique ability to summon undead mob as their “pet” to assist in combat, thus making them more likely to survive in challenging encounters compared to the rest of the Mage classes. In addition, EQII provides some flexibility in character development through the Achievement system. Along with XP gain, every character could earn Achievement Experience (known as “AA” points) during their combat and questing. These AA points could be invested in several alternate lines of abilities, thus offering a way for players to customize their character. As a result, two players with exactly the same character class (e.g., both are Wizards) at exactly the same level could represent somewhat different combinations of skills, owing to their different investment of AA points.

29 Still, even with some limited customizability through the AA system, character class dictates most of the potential abilities and skills one might acquire during the course of game play. Class is one of the two key attributes (along with character level) a player would typically use to identify others and be identified (e.g., a level 50 Paladin). As such, selecting a character class is a nontrivial decision early on in the game. It is not surprising that there are heated discussions on the “best” class within an archetype and comprehensive articles researching the pros and cons of certain classes in many EQII web forums (e.g., the class discussion forums hosted on the official EQII website: http://forums.station.sony.com/eq2/forums/list.m?category_id=41, and the combat discussion on the web forum EQ2flames: http://www.eq2flames.com/combat-discussion/) The question ultimately comes down to one’s style of game play: Classes represent different combinations of offense and defense skills, which result in different roles assumed in collective activities. Through this intricate division of labor among character classes, EQII encourages collaboration so that players can combine various skill sets to accomplish quests and fights that are too difficult to be tackled alone. Social Interactions By their very nature, MMOGs are social ventures that engage multiple players in real time in a shared virtual environment. Since the basic features of character development, combat and questing are already available in stand-alone single-player video games, what attracts tens of thousands of players to EQII is the shared experience and limitless opportunities of social contact (Ducheneaut et al., 2006). Interactions in EQII can occur in many ways. On a PvE server, EQII players could engage other players

30 through three kinds of activities. The first is player-to-player trade. In EQII players can gather and produce in-game items and trade for other desirable in-game items or exchange them for in-game currency. They can either make trades through a computercontrolled broker, which does not involve human interaction, or trade directly with another player via face-to-face item exchange, which requires the two parties to know each other’s name and meet “face-to-face” in the game to finish the transaction. The second activity is chat. Players can interact with each other by typing chat messages on the game interface. Many chat channels with varying degrees of privacy are available, including direct messages to a specific player (also known as “tell”), to a group or raid one currently belongs to, to one’s guild, to players in the same zone, to players in the same level range (e.g., level 10-20), and so on. The third and the most emblematic type of interaction is collective play in groups and raids. Two mechanisms are used to incentivize collaboration in EQII. First, as discussed in the previous section, character classes have distinct abilities that are designed to complement each other. A typical combat group normally consists of at least one tank, at least one healer, and a few others who could deal damage within close proximity (melee damage) and from afar (ranged damage). A “tank,” usually played by a Fighter class, refers to a character who has good health and can withstand severe damage from the enemy, so a tank assumes the responsibility of irritating the mobs and focusing their aggression on the tank alone. A “healer,” usually played by a Priest class, refers to a character who could restore health of group members during combat. Melee damage, usually done by a Scout class, and ranged damage, usually done by a Mage class, are

31 important offense forces to defeat the enemy. It is essential, therefore, to recruit players of diverse classes to form well-balanced combat groups. Second, for many high-level mobs and quests, the only way to tackle them is through grouping and raiding. Even though less challenging encounters could be successfully accomplished by solo players, collaboration could significantly expedite the combat process and reduce the level of repetitive labor (e.g., killing many low level monsters before gaining access to a rare mob), a non-entertaining but mandatory game element known as “the grind” or “treadmill” (Taylor, 2006; Yee, 2006b). One prerequisite of grouping is that the players who form a group have to be at similar character levels. If this condition is not met, then higher-level players need to mentor lower-level players, otherwise they absorb most of the experience gains and leave lower-level players with little reward. The experienced character pays for this by earning a discounted number of points from play. One can only mentor one other character at a time, but a character could have multiple mentors. Collaborative play can be organized in many forms. EQII offers built-in functionalities for players to form groups and raids of varying number of players. Pairs represent the smallest configuration of groups. A pair is a simple team comprising the self and another player, usually of complimentary classes (e.g., a Fighter class and a Priest class). Pickup groups, or “pugs,” are ad-hoc groups consisting of players who typically do not know each other very well, if at all. They congregate in the game by advertising in the public chat channels or using the built-in “Looking For Group (LFG)” or “Looking For More (LFM)” tools to search for appropriate groups to join or the ideal groupmates to recruit. By contrast, friend groups consist of players who know each other

32 fairly well, either through existing offline connections or through repeated interactions in the game. Similarly, guild and ally groups are subsets of players drawn from the same guild (explained below) or ally guilds. Even though players may not know each other personally or have prior collaboration history, they share the same guild or ally affiliation, which may increase the level of trust and accountability. A simple example of a 6-person group could illustrate how collaborative play operates in EQII through three stages. The first stage is group formation. On a weekday evening at 10pm Eastern, Quizno, a level 22 Wizard (a Mage class), wanted to find fellow players to run the Stormhold dungeon together. She first used the “Looking for Group (LFG)” function to search for suitable groups and also tagged herself as “LFG” in order to signal her availability for group play. Because there were no level-appropriate groups to join at that time, she also advertised in the public chat channels. Soon, two players saw her “LFG” tag or heard the public advertisement and sent her private chat messages (or “tells”) to inquire about the mission. She explained to them the zone and the appropriate level so that they could make an informed decision. The group started to form up as these two (a Scout and a Priest) joined the quest. Since all three of them were scattered in different places in the enormous universe of EQII, they started travelling to the Stormhold dungeon, which might take up to 15 minutes or even longer depending on where they were, whether teleportation devices were available and what monsters they might have to undertake along the way. In the meantime, as this nascent group did not have any tank, the three group members decided to recruit more groupmates through public search, friend referral, and

33 guild chat. As an existing group, they started the search using the “Looking For More (LFM)” function (as shown in Figure 2), specifically asking for tanks. Although there were several players available, they were either not at comparable levels or could not perform the responsibility desired. Nonetheless, several of them were contacted about their interest in joining the group. Quizno also asked in her guild chat channel whether anyone was interested in running the Stormhold zone together with the existing group. Two guildmates offered their help and one of them also brought along his sister. At this point the group was officially full, with one Fighter (the primary tank), one Priest (the primary healer), and the rest being Scouts and Mages. Group members greeted each other through the group chat channel and talked about their concerns and experiences with this specific dungeon. By the time every group member traversed vast virtual plains, oceans and mountains to finally arrive at the entrance of the zone, ready to fight, 50 minutes had already passed since the inception of the group.

34 Figure 2. “Looking for More (LFM)” Interface2

2

All the character names are masked to protect (virtual) privacy.

35 The second stage is the actual combat. In this case, since most of the group members had not played with each other before, it was crucial to negotiate the common rules and expectations in order to prevent misunderstandings and conflicts. Some basic division of labor and combat tactics were decided at the beginning. For example, the main tank and the secondary tank were identified before combat. Also, the group chose “Need Before Greed” as their reward distribution scheme, which stipulates that the loot dropped from mobs is first assigned to those who need it (i.e., who select “Need” in the loot window) before giving it to those who desire it (i.e., who select “Greed” in the loot window). Other strategies and directions were negotiated on the fly through conversations in the group chat channel, or if available to all, a voice chat software3. For example, upon entering an alcove, a group member who had run the same dungeon before would advise people not to touch the flaming sword because “bad things would happen.” And, if a monster appeared to be insurmountable, the group leader (in this case, the tank) would urge everyone to take off rather than lingering for a hopeless fight. Encounters with mobs usually followed a general sequence: At first, everyone gathered in a relatively safe place adjacent to the mob area, ready to perform their respective duties. After a 10-second count down, the tank ventured into the mob area, attempting to aggravate the mob and “pull” it to where the group had decided to battle. During the course of the fight, the tank was responsible for holding the mob’s hate or aggression on himself alone, while everyone else should try not to irritate the mob, so that the whole group (except the tank) were protected from massive damage. The healer

3

At the study period (January to September 2006) EQII did not provide an in-game voice chat system, but some players used third-party voice chat software.

36 kept people alive by restoring health for the tank as well as other group members, and the rest of group were essential in dealing damage to kill the mob. As such, everybody’s responsibilities were tightly coupled and one weak link, such as the accidental death of the healer, would potentially lead to group failure (known as “wipe” or “wipeout”). Sometimes wipeouts were unavoidable, and each time a lesson was learned to improve the current combat strategy. After a successful battle, valuable items (the “loot”) were dropped from the dead mob and were distributed within the group following the predetermined reward distribution scheme. The third and the last stage is group dissolution. In this example, dissolution happened after about three hours of game play, at around 3am Eastern time, when one group member located in the Eastern time zone said that it was time for bed. Although the Stormhold dungeon had not been completely conquered yet, the rest of the group agreed that they had a good fight and thanked each other for their effort. Some stated their play schedule for the next day and expressed willingness to group again in the future. Besides these inevitable real-life engagements, such as sleep, work, and school, another natural termination of group play is the successful completion of the group’s original mission, such as killing the named boss (the most difficult monster in a zone). Given the considerable overhead investment in forming a group, deciding where to hunt, travelling to the spot, and establishing shared rules and expectations, it is not surprising that most members of a group would wish to stay grouped throughout their game session. For example, sometimes a group member has to be away from keyboard (known as “AFK”) to take a bathroom break or to make a phone call, it is not uncommon

37 for the entire group to hold off all actions and wait patiently till that person returns. By the same token, grouped individuals are also expected to contribute to the collective for as long as they can. Breaking from the group unilaterally after a bad fight is considered selfish and inconsiderate, because the group is forced to either painstakingly find a replacement or continue their battle nonetheless with a less competitive team. As the player Ethec (2004) posited, grouping is an important component of one’s EQII etiquette and also a precursor of community building : “Disbanding in frustration after a wipe (a wipe occurs when the entire group is killed) may feel good short term, but you won't build much trust or affection on the part of your groupmates… On the flip-side, finding a suitable replacement for yourself when you leave a long-duration group is the height of class. ” Groups formed in EQII may contain a minimum of 2 and a maximum of 6 members. Yet, some high-level dungeons and mobs are so challenging that they could easily vanquish any well-balanced 6-person group by a single strike. This is where complex, multigroup collaborative configurations, known as “raids,” become necessary. Raiding is a regular activity for players at high levels. It is also the sole activity for many people who have reached their maximum level (level 70 for the study period). These players have nothing else left to accomplish but “the end game,” which refers to the extremely challenging (as well as rewarding) game content designed exclusively to keep these maximum-level gamers engaged. EQII allows raids to form with two, three or four groups (known as X2, X3, and X4 raids). Figure 3 shows a screenshot taken during one X3 raid (18 people) in the Protector’s Realm. At the upper right corner is a raid window

38 displaying all 18 participants (organized in three groups) and their respective roles. The lower left corner is a raid chat window (in purple) where all the raid members could communicate in text-based chat. Figure 3. Raiding in the Protector’s Realm

Like groups, some raids form around members of the same guild or ally guilds (discussed in the next section). Guilds with a large player base (e.g., over 100) are often able to generate enough participants at comparable character levels to tackle difficult zones and mobs. In order to ensure that people are logged into the game at the same time, many guilds post a regular raid schedule to coordinate. For example, the guild “Guardians of Honor” organizes raids on Wednesday and Sunday evenings at 7:30 Eastern Time, and all potential participants are required to check in at least 30 minutes before the official raid starts to allow sufficient time for organization and deployment. “Guardians of Honor” also

39 has an ally guild to raid together regularly to guarantee that the available player pool is large enough to fight difficult foes. The opposite case to guild raids and ally raids is pickup raids. Without much advanced planning, they form spontaneously through a combination of built-in player search functions, advertisements in chat channels, referrals or word-of-mouth. Pickup raids are quite uncommon because it is extremely difficult to assemble up to 24 players with complementary expertise and appropriate character levels. Most of the mechanisms mentioned in the preceding discussion of groups apply equally to raids, except that the coordination cost associated with raiding increases dramatically, mainly for two reasons. First, a raid may consist of up to 24 people who are at comparable levels and have complementary skills and abilities. The actual raiding battles could only take place after a series of organizational chores, which grow proportionally with group size: deciding where to raid, identifying a pool of eligible players, screening these players and making decisions about whom to include, resolving conflicts over inclusion and exclusion, organizing players into groups, waiting for all the players to travel to the raiding zone with fully-functional gear and consumables (such as potions and repair kits), selecting a loot distribution scheme, establishing rules and common understandings, assigning individual roles and responsibilities (e.g., who are the tanks), and so forth. For example, the guild “Despair” has established an elaborate list of raid rules to help streamline the preparation before raids, including specific instructions for every single class to create and carry a “raid bag” with essential items before every battle (Despair, 2010b). With the daunting coordination work, it is not uncommon that some raids may take hours to get organized. Even for pre-scheduled raids that have

40 players signed-up well ahead of time, most would require participants to check in at least 30 to 45 minutes before the raid starts to allow sufficient time for preparation. Second, the coordination cost is high because raiding battles are among the most difficult in EQII. The likely consequence of wipeout necessitates extremely sophisticated planning and deployment of combat strategies. For the raid group as a whole, a successful battle is usually preceded by lengthy discussions of tactics, extensive research online of the mobs, numerous failed attempts, as well as reflections and improvement as a result of these failures. For every participant, discipline and teamwork surpass individuality and personal preferences. They are expected to perform exactly as instructed, act in the best interests of the raid and align their actions with those of others. “Wandering off,” not doing the assigned job, using less-than-optimal spells, or even taking an unannounced bathroom break could easily lead a several-hour long team effort to futility. Perhaps because raids are so difficult to organize, the opportunity to participate in a raid, especially those with good leaders and well-balanced team members, is often the subject of fierce competition. For example, in some guild-based raids, many qualified players would have to contend for the same spot (e.g., the primary tank). The raid leader has to make a difficult decision as to whom to include, based on a multitude of factors such as experience, seniority, competence, team spirit, among others. Because of the high coordination cost and the high stakes involved in raiding battles, it is not unusual for leaders to select the people who they know personally or have raided with together before, not only because these people are capable of performing the assigned duties, but also because the shared history provides a basis for building a reliable, mutually-trusted and

41 cohesive team. Naturally, membership in these temporary raiding groups solidifies over time as people deepen their ties through repeated collaboration. Player Guilds While groups and raids are short-term social ventures, guilds represent a more stable type of formal player association with a hierarchical leadership structure to coordinate in-game actions and accomplish joint tasks in EQII (Ducheneaut, Yee, Nickell, & Moore, 2007; Taylor, 2006; Williams et al., 2006). Guilds can range in size from several players to a couple hundred or even more. Players can belong to only one guild but are allowed to quit one guild and join another. Like many other MMOGs on the market, EQII provides built-in functionalities to assist guild formation and management. All guild members carry a “tag” on their character name identifying their guild affiliation. Guilds have access to a private chat channel dedicated to facilitate communication among guild members, as well as “messages of the day” that all members would receive automatically upon logging in. Other features include a specific guild window showing a guild roster, individual and guild events, the guild bank logs, guild leadership, among others. As shown in Figure 4, the guild window displays the eight ranks and their respective permissions in the guild “Guardians of Honor.” In the lower left corner is a guild chat window (in green) where members of the guild communicate, with someone saying “Hello” after logging in and another person congratulating a guild member for obtaining a legendary loot by saying “Gratz” (short for “Congratulations”). In addition to in-game tools, some guilds also maintain a website and/or a forum hosted on the official

42 EQII website with a small fee, while others may build their web presence on third-party websites (e.g., guildportal.com). Figure 4. Guild Interface

Guilds are persistent groups created by players and for players. Just like individual characters, guilds gain experience levels when their members complete guildoriented tasks (known as “writs”) and special quests (known as heritage quests). The higher the level a guild gains, the more amenities it is able to provide for its members, including access to special in-game items that require certain guild levels to unlock (e.g., some virtual “mounts” such as floating disks), as well as a customizable guild hall filled with useful teleportation devices, crafting stations, NPC merchants and trainers only to

43 serve the needs of guild members. These practical benefits are only part of the reason why a player would join a guild, however. As many have suggested, guilds provide a stable backdrop upon which community life could be constructed in the game (Ducheneaut, Yee et al., 2007; Williams et al., 2006). The existence of guilds ameliorates the aforementioned coordination costs associated with groups and raids and helps to streamline the formation and maintenance of collaborative activities such as fighting mobs. Moreover, guilds also provide a more intimate social space (as represented by the guild chat channel, the guild hall, and/or the guild website) within the larger game for members to carry out small talk, hang out, and develop meaningful social relationships as well as communal identities. As shown in Figure 4, people exchanged friendly greetings in the guild chat channel as one guildmate logged in and another obtained a piece of legendary weapon, even though the players involved in these exchanges may not know each other personally. These social rituals are part of a set of shared norms and values among guildmates, developed over time and through repeated interactions. Some guilds have codified these shared norms into “guild charters.” For example, the guild charter of “Blackhawks” states that the basic principles of the guild are “helping others,” “have fun” and “don’t be a jerk,” and defines the appropriate behaviors in combat and guild chat (Blackhawks, 2010). In addition, the EQII website tracks a guild’s progress (e.g., recruitment, leveling, NPC kills) and also maintains a dynamic ranking of all the guilds according to several game statistics, such as the size of membership, the aggregated NPC kills, the amount of total wealth, and the kill versus death ratio

44 (http://everquest2.com/leaderboards/guild_leaderboard_index.vm). These accessible statistics facilitate the development of a public reputation system for guilds, through which players are able to keep track of and take pride in their collective achievements. Guilds vary widely with regard to their mission, size, structure, and life cycle (Chen, Sun, & Hsieh, 2008; Williams et al., 2006). Although researchers have suggested multiple typologies to classify guilds (e.g., Taylor, 2006; Williams et al., 2006), the extent to which a guild attaches importance to high-end raiding (as opposed to casual gaming and social engagement) is arguably the most distinguishing dimension of all guilds. On the one end, raiding guilds are the most hardcore guilds (sometimes known as “elite,” “leet,” or “uber” guilds) with an exclusive commitment to the organization of raids. They strive to excel in tackling the most challenging content in EQII. Raiding guilds tend to have a raiding schedule and impose strict play requirements. They do not necessarily have a large member base since they are very selective in admitting new members with regard to their level, class, equipment and raiding availability. For example, the high profile raiding guild “Despair” has strict level and class requirements and are only accepting new applications from five character classes (out of a total of 24 classes available in EQII) as of July 2010 (Despair, 2010a). On the other end of the spectrum, social guilds (sometimes known as family guilds or casual guilds) emphasize playful engagement and the social bonds created as a result of the game. Fighting and questing become secondary compared to the communities they help to foster. Social guilds normally have no minimum play requirements and welcome players regardless of their class and level. Admission into a

45 social guild tends to be less stringent—many encourage existing members’ real world contacts to join, such as family members and friends, as well as their alternative characters (known as “alts”). For example, the casual guild “Executioners” describes their approach as follows: “This is an open guild so any one can join us or leave us as they please. The guild was created for the casual player to quest, chat, or just craft with no strict rules that the hardcore guilds use. People do not have to sign up to join us, all they need to do is just ask. ” (Executioners, 2010). The guild “Blackhawks” articulates their value for community: “The Blackhawks are dedicated to mutual advancement, fun, profit, and adventure. We exist for the camaraderie of playing on a solid team and to maximize the enjoyment of the game…We don’t care about your level or your profession. To us, its[sic] about the player, not the character.” (Blackhawks, 2010). Just like in other social situations, the best criteria for selecting a guild to join are centered on the “fit” between the player and the guild. With a wide spectrum of guilds to choose from, it is important that individual players pick a guild that suits their playing style. A casual gamer who just wants to hang out with friends and have some fun would have a hard time mingling with goal-oriented power gamers or keeping pace with the demanding play schedule. Similarly, a hardcore gamer who pursues the toughest game content would be very disappointed and frustrated by the loose organization and lighthearted game play in a social guild. Guilds often issue recruitment advertisements, mission statements, and other documents such as guild charters to describe their philosophy and code of conduct. In many cases, applicants are required to talk to guild officers, have an existing guild member to serve as a sponsor, or pass a probationary

46 period, such as a series of grouping events with existing guild members, during which the guild evaluates the applicant’s competence and behavior. These application procedures serve as a mutual selection mechanism, so that only the players who are compatible with the guild and who share the same values and understandings would be inducted into the guild. As such, guilds provide one with access to a social network of likeminded others, dissecting the large virtual world of EQII into narrower but more intimate subunits. Network Patterns The social and collaborative nature of MMOGs has long been recognized by scholars in the areas of computer and social sciences (Bainbridge, 2007; Denning, Flores, & Luzmore, 2010; Ducheneaut et al., 2006). Most of the game activities offered in MMOGs are already available in standalone video game applications. Therefore, an important contributor to the massive popularity of MMOGs is the fact that they allow players to interact with fellow players, in addition to artificial NPCs controlled by the computer. This observation is corroborated by empirical data, as a survey of MMOG players from 45 countries shows that “the social factor” contributes significantly to the enjoyment of gaming (Cole & Griffiths, 2007). With an ever-increasing player base, it becomes even more imperative to explore the patterns of social interactions in these worlds and their potential impacts. This issue may be understood in the broad context of the declining community engagement in America (Putnam, 2000). Some scholars argue that time spent online contributes to (rather than alleviates) social isolation and the decline of community life (e.g., Nie, 2001), while others see new media and online spaces as vibrant venues to foster social and civic

47 engagement across spatial and temporal boundaries (e.g., Wellman & Haythornthwaite, 2002). Advocates of networked gaming communities also extol their potential in fostering informal learning (Steinkuehler, 2004) and providing fruitful training grounds for building agile and diversified virtual teams in response to complex tasks (Brown & Thomas, 2008; Denning et al., 2010). However, before exploring in detail the social impacts of MMOGs (see the section “Network Effects”), it is important, as a first step, to systematically evaluate the social experiences of players. The preceding description of MMOG communities in general and EQII in particular demonstrates the many possible ways in which social interactions may arise in these virtual worlds, but such description does not provide any empirical basis with regard to the actual level of sociability. To what extent is EQII a social space and what are the factors associated with informal sociability? This section sets out to establish a general account of the social life in EQII, which sets the stage for further analyses of the effects and evolution of player social networks (see the sections “Network Effects” and “Network Evolution”). Virtual Third Places Computer-mediated social relationships in MMOGs suggest new ways of conceptualizing the relationship between “place” and communities. One the one hand, the power of physical location has gradually waned with the advancement of transportation and communication technologies that have facilitated the formation and maintenance of social connections among spatially dispersed individuals (Mok & Wellman, 2007). Just like Licklider and Taylor (1968) predicted, “…life will be happier for the on-line

48 individual because the people with whom one interacts most strongly will be selected more by commonality of interests and goals than by accidents of proximity.” As a result, communities online and MMOGs are formed not around neighborhoods, but around social networks comprising kin, friends, and more importantly, people of similar interests (Wellman & Gulia, 1999). In other words, shared physical “place” has ceased to be the central element in creating communities. On the other hand, in the absence of “accidents of proximity,” the Internet and MMOGs appear to embody a new type of gathering place in cyberspace, where likeminded people come together and socialize in a relaxed and welcoming atmosphere. Oldenburg’s notion of “third places” has provided a useful conceptual framework to guide the understanding of the social value of virtual worlds. In Oldenburg’s original formulation, “third places” describe a certain class of public settings that are different from one’s home or workplace, such as the bars, coffee shops, and general stores. As Oldenburg described, Third places exist on neutral ground and serve to level their guests to a condition of social equality. Within these places, conversation is the primary activity and the major vehicle for the display and appreciation of human personality and individuality. Third places are taken for granted and most have a low profile. Since the formal institutions of society make stronger claims on the individual, third places are normally open in the off hours, as well as at other times. The character of a third place is determined most of all by its regular clientele and is marked by a playful mood, which contrasts with people's more serious involvement in other spheres. Though a radically different kind of setting from the home, the third place is remarkably similar to a good home in the psychological comfort and support that it extends. (1997, p. 42) Noting the numerous similarities between MMOGs and physical third places such as bars and coffee shops, many scholars of new media and MMOGs have suggested that

49 the Internet and MMOGs in particular can be understood as virtual “third places” that have emerged amid the eclipse of physical third places (Ducheneaut, Moore et al., 2007; Horrigan et al., 2001; Steinkuehler & Williams, 2006). In studying the early text-based online community the WELL, Rheingold (1993) observed the striking resemblance between the WELL and Oldenburg’s third places, suggesting cyberspace as an informal public “place” where the revival of community life is possible. Game researchers argue that MMOGs differ from earlier online social spaces, such as text-based chat rooms, in that they provide “physicality” realized through a rich 3-D environment (Ducheneaut, Moore et al., 2007). The sense of “place”, albeit virtual, is particularly salient in these worlds, as social life in MMOGs is consistently framed around and attached to specific in-world locations. Wizard spires and Druid rings, for example, used to be stations of “teleportation” for the Wizard and Druid classes in the game EverQuest. Because many players would gather around these stations and ask any Wizard or Druid for a teleportation service, these stations had grown to be vibrant public spaces in the EQ universe (Taylor, 2006). Steinkuehler and Williams (2006) argue that MMOG worlds satisfy all the defining criteria of third places: they are low-key, accessible places with a playful atmosphere, which provides a neutral ground for the regulars as well as newcomers to mingle, mainly through conversation. Also consistent with the notion of third places is their observation that MMOGs are particularly conducive for the development of weak connections with diverse populations, or bridging ties, but strong and substantive relationships, or bonding ties, can be less common.

50 Prior Research Due to the social and interactive nature of MMOGs, the “social factor” is often difficult to eschew in MMOG research, and it has been studied from a diverse array of angles. For example, on the individual level, studies have examined the goals and motivations to play MMOGs, suggesting that a significant number of gamers are attracted to various games with the intent for social interactions outside of their normal daily routines (e.g., Bartle, 1996; Yee, 2006c). Others have focused on the socioemotional and task communication (e.g., Pena & Hancock, 2006), the dynamics of team coordination (e.g., Chen, 2009), the emerging reward distribution systems in group play (e.g., Malone, 2009), as well as the practices of informal learning in user communities (e.g., Fields & Kafai, 2010). Although these studies have collectively contributed to the understanding of the “social factor,” they all have different primary focuses and thus are beyond the scope of this dissertation. The following review of prior empirical research, therefore, concentrates only on those studies that examined the forms and enactment of sociability in MMOGs. The conceptualization of “virtual third places” signals MMOGs’ potential for informal sociability, yet empirical evidence is far from conclusive. Based on their methodological approaches, empirical studies can be classified into two streams. Studies employing qualitative methods, such as interview and participant observation, have offered rich descriptions of how social interactions could be carried out in various MMOGs. For example, through her ethnographic study of EverQuest, the prequel of EQII and one that influenced a generation of MMOGs, Taylor (2006) provided an in-

51 depth examination of the social architecture that undergirds the formation of relationships among gamers. She chronicled the many forms of sociality in EverQuest, such as groups, raids, and guilds. She argued that worlds like EverQuest are fundamentally social spaces, where even the most instrumental and goal-oriented “power gamers” are not stereotypical social isolates but instead rely deeply on social networks and resources to succeed. Similarly, Nardi and Harris (2006) discussed a multiplicity of collaborative play in another popular MMOG, World of Warcraft (WoW), ranging from lighthearted random acts of kindness, collaboration with complete strangers as well as offline friends and close ties, to highly organized play in structured groups. Interviews with WoW players also revealed that the depth of social connections varied widely: roughly a third of players used the game as an additional channel to strengthen and maintain pre-existing offline friendship, while the majority (a third to half) interviewed merely used their guilds as a casual third place to make diverse but weak connections (Williams et al., 2006). While qualitative studies are able to unpack MMOG sociability into diverse strata of social experiences, they fall short of providing generalizable insights. A second stream of studies employs quantitative methods. For example, Yee (2006a) found that players derive meaningful relationships from game play, as 39.4% of male players and 53.5% of female players consider their MMOG friends comparable or even better than real-life friends. Cole and Griffith (2007) also found that about two thirds of MMOG players had made good friends in game worlds, and many feel emotionally attracted to their game friends and discuss sensitive issues with them.

52 Yet, in all of these cases the work has been limited by lack of access to comprehensive game populations. Survey studies often rely on convenience samples that may entail severe self-selection bias. Very few studies are based on data from the games themselves. However, it is the unobtrusively collected data from game servers that could reveal the most accurate social experiences of MMOG players, and studies based on such data have already yielded differing conclusions. One study systematically examined player behaviors in WoW, one of the largest MMOGs on the market, and found that the extent of social activities might have been over-estimated in previous accounts (Ducheneaut et al., 2006). Instead, in character selection players display a strong preference for “solo-able” classes which stand a better chance of survival independently than other classes, and on average, players spend much less time collaborating with others than playing solo. They described this rather surprising phenomenon as “alone together”—many players opt to play alone despite the thousands of fellow players surrounding them in the vast online space. Mapping the Social World of EQII To date, Duchenaut et al.’s (2006) study is the only systematic examination of MMOG sociability, using behavioral data collected unobtrusively from game servers. Since their findings paint a somewhat different picture than what the qualitative studies and self-reported surveys suggest, it is important to ask whether the “alone together” observation is applicable to WoW only or across different online gaming worlds. With all the embedded social mechanisms and thousands of fellow players, to what extent are MMOGs truly social spaces? The first part of this dissertation, therefore, aims to explore

53 how EQII functions as a social world and the factors associated with individuals’ social experiences. Using behavioral server logs provided directly from the game operator, this study adds to the scarce but growing literature on the social dynamics in MMOGs. The field of MMOG research has been stymied by the lack of systematic and generalizable assessment of social life in MMOGs (see Ducheneaut et al., 2006 for an exception), and this study helps to establish a much-needed benchmark. It also provides a solid empirical foundation to better understand the effects and evolution of player social networks (reported in sections “Network Effects” and “Network Evolution”). In addition, this descriptive study of EQII is a necessary first step in facilitating comparisons across different MMOGs. Such comparisons, combined with the features of social architecture of different game worlds, could lay the groundwork for studies exploring the causal impact of game mechanics and architecture (Lessig, 2006). The study starts with a preliminary overview of the various social activities in EQII. As discussed previously, sociability in EQII is practiced in a diversity of ways, including trade, chat, and collaboration. Through these dynamic activities, individual characters are able to connect with other players and weave an ego-centric social network. While trade, chat, and collaboration represent dynamic and ephemeral connections, guilds, on the other hand, are persistent groups created and maintained by players to coordinate in-game actions and accomplish joint tasks. Research shows that guilds provide a social backdrop to many game activities (Ducheneaut, Yee et al., 2007; Williams et al., 2006). Guild membership is thus another indicator of social engagement.

54 The extent to which people are involved in dynamic interactions as well as permanent association speaks directly to the overall sociability of the online world. As such, EQII provides a direct (but not fully equivalent) comparison to Ducheneaut et al.’s (2006) study of WoW. Following the preliminary overview, a series of factors are explored for their association with sociability. Demographic indicators such as age and gender have long been used to construct the gamer stereotype. The typical image of a young male gamer has dominated popular as well as academic discourse. Yet, research has consistently debunked the myth that gamers (especially MMOG gamers) are predominantly adolescents and young adults (Williams et al., 2008). For example, an online survey of EverQuest players shows that the majority of players are adults, with an average age of 27.9 years (Griffiths et al., 2004). Another comprehensive survey of players from various MMOGs corroborates this finding, showing an average player age of 26.57 and only a quarter of respondents being teenagers (Yee, 2006a). These studies also suggest that MMOGs are being played by a wide range of age groups. Given the inherent complexity and the multitude of activities available in these gaming worlds, it is very likely that MMOGs appeal to different age cohorts for different reasons. The social circumstances of a teenager, a college student, a working professional, and an old retiree can be drastically different, leading to differing motivations to seek community life in online worlds. Even though the extant literature does not offer any direct test of the relationship between age and social engagement in MMOGs, it is reasonable to include age as a control variable. Therefore:

55 H1: Age is related to players’ level of engagement in social networks. Studies have shown consistently that gender representation in MMOGs is heavily skewed, with males outnumbering females about four to one (Griffiths et al., 2004; Williams et al., 2008; Yee, 2006a). Taylor’s (2006) ethnographic study of female gamers reveals that they play for achievement, competition as well as social reasons. Yet, quite a few studies suggest systematic behavioral differences between the two genders (Cole & Griffiths, 2007; Williams, Consalvo, Caplan, & Yee, 2009), providing solid evidence for gender role theory that people internalize cultural expectations associated with their gender and tend to perform as predicted by gender stereotypes (Kidder, 2002). Among EQII players, female players were motivated more by social reasons, but men were motivated more for achievement-related reasons (Williams et al., 2009). A majority of female players (62%) used EQII as a relationship-oriented space to play with a romantic partner (Williams et al., 2009). Females were also more likely to play with a family member (Yee, 2006a). These studies suggest that male and female players can have distinct patterns of social interaction in MMOGs, leading to the following hypothesis: H2: Gender is related to players’ level of engagement in social networks. A consistent finding from existing literature is that game mechanics and social architecture play a substantial role in shaping players’ social experiences (Taylor, 2006; Williams et al., 2006; Yee, 2009). If MMOGs function as virtual “third spaces”, they too, are governed by the rules as much as brick-and-mortar third spaces such as bars and coffee shops are regulated by laws and implicit social norms (Oldenburg, 1997). The rules and norms in online gaming worlds are in part encoded in their technical and social

56 design (Lessig, 2006). Two aspects of game design of EQII are particularly powerful mechanisms to motivate and structure social interaction (especially collaboration). First, many mobs and areas are simply too difficult to be tackled by solo players. This situation is further exacerbated as players advance in character level. In order to access the difficult but rewarding game content, players are increasingly motivated to seek help from others to collaborate in groups and raids. Therefore, it should be expected that character level positively predicts the degree of social engagement. Second, different character classes are equipped with distinct skill sets of defense, self-preservation, and offense. Although EQII has made an effort in ensuring no class is significantly stronger than others, it is not surprising that some well-rounded classes can stand a better chance of survival for solo play, while others tend to be more specialized (and also “handicapped”), thus are more dependent on co-players. Class choices, therefore, may have significant implications for the degree of social involvement. Based on the above argument, the following hypotheses are proposed: H3: Character level is positively related to players’ level of engagement in social networks. H4: Character class is related to players’ level of engagement in social networks. Finally, one’s membership in permanent player associations—guilds—may also influence the level of sociability. As discussed previously, guilds provide a stable social backdrop for many in-game activities. One of the major functions of guilds is to organize collective events such as groups and raids, making collaboration less costly. Players affiliated with the same guild tend to play longer and collaborate more often (Ducheneaut

57 et al., 2006). In addition, because guild recruitment and admission processes function as a mutual-selection mechanism, people admitted into the same guild tend to have compatible values and play styles. As players gain access to an existing social network of guildmates with shared collective identity, they are more likely to develop trust and friendship through repeated collaboration and other forms of social contact in the guild chat channel, guild hall, guild website, and the like. Therefore, H5: Guild membership is positively related to players’ level of engagement in social networks. Network Effects Social Impacts of Internet Use The preceding discussion conceptualizes MMOG worlds as virtual third places and explores the ways in which sociability is practiced and structured. But, what exactly does participation in online gaming communities do to the users? What are its implications on social and civic engagement? These questions form the core of an ongoing controversy in both academic and popular discourses surrounding the social consequences of new media technologies. On one hand, Internet and various social media applications such as emails, online forums, Social Network Sites, and MMOGs provide unprecedented functionalities to help maintain existing social relations and build new ones (e.g., Ellison et al., 2007; Wellman & Haythornthwaite, 2002). Together, they represent the culmination of a critical process of decoupling between sociability and locality: proximity-bound social interactions has given way to relationships created and maintained by choice (Licklider & Taylor, 1968; Wellman & Gulia, 1999). On the other

58 hand, critics, reporters, and academic researchers argue that the vast diffusion of Internet is leading to social isolation, to a continuing disintegration of communities and societies, as random encounters online substitute for more meaningful social connections in reallife settings (Nie, 2001). This controversy is hardly new, for similar questions of media effects have been asked time and again as each new medium emerged on the horizon. David Nye (2004) argues that almost all technological advancements have been interpreted in the dichotomy of utopian or dystopian narratives. Utopian narratives see technologies as natural outgrowth of society. They transform social realities by improving our everyday life. Dystopian narratives, by contrast, portray technologies as agents of hegemony and regression, as new inventions would change our everyday life to the worse despite their expected positive outcomes. Reviewing a history of mass media research, Wartella and Reeves (1985) observed that media research has been characterized by recurring research topics about different media, such as film, radio, and television, suggesting predictable cycles of public and scholarly controversies about the adoption of new media technology into everyday life. Such cycles usually start with a feeling of distrust and fear that the use of the new medium would substitute for more acceptable behaviors. The concerns of looming health threats, violence, deviance, and social disintegration arise in public discourse soon after the advent of a new media technology and do not recede before the medium becomes entirely commonplace. The controversy surrounding Internet effects should also be situated within a long-lasting tradition in social theory regarding the loss of community and the decline of

59 social and civic engagement in modern societies (Fischer, 1982; Norris, 2002; Putnam, 2000; Simmel & Hughes, 1949; Tönnies, 1974). As Claude Fischer noted, Few ideas saturate Western thought as does the conviction that modern life has destroyed “community”. Virtually taken for granted by philosophers and citizens alike is the belief that modern society has disrupted people’s natural relations to one another, loosened individuals’ commitments to kin and neighbors, and substituted shallow encounters with passing acquaintances. (1982, p. 1) This recurring theme has entered public spotlight with Robert Putnam’s “Bowling Alone” hypothesis (2000), in which he argued that collective activities, from family dinners to participation in bowling clubs, are in a marked decline in American society. This is due in part to the massive adoption and diffusion of media technologies, the television in particular, that have weakened social bonds as people are watching more TV (especially entertainment programs), watching it more habitually, pervasively, and alone. Data and arguments presented in “Bowling Alone,” however, did not provide direct insights with regard to the social implications of the Internet, at that time a nascent communication medium less saturated than television (Pew Internet and American Life Project, 2010; Putnam, 2000). Putnam’s stance on the Internet was ambivalent, so are the empirical studies conducted to test the effects of Internet use (Shklovski et al., 2006). Two perspectives are noteworthy. On the one hand, some scholars are concerned that Internet use would encroach on time that was previously spent with family and close friends and leave many feeling lonelier (Nie, 2001; Nie & Erbring, 2002), a finding that has received some empirical support (Kraut et al., 1998; Williams, 2007). They view online and offline activities as a zero-sum equation. Because of the “inelasticity of time” (Nie, 2001, p. 420), individuals spend time online at the expense of reducing their time

60 on offline social activities. Therefore, the Internet inevitably makes people more isolated and reduces social capital. This approach, however, assumes that Internet use is not social, or at least less social than other activities. This assumption has been criticized because it fails to recognize that the Internet could serve as a site for some degree of social and communicative activities (Williams, 2006a). On the other hand, the social augmentation perspective praises the potential of the Internet to supply an additional avenue of social interaction. In this approach, not only could the Internet enhance one’s everyday communication with family and friends locally and over a distance (e.g., Wellman, 2001), it could also enlarge one’s existing social network by bringing together people with shared interests and values in virtual communities (e.g., Horrigan et al., 2001). The social augmentation perspective has also received consistent support from empirical studies (Boase et al., 2006; Cole et al., 2000; Katz & Aspden, 1997; Katz & Rice, 2002; Kraut et al., 2002; Quan-Haase et al., 2002). What is Social Capital? In new media effects research, the concept of “social capital” has been adopted as an umbrella term to synthesize research in this area (Quan-Haase & Wellman, 2004; Williams, 2006a). Existing studies have investigated the social effects of Internet on slightly different dependent variables, such as sociability (Nie & Hillygus, 2002), social interactions (Shklovski et al., 2006), social involvement (Kraut et al., 1998), civic engagement and sense of community (Hampton & Wellman, 2003; Quan-Haase et al., 2002). Ironically, the fact that these dependent variables could be subsumed under the

61 broader rubric of “social capital” speaks to the inclusiveness, or ambiguity, of the term itself. Indeed, social capital is one of those catch phrases that invoke different meanings in different groups (for overviews, see Adler & Kwon, 2002; Lin, Cook, & Burt, 2001). Various definitions of social capital exist in social science literature. Burt defines social capital as “the advantage created by a person's location in a structure of relationships” (2005, p. 5). Coleman defines social capital as a function of social structure producing advantage: “It is not a single entity but a variety of different entities, with two elements in common: they all consist of some aspect of social structures, and they facilitate certain actions of actors-whether persons or corporate actors-within the structure.” (1988, p. S98) Putnam defines social capital as “social networks and the norms of reciprocity and trustworthiness that arise from them” (2001, p. 19). One notable difference between theses conceptualizations is whether social capital is an asset of the individual or the public. Assuming individuals are self-interested and rational, Burt’s definition emphasizes individual benefits created by structural locations. Coleman and Putnam, on the other hand, recognize that social capital is both individual and collective—it not only creates benefits for individuals in the social structure, but also brings positive externalities for the community in the form of trust, norms, generalized reciprocities, and collective actions (Coleman, 1988; Putnam, 2000). Despite their different emphases, these definitions of social capital all constitute both social structure and some value produced by the structure. In other words, social capital is both the network and the effects of the network.

62 Compared to other types of capital, social capital has several notable characteristics. Like physical capital, such as tools, and human capital, such as a person’s educational attainment, social capital is an asset into which other types of resources could be invested with the expectation of future returns. Through investing time and energy in building and maintaining networks of kin, friends, colleagues, and neighbors, individuals and collectives could enhance their social capital and gain benefits (Adler & Kwon, 2002). Such benefits could take many different forms, such as power, knowledge, solidarity, and social support, because social capital is “appropriable” and could be used for various purposes (Coleman, 1988). But social capital also differs fundamentally from other types of capital in that it resides not in the objects themselves (e.g., people) but in their relations with other objects. Human capital, for example, represents individual attributes and characteristics, such as attractiveness, intelligence, and educational attainment. These assets are possessed by individuals. By contrast, social capital is embedded in the relationships among individuals. “No one player has exclusive ownership rights to social capital. If you or your partner in a relationship withdraws, the connection dissolves with whatever social capital it contained" (Burt, 1992, p. 58). Social Capital: Source or Outcome? The above conceptualizations have informed Internet-focused social capital research. In particular, drawing from Putnam’s and Coleman’s work, Williams (2006a) developed the Internet Social Capital Scales (ISCS), which have since been widely adopted by researchers of Internet effects and greatly facilitated empirical studies in this area (e.g., Ellison et al., 2007; Steinfield, Ellison, & Lampe, 2008). However, as

63 Williams (2006a) noted, the ISCS operationalize and measure social capital as an outcome rather than the network itself. As such, this approach represents a deliberate choice to simplify the conceptual conundrum confronted by researchers that social capital represents both the source and outcome. This outcome-oriented approach is attractive for a couple of reasons. Conceptually, it fits well with the media effects research tradition which has strongly influenced the scholarship on Internet use and its consequences. Methodologically, with the development of ISCS and similar scales, social capital can be conveniently measured and tested using survey-based data, a popular approach in communication research. This outcome-oriented approach to study online social capital has stood in stark contrast with traditional social capital research (e.g., Adler & Kwon, 2002; Inkpen & Tsang, 2005; Oh, Labianca, & Chung, 2006), which has long recognized that the structural properties of social networks are an indispensable dimension of the concept and measurement of social capital. Taking a different route, this dissertation attempts to examine social capital online from its sources—the social networks that produce individual and collective benefits. In so doing, this dissertation contributes to Internet effects literature by recognizing the structural dimension of social capital. This structural approach is a valuable complement to the popular outcome-based approach, for several reasons. First, it recognizes that networks are the causal agents of social capital measured by ISCS and similar scales (Williams, 2006a). The network-based approach allows for a better understanding of the causal mechanisms leading to observed social consequences brought by Internet use and

64 MMOG participation. It provides a basic validity check for the outcome-based approach as well as an opportunity to probe the processes of social capital creation. As suggested by its conceptualization, social capital produces benefits or outcomes for individual and collective actors, and these benefits should match the respective structural sources that generate them (Burt, 1992). The process of social capital creation can be further explicated by including process variables and contingency factors. Second, because it focuses on the causal mechanisms, this network-based approach could generate valuable insights for the design of better social worlds online. As the “code is law” argument points out, technological features can significantly impact how online relationships are structured (Lessig, 2006). Network structures associated with different types of social capital can be individually identified, examined and engineered (e.g., designing an automatic “recommend to connect” system), thus making it possible to reach desired social outcomes through improved design of the social architecture. Bridging and Bonding, Brokerage and Closure As reviewed previously, empirical research on the social impacts of Internet use has produced conflicting findings, as some studies support the time displacement perspective while others confirm the social augmentation effects of the Internet. The conflicting findings call for more refined conceptualization and measurement of both the predictor (Internet use) and the outcome variable (social capital). Most of the existing studies treat “Internet use” as a single entity and make strong assumptions about whether “Internet use” is social or not social. Some regard the Internet as an alienating medium (Nie & Erbring, 2002), while others consider it as facilitating social communication, even

65 though their results suggest otherwise (Kraut et al., 1998). A meta-analysis of studies on this subject showed that the effect size of Internet use is so small that Internet has shown hardly any impact on social capital at all (Shklovski et al., 2006). The reason might be that social capital-enhancing and –weakening activities simply cancel each other out. For example, one might use the Internet to rekindle old friendship through interactions on social network sites and personal blogs, which enhances social involvement, but at the same time one may also spend considerable time playing an online game alone, which displaces interactions with the existing social network offline, resulting in a very small net effect from both types of Internet use. The Internet is a malleable medium that allows for both social and asocial use. People are not just passively affected by technology, but also actively appropriate technology to reach their own goals (Bargh & McKenna, 2004). Some recent studies start to employ a differentiated approach. For example, Facebook use in particular was found to affect college students’ social capital (Ellison et al., 2007). Another study found that MMOG play with family and friends enhanced family communication and reduced loneliness, but interacting with strangers did not (Shen & Williams, in press). Perhaps more importantly, online activities not only enhance and weaken social capital simultaneously, they also enhance and weaken different types of social capital, leading to a dynamic process of reconfiguration, an issue less discussed in new media effects research until recently (Norris, 2002; Williams, 2006a). Various types of social networks and interactions can be qualitatively different, resulting in diverse social outcomes. Just like the same amount of time spent online can lead to different outcomes

66 depending on what that time is used for, participation in the same online community could also generate different types and levels of social capital, depending on the structure of the network and the quality of the relationships. In particular, two types of social capital are identified: bridging and bonding (Burt, 2000b; Coleman, 1988; Putnam, 2000). In current scholarship of online social capital, bridging and bonding are operationalized and measured as two distinct, but related, social outcomes (Williams, 2006a). As the discussion below demonstrates, each type of social capital is rooted in a specific network structure. According to Putnam’s formulation (2000), bridging social capital is inclusive. These relationships tend to be episodic, ephemeral, and singlestranded, the so-called “weak ties.” Bridging social capital occurs when individuals from diverse backgrounds reach out to make connections beyond their own homogenous groups. As a result, bridging social capital leads to a broad worldview, diversity in opinions and resources, and information diffusion. By contrast, bonding social capital is exclusive. These relationships tend to be repeated, long-lasting, and multistranded, the so-called “strong ties.” Bonding social capital occurs when individuals reinforce exclusive identities and create dense networks within homogenous groups. As a result, bonding social capital undergirds reciprocity and solidarity, builds trust within the group and provides substantive and emotional support. These qualitative differences mentioned above form the conceptual basis of much of the current scholarship on online social capital. The ISCS and similar scales measure bridging and bonding social capital as outcomes, i.e., the effects they create, such as

67 broad worldviews or emotional support. What they fail to explore, however, is the source of these qualitative differences between bridging and bonding. Putnam’s formulation of bridging and bonding social capital can be traced back to the seminal work of Mark Granovetter (1973). In examining people looking for employment, Granovetter illustrated that there were two kinds of social relationships: weak ties and strong ties. Contrary to popular belief, Granovetter found that the most successful job seekers were not those with the strongest ties. On the contrary, because weak ties with acquaintances provide a broader set of information and opportunities, they are more helpful during people’s job search than strong ties with family and friends. Granovetter’s study may suggest that tie strength is an important structural cause for differing social capital outcomes. But tie strength is not the complete story—it also represents a proxy measure for another structural property: the proportion of shared contacts. For any two arbitrarily selected individuals A and B, S is the set of any persons with ties to either or both of them. As Granovetter argued, “…the stronger the tie between A and B, the larger the proportion of individuals in S to whom they will both be tied, that is, connected by a weak or strong tie” (Granovetter, 1973, p. 1362). In other words, weak ties are the people with whom one has few common contacts, but strong ties are the people with whom one shares many friends. Different degrees of network overlap would lead to differing social outcomes. Granovetter’s argument, along with the works of many other sociologists (Coleman, 1988; Cook, Emerson, & Gillmore, 1983; Lin, 1999) fashioned the foundation of Burt’s formalized theory of network brokerage and closure (1992, 2000b,

68 2005). Burt defines structural holes as the lack of connection between any pair of individuals in the network. A structural hole exists if individuals A and B are not linked directly. Structural holes distinguish the two general sources of social capital: brokerage and closure. Network brokerage refers to the social structure where the focal person builds connections across structural holes (Burt, 2005). That is, brokerage occurs when the focal person, or the broker, links otherwise disconnected individuals. For example, a network broker C independently connects to individuals A and B, but there is no direct link between A and B. Therefore, as a broker, C facilitates an indirect connection between A and B. Compared with already connected people, disconnected people are more likely to have different ideas and resources. The more disconnected the contacts in the network, the more the focal person is exposed to diverse opinions and practices. This structural position provides benefits (in Granovetter’s case, job opportunities) for C, as A and B are likely to supply nonredundant information and resources. The effects of network brokerage are precisely bridging social capital—broadened worldviews, heterogeneous opinions and resources, and information diffusion (Putnam, 2001; Williams, 2006a). Network closure, on the other hand, refers to the social structure where the focal person stays on their side of the structural hole. In other words, closure occurs when the focal person links already connected individuals. As connected people tend to share similar views and resources, closure shields the focal person from variations in opinion and behavior. Closure measures the extent to which everyone is connected to everyone else in the network. Using the same example, the focal person C links to A and B, while

69 A and B are already directly connected. C is likely to get redundant information from A and B, which would not provide C with much advantage in a job search. But with everyone knowing everyone else, such a closed network could reinforce existing knowledge and opinions. Also, since bad behaviors would be easily caught by the group, people are unlikely to perform them, thus breeding trust and group cohesion within. Therefore, the effects of network closure are consistent with bonding social capital— deepened beliefs and practices, homogeneity within the group, trust, and cohesion. Collectively, brokerage and closure define the tension between two opposing extremes of network structure. Their effects can be interpreted as either beneficial or detrimental, depending on the times and circumstances. Brokerage brings novel information and opportunities, but the connections are too weak to provide emotional and substantive support. Closure builds a strong and cohesive support network one could count on during adverse times, but it also confines one to what is already known. As Putnam puts it, brokerage is good for “getting ahead” while closure is useful for “getting by” (2001, p. 23). In recognizing brokerage and closure as the structural antecedents of bridging and bonding social capital, this network-based approach goes beyond the two persons involved in a social relationship. Focusing on the dyad itself does not reveal the whole picture. Instead, the complete networks of the two persons need to be included because they affect the uniqueness (or redundancy) of a particular relationship. And it is the very uniqueness (or redundancy) of contacts that determines the social consequences.

70 The theory of brokerage and closure has received consistent support from empirical research. Besides Granovetter’s original finding that weak ties (brokerage) enhance job search success, a great number of studies has confirmed the link between network brokerage and superior performance. For example, in economic networks, producers that broker more structural holes were found to make more profits from negotiating more favorable transactions with suppliers and customers (Burt, 1992). Within organizations, individuals’ mobility is enhanced by having an informational network rich in structural holes (Podolny & Baron, 1997). In less hierarchical organizations such as TV project teams, spanning more structural holes was found to induce better performance in terms of producing more popular TV programs (Zaheer & Soda, 2009). Empirical studies also confirm network closure’s effects in creating trust and bonding groups and communities. Coleman (1988) found that high-school adolescents are less likely to drop out of school if they live within more dense networks of adults, because people in these neighborhoods tend to know each other and collaborate in the supervision of children. As shown in Burt (2005), the closure effects were tested in two populations: senior managers in a computer manufacturing company and staff officers in financial services. In both populations, two people who share common third party ties are more likely to trust each other. The effects of network brokerage and closure have rarely been examined in online worlds. As discussed previously, most Internet-focused social capital research takes the outcome-based approach. Among the few network-based studies, Ganley and Lampe

71 (2009) studied brokerage and closure as generative mechanisms for online social capital in a large online community, Slashdot. They found that users’ reputation in the community, as measured by an embedded system ranking called “Karma,” is positively associated with the degree of network closure. Burt (2010) presents a comprehensive test of brokerage and closure in a large virtual world, Second Life. Results confirm the predicted effects of both structures, as brokers who span more structural holes in their online networks tend to found more groups and found more groups that remain active, and people who are embedded in closed networks tend to trust each other more. Based on the above review, it is expected that brokerage and closure would also produce similar effects in online gaming communities such as EQII. Therefore, this dissertation contributes to the online social capital literature by using a network-based approach rather than an outcome-based approach. It also represents a validity test of the theory of brokerage and closure in another variety of online environment: MMOG communities. Figure 5. A Simple Structural Illustration of Brokerage and Closure C

C

A

B

Brokerage: The focal person C links A and B, while there is no direct connection between A and B. The ego-network of C consists of two dyads (AC and BC).

A

B

Closure: The focal person C links A and B, while A and B are already directly connected. The ego-network of C consists of a closed triad (ABC).

72 In EQII, network brokers who span structural holes, i.e., those who connect otherwise unconnected characters, are exposed to diverse information and resources. To put this argument in context, brokering happens in EQII when people engage in social interaction with other players, especially when they reach out to meet new people outside of their own little cliques. Structurally, in a three person network, the focal person C plays with A as well as B. But A and B are not directly connected in the game. In other words, a broker’s ego-network consists of dyads but few closed triads (see Figure 5 for a simple illustration). As mentioned earlier, by design, the game mechanics of EQII encourage collaborative play among characters with different specialties. As players level up, they are increasingly confronted with challenging monsters and quests that can only be tackled successfully if they amass a variety of resources (e.g., epic weapons and armors), build a repertoire of knowledge and skills (e.g., the skill to mine precious ores, or the ability to resurrect team members immediately after death) through active networking with players of different classes and specialties, and devise sophisticated strategies for collaborative combat. Steinkuehler (2004) argues that MMOGs are essentially “communities of practice” (Wenger, 1998) where learning occurs socially. Gaming is not mastered through explicit instruction of codified knowledge (although reading manuals can be helpful, especially for novices) but rather through various social interactions with more experienced or knowledgeable others. As such, through their diffuse connections, brokers are exposed to nonredundant knowledge, skills, resources, and strategies, all contributing to more successes in combat and questing than their peers who are less connected and who only interact with a couple of regular playmates. Brokers

73 are able to get things done and get them done faster. Observing the behavioral patterns of some of the best players on EverQuest, known as “the power gamers,” Taylor noted: “The reliance on, and involvement with, social networks and resources—Web information and bulletin boards, guilds, and off- and online friendship networks—indeed reveals power gamers to be some of the most socialized players in MMOGs” (2006, p. 81). Therefore, the following hypothesis is proposed: H6: Brokerage in players’ social networks is positively associated with their task performance. While brokering structural holes helps provide diverse information and resources, network closure helps create homogenous groups, enhances solidarity and mutual trust. Closure happens when EQII players interact only with a small but dense network of playmates. In their ego-networks, individuals are surrounded by tightly-connected and redundant ties (as opposed to diffuse and nonredundant ties). As shown in Figure 5. Figure 5, the focal person C plays with A as well as B, while A and B are already directly connected in the game. In other words, the focal person’s ego-network is closed if it consists of triads (i.e., alters are themselves directly connected). In a closed network, people tend to do things together and do so repeatedly. They develop shared knowledge, routines, and strategies for game play, and in so doing create a common identity and sense of belonging. As everybody knows everybody else, a reputation system is naturally in place, promoting in-group trust and preventing deviant behaviors from occurring. Small family groups or guilds are typical examples of closed networks. Some of those inworld interactions are harbingers or extensions of substantive relationships in the offline

74 world (Nardi & Harris, 2006; Skoric et al., 2010). People are content and comfortable staying within their own little clique. Boundary-spanning behaviors are infrequent— people do not normally venture outside of the group to interact with strangers. Therefore, the groups or guilds they belong to demarcate the boundary of their play experiences. The ties that bind a closed group can be qualitatively different from the ties that broker structural holes in EQII. As Taylor (2006) observed in the world of EverQuest, power gamers, typically good brokers, tend to practice sociability with instrumental goals in mind. In small family guilds, by contrast, “collective actions arise not from an instrumental orientation…but from mutual values of one another” (2006, p. 48). Therefore, it is reasonable to expect that the degree of closure in players’ egonetworks would positively predict their trust of other players and their sense of community, two important elements of bonding social capital. Compared to the people with whom one plays on a coincidental or ad-hoc basis, the guild represents a stable social group within which one could have repeated interactions and develop meaningful social relationships (Williams et al., 2006). As guild life is fundamental to the creation of social bonds, the degree of closure in one’s social network should be positively associated with the level of trust towards one’s guildmates. Therefore, the following hypotheses are proposed: H7: Closure in players’ social networks is positively associated with their trust towards guildmates. H8: Closure in players’ social networks is positively associated with their sense of community.

75 Past research has indicated that brokerage and closure effects can be contingent upon demographic variables such as age and gender (Burt, 2005). In addition, as EQII character classes define different play experiences, it is possible that individuals’ class choice may influence their performance in the game. Therefore, a series of control variables, including age, gender, and class archetypes, are also considered in the analysis to control for extraneous variance. Network Evolution The previous discussion focuses on the patterns of interaction and the potential effects resulting from social networks within online gaming communities. A key set of questions remain: How do social networks formed in MMOGs change over time? Are online relationships merely random, short-lived encounters or lasting and substantive connections? What makes some relationships more durable than others? These questions are important because they highlight the dynamic processes of relationship formation, maintenance, and demise in online worlds, an issue rarely examined in the extant literature despite repeated calls for more research employing longitudinal analyses (Ellison et al., 2007; Harris et al., 2009; Lewis et al., 2008; Williams, 2007). Further, an understanding of the stability, or rather, fragility, of online relationships provides a crucial context to qualify the significance and generalizability of findings obtained in empirical studies relying on cross-sectional data. This dissertation is an attempt to tackle these questions in the context of social relationships formed in EQII. Evolutionary theories, as represented by Campbell’s influential work on socio-cultural evolution (Campbell, 1965), population and

76 community ecology (Aldrich, 1999; Hannan & Freeman, 1977), and the recent extensions to network evolution (Monge et al., 2008), represent a powerful framework to study the longitudinal transformation of social ties in online worlds. Evolutionary theories are ultimately about change in various social systems. Although most of the developments and applications of evolutionary theories concentrate in the study of organizations, the fundamental mechanisms of variation, selection and retention can still be usefully applied to derive hypotheses about the persistence and decay of social relationships in online worlds. Further, evolutionary theories also provide a well-established methodology to test these hypotheses (Monge, Lee, Fulk, Weber et al., in press). This section begins with an overview of the major tenets of the evolutionary framework—processes of variation, selection and retention. This framework is then applied to the social networks in EQII to derive hypotheses based on three sets of evolutionary factors: the aging process (including tie age and node age), social architecture of EQII, and homophily and proximity. Variation, Selection, Retention The starting point of an influential line of research in modern social theory can be traced back to Donald T. Campbell’s 1965 paper “Variation and selective retention in socio-cultural evolution,” in which he used evolutionary ideas, analogous to the natural selection processes in biological evolution, to explain the emergence, transformation and termination of human social systems. Campbell’s evolutionary model consists of three components. First, variations are the source of the inherent differences among social entities. Most of these variations are not designed purposefully, but are random, blind,

77 and haphazard. Second, based on the rich variety of alternatives generated by the blind variation process, selection is the process of “winnowing” out some variations but accepting others (e.g., the differential box office responses to certain Hollywood movie formulas). Finally, retention is the process of solidifying and maintaining selected variations through preservation, duplication, or diffusion in the social system (e.g., the replication of movies that achieved commercial success). Campbell’s evolutionary framework rests on the premise that the environment resources which sustain those social entities are ultimately scarce. In other words, the environment could not accommodate every variation, thus there is a constant struggle over scarce resources. The selection criteria are consistent and ensure “evolution in the direction of better fit to the selective system” (Campbell, 1965, p. 27). This evolutionary framework is intentionally generic and has been extended and applied to the study of many social phenomena including organizations (Aldrich, 1999; Carroll & Hannan, 2000; Carroll, 2004; Hawley, 1986; Nelson & Winter, 1982). Viewed through an evolutionary lens, organizational entities depend on environmental resources (e.g., market share) to survive, just like living organisms. Many possible variations may exist, and they are eliminated or selectively reproduced based on whether a variation contributes to or undermines the fitness of the organization. It is through competition processes that the less “fit” entities are winnowed out, retaining and solidifying the existence of those entities that are better suited for such environmental conditions (Aldrich, 1999; Carroll & Hannan, 2000; Hannan & Freeman, 1977). The evolutionary framework has explained various mechanisms of organizational change, including (but

78 not limited to) demographic processes of size, age, and density dependence (Hannan & Freeman, 1977, 1984), resource-partitioning processes of market concentration and specialization (Carroll, 1985; Carroll, Dobrev, & Swaminathan, 2002), structural inertia, learning, and imitation (Hannan & Freeman, 1984), and ecological processes of commensalism and symbiosis (Astley, 1985; Hunt & Aldrich, 1998). Together, these mechanisms constitute the V-S-R processes organizational entities experience as they respond to resource constraints and environmental pressures. The generality of the evolutionary framework is evident in that the mechanisms of variation, selection and retention are applicable across different levels of analyses: routines and competencies, groups, organizations, populations and communities. Among these units, “population” and “community” are two key constructs in the evolutionary framework. Population is defined as a set of organizations “which are relatively homogenous in terms of environmental vulnerability” (Hannan & Freeman, 1977, p. 934). Populations are also distinguished based on their shared features and properties, which is jointly referred to as organizational forms (McKelvey, 1982). In other words, a population is a set of similar organizations. Various organizational populations are identified and studied in the organization literature, such as daycare centers (Baum & Oliver, 1992), farm wineries (Swaminathan, 1995), gas stations (Usher & Evans, 1996), and savings and loan unions (Haveman, 1992). An even higher level of analysis is the community, which is defined as a “set of coevolving organizational populations joined by ties of commensalism and symbiosis through their orientation to a common technology” (Hunt & Aldrich, 1998, p. 272).

79 Commensalist relations typically occur between populations which exist in similar resource spaces and can range from cooperative to competitive, while symbiotic relations are mutually beneficial and usually occur between populations in different resource niches (Aldrich, 1999; Hawley, 1986). In other words, while analysis at the population level considers the evolutionary change between similar organizations in an established population, community ecology focuses on the competitive and cooperative dynamics among interacting and heterogeneous populations (Astley, 1985). Evolutionary processes have been explored in various communities, such as the World Wide Web community which includes 17 populations such as browser developers, standard-setting bodies, internet service providers, and regulatory agencies (Hunt & Aldrich, 1998), as well as the biotechnology community which comprises dedicated biotechnology firms, venture capital firms, pharmaceutical companies, universities, research organizations, and government regulators (Powell, White, Koput, & Owen-Smith, 2005). Indeed, empirical studies have explored evolutionary processes across different units of selection. At the intra-organizational level, for example, Schultz (1998) explored the generation and retention of bureaucratic rules in a large university over time. Consistent with the evolutionary prediction that the density of entities in a given environment induces competition, he found that rule production declines with the number of existing rules in a specific domain. At the organizational level, Haveman (1993) found that organizational decisions to enter into new markets are influenced by the large and profitable players in the same population. In other words, decisions of diversification are selected on the basis of “following the leaders”. At the population level, Usher and Evans

80 (1996) examined the transformations of gas stations in a Canadian city over a thirty-year period and found interesting variations of organizational forms. Besides selling gas, different types of gas stations can also offer service stations, gardening equipment, and car wash service. These organizational forms evolve as some forms grow and others decline. And finally, at the community level, Dimmick (2003) examined the changing contour of media industries in the United States. He showed that coexistence of various media outlets in the same market (e.g., the daily news media in Columbus, Ohio) is possible because each provides somewhat differentiated cognitive and affective gratifications to the viewers. It is clear that an evolutionary perspective (especially at the community level) recognizes the inherent interdependence among organizational entities. The birth, growth, transformation, and death of organizations are understood in the context of their dynamic interactions, either competitive or collaborative, with other organizations located in the same resource environment (Carroll & Hannan, 2000). Populations and communities are about objects, such as groups and organizations, as much as they are about the relations that bind these objects. This suggests that organizational communities are essentially networks, which are defined as a set of objects (nodes) tied together by a set of relations (Wasserman & Faust, 1994). Therefore, DiMaggio (1994) encourages scholars to “apply population models not to organizations, but to relations among organizations in different populations or networks” (p. 447). A valuable extension of the V-S-R processes to networks (instead of nodes) is articulated in Monge, Heiss and Margolin (2008). In their framework, nodes represent

81 organizations or populations of organizations in a community, and ties can be considered as mechanisms of resource exchange among these nodes. Following the evolutionary argument, an organizational community is constrained in its capacity to sustain the nodes as well as the ties. Monge, Heiss and Margolin (2008) label them as: 1) a member carrying capacity, which defines the upper limit of nodes the community can support, and 2) a relational carrying capacity, which defines the upper limit of ties the community can support. The relational carrying capacity adds another layer of constraints imposed by the social system. Here, resource is conceptualized as the capability to initiate and maintain ties. For a given node in the network, its capability to make connections is finite and cannot accommodate all the possible linkages. For example, in a network of a thousand nodes, the possible number of linkages for any node is 999 and the largest possible network density—defined as the total number of existing ties divided by the total number of possible ties—is one, with every node connected to every other node (a fully connected network). Yet, because of resource constraints, the number of connections of a single node is often much smaller than the number of possible connections, and the overall density observed in real world networks is usually considerably lower than one. Member carrying capacity and relational carrying capacity can be related, but such relationship is not necessarily linear. Studies of empirical networks show that the density of links may grow with the density of nodes, but the growth rates vary greatly across different networks (Leskovec, Kleinberg, & Faloutsos, 2007). This indicates that

82 relational carrying capacity should not be discussed in the abstract, but grounded in the context of a specific network. The framework proposed by Monge, Heiss and Margolin (2008) applies the V-SR mechanisms to network ties. Variations can be considered as the numerous possibilities of tie formation among all the available nodes in the network. Just like nodes, ties are also subject to selection pressures. While nodes are eliminated or retained based on their level of fitness, ties are selected, too, based on their fitness, defined as the “the propensity for a relationship to sustain itself, that is, to survive or to reproduce itself.” (Monge et al., 2008, p. 462). A tie can survive only if both nodes it connects survive. A tie is fit if it is easy to sustain and it provides important resources to one or both partners involved in the relationship. Many studies of interorganizational networks have examined the various factors affecting the V-S-R processes of network ties. For example, through an examination of the children’s television community from 1953 to 2003, Bryant and Monge (2008) found that major environmental events, such as technological innovations and new regulatory regimes, led to initiation of ties between populations. A comprehensive study of the biotechnology industry, comprising populations of research institutions, pharmaceutical companies, venture capital firms, and government regulators, demonstrated that the evolution and dynamics of the community are shaped by various logics of tie formation: accumulative advantage, homophily, follow-the-trend, and multiconnectivity (Powell et al., 2005). Finally, Shumate, Fulk and Monge (2005) found that International Non-Governmental Organizations (INGO) in the HIV/AIDS sector

83 were more likely to establish relations with other INGOs if they shared partners with the same International Governmental Organizations. The Evolution of Personal Ties in EQII The evolutionary perspective reviewed above provides a useful framework to understand change in online social networks. The unit of analysis, then, is the tie between two nodes (individuals) instead of the node itself. It is necessary to contextualize this framework with the specifics of the relations within EQII, as substantive differences of relations and networks can have tremendous implications on the evolutionary dynamics (Leskovec et al., 2007; Monge et al., 2008). Variations of social ties in EQII refer to the many potential opportunities for new relations to form. These variations originate from diverse factors that bring people together. Most of these factors are exogenous to the two individuals involved. For example, one could only connect to those characters currently logged in the game (and for most activities, on the same server). To collaborate in impromptu groups or raids, solo characters who are “LFG” (looking for group) and existing groups which are “LFM” (looking for more) may connect with each other only if they happen to fit the search criteria, such as the character level (e.g., a level 30 character joining an level 28-32 group) and specialty (e.g., a Paladin who wishes to serve as a “tank” joining a group looking for tanks). Even the one-to-many broadcast chat messages have a limited reach. They can be heard if the recipient subscribes to certain chat channels, such as the level 10-20 channel and the wizard channel, or happen to be in the (virtual) vicinity. Although some variations can be intentional, such as interacting with a specific character who is a real-

84 world friend, most of the new opportunities for social interactions in EQII are random, blind, and haphazard. One character plays with another because they happen to be logged on at the same time, on the same server, serve as a valuable partner and/or good company for each other. In online worlds, the number of personal ties one player could hold is also limited. The notion of relational carrying capacity recognizes the cost of relations, a subject not fully discussed in existing network research (Monge et al., 2008). Indeed, ties are costly to initiate and to maintain over time. Two types of cost associated with interpersonal relations need to be considered: time and cognitive effort (Donath, 2007; Dunbar, 2008; Roberts, Dunbar, Pollet, & Kuppens, 2009). First, interactions themselves take time to complete. While weak and episodic ties may only take a nominal amount of time, building strong and trusting relations is often a long-term effort. Second, individuals need to be able to keep track of the status of the relations, the signals of the characteristics of their contacts (Donath, 2007) and the perception of the changing social structure as a whole (Krackhardt, 1987). The amount of required time and cognitive effort to sustain relations varies according to the type of network content, with the strongest and most intense relations most costly to maintain. Empirical evidence has largely supported this argument. Many studies have shown that the size of most human social networks has an upper limit of around 200 people, referred to by many as the “Dunbar number” (Dunbar, 1996). Also, research has found that there is an upper limit to the total network size, and the average emotional closeness of the network is inversely correlated with its size (Roberts et al., 2009).

85 The number of ties one could actually initiate and maintain in EQII, therefore, is curtailed by one’s limited relational carrying capacity. Selection and retention processes are in place to eliminate excessive ties based on their fitness—their propensity to sustain themselves and the extent to which they can provide benefits to the players involved. It should be noted that the evolutionary framework presented in Monge, Heiss and Margolin (2008) is developed for network ties within organizational communities. Their model conceptualizes ties primarily as conduits for resource acquisition, such as manufacturer-retailer relationships, strategic partnerships, etc. In the social world of EQII, however, this resource-oriented view does not reveal the whole picture. Certainly, the acquisition of resources (e.g., knowledge, skills, ores and armors) is an important function of EQII ties, as the game is indeed designed to promote player interaction through an intricate division of labor. Social and collaborative play helps one finish quests, kill monsters, and develop one’s character more rapidly and successfully. However, in-game achievement and character development are not the only motivations for connecting. Players come together also because they enjoy each other’s company. This aspect is unique for personal networks, but is rarely discussed in the context of organizational networks where instrumental motivations and behavior dominate. As the sociologist George Simmel wrote, “Sociability is the art or play form of association, related to the content and purposes of association in the same way as art is related to reality. While sociable interaction centers upon persons, it can occur only if the more serious purposes of the individual are kept out…” (Simmel & Hughes, 1949, p. 254). In EQII and other MMOGs, connections with other players offer the resources to reach

86 instrumental goals as well as the pleasure of the company of others. As such, extending the argument in Monge, Heiss and Margolin (2008), the selective retention of ties is determined by their propensity to sustain themselves as well as the extent to which they can be valuable in supplying instrumental benefits as well as good company, such that the ties that are less costly to maintain and are more beneficial to the two nodes involved are more likely to survive. In summary, under the evolutionary framework, personal ties in EQII undergo variation, selection and retention processes. Variation refers to the many blind and haphazard opportunities to establish new connections. Because individuals have limited relational carrying capacity, these new connections (variations) have to compete for opportunities of survival, driven by various selection forces based on both the potential benefits of these ties and the cost of tie maintenance. Taken together, the evolutionary perspective explains how and why particular ties are preserved while others are eliminated. In the following, three sets of selection criteria are discussed in relation to tie survival and decay: aging and inertia, social architecture, and homophily and proximity. Aging and Inertia As the evolutionary perspective suggests, many individual relations in EQII originate from random chance and haphazard encounters. One character plays with another because they happen to be logged on at the same time, on the same server, at similar character levels, located in the virtual vicinity, or have heard each other’s cry for help. These random factors bring people together regardless of individual preferences. Therefore, relationships generated by random variations will often connect individuals

87 who would not otherwise enjoy each other’s company or collaborate well. Players of EQII and other MMOGs usually have diverse motivations and playing styles (Bartle, 1996; Yee, 2006c), The possibility of finding the ideal sidekick or even building the perfect alliance out of strangers is extremely slim. Thus, when people discover that their relations are less satisfactory in realizing their instrumental and social goals, it is natural to disengage in favor of more compatible contacts. Through this selection process, more compatible players replace existing incompatible ones, leading to the strengthening of some ties and the weakening, or decay, of others. Following the logic of relational cost and relational carrying capacity, individuals are constrained in their ability to develop and maintain interpersonal connections. To achieve a certain relational goal (e.g., acquiring certain expertise from a fellow player), it is usually more costly to initiate a new relation with a previously unknown partner than to exploit existing network resources (Lazer & Friedman, 2007). Therefore, individuals are motivated to “recycle” their existing contacts over exploring new ones. So instead of assembling a new team every time to accomplish collective tasks, individuals are more likely to resort to their past ties for potential collaboration opportunities. Over time, these ties are then filtered based on their perceived benefits, with only a subset of them selected and reproduced in future interactions. As the selection and retention processes continue, those repeated ties are strengthened and take more time and network energy to maintain than one-off ties, leading to decreased ego-network size. In the organization literature, this is often referred to as “liability of newness,” which derives from the lack of time available for new or young organizations to learn

88 how to perform reliably and establish legitimacy within the community (Carroll & Hannan, 2000). New players in EQII also suffer from this learning curve disadvantage. Even if they may have played similar MMOGs before and carry with them knowledge and experiences that may be transferrable to another MMOG, EQII still represents an unfamiliar territory where rules, mechanics, and strategies all have to be learned through experimentation. Therefore, it is expected that novice players will tend to form a large number of weak ties to experiment with connections with different sets of partners. This stage of experimentation represents an important “trial-and-error” process. During any given time interval (e.g., a week), compared to more seasoned players, novice players tend to cast a wider net by engaging in social interaction with more contacts, resulting in a larger network. The promiscuity of connections tends to wane as players advance in the game and have a better understanding of the world. This “trial-and-error” process is manifested in the changing size of players’ dynamic social networks in the virtual world. Therefore, H9: The size of players’ social networks shrinks as they advance in EQII. Similarly, as players spend more time playing the game, they not only are better at identifying potential compatible partners, they are also better at keeping them by developing routines and strategies. According to Hannan and Freeman (1989), “…new organizations typically rely on the cooperation of strangers. Development of trust and smooth working relationships takes time, as does the working out of routines. Initially there is much learning by doing and comparing alternatives. Existing organizations have an advantage over new ones in that it is easier to continue existing routines than to create

89 new ones or borrow old ones (Nelson and Winter, 1982: 99–107). Such arguments underlie the commonly observed monotonically declining cost curve at the firm level, the so-called ‘learning curve’” (p.80). In interpersonal contexts such as EQII, seasoned players become better connectors, because they have developed social routines of online interactions, and are also at an advantage in identifying potential partners that are compatible. Thus, ties involving players who have spent more time playing EQII have a better chance of survival4. H10: Ties involving players who have spent more time playing EQII are less likely to decay. An example would illustrate how the processes captured in H9 and H10 occur simultaneously. A novice character at level 10 was relatively unfamiliar with the game. In order to search for the perfect collaborators, he experimented extensively with different sets of partners, often through ad-hoc groups (pugs), resulting in a large network of 200 contacts during one week of play. Due to their experimental and ad-hoc nature, these 200 ties were extremely fragile. 180 (90%) of them were discarded right after the collective task was over. Through this “trial-and-error” process, this character became a more discerning connector in that he accumulated more experience and knowledge in identifying the attributes of compatible partners for successful collaboration and friendly relationship. Therefore, when he leveled up to level 50, the size of his network shrank— he only interacted with 50 people during one week of play (as opposed to 200 when he

4

Even though “survival” and “decay” describe two distinct consequences of tie evolution, they are theoretically and analytically the same phenomenon. A better chance of survival means less likelihood of decay. In this dissertation I chose to use “decay” in the hypotheses and analyses to be consistent with prior research (Burt, 2000a).

90 was at level 10). An increasing proportion of these contacts were from regular groups or raids, which were emergent structures from repeated interactions as players “recycled” their contacts. Moreover, because of his acquired experience in selecting and retaining compatible partners, the 50 ties he maintained in the network were also more likely to persist in the future. Only 20 (40%) of them were terminated after this week of play, and the rest were repeated in future interactions. Together, H9 and H10 represent a dynamic process of tie initiation and termination as players advance in levels and spend more time in developing strategies for linking. H9 focuses on the experimentation of partner choices and the process of “recycling” and repeating past social ties. H10 captures the maturation of players during which they learn to identify compatible partners as well as the social routines to preserve them. Learning not only occurs as players accumulate experiences in the game, it also occurs when a specific relationship matures over time (Burt, 2000a; Carroll & Hannan, 2000). Burt (2000a) argued that the rate of tie decay slows over time, because the longer a tie has survived, the more likely that the two individuals involved are compatible. In addition, as two individuals are engaged in a relationship, they have more time and opportunities to learn the social routine of working together and to appreciate the benefits of the connection (Burt, 2000a). In other words, compatibility or tie fitness grows as ties age. In the context of EQII, it is difficult to complete a challenging task with “pugs,” or pick-up groups that consist of players who have never collaborated before, because there is little shared understanding of the tasks at hand, the play styles of other members, or the norms of coordination, leadership and reward distribution. Rather, players prefer to group

91 or raid with regulars who they have collaborated with before, because social routines of playing together take repeated interactions to build up. It is not surprising that many guilds or regular raid groups require a trial period, during which a new recruit has to collaborate with existing members in less important tasks before they can join collective activities of greater stake. This argument is consistent with the idea of network inertia, the phenomenon that networks tend to preserve and reproduce past structures (Gulati & Gargiulo, 1999; Ramasco & Morris, 2006; Walker, Kogut, & Shan, 1997). Similar to the ecological argument of structural inertia, which posits that organizational structures tend to stay unchanged because selection forces favor reliable and accountable systems (Hannan & Freeman, 1984), networks also tend to stay unchanged, because inert relations help provide rich and reliable information, build mutual trust and attraction, and constrain partners’ opportunistic behaviors. Inertia is essentially a natural process to preserve and accumulate social resources from existing structures (Walker et al., 1997). This phenomenon has received empirical support from studies in many contexts, including the open source project team formation (Hahn, Moon, & Zhang, 2008), the formation of biotechnology start-up networks (Walker et al., 1997), interorganizational strategic alliances (Gulati & Gargiulo, 1999) and the network of HIV/AIDS non-governmental organizations (Shumate et al., 2005). In online gaming communities, a similar process of network inertia and tie decay is expected. Therefore: H11: The longer a tie has been maintained between two players, the less likely it is to decay.

92 Social Architecture An important set of selection forces are external to the entities vying for survival. In organizational evolutionary literature, these forces may consist of environmental shocks, market forces, and conformity to institutionalized norms (Aldrich, 1999). For example, dramatic restructuring of the environment, in the form of technological and regulatory shifts, could produce significant differences in selective retention, resulting in the death of once-successful organizations and the rise of others (Haveman, 1992; Tushman & Anderson, 1986). Similarly, some selection forces of social ties in virtual worlds are also external to the ties and nodes involved. They stems from the opportunities and constraints inherent in the social environment of EQII. Just like any other social spaces, such as bars, plazas, and shopping malls, the social world of EQII also has norms and rules—imposed by game mechanics— that constrain certain types of interactions but incentivize others (Lessig, 2006). In fact, one of the most fundamental mechanisms to promote sociability—the division of labor among characters, as manifested in the character class system—is deeply embedded in almost all aspects of social interactions. Each character class is associated with a distinct set of skills and strength for combat. Players chose to belong to one of four general class archetypes – Fighter, Priest, Scout, and Mage – and usually assume different roles in collaborative activities: Fighters can serve as the main "tank" of the group which absorbs the attack damage from monsters and protects other group members; Priests are healers who restore the health and strength of group members; Mages and Scouts use different skills to attack monsters. EQII game mechanics

93 specifically encourage collaboration among players with different character classes. By combining skills in defense, self-preservation, and offense, players can implement different fighting strategies. Indeed, advertisements from groups “LFM” (looking for more) usually specify the skill sets and classes they are looking to fill, which are usually different from what groups already have. As dictated by game design, a tie between players of different classes brings more potential benefits than a tie between players of the same class, thus is less likely to discontinue. Therefore, H12: Ties between players of the same character class are more likely to decay. Another constraint is the level of experience. As discussed previously, the game play experience in EQII is highly structured—players usually follow the same general path of progression. As quests, monsters, and zones are all aligned by level, players would have to tackle them in a relatively linear order. This imposes a level hierarchy on the formation of social interactions. Collaborative play in groups or raids requires that all the members have similar levels, otherwise the higher level members tend to absorb most of the experience gain—a major objective for playing, which discourages collaboration among players with considerable level disparity. Individuals who have reached similar levels are usually given the same quests and tend to be exploring the same zones to accomplish these tasks. In addition, players may find others more compatible if they have relatively similar levels in EQII, because they have comparable knowledge, status, and familiarity with the game community. In other words, level similarity ensures that players are “on the same page” in their character development, which reduces potential conflicts or misunderstandings. A special exception to this level hierarchy is the mentoring

94 mechanism. In order to group or raid together, a higher level player could “mentor” a lower level player, so that the mentor temporarily lowers the character level to match the apprentice’s character level and earns experience points at a reduced rate. In other words, character level is downward compatible, but such compatibility comes at a significant cost of slower experience gain and, very likely, repeated game content for the mentor, leading to ephemeral rather than long-lasting relations. Therefore, H13: Ties between players of similar character levels are less likely to decay. Whether two players belong to the same guild could also influence tie decay. Guilds are persistent self-organized player associations. Although the level of sociability within guilds can vary widely, guilds provide a reliable way to recruit team members to tackle difficult game objectives (Ducheneaut, Yee et al., 2007; Williams et al., 2006). Compared to any random stranger, guildmates typically provide some basic familiarity, shared game ethic and goals (Williams et al., 2006). Joining a guild is a mutual-selection process, in which both the applicant and the guild evaluate their compatibility. Many guilds provide a detailed description on their website or in the recruitment advertisement. For example, the guild “Ancient Legacy” is an active raiding guild that has scheduled raids three days a week. Its description suggests that it has a high commitment to raiding: “We do not have any attendance requirements, but applicants should be interested in raiding and generally available during our raid times. We expect you to make an effort to attend our raids, but recognize that real life sometimes intervenes” (Ancient Legacy, 2010) . By contrast, the guild “Blackhawks” does not have any expectations on raiding but instead focuses on creating a casual, friendly community: “We exist for the purpose

95 of helping others - both in and out of guild. A blackhawk is someone who will race across three zones to help you kill a quest mob or offer a ressurrect[sic] to a complete stranger...The Blackhawks are dedicated to mutual advancement, fun, profit, and adventure” (Blackhawks, 2010). Admittance into the guild usually involves communicating with a guild officer, who provides detailed information about the guild and makes judgments about the applicant’s fit with the guild. Sometimes sponsorship from an existing member is also required. Therefore, through the selection and admittance into the same guilds, two players are more likely to find each other as compatible game partners. In addition, guild members have access to an array of functionalities that further support sociability. For example, the guild name is automatically attached to one’s character name, so that it is easier to identify guildmates during random encounters in the game. All guild members are automatically subscribed to the guild chat channel, in which every member can broadcast chat messages to every guild member currently logged into the game. Many guilds have a shared in-game space, the guild hall, where people go regularly to enjoy its amenities (such as teleport stations and NPC merchants) or just to hang out. Finally, many guilds also organize collective activities, such as scheduled raids, and such information is delivered to each member via in-game mailbox or is discussed through the guild chat channel. Based on the above reasoning, the following hypothesis is proposed: H14: Ties between players in the same guild are less likely to decay. Homophily and Proximity

96 Another set of selection mechanisms stems from the similarity (or difference) between the two individuals, which has long been found to exert a significant impact on tie formation and dissolution (Burt, 2000a; , 2002; McPherson, Smith-Lovin, & Cook, 2001). In the organizational evolution literature, this set of mechanisms falls into the category of internal selection forces. They represent “pressures toward stability and homogeneity” (Aldrich, 1999, p. 22). The need for consistency and cohesion within groups and organizations represents an inherent evolutionary force. As Campbell (1969) suggested, “any social organization tends to move in the direction of internal compatibility independently of increased external adaptiveness” (p. 76). In the context of social ties, the similarity (or difference) between two individuals forms the basis of their compatibility and the cost of interaction, as suggested by the theory of homophily and proximity. The word “homophily” was first coined by Lazarsfeld and Merton (1954), referring to a tendency for people to be attracted to others who have similar attitudes, beliefs, and personal characteristics. Monge and Contractor (2003) summarized two lines of theoretical underpinnings of homophily: the similarity-attraction hypothesis and the theory of self-categorization. The similarity-attraction hypothesis postulates that people are more likely to interact with those who have similar traits (Byrne, 1971). Selfcategorization theory argues that people tend to self-categorize with regard to race, gender, socio-economic status, etc., and they differentiate between similar and dissimilar others based on these attributes (Abrams & Hogg, 1999; Turner, 1987). Simply put, homophily is well illustrated by the old saying “birds of a feather flock together.”

97 Homophily, especially with regard to gender, ethnicity, and occupation, has been found as a critical factor of relationship formation in entrepreneurial teams (Ruef, Aldrich, & Carter, 2003), work team composition (Hinds, Carley, Krackhardt, & Wholey, 2000), as well as the formation of social networks in general (McPherson et al., 2001). Recent research of online social networks suggests that homophily is a strong predictor of relationship formation even if people are interacting through computer-mediated communication (Adamic, Buyukkokten, & Adar, 2003; Yuan & Gay, 2006). A special case of homophily is the mechanism of proximity. Since the 1930s, researchers have studied how spatial proximity affects constructs such as friendship, romantic relationships and other variables like amount of communication (Festinger, Schacter, & Back, 1950). When the spatial distance between workers reached 30 meters and beyond, their frequency of spontaneous communications dropped drastically (Allen, 1977). With the advent of communication technology, scholars began to research whether spatial proximity still influences computer-mediated communication (Hampton & Wellman, 2001; Kraut, Egido, & Galegher, 1988). This line of research generally looks at teams spread across geographic distances and how computers mediate this interaction. In general, it was found that individuals who are located closer to each other are more likely to communicate than individuals who are located further away, regardless of whether or not communication is face-to-face (Eveland & Bikson, 1986). Even taking into account organizational proximity and similarity in interests, it was found that individuals located closer were more likely to collaborate with each another (Kraut et al., 1988).

98 Theory of homophily posits that people of the same attributes tend to interact with each other because of their common background and shared interests, while proximity predicts that people located in the same area are more likely to communicate. Although in online gaming communities, players’ demographic attributes and location information are not immediately visible, other players can still detect behavior patterns related to their attributes and the common background. For example, male players are more likely to choose male characters and engage in combat activities. Furthermore, through online interactions and social activities, players may exchange some personal information and get more familiar with each other. Therefore, in online gaming communities, homophily and proximity can play a considerable role in tie selection. Players of the same demographic attributes and geographic location are more likely to find each other compatible, thus they tend to stay in the relationship rather than abandon it. Therefore, H15: Ties between players of the same gender are less likely to decay. H16: Ties between players of similar ages are less likely to decay. H17: Ties between players of geographic proximity are less likely to decay. All the hypotheses are summarized in Table 2.

99 Table 2. Summary of Research Hypotheses Network Patterns H1 Age is related to players’ level of engagement in social networks. H2 Gender is related to players’ level of engagement in social networks. H3 Character level is positively related to players’ level of engagement in social networks. H4 Character class is related to players’ level of engagement in social networks. H5 Guild membership is positively related to players’ level of engagement in social networks. Network Effects H6 Brokerage in players’ social networks is positively associated with their task performance. H7 Closure in players’ social networks is positively associated with their trust towards guildmates. H8 Closure in players’ social networks is positively associated with their sense of community. Network Evolution H9 The size of players’ social networks shrinks as they advance in EQII. H10 Ties involving players who have spent more time playing EQII are less likely to decay. H11 The longer a tie has been maintained between two players, the less likely it is to decay. H12 Ties between players of the same character class are more likely to decay. H13 Ties between players of similar character levels are less likely to decay. H14 Ties between players in the same guild are less likely to decay. H15 Ties between players of the same gender are less likely to decay. H16 Ties between players of similar ages are less likely to decay. H17 Ties between players of geographic proximity are less likely to decay.

100 Chapter 3: Methods Data Sources To date, researchers of communication and social networks have been relying either on estimates and sampling for large-scale research, such as field surveys (e.g., McPherson et al., 2006), or on direct observation for small-scale research, such as case studies of small networks (e.g., Mercken, Snijders, Steglich, Vartiainen, & de Vries, 2009) and controlled experiments in laboratory settings (e.g., Walther, Van Der Heide, Hamel, & Shulman, 2009). Although both approaches are useful, surveys often suffer from sampling biases. Findings from case studies and laboratory experiments may be difficult to generalize to different cases and social contexts Network data collected through self-reports are often limited by participants’ memory and cognitive capacities (e.g., sociometric surveys asking for names of 5 friends instead of all friends). In addition, as human subjects are usually aware of the research being conducted, natural attitudinal and behavioral responses collected from human subjects are often inseparable from those induced by social desirability and reactivity, known as the Hawthorne Effect (Cook & Campbell, 1979). Large scale server data from virtual worlds represent an exciting opportunity to resolve the above problems. Data of the whole population, instead of a small subset of the population, can now be collected and accessed at a relatively low cost. This provides an exciting opportunity for network research, because every node and every tie among these nodes are included in the data, making it possible to construct complete large-scale networks and to trace their evolution over time. In essence, virtual worlds are excellent

101 sites for natural observation, where researchers are nearly omniscient about every aspect of player behaviors, at least within the virtual world. In addition, this recording happens without the players’ knowledge, so demand conditions cannot be a factor (Webb, Campbell, Schwartz, & Sechrest, 1966), eliminating any Hawthorne Effects (Cook & Campbell, 1979) or social desirability issues. Practically speaking, this is an unobtrusive methodology that has not existed to this extent before. Although there have been attempts to conduct such work, researchers have been stymied by the simple problem of access. With very few exceptions (e.g., Feng, Brandt, & Saha, 2007), game companies have been largely closed to researchers. This dissertation takes advantage of a rare opportunity to combine two distinct data sources in the analyses. The game operator, Sony Online Entertainment (SOE), provided behavioral server data from the game’s large back-end databases and facilitated the distribution of a survey to a large sample of players. Therefore, the existing dataset used in this dissertation consists of two parts: 1) unobtrusively collected game-based behavioral data in the form of server logs from the EQII database, and 2) survey data from around 7000 players. These two components are connected by unique identifiers assigned to each player account as well as character. Although this dataset also has a few inherent limitations (as discussed in this chapter), it represents an early attempt to conduct communication research utilizing electronically-captured behavioral data in conjunction with self-reported survey data.

102 Behavioral Logs EQII utilizes an elaborate log system that constantly collects data on almost all individual and collective activities occurring within the game, such as economic transactions, in-game communications, questing, combating, crafting, and so forth. The total size of the original dataset was 1.14 terabytes. It was provided directly from SOE and was then stored in an Oracle relational database hosted at the National Center of Supercomputing Applications at the University of Illinois, Urbana-Champaign. The database was accessed remotely via direct SQL queries. The data base contains 23 tables, including 1 table of chat communications, 6 tables of static information on players, characters and guilds, and 16 tables of dynamic ingame activities with time stamps. Each player (account) can create multiple characters in the game up to a maximum of eight. So the character, instead of the account was selected as the main unit for most analyses in this dissertation. The time range of 22 out of the 23 tables is from January 1st, 2006 to September 11th, 2006, while the chat table has a much shorter and sporadic time range from August 27, 2006 to September 10th, 2006. The database contains complete information on players of four EQII servers, Antonia_Bayle (role-playing, PvE), Guk (PvE), Nagafen (PvP), and The Bazaar (exchange-enabled, PvE), each representing a unique server type. This dissertation chose to concentrate on player dynamics on the server Guk, because it represents the most emblematic server type of MMOGs: Player-versus-Environment. Among all the data in the behavioral server logs, five tables were used for analyses in this dissertation—three tables of dynamic activities (Economy, Experience,

103 and Chat) and two tables of static attributes (Character_Store and Demographics). The Economy table records all economic transactions conducted between players and between players and NPCs. A typical entry in the Economy table includes a timestamp and information on the buyer (player or NPC), the seller (player or NPC), the item traded, the quantity of the trade, the item price, and the type of the trade (e.g., brokered-trade, direct player trade, etc). The Experience table records information on the accrual of experience points. Every record describes an event during which a certain amount of experience points are gained. A typical entry in the Experience table includes a timestamp, the ID of the character, the character level at the time of the event, the zone in which experience points are gained, the size of the group of which the character is a member, the amount of experience points earned, and the reason for gaining the experience points (e.g., completing a specific quest). Unlike the Economy and Experience tables that are EQII-specific, the Chat table records all the chat communications among players in EQII and other SOE games, as a player could simultaneously chat with another person in a different game. A typical entry in the Chat table has a timestamp and information on the sender, the receiver, the type of chat (e.g., send an instant message to another player, check friend status, send a broadcast message, etc), and length of the message, but not the content of the message. Because the Economy, Experience and Chat tables track the dynamic behavior of players over time, each character normally has multiple records corresponding to their activities in the game.

104 Two tables in the database provide static, or cross-sectional, attributes of EQII players for data analyses. The Demographics table contains basic demographic information for each player, including their date of birth, language, the date of account creation, gender, zip code, and their country and state of residence. Information in this table was supplied by players at the time of account creation. The Character_Store table contains information on each character’s attributes in the game. A typical entry includes the ID of the character and information on its server, character class, race, current character level, current experience points accumulated, current amount of in-game currency, guild affiliation, amount of total time spent playing the particular character, current character attributes such as health and power, total number of quests completed, and so on. Unlike in tables of dynamic activities such as Economy and Experience, the Demographic and Character_Store tables only include one record per character. These attributes in the Character_Store table update as the character advances in the game, but each character only has one entry, reflecting the current status of the character instead of its development trajectory. For example, if a character reached a higher level, say, from level 20 to 21, the Character_Store table then erased its original level of 20 and updated the variable with the new value of 21. Therefore, both tables contain cross-sectional rather than longitudinal data. The attributes recorded in both tables were updated as of September 2006. The static nature of these attributes imposed certain restrictions on the analyses.

105 Player Survey The second component of the dataset is a player survey, which was administered to obtain attitudinal and demographic information about the players. In February 2007, the survey was distributed to a large sample of players through four EQII game servers (Williams et al., 2008). Although some players used multiple characters, prior research suggests that these players tend to have a “main” character that is played most often (Williams et al., 2006). The main character for each player account was identified by cross-checking the game database. These characters then populated the sampling frame and were evenly distributed in the four game servers where behavioral server logs were provided by SOE, Antonia_Bayle (RP), Guk (PvE), Nagafen (PvP), and The Bazaar (EXC). Potential respondents were invited and directed to a secure web survey if they logged into the game within the survey time window. Respondents were given a special in-game item, the “Greatstaff of the Sun Serpent,” created by SOE as a unique incentive for completing the survey. The item was desirable for all players regardless of their classes and levels because of its rarity and usefulness in combat. The survey lasted around 25 minutes. Based on previous survey studies in this area using cash incentives or no incentive (Yee, 2006c), it was expected that the planned sample size of 7,000 respondents would be reached in one or two weeks. Instead, the data collection process only took two days, proving that the special in-game item was desirable for all and an effective tool for recruitment. With these procedures, the current sample is less prone to self-selection bias typical of a convenience sample.

106 Like behavioral logs, the survey data were recorded anonymously as no realworld identifying information was attached to the data. The survey data were then transferred and stored in the same Oracle relational database. All the survey respondents were identified by their unique account number within EQII, through which the survey data were linked with behavioral log data collected from game’s back-end databases. Because the survey was taken by a fraction of the players, the sample used in those analyses combining both behavioral data and survey data is a subset of the total sample. In summary, the dataset consists of two parts: 1) behavioral logs, including three tables of dynamic activities and two tables with static attributes, and 2) survey data of player attitudes and demographics. It should be noted that these data sources differ in their respective time range (see Table 3), putting a constraint on the availability of data for certain types of analyses. The time range used for each analysis is specified in the “Hypotheses Testing” section. Table 3. Time Range of Data Sources Raw Data Time Range Experience table January 1,2006 – August 31, 2006; September 4, 2006 – September 10, 2006 Trade table January 1,2006 – August 31, 2006; September 4, 2006 – September 10, 2006 Chat table August 27, 2006 – September 10, 2006 Character_Store table September 10, 2006 Demographics table September 10, 2006 Player survey February 2007

107 Data Transformation Network Construction Time Interval The three dynamic tables in the behavioral logs, Economy, Chat, and Experience, trace player activities at a one-second temporal resolution. In order to construct networks based on these dynamic activities, it is necessary to reduce the temporal resolution and select a meaningful time interval as the basic unit of analysis. Previous research shows that participation in online gaming communities has regular cycles (Feng et al., 2007). A longitudinal study of EVE Online, a popular MMOG, found that the number of active players reaches a strong daily peak during the evening and a moderate weekly peak around Saturday and Sunday, but week-to-week variations were quite small (Feng et al., 2007). These temporal trends are not surprising, as they reflect precisely the after-work and after-school hours and weekends when most game play may occur. Because the social dynamics in EQII would vary according to the active characters in the game world, it is important to select a time interval large enough to accommodate these temporal fluctuations. Therefore, this dissertation uses the week as the basic time unit for network construction and analysis. The networks of trade, chat, grouping, mentoring, and collaboration were created by aggregating all records during every week. A quick empirical evaluation of the active characters validated this approach. As shown in Figure 6, the size of the active population on the Guk server remained relatively constant through 13 consecutive weeks (June 1, 2006 to August 30, 2006). Most week-to-week variations were less than 10%.

108 Figure 6. Active Characters on Guk from June 1 to August 30, 2006

Trade Network EQII players can gather and produce in-game items and trade for other desirable in-game items or exchange them for in-game currency. Although trade activities are supported in many ways in EQII, most of them do not involve direct player-to-player interactions. For example, players can buy and sell items at certain NPCs, known as merchants. These merchants are computer-controlled characters and such transactions only involve a single player. Another way to buy and sell items is through the consignment system, which operates through a computer-controlled broker. A player could go to a broker NPC and place items with that NPC at desired prices. Other players may visit the NPC, who lists the items for sale, and browse all the available items. If a transaction is completed, the selling player would receive the money minus a broker service fee. Although the consignment system facilitates player-to-player trade, it does

109 not involve simultaneous interaction between the buying and selling players. It is still an impersonal system where the buyer does not know the seller, and vice versa. The only genuine type of player-to-player trade happens via virtual face-to-face transaction. A player can right-click on other players and invite them to trade. Both parties need to agree upon the item and the price before completing the trade by clicking “Accept.” Therefore, virtual face-to-face transaction is based on direct social interactions between buyers and sellers. The two parties involved in a trade need to at least know each other’s names and meet virtually “face-to-face” in the game world to finish the transaction. Because this dissertation concentrates on social network dynamics of EQII players, it is appropriate to focus only on the direct interactions among players. Therefore, a trade network was constructed based on players’ virtual face-to-face transactions in EQII. A player trade relation was constructed as an undirected tie if two players have completed one or more transactions during a week. In the trade network, nodes represent individual characters and edges are their transactions. The strength of a trade relation was calculated by the number of transactions occurring during the time period between one particular dyad. The network data were prepared by searching face-to-face transactions in the Economy table, extracting seller ID and buyer ID, and aggregating the number of transactions between the same buying and selling parties within a specific week, thus creating a valued network.

110 Grouping Network In the world of EQII, players are allowed to form "groups" among players in order to complete challenging tasks and play together. The game mechanics require that the players who form a group have to be at very similar character levels in order to earn experience points together. If the character level of one player is significantly lower than others, that player cannot receive experience points or quest rewards and has to be mentored (discussed in the next section). EQII allows players to look for groups to join through the “Looking For Group (LFG)” function. Players are able to specify their responsibilities in the group, such as whether they are willing to mentor other players, and what roles they are willing to undertake (e.g., primary tank, primary healer, etc). Alternatively, existing groups are also able to advertise their current task and try to find more players through the “Looking For More (LFM)” function by specifying their needs and expectations of potential group mates. Characters have to meet physically in the game world to group. One character could send an invitation to another character to group, and once grouped, characters are also free to leave the group. Several functions are available in EQII to facilitate collaborative play in groups. For example, players have access to a group chat channel where they can freely communicate with their group mates. They can also negotiate and select a specific option to distribute loot. If two or more players group together and earn experience points in combat or quest activities such as fighting monsters, a grouping relation was constructed between every pair of players. Such relations were used to construct an undirected graph where nodes represent individual players and edges are their partnership. The grouping network

111 is essentially an affiliation network where every member is connected to everyone else in a group. To prepare the network data, all records in the Experience table were systematically examined. If several records had exactly the same timestamp, the same location in the game (as represented by location ID), the same group level, and the same reason for gaining the experience points (except mentoring), characters involved in these records were considered to be grouping together. This assumption was further validated by cross-checking the group size variable recorded in the Experience table. A grouping relation was then recorded between each pair of players in the group. The strength of the grouping relation was calculated by aggregating the number of grouping instances within a specific week. Mentoring Network Mentoring is a mechanism by which an experienced character plays with a lessexperienced character to help the less-experienced gain points to quickly progress to higher character levels. In a mentoring event, the experienced player, the mentor, lowers the original character level temporarily to match the lesser player, the apprentice. The mentor pays for this by earning a discounted number of points from play and the apprentice gains a 10% experience bonus. Mentoring is thus a form of help in which an experienced character donates time and effort to enhance a lesser character. Mentoring is a special case of grouping when two or more players want to collaborate but they have disparate character levels. If one character A mentors another character B, a directed relation was constructed from A and B. To prepare the network data, the Experience table was

112 searched for records with experience gain due to mentoring. Those characters in these records were identified as mentors. A mentor’s apprentice was identified as the character who earned experience points at exactly the same time and the same location and the record was adjacent to that of the mentor (as represented by a consecutive Sequence ID) in the Experience table. This relation was further validated by checking whether the mentor’s character level was lowered—as represented by the raw variable PC_Effective_Level in the Experience table— to match the apprentice’s character level. The strength of the mentoring relation was calculated by aggregating the number of mentoring instances between two players within a specific week. Collaboration Network As discussed above, mentoring is a special case of grouping where characters have disparate levels. Therefore, in order to examine social dynamics of player collaboration, a collaboration network was created by combining the grouping and mentoring networks within the same time interval. The collaboration network did not distinguish mentoring from grouping and treated both as instances of collaboration. The network data were prepared by simply aggregating instances of mentoring and grouping between any pair of players. Chat Network SOE provides a universal chat system for all its online games including EverQuest II. Two types of chat were recorded in the Chat table, broadcast messages and private instant messages. Broadcast messages could be sent to a specified channel, such as a character class channel (e.g., wizard) or a level channel (e.g., level 10-19), so that

113 everyone who subscribes to that channel automatically would receive the message. Broadcast messages represent one-to-many communication. They do not have a designated target and are often impersonal. Private instant messages, on the other hand, are exchanged between two specific characters. Senders and receivers of private instant messages know each other’s screen names and, possibly, their off line names. Representing one-to-one communication, private instant messages are exclusive and personal. Because this dissertation concentrates on the social dynamics of players, only private instant messages were included to construct a chat network. It should be noted that the Chat table used in this dissertation does not include group chat or guild chat. EQII deploys a different chat system to facilitate communication within groups and guilds, but such data were not available in the current dataset. Messages sent to the group or guild would reach every group or guild member who was currently logged in. Therefore, even though private chat, group chat and guild chat are all available to players who are affiliated with a group and/or a guild, only the private communication between any pair of players was observed and recorded in the current dataset. As such, the chat relations constructed here represent the most private, exclusive and intimate conversations between two players rather than the casual and inclusive chat among groups of characters. A chat relation was constructed as a directed tie if one character A sent one message to another character B during a specific time interval. To prepare network data, the Chat table was searched for private instant messages sent and received by characters on the Guk server. The character IDs of the senders and receivers were then extracted to

114 create the network. The strength of a chat relation from character A to B was calculated by aggregating the number of messages sent from A to B during a specific week. Active Players In order to examine the extent to which players are engaged in social activities, it is important to establish a population base of online players within particular time intervals. In other words, a necessary first step for understanding social dynamics is to know the total number of players in the EQII world who are available for social interaction. Such a procedure would require a dynamic count of characters logged in the game on the Guk server. However, the EQII database does not supply information about active characters. Therefore, an active character was defined as someone who had at least one entry in the Experience,.Chat, or Economy tables. Applying this logic, the list of active characters on the Guk server was created for each week, providing a basic population count. Network Measures Network Size (Degree) An important measure of the extent to which players are engaged in social interactions is the size of individual networks. In network terms, an individual’s social network is called an ego-network, consisting of an ego (the focal person) and a set of alters who have ties to ego (Wasserman & Faust, 1994). Network size was measured by counting the number of unique alters in one’s ego-network, excluding the ego. This measure is essentially the same as degree centrality, or degree, of the ego-network (Freeman, 1979; Wasserman & Faust, 1994).

115 Brokerage In a series of publications, Burt (1992, 2000b, 2005, 2010) has explicated the definition and measurement of network brokerage and closure. Brokerage, the process of building connections across structural holes, provides one with exposure to variations in information and opinions, because disconnected people (structural holes) are more likely to possess different ideas and behaviors. Therefore, brokerage measures the extent to which one is connected to otherwise disconnected people, or the extent to which alters are indirectly connected through the ego. Following the procedure in Burt (1992, pp. 51-54), brokerage was measured in this dissertation as the effective size of the ego’s network. Effective size is the count of bridge relations. Ego’s relationship with contact j is considered a bridge if j has no connection with any of ego’s other contacts. Therefore, in an ego-network, the minimum effective size is zero (all of ego’s contacts are connected among themselves) and the maximum effective size is the total number of contacts (none of the alters are connected to any other alter, thus every tie is a bridge). Further taking tie strength into consideration, the nonredundant portion of ego i’s connection with j, or the extent to which a tie is a bridge, can be illustrated the following formula:

1 − ∑ piq m jq , q ≠ i, j q

where piq is the proportion of ego i’s time and energy invested in the connection with alter q, mjq is the marginal strength of j’s relationship with q within ego’s network (explained below). Effective size was then calculated by summing the nonredundant portion (the extent to which a tie is a bridge) across all the j contacts in ego’s network, using the following formula:

116

∑ (1 − ∑ p j

iq

m jq ) , q ≠ i, j

q

A key variable in the calculation of effective size is marginal tie strength. Because the raw count of tie strength (how many times a relation occurs between two characters) can vary widely from character to character, the raw count was further normalized into a measure of marginal strength, also following procedures presented in Burt (1992, 2010). For example, a trade relationship of three trades is strong for a character who at most made four trades with anyone (three trades would be .75 marginal strength), but would be minor for a character involved in 100 trades (three trades would be .03 marginal strength). The strength of ego’s relation to alter was measured relative to ego’s strongest relation: the marginal strength is a proportion equal to the number of trades between ego and alter divided by the maximum number ego made with anyone. The marginal strength was then used to compute effective size. Closure Closure describes the process of staying within closely connected groups. As connected people tend to share the same opinion and information, closure protects individuals from diversity. Closure measures the extent to which everyone is connected to everyone else in the network. This dissertation adopted a widely-used composite measure of network closure: network constraint. This measure was computed following the same procedure presented in Burt (1992, pp. 54-65). Network constraint gauges the extent to which ego’s time and energy is concentrated on already-connected alters in the egonetwork. The constraint index was computed for each individual node respectively as the

117 sum of network constraint across all the contacts in one’s ego-network, using the following formula:

∑( p j

ij

+ ∑ piq p qj ) 2 , q ≠ i, j q

where pij is the proportion of ego i’s time and energy invested in the connection with alter j, pqj is the strength of q’s tie to j, and total network constraint in i’s ego network with j alters is defined as the sum of the individual network constraint of i across all the alters. To align the scale of this measure with other variables, the constraint score (originally ranging from 0 to 1) was multiplied by 100 and rounded to the nearest integer. A large score of network constraint signals a high degree of closure in an individual’s egonetwork. The trade and mentoring networks from January 1 to September 10, 2006 were used to compute effective size and network constraint. The decision to use these two networks during this time range was based on several considerations. First, the raw behavioral data used to create player networks were activity-based, so that the networks were only reflective of the transactions completed during a specific time interval. For example, a trade network of September 4-10, 2006 only includes those transactions during this particular week but not past transactions. Brokerage and closure, however, operate through a cumulative process, requiring a relatively complete picture of the network structure (Zaheer & Soda, 2009). A trade tie between characters A and B should be understood in the context of their alters and the connections within. Because variations in dependent variables—task performance, trust and sense of community—tend to occur over an extended time period (longer than a week), the network structures producing

118 these variations should also be considered in a longer time frame. A weekly snapshot of trade or mentoring activities, therefore, would result in disproportionally more structural holes than there actually are. A prolonged observation window, rather than the basic time interval of a week, is necessary for measuring brokerage and closure. Thus, the whole time range of the behavioral logs, from January 1 to September 10, 2006, was used. Three networks—trade, grouping and mentoring—were available based on the whole observation window. Second, although three networks are available during the prolonged time window, trade and mentoring networks were selected to generate network effective size and constraint measures because they consist of dyadic relations. A trade or an instance of mentoring happens between two characters and only between two characters. Any trade or mentoring relation between two characters is independent of other trade or mentoring relations in the network. The grouping network, however, is an affiliation network by definition. Within a group, people are connected to everyone else, thus leading to extremely high constraint scores. Therefore, effective size and network constraint scores were computed on trade and mentoring networks but not the grouping network. Third, since there were two networks, one option was to combine both the trade and mentoring networks into a comprehensive network of players and compute effective size and constraint measures. The Quadratic Assignment Procedure (QAP) was used to test the association between the trade and mentoring networks on a dyadic basis. Results showed a low correlation (Pearson’s Correlation = 0.19, p <.001), indicating that one relation was able to predict another, but the extent of tie co-occurrence was very limited.

119 Therefore, these two networks were treated separately and two sets of constraint scores were computed and used in the analyses. Other Measures Demographics Two demographic variables were included in the analyses: players’ age and biological gender (as opposed to the gender of their avatar). Information on gender and birth date was self-reported when players registered their profiles with EQII, and was available in the Demographics table in the game behavioral logs. Gender was coded as a dichotomous variable and players’ age was calculated by subtracting their year of birth from 2006. To adjust for erroneous self-reports in the database (0.4% of all cases), the age variable was capped at a minimum of 12 and a maximum of 65. EQII Play Time The behavioral log (Character_Store table) recorded the total number of seconds a character had been played since its creation, which was then transformed into the total number of hours. This variable is static, meaning that there is only one value recorded in the database. Character Level Every player’s character level was recorded in the Experience table as well as the Character_Store table. In the Experience table, each experience record has a corresponding character level, which indicates the player’s character level at the time of experience accrual. Since a player could have multiple entries in the Experience table, the value of character level is also dynamic. In addition, the Character_Store table records

120 the latest character level a player had earned by September 2006. Therefore, depending on the nature of analysis (i.e., static or dynamic), character level was extracted from either the Experience table, as a dynamic value, or the Character_Store table, as a static value. Proximity All players self-reported their state and country as part of the player profile, which was recorded in the Character_Store table. The geographic proximity between two players was measured as a dichotomous variable, with 1 indicating that two players are located in the same state, and 0 indicating that two players are located in different states. It should be noted that this was a crude measure of proximity. People located in the same state generally share the same time zone (with a few exceptions), but geographically they could still be hundreds of miles apart. Character Class In EQII, each character can choose a player character class which has a distinct set of skills and strength for combat. There are 24 classes to choose from, which can be further condensed into four archetypes, Fighter, Priest, Scout, and Mage, which usually assume different roles in collaborative activities. In most analyses character class was measured as a categorical variable with four values (Fighter, Priest, Mage and Scout). This variable was further converted into three dummy variables (Priest, Mage and Scout; Fighter was the comparison group) to be included in regression models.

121 Guild Membership The Character_Store table contains information on the name of the guild a player belonged to at the time of the study. Individual character’s guild membership was measured by a binary variable (Guilded =1, Unguilded =0). This information was also used to generate a dyadic variable describing whether two players belonged to the same guild. Task Performance Task performance was measured by players’ character level (maximum level is 70). Because character level may be partially attributable to players’ total play time (in hours). total play time was also included in the models as a statistical control. Therefore, when total play time is held constant, the higher level a character reaches, the better the task performance is. Trust Trust was measured in the player survey. Participants were asked if their guildmates in EQII “can be trusted” or if they “can’t be too careful in dealing with” their guildmates. Participants responded on a 5-point scale ranging from “Trust them not at all” (1) to “Trust them a lot” (5). Sense of Community Online The players’ sense of community online was measured by asking the participants to rate the following statement on a 3-point scale (No = 1, It depends = 2, Yes = 3): “The people I have met online give me a sense of community.”

122 Table 4. Data Sources of Selected Variables Variables Networks Chat Trade Mentoring Grouping Collaboration Network size/Degree Brokerage/Closure Demographics Gender (binary) Age (years) Character attributes Accumulated play time (hours) Character level Geographic proximity (categorical) Character class (categorical) Guild membership (binary) Online Social Capital Trust – guild members (1-5) Sense of community online (1-3)

Data Source

Data Type

Game database Game database Game database Game database Game database Game database Game database

Behavioral Behavioral Behavioral Behavioral derived derived derived

Game database Game database

Self report Self report

Game database Game database Game database Game database Game database

behavioral behavioral Self report behavioral behavioral

Survey Survey

Self report Self report

Data Analysis As the proposed hypotheses operate at the intersection of social network theory, evolutionary theory and the studies of online communities, a combination of analytic methods were utilized to test the hypotheses. An overview of analytic methods is provided in the following before discussing the specific analyses. Analysis Overview Based on the type of analyses needed to test them, the hypotheses on network patterns, effects and evolution could be broadly categorized into two groups: 1)

123 hypotheses concerning static relationships, and 2) hypotheses concerning dynamic changes over time. The first category investigates associations among variables at a single time period and relies on cross-sectional data. For example, H1 examines the relationship between the level of network brokerage and individuals’ task performance in EQII. This category of hypotheses is essentially identical to most hypotheses commonly tested in the field of communication, and ordinary regression analysis was chosen as the analytic method to test them. The second category, on the contrary, focuses on the change of player networks over time. The dependent variable captures a dynamic process of network evolution as social ties are continuously being created and dissolved among EQII players. Furthermore, factors that may contribute to network evolution also experience dynamic changes over time. For example, H10 predicts that as a tie ages, its probability of survival also increases. In other words, time becomes an inherent dimension of both independent variables and dependent variables (although in this case, time is not directly included as a casual variable), creating additional complexities for data analyses. Such complexities represent an increasingly imperative challenge faced by communication and network researchers, one that fuels the development of a series of techniques to understand and analyze dynamic and longitudinal network data. Still, network change is itself a multifaceted phenomenon that could be examined from many angles. The analytic technique is closely associated with the aspect of network change one chooses to study. In order to organize the variety of approaches applicable to the studies of network change, Feld, Suitor and Hoegh (2007) have

124 suggested a useful typology by focusing on two dimensions: 1) the level of analysis (a tie or a personal network), and 2) the type of network change (existence of ties or the nature of ties). Together, these two dimensions delineate four general areas, as shown in Table 5. This dissertation concentrates its discussion on the existence of ties. All the hypotheses about network change fall in Cells #1 and #3 and they would be tested using different methods respectively. Table 5. Types of Changes in Networks Level of analysis

Type of network change

Existence of ties A tie

Nature of ties that exist

1. which ties come and go

2. how characteristics of ties change A personal network 3. expansion and 4. change in the overall contraction of networks characteristics of networks Note. Table adapted from “Describing Changes in Personal Networks over Time” by S. L. Feld, J. J. Suitor, and J. G. Hoegh, 2007, Field Methods, 19(2), p.218-236. H9 predicts that the size of personal networks shrinks over time as players advance in EQII. This hypothesis falls into Cell #3, which concerns the expansion and contraction of networks. For this type of change, the primary dependent variable is the number of ties, regardless of the particular contacts involved in one’s network. The size of a personal network could remain constant across different time periods even though specific ties have been replaced by others. As the unit of analysis is the overall network, Feld et al. (2007) suggest that it is appropriate to use multivariate analysis with network size as the dependent variable, and other characteristics of the ego and the entire network as explanatory variables.

125 H10 to H12 fall into Cell #1, which focuses on various factors impacting the persistence and decay of ties over time. Unlike Cell #3 which operates at the network level, Cell #1 operates at the tie level. Researchers measure the existence of a tie at time 1 and then again at later time points, attempting to understand the factors contributing to tie survival. As the dependent variable is usually a dichotomous one about tie persistence or decay, Feld et al (2007) suggest using logistic regression to analyze hypotheses in this category. Logistic regression provides a general analogy to understand the many approaches available for longitudinal analyses of tie persistence and decay, but these approaches all have distinct merits and limitations. The following is a brief introduction of three analytic techniques that represent state-of-the-art development in the field of longitudinal network analysis: 1) Exponential Random Graph (ERG) Models, 2) stochastic actor-based models for network dynamics, and 3) event history analysis of tie persistence/decay. Exponential Random Graph (ERG) Models ERG models, also known as p* models, are a special class of network analysis used to examine the statistical likelihood of network configurations. They are originally presented by Wasserman and Pattison (1996) and more recently extended by Pattison, Robins and associates (Robins, Pattison, Kalish, & Lusher, 2007; Robins, Snijders, Wang, Handcock, & Pattison, 2007). The mathematical form of ERG models can be summarized in the following equation (Robins, Pattison et al., 2007):

⎧ ⎫ ⎛1⎞ Pr(Y = y ) = ⎜ ⎟ exp⎨∑η A g A ( y )⎬ ⎝κ ⎠ ⎩A ⎭

126 where Pr(Y = y) indicates the probability of the observed network y, ηA is the parameter corresponding to the configuration A, gA(y) is the network statistic corresponding to configuration A (i.e., gA(y) equals one if configuration A is observed in network y, and zero otherwise), and κ is the normalizing function included to ensure proper probability distribution across random networks. Unlike traditional logistic models which assume that the probability of any tie is independent of the existence of any other tie, ERG models account for the inherent interdependence among network structural characteristics. ERG models estimate the likelihood for a particular network configuration (e.g., triangles) to exist in a network more or less frequently than would be expected by chance alone, conditional upon other structural configurations of the network. A positive and significant parameter estimate in an ERG model indicates that a specific structure exists in the network and the value of the estimate corresponds to the intensity of the effect. Recent developments of ERG models use Markov Chain Monte Carlo (MCMC) maximum likelihood estimation method, which relies on computer simulation to generate a distribution of random networks to be compared with the observed network y (Robins, Pattison et al., 2007; Robins, Snijders et al., 2007). Through repeated simulation processes, the parameter estimates are revised until they stabilize (i.e., converge). While simple ERG models enable the analysis of interdependent relational structures, various extensions further incorporate individual attributes, actor behaviors, and additional relations in the estimation. These extensions are particularly useful for researchers who are interested in exploring the myriad of factors associated with the

127 occurrence of certain structural configurations (for applications of ERG models in communication research, see Shumate & Palazzolo, in press). It is only recently, however, that ERG models have gained popularity among social network researchers because software packages, such as the PNet suite (Wang, Robins, & Pattison, 2005) and Statnet (Handcock, Hunter, Butts, Goodreau, & Morris, 2003), have been developed to make ERG modeling more accessible. Still, these software packages have important limitations, especially for estimating longitudinal ERG models. Up till now, the only available software for such analysis is LPNet, a part of the PNet software suite (Wang et al., 2005). However, the current version of LPNet is only capable of analysis of network change between two points in time. Also, as ERG modeling in general requires significant computational resources (Shumate & Palazzolo, in press), LPNet as well as the whole PNet suite is not very scalable. The maximum number of nodes in the network is limited to roughly 500. Based on the above considerations, LPNet is not suitable for longitudinal network analysis in this research. Stochastic Actor-Based Models for Network Dynamics The second approach is stochastic actor-based models for network dynamics, implemented in various versions of the software package Siena (Snijders, Steglich, Schweinberg, & Huisman, 2007; Snijders, van de Bunt, & Steglich, 2010). Actor-based models are specifically developed to study the change of network patterns across a minimum of two time periods. Unlike ERG models that focus on the probability of certain structural configurations, the focus of stochastic actor-based models is the rate of

128 transition, or network change, across time (Snijders, Pattison, Robins, & Handcock, 2006; Snijders et al., 2010). Stochastic actor-based models rely on a statistical process called Markov-based models, which infers the probability distribution of future relations as a function of only the present state. The time variable t, is taken to be continuous but the network is observed at discrete time intervals. Stochastic actor-based models have many built-in assumptions. As its name indicates, one key assumption of this approach is that nodes in the networks, known as actors, have control over their outgoing relations, on the basis of their individual attributes, their position in the network, and their perception of the network (Snijders et al., 2010). Their actions are assumed to be purposeful and based upon information of the full network and other actors. Actors’ choice of changing network relations is represented in the objective function, describing the likelihood of the actor moving in a specific direction to maximize individual utility by changing network relations, subject to structural constraints, changes made by other actors, and random influences. It is assumed that no more than one change could occur at any given moment, which means that changes in relations are individual decisions without any coordination or negotiation among actors. Additionally, the actor-oriented models assume that network ties are defined as enduring states, rather than events. This assumption applies well to studies where relations undergo gradual changes over time, such as friendship and trust, but not to studies where relations are brief and network changes are in flux. Empirically, these assumptions can be translated into specific data requirements for successful model convergence. For example, as time is understood as continuous but

129 observed at discrete points, usually the number of observations in the network panel data for actor-based modeling is not larger than 10 to ensure that the objective function is constant across time (Snijders et al., 2010). Because actors are assumed to have full information regarding the available partners they could link to, the number of actors in the network is usually under a few hundred. Also, because the change process is gradual and follows a constant objective function, the number of changes between any pair of consecutive waves should not be too high. Snijders et al. (2010) suggest using the Jaccard index to assess the degree of tie change: N11 N11 + N 01 + N10

where N11 is the number of ties present at both waves, N01 is the number of ties created at the second wave, and N10 is the number of ties terminated. Snijders et al suggest that values higher than 0.6 are preferable for actor-based models, while values lower than 0.3 lead to doubts about meeting the assumption of gradual change and constant objective function and, thus, are problematic for model estimation. Based on these considerations, stochastic actor-based models are particularly well-suited for estimating gradual change of small- to medium-sized networks, such as the studies of co-evolution of friendship and smoking behavior among adolescents (Mercken et al., 2009) and the changing personal networks of Argentina immigrants in Spain (Lubbers et al., 2010). However, this dissertation focuses on the longitudinal dynamics of large-scale player networks in EQII over 13 weeks. Changes are rapid rather than gradual between many pairs of observation periods, as evidenced in Jaccard indices consistently below 0.3. Therefore, stochastic actor-based modeling is not chosen to

130 analyze network change in this dissertation because the network panel data available do not meet its requirements and assumptions. Event History Analysis of Tie Persistence/Decay

Event history analysis is a one of the most commonly used techniques for testing evolutionary and ecological hypotheses (Carroll & Hannan, 2000; Monge, Lee, Fulk, Frank et al., in press). In event history models, the dependent variable is the occurrence of an "event." Event history models estimate the risk or likelihood of the event occurring given a set of covariates. Many organizational phenomena can be viewed from an event history perspective, such as joining and dropping out of a strategic alliance, implementing a special technology, and organizational demise. For longitudinal network analyses, the event of interest is usually the termination of a tie (Burt, 2000a, 2002). As event history analysis deals with categorical dependent variables, it could be understood as a special class of logistic regression, but with two important features: the capabilities to adjust for right-censoring and to incorporate time-varying covariates in the model (Allison, 1984). Consider an example that studies the likelihood of organizational demise. Such a study usually has a finite—and usually arbitrary—window of observation. Therefore, those organizations still active at the point of data collection may disappear later, after the observation ends, but their disappearance is not reflected in the data analysis. In a standard logistic regression model, those organizations are simply considered as “alive,” or not experiencing the event of demise even though all organizations would die at some point, thus adding bias to the analysis. This problem is known as "right-censoring", which occurs when the observation period ends before the

131 event has occurred. In a censored observation, the risk of the event occurring is unaccounted for by the analytic method. A second problem inherent in the standard logistic regression approach is how to incorporate time-varying explanatory variables. Some predictors, such as the year in which a organization was founded, remain constant throughout the observation period. But other predictors, such as the size of the organization, may change over time. These changes are difficult to capture with standard logistic regressions5. Event history analysis consists of a suite of techniques that are developed to properly handle these two common problems: right-censoring and time-varying explanatory variables (Tuma, Hannan, & Groeneveld, 1979). An event history model evaluates cases in terms of the periods during which entities are "at risk" for an event. Event history analysis is also able to incorporate time-varying covariates. The whole observation period is divided into various “spells.” Fixed covariates such as gender stay the same across the whole observation, while time-varying variables change across different spells. In this way, event history analysis is well-suited for the analyses of dynamic and longitudinal processes, including the change of networks. For example, Burt (2000a, 2002) has used event history models to examine the persistence and decay of network ties among bankers. Note that there are two principally similar ways to examine the likelihood of an event. One could model the "risk" of an event using hazard functions, 5

It should be noted that some sophisticated forms of logistic regression, such as McFadden’s (1973, 1981) discrete-choice model, a variant of the conditional logit model, could also accommodate time-varying variables. Such a model was used in a recent study that examined the network partner choice among biotechnology firms (Powell et al., 2005). But McFadden’s model is a multinomial logistic model, instead of a standard logistic model with a binary dependent variable (death of ties), as discussed in this dissertation. Therefore, multinomial logit models, including McFadden’s, are not covered here. This dissertation only compares event history models with the standard logistic regression.

132 which describe the likelihood of an event occurring (e.g., organization death) at any particular point in time given that the observed cases are "at risk" for the event. Most studies on organizational failure and transformation have used hazard models (e.g., Amburgey, Kelly, & Barnett, 1993). Alternatively, if the process of interest is the event not occurring (e.g., organization persistence), it is typical to use survival functions. For example, Carroll and Delacroix (1982) estimated the survival probabilities of newlyfounded newspaper organizations. Burt’s approach (2000a, 2002) was used to model tie decay as a hazard function. The event of tie dissolution was specified in terms of an instantaneous rate (Allison, 1984): r j (t ) = lim[ p j (t , t + Δt ) / Δt ]

where pj is the discrete probability of tie j experiencing dissolution (failure) between t and (t + Δt), conditional upon being at risk (tie j still exists) at time t. These models were specified in the following form (Allison, 1984):

r j (t ) = exp[ β X t ] where Xt is a vector of time-varying covariates, and β is a vector of coefficients indicating the effect of each variable on the instantaneous rate of tie decay. The event history models were estimated using Stata 7.0. As discussed previously, the spell (basic time interval) used in this research is a week. For longitudinal analyses, three consecutive months of data were included: June, July and August 2006. These three months were selected because they covered summer time, when some players (e.g., college students) may have seasonal playing patterns

133 different from other months during the year. Altogether, the three months consisted of 13 spells (from June 1 to August 30, 2006). Unlike organizational events, ties observed in the same network are not independent from each other. To study tie decay over time violates the assumption of independent observations and could produce unreliable standard errors (Feld et al., 2007; Lubbers et al., 2010). Therefore, the event history model is further specified to account for the inherent interdependence of network data. Ties originating from the same individual could be considered as nested observations within egos. Therefore scholars argue that it is appropriate to use multilevel techniques or autocorrelations to adjust for standard errors (Burt, 2000a; Lubbers et al., 2010; van Duijn, van Busschbach, & Snijders, 1999). This was achieved by using the “cluster” routine in Stata to adjust for tie interdependence. In summary, hypotheses either concern 1) static associations, or 2) the dynamic changes of networks. The first category can be tested using ordinary regression analyses. The second category can be further classified using Feld et al.’s (2007) framework of longitudinal network analysis. Cell #3 considers the changing size of networks over time, and could be tackled using multivariate analysis. Cell #1 concerns the addition and deletion of ties, and could be tackled using three methods: ERG models, stochastic actorbased models for network dynamics, and event history analysis of tie persistence/decay. Considering each method’s merits and limitations, event history analysis was chosen to test hypotheses on network evolution.

134 Hypotheses Testing Network Patterns A snapshot of player data for the week of Sept 4, 2006 to Sept 10, 2006 was examined for general network patterns. This week was selected because it is the only week during which complete chat, trade, and collaboration network data as well as character attributes were all available. Since the Character_Store table included information on the active players as well as the dormant or even “dead” characters, only those players who logged in the game during this week were included in the analysis. The preliminary overview of sociability was conducted through descriptive network analysis in UCINET (Borgatti, Everett, & Freeman, 2002). Regressions (including a first set of logistic regression models, and a second set of linear regression models) were used to test hypotheses about the effects of player demographics (H1and H2), character level (H3), character class (H4) and guild membership (H5) on players’ social engagement. Network Effects Hypotheses (H6-H8) on network effects were tested using regression models. The hypothesis (H6) on task performance was tested by examining the unique effects of network brokerage (measured by effective network size) on task performance (measured by character level, with total play time included as a statistical control). The hypothesis would be supported if effective size is positively associated with task performance. The hypotheses (H7, H8) on trust and sense of community of players were tested by examining the effects of network closure (measured by network constraint) on peoples trust toward guildmates and other players in EQII as well as their sense of community

135 online. If these hypotheses were supported, network closure would have positive and significant regression coefficients on dependent variables. Network Evolution This set of hypotheses on network evolution was tested using longitudinal data from June 1 to August 30, 2006. The individual time interval used in this analysis is a week, so that this time period contains a total of 13 consecutive weeks. To observe the evolution of ties during this period, it is necessary to assume that the set of nodes (players) remains available for connections. In other words, the existence of nodes themselves is the prerequisite of the existence of ties among these nodes. However, during the 13-week observation period, a considerable proportion of nodes (players) were constantly entering and exiting the population of active players, adding an additional layer of complexity to the evolutionary process. The Jaccard index for nodes6 between any two consecutive weeks was consistently lower than 0.6, indicating a high rate of player churn between observations. Therefore, the players chosen for analysis were a subset of the total population who were active and available for connections throughout the 13 consecutive weeks. The total number of this subpopulation was 1442 players. It should be noted that the 1442 players included both connected players as well as isolates. This approach was able to control for the additional complexity brought by player churn, but it may also limit the study population to those habitual or “hardcore” gamers who logged in consistently over the 13 consecutive weeks.

6

The Jaccard index was calculated based on the number of players present at both weeks, the number of new players entering the population at the second week, and the number of players exiting. The formula is essentially the same as the Jaccard index for ties (discussed in previous section).

136 H9 predicts that network size decreases as players advance in the game. It was tested by examining the number of unique contacts in each individual’s ego networks using two way mixed-measures analysis of variance (ANOVA). Based on their character level, players were split into four groups (with a band of 20): level 1-20, 21-40, 41-60 and 61-70. This band was selected because level 20 represented a critical milestone in character development. For example, starting at level 20, players could start earning Achievement points, which could be used to customize the development trajectory of their character. Level 20 also marks the time when a player has to decide a specific tradeskill profession (e.g., alchemist). Therefore, for this analysis, the within-subject factor was time (13 weeks) and the between-subject factor was character level (four groups). To test Hypotheses H10-H17, three discrete-time parametric models were estimated for tie dissolution. As the longitudinal network data were collected every week for a total of 13 consecutive weeks, event history data were constructed by populating 13 time-event records every week for each tie during its existence. Altogether, there were 19384 relations and 35271 total observations (one observation means one tie-week record) and 17674 ties failed during the 13 weeks. The first model included total play time (H10) and tie duration (H11). Because the collaboration networks examined here are symmetric, i.e., two players are connected as a result of their mutual choice, the mean of their total play time (originally measured at the individual level) was used as a dyadic variable to test H10. The second model added variables measuring whether players belonged to the same character class (H12), their

137 difference in character level (H13) and whether they belonged to the same guild (H14). The third model further added gender homophily (H15), age difference (H16) and geographic proximity (H17). All the test statistics were adjusted for autocorrelation at the individual player level (i.e., the same individual involved in more than one tie) using the “cluster” routine in Stata.

138 Chapter 4: Results Network Patterns Preliminary Analysis A snapshot of player data for the week of September 4, 2006 to September 10, 2006 was examined for general network patterns. Among these characters who reported valid demographic information in the game database, 17% were female players and 83% were male players. The median age was 33 years old (Mean = 34.25, SD = 9.65, capped at minimum of 12 and maximum of 65). The basic demographic profile reported by these players at the time of game registration was fairly consistent with the self-reports of gender (19.20 % female) and age (Median = 31, SD = 9.65) from the online survey (Williams et al., 2008). The median character level of these players was 33 (Mean = 35.40, SD = 23.31). Because EQII had a level cap of 70 at the time of study, the number of characters with the highest level was disproportionally high (N = 1370, 14.45% of all characters). As shown by the accumulated play time recorded in the game database, EQII requires significant time commitment. On average, a character spent a total of 699.37 hours in the game (SD = 1079.84), with the longest playing character spending 9638.61 hours. As shown in Figure 7, the accumulated play time rose approximately linearly with character level in the lower levels, and the curve became steeper when characters advanced further (e.g., around and after level 45). This indicates that the EQII leveling system was designed to increase difficulty gradually over time, which was similar to the observed

139 game mechanics of another popular MMOG, World of Warcraft (Ducheneaut et al., 2006). Figure 7. Accumulated Play Time by Character Level

2,500

2,000

1,500

1,000

500

0 0

5

10

15

20

25

30

35

40

45

50

55

60

65

70

Character Level

The distribution of character class was also examined (see Figure 8). EQII players clearly favored some character classes more than others. Among the 9843 characters, the most popular character classes were Wizard (609), Necromancer (599), Monk (593) and Conjurer (589). The popularity of classes, especially Necromancer, Monk and Conjurer, corresponded well to their “solo-ability.” As discussed in many online forums and expert reviews, Necromancer and Conjurer are popular because they carry a pet around, which could serve as an inherent groupmate. Monks are also among the best choices for a

140 “career soloer”, because they are able to “do a fair amount of damage for a fighter and have the handy ability to heal and feign which allows them to avoid death when others find it inevitable” ("Soloing in EverQuest ii", 2010). These design features were clearly reflected in players’ choice of character class. Condensed into the four class archetypes, the most popular archetype was Fighter (30%), followed by Mage (27%), Priest (23%) and Scout (20%). Male and female players differed in their preferences for character class (Pearson Chi-Square =104.32, df = 3, p <.001). As shown in Figure 8, compared with males, female players were much more likely to play Priests, which have superior healing power and tend to take auxiliary roles in combat. By contrast, male players were more likely to play Scout and Fighter classes.

141 Figure 8. Character Class by Gender

142 Figure 8, Continued

Who is connected? To examine the general patterns of sociability, social networks of chat, trade, and collaboration were examined (see Table 6). Only about half of all the active characters were engaged in chat, trade, or collaboration activities during the week, while the rest played alone even though the game clearly rewards collaboration. This finding resonated with Ducheneaut et al.’s (2006) WoW study, where they found that around 40% of the play time was spent in groups. It should be noted that this study used a different metric than the proportion of play time spent in groups to measure the extent of sociability. While not a direct comparison, the metric used here represented a more relaxed measure of social engagement than Ducheneaut et al.: if a player grouped once during the entire week, that person would be considered as “connected” rather than solo. Therefore, while

143 the numbers reported here were similar to those in Ducheneaut et al. (2006), the substantive interpretation painted a bleaker picture of social engagement in general—only about half of all the active players in the week of September 4 to September 10, 2006 ever talked to others, traded with others, or collaborated with others. Table 6. Descriptives of Chat, Trade and Collaboration Networks N % of Total Active Density Median of Characters Degree Chat 5100 53.78 0.02 32 Trade 4069 42.91 0.001 3 Collaboration 4619 48.70 0.01 10 Note. N only includes connected characters (i.e., non-isolates).

SD of Degree 275.45 10.97 276.32

Characters’ choice to play alone or to engage in chat, trade or collaboration activities was closely associated with their experience in the game. In general, characters tended to become more socially connected as they leveled up (see Figure 9).

144 Figure 9. Percentage of Connected Characters by Level

The overlap of the chat, trade, and collaboration networks was examined. Taken together, out of 9483 active characters, there were 6672 characters that engaged in any form of chat, trade, or collaboration activities, and 2469 of them appeared in all three networks. The overlap of network ties was measured by QAP correlations of the three networks. Results show that these networks were significantly correlated, but the level of correlation was fairly low (see Table 7). This indicated that chat, trade, and collaboration often do not co-occur. Individuals tended to have distinct partners in chat communication, transaction, and collaboration. One caveat worth nothing is that the chat communication captured in this analysis only included private “tell”, or instant messages, between two

145 characters. The chat messages in group, raid, or guild channels were not captured in the dataset, thus not included in the analysis. Table 7. Density, SD and QAP Correlations of Chat, Trade and Collaboration Ties Density SD of Degree Chat Trade Chat 0.02 1.62 Trade 0.001 0.14 0.06 (p <.01) Collaboration 0.02 2.48 0.05 (p <.01) 0.20 (p <.01) Guild Membership Guilds are relatively stable social groups within the game world. Among the 9483 characters, 66% were affiliated with guilds. Female players were more likely to join guilds than male players (Pearson Chi-Square = 39.08, df = 1, p =.004). Guild membership also differs across character classes, with Priests the most likely to join guilds (see Figure 10). Given that Priests often assume the role of “healers” who are responsible for restoring health in collaborative combat, it is not surprising that they tended to be more “handicapped” and have to rely on co-players in order to succeed. Also, as shown previously, Priests were often in short supply compared to the other character classes, but their healing function was indispensable in combat. Therefore, guilds may need Priests just as much as Priests rely on guilds.

146 Figure 10. Guild Membership by Character Class

Guild membership also varied across different character levels. As shown in Figure 11, the proportion of characters in guilds increased with level advancement. When players reached level 20 and above, over 50% of all characters were affiliated with guilds. Almost all the players level 65 and higher participated in guilds.

147 Figure 11. Guild Membership by Character Level

Factors Associated with Social Engagement To address H1to H5, correlation and regression analyses were performed to examine the impact of demographics, character level, character class and guild membership on players’ level of social engagement (correlation coefficients are shown in Table 8). Two sets of regression analyses were performed. First, a set of logistic regression models (Table 9) were run to predict the likelihood of inclusion in chat, trade and collaboration networks. The dependent variable for these models was binary (connected or solo). Second, among those characters who were socially engaged in EQII, a set of linear regression models (Table 10) were run to predict the size of chat, trade and collaboration networks. For the linear regression models, because the dependent variables were both positively skewed (all skewness statistics were larger than 6) and leptokurtic

148 (all kurtosis statistics were larger than 70), the size of chat, trade and collaboration networks were log-transformed before analysis.

149 Table 8. Correlations Among Study Variables 1 2 3 4 5 6 7 8 9

Female (D) Age Mage (D) Priest (D) Scout (D) Guild (D) Character Level Chat Degree Trade Degree    

Collaboration Degree

1

2

3

4

5

6

7

8

0.06** -0.02 0.15** -0.03 -0.03 -0.04 0.02 0.08**

-0.03 0.05* -0.01 0.06** 0.02 -0.13** -0.08**

-0.35** -0.29** 0.00 0.00 0.00 -0.03

-0.29** 0.04 0.04* -0.03 0.05**

-0.01 -0.01 0.03 -0.05**

0.38** 0.21** -0.08**

0.39** 0.09**

0.16**

0.02

-0.04

0.00

0.00

-0.02

-0.05**

-0.27**

0.01

9

0.21**

Note. D indicates dummy variable; Chat, trade and collaboration degrees were log-transformed to adjust for skewness; * p <.05; ** p <.01.    

149

150 Table 9. Logistic Regression Models Predicting the Likelihood to Chat, Trade and Collaboration (N = 8074) Chat Trade Collaboration Age 0.99** 0.98** 0.98** Female (D) 1.12 1.24** 1.04 Level 1.04** 1.03** 1.02** Mage (D) 0.83** 0.93** 0.97 Priest (D) 0.97 1.18** 1.36** Scout (D) 1.09 0.96 1.03 Guild (D) 1.46** 1.34** 1.62** Chi-Square 1280.37** 971.94** 636.50** Cox & Snell R2 0.14 0.11 0.08 Note. D indicates dummy variable (male characters and the Fighter class are the comparison group); Numbers represent odds-ratios. * p <.05; ** p <.01. Table 10. Regression Models Predicting Degree in Chat, Trade and Collaboration Networks Collaboration Degree Chat Degree (log) Trade Degree (log) (log) Age -0.02** -0.01** -0.01** Female (D) 0.35** 0.13** 0.04 Level 0.03** 0.01** -0.01** Mage (D) -0.02 -0.01* 0.03 Priest (D) -0.11 0.04 0.04 Scout (D) 0.01 -0.10** -0.06 Guild (D) 0.34** -0.24** 0.35** R2 0.20** 0.03** 0.01** SE 1.51 0.78 1.74 N 5077 3703 4256 Note. D indicates dummy variable (male characters and the Fighter class are the comparison group); Chat, trade and collaboration degrees were log-transformed to adjust for skewness and kurtosis; Numbers represent unstandardized regression coefficients. * p <.05; ** p <.01.

151 Age had a negative impact on the likelihood of inclusion in chat (OR = 0.99, p <.001), trade (OR = 0.98, p <.001), and collaboration networks (OR = 0.98, p <.001). For players who were engaged in chat, trade, or collaboration activities, age also showed a negative impact on the size of networks (b = -0.02, p <.001; b = -0.01, p <.001; b = -0.01, p <.001), providing support for H1. Younger players are more likely to be socially connected in EQII, weaving larger ego-networks through trade, chat, and collaboration activities. But the size of the effect was very small. Compared with male players, female players were significantly more likely to be included in trade networks (OR = 1.24, p <.001), but not in chat (OR = 1.12, p =.08) or collaboration (OR = 1.04, p = .51). Among those connected players, female players have 42% more chat partners (b = 0.35, exp (0.35) = 1.42, p <.001) and 14% more trade partners (b = 0.13, exp (0.13) =1.14, p <.001), compared to male counterparts, but the size of their collaboration network is not significantly different from that of males (b = 0.04, p = .60). Therefore, the results provided support for H2. H3 predicted that social engagement would rise with character level. In particular, results showed that high level characters were slightly more likely to engage in chat (OR = 1.04, p <.001), trade (OR = 1.03, p <.001), and collaboration (OR = 1.02, p <.001). Among those connected, high level characters also tended to have larger chat networks (b = 0.03, p <.001) and trade networks (b = 0.01, p <.001). This finding might be a result of accumulated social contacts throughout one’s career in the game. As one spends more

152 time in the game, the probability of interacting with others through chat and trade increases. These contacts are then accumulated and become “regulars” in one’s chat and trade network. On the other hand, character level showed a negative relationship with the size of collaboration network but the coefficient was very small (b = -0.01, p <.001).This may appear to contradict the game design that higher level players are increasingly confronted with more challenging mobs, making collaboration a must. However, such collaborations are not necessarily carried out with more partners. In other words, the probability of grouping would increase with character level, but the size of one’s collaboration network would shrink (also see the results for H9). H4 predicted that the level of social engagement would vary with character class. As shown in Table 9, compared to the Fighter class, Mage, Scout and Priest classes exhibited somewhat different patterns in their inclusion in chat, trade and collaboration networks, supporting the prediction of H4. Particularly, the Priest archetype had 36% more chance than Fighters to participate in groups and raids (OR = 1.36, p <.001). This provides solid evidence for the effects of EQII social architecture, as Priests tended to play supporting roles and largely rely on co-players to succeed. When comparing the network size of connected players, character classes did not seem to matter too much— compared with Fighters, Priests, Mages, and Scouts tended to have a relatively similar number of alters in their chat networks (b = -0.11, p = .06; b = -0.02, p = .66; b = 0.01, p = .93) and collaboration networks (b = 0.04, p = .62; b = 0.03, p = .71; b = -0.06, p = .47).

153 But character class showed a small effect on the size of trade networks, with Mages (b = 0.08, p = .02) and Scouts (b = -0.10, p = .01) having slightly fewer trade partners but not the Priests (b = 0.04, p = .28). H5 stated that joining a guild increases the level of social engagement, which is also strongly supported by the data (see Table 9 and Table 10). In the logistic regression models, guilded players were 46% (p <.001), 34% (p <.001), and 62% (p <.001) more likely than unguilded players to participate in chat, trade and collaboration activities. Among connected individuals, the number of partners in the chat and collaboration networks also increase significantly as a result of guild membership: guilded players tend to have 40% (b = 0.34, exp (0.34) =1.40, p <.001) and 42% (b = 0.35, exp (0.35) = 1.42, p <.001) more contacts in the chat and collaboration networks. The only exception is the trade network. Guild membership decreases the size of the trade network by 21% (b = -0.24, exp (-0.24) = 0.79, p <.001). This phenomenon could be the result of the internal share and exchange mechanisms available within guilds, such as the guild bank, which provides guild members a communal repository to store as well as trade items so that there is little need to go through other channels.

154 Network Effects Brokerage H6 predicted that brokerage is positively associated with players’ task performance. This hypothesis was tested using regression models (see Table 11 for correlations and Table 12 for regression results). Brokerage was measured by the effective network size of players’ mentoring and trade networks, respectively. Controlling for total play time, age, gender, and character class, the effective size of players’ mentoring networks (b = 0.60, p <.001) and trade networks (b = 0.09, p <.001) were found to have a significant and positive impact on character level. In other words, when total play time, age, gender and character class are held constant, players could reach higher levels if the effective sizes of their mentoring and trade networks are larger. Therefore the results provided strong support for H6.

155 Table 11. Correlations Among Variables in Regression Models Predicting Character Level (N = 27770) Mean 1 2 3 4 5 6 7 8 9

Total play time Female (D) Age Mage (D) Priest (D) Scout (D) Mentoring effective size Trade effective size Charater level

466.29 0.17 33.10 0.28 0.22 0.20 5.83 74.94 32.94

SD

1

2

849.94 0.38 0.01 9.60 0.06 ** 0.09 ** 0.45 -0.04 ** 0.00 0.41 0.06 ** 0.08 ** 0.40 -0.03 ** -0.03 ** 10.94 0.19 ** 0.01 91.44 0.51 ** 0.02 ** 18.84 0.64 ** -0.05 **

3

-0.004 0.05 -0.03 -0.01 0.04 0.01

4

** -0.33 ** ** -0.31 ** -0.02 ** ** -0.01 * 0.01

5

6

7

8

-0.26 ** 0.03 ** -0.03 ** 0.01 -0.003 0.44 ** 0.05 ** -0.04 ** 0.46 ** 0.64 **

Note. D indicates dummy variable (male characters and the Fighter class are the comparison group); * p <.05; ** p <.01.

155

156 Table 12. Regression Models Predicting Character Level (N = 27770)

Total play time Female (D) Age Mage (D) Priest (D) Scout (D) Mentoring effective size Trade effective size 2

R

SE R2 change

Mentoring Network Model 1 Model 2 0.01 ** 0.01 ** -2.60 ** -2.77 ** -0.04 ** -0.02 ** 1.77 ** 2.15 ** 1.30 ** 1.23 **

-0.33

0.41 ** 14.46

0.29 0.60 ** 0.53 ** 12.93 0.12 **

Trade Network Model 1 Model 2 0.01 ** 0.01 ** -2.60 ** -2.98 ** -0.04 ** -0.04 ** 1.77 ** 1.67 ** 1.30 ** 1.68 ** -0.33 -0.47 *

0.41 ** 14.46

0.09 ** 0.55 ** 12.66 0.14 **

Note. D indicates dummy variable (male characters and the Fighter class are the comparison group); Numbers represent unstandardized regression coefficients. * p <.05; ** p <.01.

157 Table 13. Correlations Among Variables in Regression Models Predicting Trust and Sense of Community (N = 802)

1 2 3 4 5 6 7 8 9 10

Total Female (D) Age Mage (D) Priest (D) Scout (D) Mentoring constraint Trade constraint Trust toward guildmates Sense of community online

Mean SD 1193.20 1216.53 0.20 0.40 34.01 9.38 0.26 0.44 0.25 0.44 0.16 0.37 63.64 32.81 17.68 28.03 3.37 0.72 2.46 0.70

1

2

0.003 0.01 -0.08 ** 0.05 -0.02 -0.07 * -0.31 ** -0.02 0.10 *8

0.11 ** 0.02 0.14 ** -0.05 -0.07 * -0.02 -0.01 0.09 **

3

4

0.001 0.07 * -0.35 ** -0.02 -0.26 ** 0.03 0.003 -0.02 0.03 0.02 0.02 -0.02 0.03

5

6

-0.26 ** 0.01 0.02 0.02 -0.03 -0.02 -0.01 -0.02 -0.06

7

0.36 ** -0.03 -0.05

8

-0.01 0.03

9

0.25 **

Note. D indicates dummy variable (male characters and the Fighter class are the comparison group); * p <.05; ** p <.01.

157

158 Closure H7 and H8 predicted that network closure is positively associated with players’ trust towards their guildmates and their sense of community online. To address H7 and H8, regression models predicting trust towards guildmates and sense of community online were tested (see Table 13 for correlations and Table 14 for regression results). Network constraint in the mentoring and trade network was not found to have any impact on the level of trust towards guildmates (b = -0.001, p = .41; b = -0.001, p = .63), neither did it show any positive influence on players’ sense of community online (b = 0.002, p = .01; b = 0.001, p = .12). Therefore, H7 and H8 did not receive empirical support. Table 14. Regression Models Predicting Trust and Sense of Community (N = 802)

Total play time Female (D) Age Mage (D) Priest (D) Scout (D) Mentoring constraint Trade constraint 2

R

SE

Trust toward guildmates 0.00 0.03 0.001 -0.02 -0.06 0.06 -0.001

0.00 -0.02 0.001 0.03 -0.04 -0.03 -0.001

0.004 0.72

0.003 0.723

Sense of community online 0.0001 ** 0.0001 ** 0.15 * 0.17 ** -0.01 -0.01 * -0.05 0.00 -0.11 -0.09 -0.09 -0.11 -0.002 * 0.001

0.04 ** 0.70

0.03 ** 0.71

Note. D indicates dummy variable (male characters and the Fighter class are the comparison group); * p <.05; ** p <.01.

159 Network Evolution Size of Ego Networks To address H9, a two-way mixed-measures ANOVA was conducted to evaluate the effect of time and the character level at the beginning of the observation (week 1) on the size of individuals’ collaboration network. The within-subjects factor of time has 13 levels (repeated measures of network size from week 1 to week 13). The betweensubjects factor of character level included four levels, with cut points at level 20, 40, and 60 respectively. The time main effect was tested using the multivariate criterion of Wilks’ lamda (Λ) because the sphericity assumption was not met (Mauchly’s W = .00, df = 77, p <.001). Results indicated a significant within-subjects time main effect, Λ = .97, F(12, 1427) = 3.57, p <.001, partial eta squared = .03, a significant between-subjects character level main effect, F(3, 1438) =48.71, p <.001, partial eta squared = .09, as well as a significant interaction (time x character level) effect, Λ = .92, F(36, 4216.96) =3.45, p <.001, partial eta squared = .03, but all the effect sizes were quite small. Post hoc analysis of the four groups based on character level using the Scheffe test show that all the comparisons are significant except between the 21-40 group and the 41-60 group (see Table 15). Clearly, as characters advance in level, their network size decreases.

160 The interaction effects of time and character were less straightforward and suggested that the longitudinal trend differed for the four groups. As shown in Figure 12, over the 13 weeks of observation, the “beginners” with the lowest character level expanded their social networks, while the other more advanced groups, especially those over level 60, tended to keep their network size stable (but not necessarily the alters). Taken together, these results provide partial support for H9. An additional exploratory analysis tested whether the networks themselves stabilize as players level up (i.e., less churn of alters). The results showed that the players of the four level groups did not display a consistent trend with regard to the stability, or churn, of their networks over the course of 13 weeks. Rather, the rates of overall tie decay during the 13 weeks were all quite high for the four groups (90%, 90%, 92% and 91%, respectively). Table 15. Post Hoc Comparisons Between Character Level Groups Level Mean Network Size 1a 2a 3a 1 1-20 185.79 2 21-40 121.38 64.42(18.46)** 3 41-60 94.58 91.21(16.99)** 26.79(12.67) 4 6126.37 159.42(16.43)** 95.00(11.91)** 68.21(9.47)** Note. a. value represents the difference in mean and standard error (in parentheses). * p <.05; ** p <.01.

161 Figure 12. Estimated Marginal Means of Network Size Over 13 Weeks

250

C haracter Level at Week 1 Level <= 20 20 < Level <= 40 40< Level <= 60 Level > 60

E stimated Marginal Means

200

150

100

50

0 1

2

3

4

5

6

7

8

9

10

11

12

13

time

Tie Decay To address Hypotheses H10-H17, three nested models were estimated. The first model included player experience (H10) and tie duration H11. The second model added players’ character class (H12), character level (H13) and guild membership (H14). The third model further added gender (H15), age (H16) and geographic proximity (H17). Since these models were nested, Chi-Square tests were conducted between Model 1 and Model 2, and between Model 2 and Model 3. Results showed that the log

162 likelihood increased significantly as more independent variables were added into Model 2 and Model 3 (p <.001 for both tests). Therefore, since Model 3 includes the most comprehensive list of independent variables, coefficients in Model 3 were interpreted as the final results (see Table 16). Altogether, there were 19,384 relations and 35,271 total observations (one observation means one tie-week record). Altogether, 17,674 ties (91.18%) failed during the 13 weeks, suggesting that ties among EQII players were remarkably transient. H10 hypothesized that a tie is less likely to decay between two players with greater total play time. This hypothesis was not supported. Although in Model 1, average play time had a significant and negative coefficient (coefficient = -0.0002, p <.001), indicating that play time decreased the likelihood of tie decay, this effect disappeared in subsequent Models 2 & 3 (coefficient = -0.00002, p = .21; coefficient = -0.00002, p = .20). H11 predicted that the longer a tie has been maintained between two players, the less likely it is to discontinue. This hypothesis received strong support. Results showed that tie duration had a significant and negative effect on the hazard of tie decay (coefficient = -0.17, p <.001). Substantively, with every additional week during which two players stay connected, the likelihood of tie decay in the future decreased by 16% (exp (-0.17) = 0.84).

163 Table 16. Event History Models Predicting Tie Decay Model 1 Model 2 Constant -0.03 * -0.06 Average play time -0.0002 ** -0.00002 Tie duration -0.25 ** -0.17 Same character class (D) -0.002 Level difference 0.003 Same guild (D) -0.72 Same gender (D) Age difference Same state (D) df 2 5 Log likelihood -23402.33 -22625.10 Chi-square 3824.76 ** 5379.22

** ** ** **

**

Model 3 -0.08 -0.00002 -0.17 -0.003 0.002 -0.71 0.02 0.001 -0.15 8 -22585.40 5410.11

Hypothesis ** ** ** **

**

H10 H11 H12 H13 H14 H15 H16 H17

**

Note: D indicates dummy variable. Numbers represent logit coefficients predicting tie decay, adjusted for autocorrelation between relations involving the same individual. All models are estimated across the 19,384 relations (35,271 total observations). * p <.05; ** p <.01. H12 predicted that players of the same character class are more likely to discontinue their ties. On the contrary, the results showed that players of the same character class were not prone to tie decay (coefficient = -0.003, p =.89). Therefore, H12 did not receive empirical support. H13 predicted that players of similar levels tend to keep their connections. This hypothesis would receive support if the variable measuring the difference of character levels between two players has a positive and significant impact on tie decay. The results showed support for H13 (coefficient = 0.002, p = .004), but the effect was quite small as

164 a one unit increase of level difference between two players only added 0.2% greater chance of tie decay (exp(0.002) = 1.002). H14 predicted that players who belong to the same guild are more likely to keep their connections over time. This hypothesis received strong support (coefficient = -0.71, p <.001). In other words, belonging to the same guild decreases the hazard rate of tie decay by more than 50% (exp (-0.71) =0.49). Figure 13 shows the smoothed hazard rate estimates for the ties connecting guildmates as well as the ties connecting people from different guilds or unguilded players. This graph clearly visualizes the marked difference between these two groups. Ties between guildmates have a substantially lower hazard of tie decay. Figure 13. Hazard Estimates of Tie Decay by Guild Membership.

165 Finally, a set of variables was added to the model to test the impact of demographic similarity (or difference) on the persistence or fragility of ties. H15 predicted that people of the same gender are more likely to keep their connections. This hypothesis did not receive empirical support, as the variable on gender homophily did not generate a significant coefficient (coefficient = 0.02, p = .14). H16, which predicted the impact of age homophily, was not supported either. The age difference between two players did not exert any significant influence on the prospect of their connection (coefficient = 0.001, p = .40). H17 predicted that ties between people located in geographic vicinity (measured by whether two players are from the same state) are more likely to survive. The results showed a negative and significant coefficient produced by geographic proximity (coefficient = -0.15, p<.001). Being in the same state attenuated tie decay by 14% (exp (-0.15) = 0.86), thus this hypothesis was supported. Table 17 provides a summary of all the hypothesis-testing results presented in this chapter. Overall, this study provides a critical test in three aspects of player social networks in online gaming communities. Distinct patterns of social interactions were observed for different demographic groups, and the social architecture of the game was found to be strongly associated with the degree of sociability. The properties of egonetworks, especially the extent to which a player serves as a broker in connecting otherwise unconnected individuals, were instrumental for obtaining better task performance, but network closure was not found to influence trust or sense of

166 community. Finally, an examination of longitudinal network development over 13 consecutive weeks provided support for the evolutionary argument. In general, player social ties were very fragile. A durable relationship was often the result of consistent maintenance in the past, so that the longer a tie has survived, the better the chance it would persist in the future. Membership in the same player guild and geographic proximity also reduced the possibility of tie decay.

167 Table 17. Summary of Hypothesis-Testing Results Hypotheses Network Patterns H1: Age is related to players’ level of engagement in social networks. H2: Gender is related to players’ level of engagement in social networks. H3: Character level is positively related to players’ level of engagement in social networks. H4: Character class is related to players’ level of engagement in social networks. H5: Guild membership is positively related to players’ level of engagement in social networks. Network Effects H6: Brokerage in players’ social networks is positively associated with their task performance. H7: Closure in players’ social networks is positively associated with their trust towards guildmates. H8: Closure in players’ social networks is positively associated with their sense of community. Network Evolution H9: The size of players’ social networks shrinks as they advance in EQII. H10: Ties involving players who have spent more time playing EQII are less likely to decay. H11: The longer a tie has been maintained between two players, the less likely it is to decay. H12: Ties between players of the same character class are more likely to decay. H13: Ties between players of similar character levels are less likely to decay. H14: Ties between players in the same guild are less likely to decay. H15: Ties between players of the same gender are less likely to decay. H16: Ties between players of similar ages are less likely to decay. H17: Ties between players of geographic proximity are less likely to decay.

Results Supported Supported Partially Supported Supported Partially supported Supported Not supported Not supported Partially supported Not supported Supported Not supported Supported Supported Not supported Not supported Supported

168 Chapter 5: Discussion and Conclusion Discussion This study examined a series of hypotheses related to three major aspects of player social networks in the online gaming world of EQII. Results reported in the preceding chapter provide a descriptive overview of the general patterns of social interaction, test the effects of player social networks on individuals’ performance in the game and the trust and sense of community they derive, and trace the longitudinal development of these social connections. Network Patterns Using behavioral data accumulated during a week in September 2006, a systematic examination of network patterns revealed several interesting trends of EQII dynamics. The basic demographics of players are consistent with previous reports in similar MMOGs (Griffiths, Davies, & Chappell, 2003; Yee, 2006a). This study provides one of the closest comparisons to Duchenaut et al.’s (2006) study of the social dynamics in World of Warcraft, in which they found that players spent only about 30% to 40% of their total play time interacting with other players. Similarly in EQII, only half of the players were found to ever chat, trade, or collaborate with anyone during the week, and the remaining half preferred to be left alone instead. It should be noted that this study used a different metric than the proportion of play time spent in groups used by Duchenaut et al. to measure the extent of sociability. The metric used here was more

169 liberal in measuring social engagement: if a player chatted, traded, or grouped once during the entire week, that person would be considered as “connected” in the chat, trade or grouping network rather than playing solo. Therefore, if employing a more precise time-based metric, the actual level of social engagement in EQII could be even lower than what Duchenaut et al. found in WoW. Further, results showed that the most soloable classes, such as Necromancer and Conjurer, were also among the most popular class choices. This finding is also consistent with the World of Warcraft study, providing further support that these game worlds may not be as social as scholars have assumed. Taken together, this study supports the “alone together” finding obtained from another popular MMOG, thus providing some confidence that this phenomenon may be relevant across different gaming communities. This does not suggest, however, that MMOGs are asocial places. They are still vibrant sites of social interaction. But it is important to not make the assumption that every player is drawn to the game because of the social factor. A significant portion of the players choose not to engage in any social activities, even though game mechanics are in place to reward these behaviors. There are two plausible explanations for the behavior of solo players. First, Duchenaut et al. (2006) suggest that being surrounded by others in the MMOG provides a sense of social presence. This perspective does not contradict the conceptualization of MMOGs as virtual “third places” (Oldenburg, 1997). Rather, patrons at physical third places such as bars and coffee shops may not necessarily seek to interact with other

170 customers. Many could be attracted to a coffee shop just to enjoy its comforting ambience and to be surrounded by other customers. By analogy, playing EQII alone in the presence of others is like reading a book in a crowded coffee shop or having a drink in a busy bar. People may not engage in any direct social interactions, but the mere presence of others and the possibility to strike a conversation if they desire make these experiences more social than if carried out in solitude. In addition, the crowds in MMOGs are also sources of entertainment, just like the entertainment one may derive from people-watching in a coffee shop. In this sense, for some players, MMOGs are public social spaces where they choose to conduct individual activities. The type of social engagement is passive rather than active. This calls for a more nuanced understanding of the different layers of sociability in online spaces and the need to conduct research on their respective characteristics and effects. A second explanation takes into account the effect of EQII experience. As shown in Figure 9, players’ choices to play alone or to engage in social activities were closely associated with their experience (character level) in the game. As players became more advanced in the game, the more likely they were to interact with others. This may reflect a process whereby new players are “socialized” into the community as they accumulate experiences in the game. On the one hand, following the linear order of progression in the game, players are presented with quests and challenges with gradually increasing difficulty. Therefore, players become more and more motivated to seek collaboration in

171 order to keep up with the game. So the linear curve in Figure 9 is another example of the effective deployment of game mechanics. On the other hand, the relationship between social engagement and character level could be a result of player survival and attrition. EQII is a demanding game and not every novice player is expected to reach level 70. Those who engage in collaborative play not only progress faster, but also assemble a personal community of players during the course of participation. Such a community may provide camaraderie, social support and enjoyment that sustains continued participation in the game. By contrast, solo players are more likely to drop out along the way because they have not established meaningful social connections in the game. Future research is required to further explore this hypothesis by examining the attrition and survival rate together with the level of social engagement. Several factors were found to influence the level of social engagement. Demographically, younger players were more likely to be socially engaged, but the effect size was negligible. Female players tended to have much larger networks of chat and trade, which provided empirical support for gender stereotypes that women are more motivated to play for social reasons (Williams et al., 2009). Thus, it was not surprising to find that women’s character class choice was heavily skewed towards Priests, the leastsolo-able archetype. Consistent with past studies (Ducheneaut et al., 2006; Williams et al., 2006), game mechanics were found to be important predictors of players’ social engagement. In

172 particular, character class was found to significantly impact the level of social engagement, with the less “solo-able” players more connected in chat, trade and collaboration networks. With increased difficulty at higher levels, players were also more likely to socialize and collaborate with others. The findings provide solid evidence that these social mechanisms embedded in the virtual world were quite successful in framing players’ social experiences as designed. On the other hand, also noteworthy is the fact that players favored more “solo-able” classes, and that around half of them did not participate in chat, trade or collaboration at all during the week. This indicates that game mechanics are only powerful to a certain extent—players also make conscious choices and adhere to their own play styles and objectives. In addition, guild membership greatly increased the likelihood of social interaction and for those connected players, expanded chat and collaboration networks. The size of trade networks became smaller for guilded players, perhaps due to the fact that most trade activities could be conducted through internal exchange mechanisms within guilds, such as the guild bank. This indicated that these spontaneous player associations, despite their own fragility and high churn rate (Chen et al., 2008; Ducheneaut, Yee et al., 2007), could serve important social functions in creating tightly-knit subunits within a virtual world. Taken together, this study demonstrated that the patterns of social interaction in EQII are shaped collectively by the computer “code” that devises the social architecture (Lessig, 2006) as well as by the players’ goals and styles.

173 Network Effects Drawing on the literature of network brokerage and closure, this study proposed an alternative way of conceptualizing online social capital of MMOGs players and tested network effects on task performance, trust, and sense of community. Based on the trade network and mentoring network over the course of nine months, effective size and network constraint scores were calculated for every character. Effective size describes the extent to which a focal player brokers otherwise unconnected individuals in the network. A high effective size in one’s ego-network indicates abundant variations in information, ideas and resources one could access, thus leading to better task performance. Network constraint describes the extent to which a player connects with other players who are already connected among themselves. A high constraint score indicates redundant and consistent information and opinions, thus it may lead to group solidarity and trust. As such, the tension between network brokerage and closure maps well to the distinction between bridging and bonding social capital. Results showed that brokerage (as measured by effective size) had a strong impact on task performance. Controlling for demographics, character class and total play time, players who span many structural holes tended to reach a higher character level than those who were constrained by redundant (interconnected) contacts in the trade and mentoring networks. This finding is consistent with previous studies in offline worlds (e.g., Podolny & Baron, 1997; Zaheer & Soda, 2009) . It also resonates with the finding

174 from a similar study conducted in another popular virtual world, Second Life (Burt, 2010), which found that brokers were more likely to found groups (both invitation-only groups and groups open to the public) and the groups they founded were more likely to survive and attract more people as members. Unlike Second Life which represents an unstructured sandbox-type virtual world, EQII represents another genre of virtual worlds with a relatively fixed story line and well-defined character development trajectories. The current study thus provides additional empirical evidence for the achievement-increasing effect of network brokerage in a different type of virtual world. Taken together, it suggests that relations established in virtual worlds still demonstrate the same network brokerage effects initially found in offline social worlds. This is initial construct-validity evidence that virtual world networks can be used as parallel for understanding real world social dynamics. However, the closure effects did not receive empirical support. Players who are socially embedded in closed networks do not show a heightened trust towards guildmates or other players in the game, nor do they feel an increased sense of community online. There are two plausible explanations for this unexpected finding. The first possibility is that the dependent variables, trust and sense of community, are measures of individuals’ social engagement as a result of their overall relationships in EQII and online. But the theory of network closure predicts that the players in a constrained relationship tend to trust each other. In other words, the dependent variables (network level) are crude

175 approximations of the predicted effect (dyadic level), which may have contributed to the unexpected finding. The second explanation paints a more dystopian picture of the impact of online social relationships. Even though players may become deeply embedded in online gaming worlds, the virtual interactions they have may not necessarily lead to better social outcomes. This explanation suggests that close networks in EQII may not be effective in generating additional bonding social capital alone. To further investigate the viability of the two possible explanations, future research needs to use more refined measures of dyadic trust and group solidarity, and to employ a more rigorous study design (e.g., experiment) to examine the causal relationships between players’ pre-existing social resources, network closure, and bonding social capital. The current study represents one of the first attempts to test a structural conceptualization of online social capital. This network-based approach complements the outcome-based approach prevalent in current scholarship, because it offers a better understanding of the casual mechanisms leading to observed social consequences. As such, this approach is particularly valuable in highlighting social capital creation—how sources (network structure) are connected to outcomes (social capital), and what contingency variables may affect this process. An example of a contingency variable is the content of the network (Burt & Schott, 1985). As shown in Table 11, other things being equal, one unit increase in the

176 effective size of one’s mentoring network contributed 0.6 unit increase in character level, while one unit increase in the effective size of one’s trade network only contributed 0.09 unit increase in character level. Both effective network sizes were found to have significant impact on task performance, but one was much more substantial than the other. A plausible explanation is that compared to trade relationships, mentoring relationships entail greater commitment—mentoring happens during a group or raid event, with a more advanced character temporarily lowering in level to match a lesser character. Mentoring is also more exclusive (one apprentice at a time) and substantive, as it may last through a several-hour long play session filled with multiple deaths as well as successful battles. Because of the difference in the nature of these networks, it is not surprising that players typically have fewer mentoring ties (M = 5.83) than trade ties (M = 74.94). The discrepancy in network content could have contributed to their differing effects— bridging relations in the mentoring network are more valuable than those in the trade network in providing knowledge, resources, and combat tactics, and thus are more effective in enhancing task performance. A promising future research direction, therefore, is to systematically test the effects of social structure in different types of networks (e.g., chat, trade, collaboration). In sum, this example clearly demonstrates the advantage of the structural approach, as it focuses on the causal agents leading to the creation of online social capital.

177 In addition, the current study also offers some evidence that the social capital effects (especially the achievement-enhancing effects of brokerage) observed in the offline world may be equally valid in virtual worlds. It suggests that virtual world networks could be viable parallels to real world networks and may provide an additional avenue to generate insights on social dynamics in the offline world. This study joins recent research efforts (Burt, 2010; Lazer et al., 2009) in establishing the validity of virtual worlds as legitimate and productive sites for social science research. Network Evolution An ecological framework was employed to examine the variation, selection and retention processes of player collaboration ties over a period of 13 weeks. Individuals are limited in their carrying capacity to create and maintain ties. The numerous ties generated through random and haphazard encounters in the EQII world are then subject to multiple selection forces. For individual ties, three sets of selection mechanisms were tested: aging and inertia, social architecture, and homophily and proximity. The longitudinal network data provide an illustration of how variation, selection and retention processes of network ties unfold as players advance in the game. The beginners were “hyper-connectors” who interacted with a large number of players to experiment with different strategies of connection and different sets of partners, but these connections were extremely volatile and ephemeral. As players reached higher levels, the promiscuity in connections also waned. They accumulated more experience in identifying

178 compatible partners, resulting in smaller and more selective networks. Contrary to expectation, the total time spent playing EQII did not show a significant effect on tie decay. Therefore, character level (instead of total play time) seemed to be a better indicator of individuals’ learning and maturation processes. Another important factor of aging is the duration of ties. The longer a tie had been maintained between two players, the less likely it would discontinue. Together, the findings were indicative of two types of learning, taking place concurrently as players experimented with different networking strategies and aligned their goals and objectives with their ensemble of network partners: 1) learning to identify compatible partners, and 2) learning how to maintain social ties. The effects of aging and inertia on the evolutionary trajectory of player networks are not very different from longitudinal studies conducted in offline settings. Although it is difficult to compare the degree of network turnover across different research contexts, social ties established in the offline world are also quite transient. For example, a longitudinal study of 33 residents in Toronto showed that only 27% of intimate ties persisted after a decade (Wellman, Wong, Tindall, & Nazer, 1997). Burt’s (2000a) systematic study of tie decay based on a four-year panel study of investment bankers also showed that three in four relationships disappear one year later. Similar to what was observed in the online world of EQII, aging and inertia processes constituted an important set of predictors of tie persistence and decay among bankers. Both node age, as

179 measured by bankers’ age, seniority, and job rank, and tie age, as measured by the duration of ties, significantly slowed down the speed of tie decay (Burt, 2000a). The results also highlighted the important role of social architecture in retaining collaborative ties. Difference in character level contributed to the likelihood of tie decay, but the effect was quite small. This finding was not surprising, given that EQII imposes a penalty on collaboration between players with disparate levels: the higher level character would absorb most of the experience points, leaving the lower level player emptyhanded. Because people may differ in their time commitment to EQII, it is quite likely that two friends who started together found themselves not able to collaborate after a month, due to their level difference. The mentoring system could partially mitigate this situation, but the mentor has to bear the cost of lowered character level and some repeated game content. Clearly, these elements of the collaboration mechanism in EQII hinder the creation and development of sustained social networks by effectively restricting the pool of players available for connection. When a player is trying to look for a group or when a group is trying to look for more members, character level becomes one of the most important parameters to “filter” possible candidates, which significantly limits the potential opportunities for social interaction. This also suggests that empirical studies examining MMOG sociability should not overestimate the population size of available players. In other words, at a specific time point, if there are 1000 active characters on a specific server, for each character, the pool of potential playmates is often

180 significantly smaller. Taking into consideration the different real life schedules and the potential time commitment for a specific group task, it is not uncommon that players can have a hard time finding suitable others to play together, thus “soloing” becomes the only choice. Belonging to the same guild reduced the risk of tie decay between two players by half (see Figure 13 for an illustration). This suggests that much of the collaboration is structured around guild members. As player-created social institutions, guilds help partition the vast social space of EQII into smaller and more intimate units. Just like neighborhoods provide a basis for place-based interactions (Wellman & Gulia, 1999), guilds provide a virtual “locality” for like-minded players to socialize and collaborate. Feld’s (1981) concept of “social foci” offers a useful way to understand the social function of guilds. Social foci are anything that co-locates people, creating opportunities for relationship formation and development. Examples are a classroom co-locating children, a research project co-locating scholars, and a parent-teacher association at a local school co-locating parents of children attending the same school. Guilds could also be conceptualized as social foci, which bring together players who share the values and play styles compatible with those of the guild. This finding is particularly interesting given that guilds are quite unstable and often involve a high level of churn (Chen et al., 2008; Ducheneaut, Yee et al., 2007). Also, entering and leaving a guild incurs little cost to the player (Galston, 2000). Nonetheless, even a social institution with very low exit

181 and entry cost can prove to be effective in strengthening ties and building cohesive networks among players. An interesting future research direction is to examine how guild turnover, guild stability, and guild internal dynamics (e.g., management style) influence the longevity of interactions among guildmates. Finally, the effects of homophily and proximity on tie decay were tested. Age homophily and gender homophily did not have a significant effect on tie decay. This finding was not surprising, as demographic attributes such as age and gender are not immediately visible in virtual worlds. Also, several studies have shown that a significant portion of female players started participating in MMOGs to spend time with their romantic partner (Williams et al., 2009; Yee, 2006a), which may explain why gender homophily was not observed. By contrast, proximity was found to be a basic logic determining the risk of tie decay. This finding is particularly interesting because virtual worlds are supposed to mitigate the clustering effects of location on social networks. Two mechanisms may be at play. First, although location details are not immediately visible to others in EQII, players can communicate with each other about basic information and decide whether to maintain the relationship. Second, individuals may have already known each other offline and migrate their offline relations into virtual worlds. These migrated relationships, with their pre-existing history in the offline world, are stronger and more likely to persist. It should be noted that this dissertation used a crude measure of proximity (same state). Future research should adopt more nuanced measures, such as the

182 actual distance between two players, so that it may offer more insights about how offline proximity functions in online worlds. Limitations and Future Research Limitations This study has a number of limitations due to the nature of the data. First, as the data used in the analyses come exclusively from behavioral server logs and user surveys of the online gaming community, EverQuest II, this study is better considered as a case study of one specific MMOG, thus generalizing the findings here to other online MMOGs or gaming communities should be taken with caution. Although the format and features of EQII are fairly representative of the MMOG genre, the general population engaged in EQII may be distinct from participants in other MMOGs. For example, because EQII is a sequel of the MMOG EverQuest, many EQII players have been attracted to EQII because of their prior experience with the original EverQuest. As a result, a large segment of EQII participants have had substantive history with MMOGs, and are more likely to be hardcore MMOG players. Also, the game content and mechanics of EQII are considered to be quite challenging, with its steep learning curve and stringent structure of character advancement. It is not uncommon to see players complaining in Internet forums about how difficult and time-consuming it is to play EQII. This is in stark contrast with some other MMOGs on the market, such as World of Warcraft, which actively attempts to make the game more accessible to casual players by

183 demanding less “grinding,” a non-entertaining but necessary process for game progression (Ducheneaut et al., 2006). Therefore, it is reasonable to expect that the EQII population may be inherently different from gamers of other MMOGs in some aspects (e.g., more dedicated and more tolerant of demanding game content), a speculation that warrants further investigation, especially systematic comparisons across different MMOG worlds. Second, although the unique combination of behavioral logs and self-reported survey data is a critical feature of this study, each type of data source has its own limitations. One important limitation of the behavioral server logs is the lack of chat communication conducted in group, raid, and guild chat channels. In EQII, however, group, raid, and guild chat channels are often used in conjunction with private tells (included in the behavioral logs). Most coordination during a group combat, for example, occurs in the group chat channel. Similarly, the guild chat channel serves as a central communal space where a guild conducts its general affairs (e.g., announcement of an upcoming guild raid) and social interactions among various guildmates may occur. Private tells, on the other hand, are only between two players, and are sometimes used as a substitute for group or guild chat channels when the conversation involves mature language, inappropriate topics (e.g., recruiting for another guild), or heated arguments (Blackhawks, 2010). As a result, the chat network used in this study is narrowly focused

184 on private chat and does not include networks resulting from communication in shared channels. The second type of data—a large scale EQII user survey—was collected from a total of 7000 individual players (only a portion was used in this study) by offering all respondents a special in-game item. Even though this incentive proved to be a very effective tool for recruitment and the planned sample size was reached much faster than expected, this sample is still not a true random sample in its most strict sense, thus it may suffer from self-selection bias. In addition, the survey was conducted in January 2007, four month later than the most recent records available in the behavioral logs (September 2006). Such a time gap again warrants cautious interpretation of the results, which might have been subject to the influence of various unobserved factors during this time period. Third, like the majority of studies on virtual world behaviors, this study examined “in-world” interactions only. Interactions via other means are not reflected in the analyses. This is an important limitation because interpersonal relationships are often sustained using multiple media and across various communication contexts (Haythornthwaite, 2002; Parks & Floyd, 1996). Not only are some players using MMOGs as an additional venue to interact with existing friends and family offline (Nardi & Harris, 2006; Yee, 2009), they are also found to engage each other, including those originally met online, in various communication channels, including voice chat, web forums, and offline gatherings (Griffiths et al., 2004; Skoric et al., 2010; Williams,

185 Caplan, & Xiong, 2007). Therefore, because of these unobserved behaviors outside of EQII, network patterns uncovered in this study may be considered as a somewhat conservative representation of the actual social interactions. Future research is warranted to employ more comprehensive data collection methods, such as qualitative interviews, surveys, content analysis of web forums, time diaries and Internet activity logs, and to cover multiple communication channels. There are some methodological limitations as well. First, most of network analyses (except brokerage and closure measures) conducted in this study did not take relationship strength into consideration. In other words, networks of chat, trade and collaboration (including grouping and mentoring) were constructed from individual events, but the original counts of events during the observation period were not given any weight. This was due in part to the lack of theoretical or empirical basis to guide the selection of an appropriate cut-point. Also, most of the longitudinal network analysis approaches (see the review of available methods in Chapter 3) are not able to incorporate tie strength into the calculation. Clearly, more sophisticated methods and tools are needed to better analyze weighted networks. Second, this study did not have an accurate measure of actual play time during the observation period, so every active user during a week was assumed to have logged the same amount of play. Further, this approach assumed that all active characters during a week were available for collaboration, while it did not consider that these characters may have been active during different periods (e.g., one character

186 playing on Monday morning and another on Tuesday evening), rendering collaboration impossible. To generate a dynamic list of available characters at any time point requires complete records of individual play sessions. The behavioral logs do not provide records of individual user’s play sessions. An approximate play session could be calculated, however, by tracking every user’s dynamic activities throughout the observation period and dissecting these activities into play sessions based on a series of criteria. Despite its intensive computational requirement, this approach could be a valuable addition to future analyses of the dataset. Lastly, each account in EQII is allowed to create up to 8 characters. Due to the nature of most analyses, the character (instead of the account) was selected as the main unit, suggesting a case of nested data. A closer examination of the complete dataset showed that every account had an average of 1.29 characters. For weekly data, the average number of characters per account was even smaller (M = 1.01), perhaps because players tended to concentrate their effort on one character at a time. Therefore, the multi-level nature of the data did not have a significant implication on the analyses. Directions for Future Research The theories and empirical findings presented in this study suggest a wide range of directions for future research, which are discussed in detail below.

187 Different Forms of Sociability In examining the general patterns of social interaction in EQII, the findings of this study resonate with Ducheneaut et al.’s study of World of Warcraft. They referred to this in-world phenomenon as “alone together.” This study buttressed the previous finding through a systematic examination of EQII, also using behavioral data obtained directly from the game servers. Combined, both studies suggest that sociability in MMOGs is rather diffuse and multifaceted. Assuming that all the MMOG players are actively engaged in community life is to perpetuate another stereotype, one that counters the myth of “isolated gamers” but is equally misleading. A fruitful future research direction, therefore, is to unpack the aggregate concept of sociability or “the social factor” and pay greater attention to its many manifestations, including passive as well as active forms. Qualitative research (e.g., interviews) could be especially helpful in uncovering the many dimensions of sociability, providing the basis for scale development. Field surveys are able to connect these dimensions of sociability with personal attributes and specific behaviors in the game. In doing so, we would be able to critically evaluate the benefits, as well as costs, associated with the social life carried out in these online spaces. Social Networks of Competition and Conflict A less considered possibility in social network research is that relationships may be negative or detrimental. It should be noted, however, that relationships and networks are analytic constructs. What constitutes a tie, therefore, is rather arbitrarily defined and

188 operationalized by scholars for the purpose of network analysis (Burt & Schott, 1985). An example is the decision in this study to concentrate on the chat, trade and collaboration relations among players on the Guk server (PvE), thus intentionally excluding relationship dynamics on PvP servers where people may engage in competitive rather than collaborative interactions. With the extensive data on Player-versus-Player encounters, MMOGs represent an excellent venue to investigate competition and conflict in networks, an aspect rarely examined in the research of personal networks. For example, two opposing guilds may engage in a direct battle, the result of which provides a direct measure of relative fitness. Players and play associations may also collaborate, form alliances, and develop dynamic relationships of symbiosis and commensalism with each other in order to obtain a competitive edge. From an evolutionary perspective, this provides an opportunity to test evolutionary predictions about competition and fitness, thus generating insights about effective strategies for linking and group assembly. Participation Life Cycles One key finding from this study is that social relationships established in online gaming communities are remarkably transient. This finding casts some doubt on the validity and reliability of previous studies on relationship formation in virtual worlds employing cross-sectional data. It also suggests the necessity of incorporating the temporal dimension in the design and implementation of future studies. The temporal dimension should recognize that participants (nodes), ties among participants (ties), as

189 well as the formal player organizations (guilds) all have their respective life cycles. The social dynamics observed within these worlds are collectively shaped by the entering and exiting player cohorts, formation and dissolution of relations, and the creation, competition and demise of player organizations. As some scholars suggested, future research should consider not the stagnant node, relationship, or player association, but the entire ecosystem that constitutes evolving players, relationships as well as guilds (Chen et al., 2008). Comparisons Across MMOGs and Other Communities As noted in the discussion of limitations, this study is better considered as a case study of a specific online gaming community, EQII. In the early stage of the field when empirical data remain scant, the present study provides a much needed benchmark on the social experiences and their effects in MMOG worlds. Yet still, the findings and insights generated in this study beg the ultimate question of generalizability. A cursory look at the existing scholarship of MMOGs shows that only a very limited pool of online worlds has been investigated so far. Most current work concentrates on World of Warcraft (e.g., Chen et al., 2008), while several earlier pieces examined EverQuest (e.g., Taylor, 2006) that predates EQII. The over-representation of a few virtual worlds poses a potential danger in equating MMOG research to WoW or EQ research, leading to the dominance of the most populated MMOG in research literature. To address the question of

190 generalizability, scholars are encouraged to conduct replication and comparison studies in different MMOGs and online communities. Virtual Worlds as Testbeds for Social Science Theory Lastly, this study demonstrates the utility of social science theory (i.e., theories of social capital and network evolution) in explaining and predicting behaviors in the virtual world. The results signal that networks among virtual avatars are not unlike social networks among people in the offline world, and more importantly, these networks bring the same influence and undergo the same stages of evolution. This represents an important premise of validity that some behavioral mechanisms are equivalent in offline and online worlds. Not only could social theories, originally developed to explain offline behavior, be applied to account for phenomena in online worlds, online behaviors could also be used to extend and refine these theories. For example, there is some empirical evidence that economic behaviors in virtual worlds follow real-world patterns (Castronova et al., 2009). Compared to large-scale experiments in real markets, the cost of the same test in virtual worlds is much lower. Virtual worlds could also offer timestamped, high-fidelity data at a precision unmatched in most offline research contexts. Therefore, testing the activities and social dynamics in MMOGs could offer valuable insights about those in the real world. This opens up a whole new research opportunity to use virtual worlds as productive venues for theory development, empirical tests and

191 training, provided that the rules and parameters in one can map onto the other (Williams, in press). Conclusion This dissertation presents a critical examination of the social interactions among MMOG participants by focusing on network patterns, effects and evolution. It is situated in a popular MMOG, EverQuest II (EQII), drawing on a combination of unobtrusively collected behavioral server logs and a comprehensive survey conducted with the players directly through the game engine. An exploratory analysis of network patterns revealed that the social architecture of the world was quite effective in shaping the structure of interaction, as the involvement in various social networks was influenced by class choice and character level. However, sociability among players was quite diffuse, with a sizable number of players opting to play solo despite the built-in mechanisms that encourage collaborative play. Second, drawing on the theory of social capital, this study tested the effects of different structural properties of player social networks. Players who bridged diverse, otherwise unconnected partners were rewarded with better task performance in EQII. But contrary to expectation, players located in dense and closed cliques did not show higher level of trust towards playmates or sense of community. Lastly, a longitudinal analysis of tie persistence and decay demonstrated the transient nature of social relationships in EQII. But these ties became considerably more durable over time. Also, geographic proximity, level

192 similarity and shared guild membership were powerful mechanisms in preserving social relationships.

193 Bibliography Abrams, D., & Hogg, M. A. (1999). Social identity and social cognition. Oxford ; Malden, Mass.: Blackwell. Adamic, L., Buyukkokten, O., & Adar, E. (2003). A social network caught in the web. First Monday, 8(6). Adamic, L. A., & Glance, N. (2005). The political blogosphere and the 2004 U.S. Election: Divided they blog. In Proceedings of the 3rd international workshop on link discovery (pp. 36-43): ACM. Adler, P. S., & Kwon, S.-W. (2002). Social capital: Prospects for a new concept. The Academy of Management Review, 27(1), 17-40. Aldrich, H. (1999). Organizations evolving. London ; Thousand Oaks: Sage Publications. Allen, T. (1977). Managing the flow of technology. Cambridge, MA.: MIT Press. Allison, P. D. (1984). Event history analysis : Regression for longitudinal event data. Beverly Hills, Calif.: Sage Publications. Amburgey, T. L., Kelly, D., & Barnett, W. P. (1993). Resetting the clock: The dynamics of organizational change and failure. Administrative Science Quarterly, 38(1), 5173. Ancient Legacy. (2010). Guild description. Retrieved July 1, 2010, from http://www.guildportal.com/Guild.aspx?GuildID=42776&TabID=377451 Anderson, C. A. (2004). An update on the effects of playing violent video games. Journal of Adolescence, 27(1), 113-122. Astley, W. G. (1985). The two ecologies: Population and community perspectives on organizational evolution. Administrative Science Quarterly, 30(2), 224-241. Bagozzi, R. P., & Dholakia, U. M. (2006). Open source software user communities: A study of participation in linux user groups. Management Science, 52(7), 10991115. Bainbridge, W. S. (2007). The scientific research potential of virtual worlds. Science, 317(5837), 472-476.

194 Bargh, J. A., & McKenna, K. Y. A. (2004). The Internet and social life. Annual Review of Psychology, 55(1), 573-590. Bartle, R. (1996). Players who suit muds. Journal of MUD research, 1(1). Bartle, R. (2003). Designing virtual worlds. Indianapolis, IN: New Riders Games. Baum, J. A. C., & Oliver, C. (1992). Institutional embeddedness and the dynamics of organizational populations. American Sociological Review, 57(4), 540-559. Blackhawks. (2010). FAQ and guild charter. Retrieved July 1, 2010, from http://www.blackhawksguild.com/index.php?name=FAQ&cat=1 Boase, J., Horrigan, J., Wellman, B., & Rainie, L. (2006). The strength of Internet ties: The Internet and email aid users in maintaining their social networks and provide pathways to help when people face big decisions. Washington DC: Pew Internet & American Life Project. Borgatti, S. P., Everett, M. G., & Freeman, L. C. (2002). Ucinet for windows: Software for social network analysis. Harvard, MA: Analytic Technologies. boyd, d. m., & Ellison, N. B. (2007). Social Network Sites: Definition, history, and scholarship. Journal of Computer-Mediated Communication, 13(1), 210-230. Brown, J. S., & Thomas, D. (2008). The gamer disposition. Harvard Business Review. Bryant, J. A., & Monge, P. (2008). The evolution of the children’s television community, 1953-2003. International Journal of Communication, 2, 160-192. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, Mass.: Harvard University Press. Burt, R. S. (2000a). Decay functions. Social Networks, 22(1), 1-28. Burt, R. S. (2000b). Network structure of social capital. In B. M. Staw & R. I. Sutton (Eds.), Research on organizational behavior (pp. 345-423). Amsterdam; London and New York: Elsevier Science JAI. Burt, R. S. (2002). Bridge decay. Social Networks, 24(4), 333-363. Burt, R. S. (2005). Brokerage and closure: An introduction to social capital. Oxford; New York: Oxford University Press.

195 Burt, R. S. (2010). Structural holes in virtual worlds: University of Chicago. Burt, R. S., & Schott, G. (1985). Relation contents in multiple networks. Social Science Research, 14(4), 287-308. Byrne, D. E. (1971). The attraction paradigm. New York: Academic Press. Campbell, D. (1969). Variation and selective retention in socio-cultural evolution. General Systems, 14, 69-85. Campbell, D. T. (1965). Variation and selective retention in socio-cultural evolution. In H. R. Barringer, G. I. Blanksten & R. W. Mack (Eds.), Social change in developing areas: A reinterpretation of evolutionary theory (pp. 19-48). Cambridge, MA: Schenkman. Carroll, G. R. (1985). Concentration and specialization: Dynamics of niche width in populations of organizations. American Journal of Sociology, 90(6), 1262-1283. Carroll, G. R., Dobrev, S. D., & Swaminathan, A. (2002). Organizational processes of resource partitioning. Research in organizational behavior, 24, 1-40. Carroll, G. R., & Hannan, M. T. (2000). The demography of corporations and industries. Princeton, N.J.: Princeton University Press. Carroll, J. (2004). Literary darwinism : Evolution, human nature, and literature. New York: Routledge. Castronova, E. (2005). Synthetic worlds: The business and culture of online games. Chicago: University of Chicago Press. Castronova, E. (2007). Exodus to the virtual world: How online fun is changing reality. New York: Palgrave Macmillan. Castronova, E., Williams, D., Shen, C., Ratan, R., Xiong, L., Huang, Y., et al. (2009). As real as real? Macroeconomic behavior in a large-scale virtual world. New Media & Society, 11(5), 685-707. Center for the Digital Future. (2009). The 2009 digital future report: Surveying the digital future - year eight. Los Angeles, CA: USC Annenberg School Center for the Digital Future.

196 Chen, C.-H., Sun, C.-T., & Hsieh, J. (2008). Player guild dynamics and evolution in massively multiplayer online games. CyberPsychology & Behavior, 11(3), 293301. Chen, M. G. (2009). Communication, coordination, and camaraderie in World of Warcraft. Games and Culture, 4(1), 47-73. Cole, H., & Griffiths, M. D. (2007). Social interactions in massively multiplayer online role-playing gamers. CyberPsychology & Behavior, 10(4), 575-583. Cole, J. I., Suman, M., Schramm, P., Bel, D. V., & Aquino, J. S. (2000). The UCLA Internet report: Surveying the digital future. Los Angeles: UCLA Center for Communication Policy. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95-S120. Cook, K. S., Emerson, R. M., & Gillmore, M. R. (1983). The distribution of power in exchange networks: Theory and experimental results. American Journal of Sociology, 89(2), 275-305. Cook, T. D., & Campbell, D. T. (1979). Quasi-experimentation: Design & analysis issues for field settings. Boston: Houghton Mifflin Company. Culnan, M., & Markus, M. (1987). Information technologies. In F. Jablin, L. Putnam, K. Roberts & L. Porter (Eds.), Handbook of organizational communication: An interdisciplinary perspective (pp. 420-443). Newbury Park, CA: Sage. Daft, R. L., & Lengel, R. H. (1986). Organizational information requirements, media richness, and structural determinants. Management Science, 32, 554-571. De Souza, C. S., & Preece, J. (2004). A framework for analyzing and understanding online communities. Interacting with Computers, 16(3), 579-610. Denning, P. J., Flores, F., & Luzmore, P. (2010). Orchestrating coordination in pluralistic networks. Communications of the ACM, 53(3), 30-32. Despair. (2010a). Home. Retrieved July 1, 2010, from http://www.despaireq2.com/index.php?gid=150389&page_id=63442 Despair. (2010b). Raid rules. Retrieved July 1, 2010, from http://www.despaireq2.com/custom.php?page_id=63442&gid=150389

197 DiMaggio, P. J. (1994). The challenge of community evolution. In J. A. C. Baum & J. V. Singh (Eds.), Evolutionary dynamics of organizations (pp. 299-319). New York: Oxford University Press. Dimmick, J. W. (2003). Media competition and coexistence: The theory of the niche. Mahwah, N.J.: L. Erlbaum Associates. Donath, J. (2007). Signals in social supernets. Journal of Computer-Mediated Communication, 13(1), 231-251. 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. Ducheneaut, N., Yee, N., Nickell, E., & Moore, R. (2006). "Alone together?" Exploring the social dynamics of massively multiplayer online games. In Proceedings of the SIGCHI conference on human factor in computing systems (pp. 407-416). New York, USA: ACM Press. 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. Paper presented at the SIGCHI conference on Human factors in computing systems, San Jose, California, USA. Dunbar, R. I. M. (1996). Grooming, gossip and the evolution of language. London: Faber and Faber. Dunbar, R. I. M. (2008). Cognitive constraints on the structure and dynamics of social networks. Group Dynamics, 12(1), 7-16. 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. Ethec. (2004). EQ2 etiquette. Retrieved July 1, 2010, from http://www.tentonhammer.com/node/54445 Eveland, J. D., & Bikson, T. K. (1986). Evolving electronic communication networks: An empirical assessment. Paper presented at the Conference Name|. Retrieved Access Date|. from URL|.

198 Executioners. (2010). Guild charter. Retrieved July 1, 2010, from http://www.guildportal.com/Guild.aspx?GuildID=153891&TabID=2328496 Feld, S. L. (1981). The focused organization of social ties. The American Journal of Sociology, 86(5), 1015-1035. Feld, S. L., Suitor, J. J., & Hoegh, J. G. (2007). Describing changes in personal networks over time. Field Methods, 19(2), 218-236. Feng, W.-c., Brandt, D., & Saha, D. (2007). A long-term study of a popular MMORPG. In Proceedings of the 6th ACM SIGCOMM workshop on network and system support for games (pp. 19-24). New York, USA: ACM Press. Festinger, L., Schacter, S., & Back, K. (1950). Social pressures in informal groups ([1st ed.). Palo Alto, CA: Stanford University Press. Fields, D. A., & Kafai, Y. B. (2010). Knowing and throwing mudballs, hearts, pies, and flowers. Games and Culture, 5(1), 88-115. Fischer, C. S. (1982). To dwell among friends: Personal networks in town and city. Chicago: University of Chicago Press. Freeman, L. (1979). Centrality in social networks: Conceptual clarification. Social Networks, 1, 215-239. Galston, W. A. (2000). Does the Internet strengthen community? National Civic Review, 89(3), 193-202. Ganley, D., & Lampe, C. (2009). The ties that bind: Social network principles in online communities. Decision Support Systems, 47(3), 266-274. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360-1380. Griffiths, M. (1997). Computer game playing in early adolescence. Youth & Society, 29(2), 223-237. Griffiths, M. D., Davies, M. N. O., & Chappell, D. (2003). Breaking the stereotype: The case of online gaming. CyberPsychology & Behavior, 6(1), 81-91.

199 Griffiths, M. D., Davies, M. N. O., & Chappell, D. (2004). Demographic factors and playing variables in online computer gaming. CyberPsychology & Behavior, 7(4), 479-487. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? The American Journal of Sociology, 104(5), 1439-1493. Hahn, J., Moon, J. Y., & Zhang, C. (2008). Emergence of new project teams from open source software developer networks: Impact of prior collaboration ties. Information Systems Research, 19(3), 369-391. Hampton, K., & Wellman, B. (2001). Long distance community in the network society: Contact and support beyond netville. American Behavioral Scientist, 45(3), 476495. Hampton, K., & Wellman, B. (2003). Neighboring in Netville: How the Internet supports community and social capital in a wired suburb. City and Community, 2(4), 277311. Handcock, M. S., Hunter, D. R., Butts, C. T., Goodreau, S. M., & Morris, M. (2003). statnet: Software tools for the statistical modeling of network data (version 2.0). from http://statnetproject.org Hannan, M. T., & Freeman, J. (1977). The population ecology of organizations. American Journal of Sociology, 82(5), 929-964. Hannan, M. T., & Freeman, J. (1984). Structural inertia and organizational change. American Sociological Review, 49(2), 149-164. Hannan, M. T., & Freeman, J. (1989). Organizational ecology. Cambridge, Mass.: Harvard University Press. Harris, H., Bailenson, J. N., Nielsen, A., & Yee, N. (2009). The evolution of social behavior over time in second life. Presence: Teleoperators and Virtual Environments, 18(6), 434-448. Haveman, H. A. (1992). Between a rock and a hard place: Organizational change and performance under conditions of fundamental environmental transformation. Administrative Science Quarterly, 37(1), 48-75. Haveman, H. A. (1993). Follow the leader: Mimetic isomorphism and entry into new markets. Administrative Science Quarterly, 38(4), 593-627.

200 Hawley, A. (1986). Human ecology: A theoretical essay. Chicago: University of Chicago Press. Haythornthwaite, C. (2002). Strong, weak, and latent ties and the impact of new media. The Information Society, 18(5), 385-401. Herring, S. (2004). Slouching toward the ordinary: Current trends in computer-mediated communication. New Media & Society, 6(1), 26-36. Hiltz, S. R., & Wellman, B. (1997). Asynchronous learning networks as a virtual classroom. Communications of the ACM, 40(9), 44-49. Hinds, P. J., Carley, K. M., Krackhardt, D., & Wholey, D. (2000). Choosing work group members: Balancing similarity, competence, and familiarity. Organizational Behavior and Human Decision Processes, 81(2), 226-251. Hirschman, A. O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organizations, and states. Cambridge, Massachusetts: Harvard University Press. Horrigan, J. B., Rainie, L., & Fox, S. (2001). Online communities: Networks that nurture long-distance relationships and local ties. Washington DC: Pew Internet & American Life Project. Hunt, C. S., & Aldrich, H. E. (1998). The second ecology: Creation and evolution of organizational communities. Research in Organizational Behavior, 20, 267-301. Inkpen, A. C., & Tsang, E. W. K. (2005). Social capital, networks, and knowledge transfer. Academy of Management Review, 30(1), 146-165. Kalman, M. E., Monge, P. R., Fulk, J., & Heino, R. (2002). Motivations to resolve communication dilemmas in database-mediated collaboration. Communication Research, 29(2), 125-154. Katz, J. E., & Aspden, P. (1997). A nation of strangers? Communications of the ACM, 40(12), 81-86. Katz, J. E., & Rice, R. E. (2002). Social consequences of internet use: Access, involvement, and interaction. Cambridge, Massachusetts: The MIT Press.

201 Katz, J. E., Rice, R. E., Acord, S., Dasgupta, K., & David, K. (2004). Personal mediated communication and the concept of community in theory and practice (chapter 9). In P. Kalbfleisch (Ed.), Communication yearbook (Vol. 28, pp. 315-371). Mahwah, NJ: Lawrence Erlbaum. Kidder, D. (2002). The influence of gender on the performance of organizational citizenship behaviors. Journal of Management, 28(5), 629-648. Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological aspects of computer-mediated communication. American Psychologist, 39, 1123-1134. Klemm, P., Bunnell, D., Cullen, M., Soneji, R., Gibbons, P., & Holecek, A. (2003). Online cancer support groups: A review of the research literature. Computers, informatics, nursing : CIN, 21(3), 136-142. Krackhardt, D. (1987). Cognitive social structures. Social Networks, 9(2), 109-134. Kraut, R., Egido, C., & Galegher, J. (1988). Patterns of contact and communication in scientific research collaboration. Paper presented at the ACM conference on Computer-supported cooperative work, Portland, Oregon, United States. Kraut, R., Kiesler, S., Boneva, B., Cummings, J., Helgeson, V., & Crawford, A. (2002). Internet paradox revisited. Journal of Social Issues, 58(1), 49-74. Kraut, R., Patterson, M., Lundmark, V., Kiesler, S., Mukopadhyay, T., & Scherlis, W. (1998). Internet paradox: A social technology that reduces social involvement and psychological well-being? American Psychologist, 53(9), 1017-1031. Lastowka, G. (2009). Planes of power: EverQuest as text, game and community. Game Studies, 9(1). Lazarsfeld, P. F., & Merton, R. K. (1954). Friendship as a social process: A substantive and methodological analysis. In M. Berger, T. Abel & C. H. Page (Eds.), Freedom and control in modern society. New York: Van Nostrand. Lazer, D., & Friedman, A. (2007). The network structure of exploration and exploitation. Administrative Science Quarterly, 52(4), 667-694. Lazer, D., Pentland, A., Adamic, L., Aral, S., Barabasi, A.-L., Brewer, D., et al. (2009). Social science: Computational social science. Science, 323(5915), 721-723.

202 Lenhart, A. (2009). Adults and social network websites Washington DC: Pew Internet & American Life Project. Lenhart, A., Jones, S., & Macgill, A. R. (2008). Adults and video games. Washington DC: Pew Internet & American Life Project. Leskovec, J., Kleinberg, J., & Faloutsos, C. (2007). Graph evolution: Densification and shrinking diameters. ACM Transactions on Knowledge Discovery from Data, 1(1), 2. Lessig, L. (2006). Code: Version 2.0. New York: Basic Books. Lewis, K., Kaufman, J., Gonzalez, M., Wimmer, A., & Christakis, N. (2008). Tastes, ties, and time: A new social network dataset using facebook.Com. Social Networks, 30(4), 330-342. Licklider, J. C. R., & Taylor, R. W. (1968). The computer as a communication device. Science and Technology Lin, N. (1999). Social networks and status attainment. Annual Review of Sociology, 25, 467-487. Lin, N., Cook, K., & Burt, R. S. (2001). Preface. In N. Lin, K. Cook & R. S. Burt (Eds.), Social capital: Theory and research (pp. vii-xii). New York: Aldine De Gruyter. Lubbers, M. J., Molina, J. L., Lerner, J., Brandes, U., Avila, J., & McCarty, C. (2010). Longitudinal analysis of personal networks. The case of Argentinean migrants in Spain. Social Networks, 32(1), 91-104. Malone, K.-L. (2009). Dragon kill points: The economics of power gamers. Games and Culture, 4(3), 296-316. Marx, G. T. (1999). What's in a name? Some reflections on the sociology of anonymity. The Information Society, 15(2), 99-112. McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), Frontiers in econometrics (pp. 105-142). New York: Academic Press. McFadden, D. (1981). Econometric models of probabilistic choice. In C. Manski & D. McFadden (Eds.), Structural analysis of discrete data: With econometric applications (pp. 198-272). Cambridge, MA: MIT Press.

203 McKelvey, B. (1982). Organizational systematics--taxonomy, evolution, classification. Berkeley: University of California Press. McKenna, K. Y. A., & Bargh, J. A. (2000). Plan 9 from cyberspace: The implications of the Internet for personality and social psychology. Personality and Social Psychology Review, 4(1), 57-75. McPherson, M., Smith-Lovin, L., & Brashears, M. E. (2006). Social isolation in America: Changes in core discussion networks over two decades. American Sociological Review, 71(3), 353-375. McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415-444. Mercken, L., Snijders, T. A. B., Steglich, C., Vartiainen, E., & de Vries, H. (2009). Dynamics of adolescent friendship networks and smoking behavior. Social Networks, 32(1), 72-81. Mok, D., & Wellman, B. (2007). Did distance matter before the Internet?: Interpersonal contact and support in the 1970s. Social Networks, 29(3), 430-461. Monge, P., Heiss, B. M., & Margolin, D. B. (2008). Communication network evolution in organizational communities. Communication Theory, 18(4), 449-477. Monge, P., Lee, S., Fulk, J., Weber, M., Shen, C., Schultz, C., et al. (in press). Research methods for studying evolutionary and ecological processes in organizational communication. Management Communication Quarterly. Monge, P. R., & Contractor, N. S. (2003). Theories of communication networks. New York: Oxford University Press. Monge, P. R., Lee, S., Fulk, J., Frank, L. B., Margolin, D., Schultz, C., et al. (in press). Evolutionary and ecological models for organizational communication. Management Communication Quarterly. Nardi, B. (2010). My life as a night elf priest: An anthropological account of World of Warcraft. Ann Arbor, MI: University of Michigan Press. Nardi, B., & Harris, J. (2006). Strangers and friends: Collaborative play in world of warcraft. In Proceedings of the 2006 conference on computer supported cooperative work. New York: ACM Press.

204 Nelson, R. R., & Winter, S. G. (1982). An evolutionary theory of economic change. Cambridge, Mass.: Belknap Press of Harvard University Press. Nie, N. H. (2001). Sociability, interpersonal relations, and the Internet: Reconciling conflicting findings. American Behavioral Scientist, 45(3), 420-435. Nie, N. H., & Erbring, L. (2002). Internet and society: A preliminary report. IT & Society, 1(1), 275-283. Nie, N. H., & Hillygus, D. S. (2002). The impact of internet use on sociability: Timediary findings. IT & Society, 1(1), 1-20. Norris, P. (2002). The bridging and bonding role of online communities. Harvard International Journal of Press/Politics, 7(3), 3-13. Nye, D. E. (2004). Technological prediction: A promethean problem. In M. Sturken, D. Thomas & S. J. Ball-Rokeach (Eds.), Technological visions: The hopes and fears that shape new technologies (pp. 159-176). Philadelphia, PA.: Temple University Press. Oh, H., Labianca, G., & Chung, M.-H. (2006). A multilevel model of group social capital. Academy of Management Review, 31(3), 569-582. Oldenburg, R. (1997). The great good place: Cafés, coffee shops, community centers, beauty parlors, general stores, bars, hangouts, and how they get you through the day. New York: Marlowe & Company. Parks, M. R., & Floyd, K. (1996). Making friends in cyberspace. The Journal of Communication, 46(1), 80-97. Pena, J., & Hancock, J. T. (2006). An analysis of socioemotional and task communication in online multiplayer video games. Communication Research, 33(1), 92-109. Pew Internet and American Life Project. (2010). Internet adoption. Retrieved Aug 10, 2010, from http://www.pewinternet.org/Trend-Data/Internet-Adoption.aspx Podolny, J. M., & Baron, J. N. (1997). Resources and relationships: Social networks and mobility in the workplace. American Sociological Review, 62(5), 673-693. Powell, W. W., White, D. R., Koput, K., & Owen-Smith, J. (2005). Network dynamics and field evolution: The growth of interorganizational collaboration in the life sciences. American Journal of Sociology, 111(5), 1463-1568.

205 Preece, J., & Maloney-Krichmar, D. (2005). Online communities: Design, theory, and practice Journal of Computer-Mediated Communication, 10(4), http://jcmc.indiana.edu/vol10/issue14/preece.html. Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. New York: Simon & Schuster. Putnam, R. D. (2001). Bowling alone : The collapse and revival of American community (1st Touchstone ed.). New York: Simon & Schuster. Quan-Haase, A., & Wellman, B. (2004). How does the Internet affect social capital? In M. Huysman & V. Wulf (Eds.), Social capital and information technology. Cambridge, Massachusetts: MIT Press. Quan-Haase, A., Wellman, B., Witte, J. C., & Hampton, K. N. (2002). Capitalizing on the net: Social contact, civic engagement, and sense of community. In B. Wellman & C. Haythornthwaite (Eds.), The Internet in everyday life (pp. 291-324). Oxford, UK: Blackwell. Ramasco, J. J., & Morris, S. A. (2006). Social inertia in collaboration networks. Physical Review E (Statistical, Nonlinear, and Soft Matter Physics), 73(1), 016122016127. Rheingold, H. (1993). The virtual community: Homesteading on the electronic frontier. Reading, MA: Addison-Wesley. Roberts, S. G. B., Dunbar, R. I. M., Pollet, T. V., & Kuppens, T. (2009). Exploring variation in active network size: Constraints and ego characteristics. Social Networks, 31(2), 138-146. Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction to exponential random graph (p*) models for social networks. Social Networks, 29(2), 173-191. Robins, G., Snijders, T., Wang, P., Handcock, M., & Pattison, P. (2007). Recent developments in exponential random graph (p*) models for social networks. Social Networks, 29(2), 192-215. Ruef, M., Aldrich, H. E., & Carter, N. M. (2003). The structure of founding teams: Homophily, strong ties, and isolation among U.S. Entrepreneurs. American Sociological Review, 68(2), 195-222.

206 Schulz, M. (1998). Limits to bureaucratic growth: The density dependence of organizational rule births. Administrative Science Quarterly, 43(4), 845-876. Shen, C., & Williams, D. (in press). Unpacking time online: Connecting Internet and MMO use with psychosocial well-being. Communication Research. Shklovski, I., Kiesler, S., & Kraut, R. (2006). The Internet and social interaction: A metaanalysis and critique of studies, 1995-2003. In R. Kraut, M. Brynin & S. Kiesler (Eds.), Computers, phones, and the Internet: The social impact of information technology (pp. 210-218). Oxford: Oxford University Press. Short, J., Williams, E., & Christie, B. (1976). The social psychology of telecommunications. London: Wiley. Shumate, M., Fulk, J., & Monge, P. (2005). Predictors of the international HIV&AIDS INGO network over time. Human Communication Research, 31(4), 482-510. Shumate, M., & Palazzolo, E. T. (in press). Exponential random graph (p*) models as a method for social network analysis in communication research. Communication Methods and Measures. Simmel, G., & Hughes, E. C. (1949). The sociology of sociability. The American Journal of Sociology, 55(3), 254-261. Skoric, M. M., Tang, S. W. H., Liao, Y., & Poor, N. D. (2010). Technology as place. In S. Allan (Ed.), Rethinking communication: Keywords in communication research (pp. 151-164). Cresskill, NJ: Hampton Press. Snijders, T. A. B., Pattison, P. E., Robins, G. L., & Handcock, M. S. (2006). New specifications for exponential random graph models. Sociological Methodology, 36(1), 99-153. Snijders, T. A. B., Steglich, C. E. G., Schweinberg, M., & Huisman, M. (2007). Siena 3.1. http://stat.gamma.rug.nl/siena.html. Snijders, T. A. B., van de Bunt, G. G., & Steglich, C. E. G. (2010). Introduction to stochastic actor-based models for network dynamics. Social Networks, 32(1), 4460. Soloing in EverQuest ii. (2010). Retrieved July 1, 2010, from http://www.tentonhammer.com/node/55334

207 Sproull, L., & Kiesler, S. (1991). Connections: New ways of working in the networked organization. Cambridge, MA: MIT Press. Steinfield, C., Ellison, N. B., & Lampe, C. (2008). Social capital, self-esteem, and use of online social network sites: A longitudinal analysis. Journal of Applied Developmental Psychology, 29(6), 434-445. Steinkuehler, C. (2004). Learning in massively multiplayer online games. In Y. Kafai, W. Sandoval, N. Enyedy, A. Nixon & F. Herrera (Eds.), Proceedings of the sixth international conference of the learning sciences (pp. 521-528). Mahwah, NJ: Erlbaum. 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. Swaminathan, A. (1995). The proliferation of specialist organizations in the American wine industry, 1941-1990. Administrative Science Quarterly, 40(4), 653-680. Taylor, T. L. (2006). Play between worlds: Exploring online game culture. Cambridge, Massachusetts: The MIT Press. Tönnies, F. (1957). Community and society (gemeinschaft und gessellschaft) (G. Simpson, Trans.). East Lansing: The Michigan State University Press. Tönnies, F. (1974). On gemeinschaft and gesellschaft. Retrieved Oct 1, 2008, from http://media.pfeiffer.edu/lridener/courses/GEMEIN.HTML Tuma, N. B., Hannan, M. T., & Groeneveld, L. P. (1979). Dynamic analysis of event histories. The American Journal of Sociology, 84(4), 820-854. Turner, J. C. (1987). Rediscovering the social group : Self-categorization theory. Oxford, UK ; New York, NY: B. Blackwell. Tushman, M. L., & Anderson, P. (1986). Technological discontinuities and organizational environments. Administrative Science Quarterly, 31(3), 439-465. Usher, J. M., & Evans, M. G. (1996). Life and death along gasoline alley: Darwinian and lamarckian processes in a differentiating population. The Academy of Management Journal, 39(5), 1428-1466.

208 van Duijn, M. A. J., van Busschbach, J. T., & Snijders, T. A. B. (1999). Multilevel analysis of personal networks as dependent variables. Social Networks, 21(2), 187-210. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8(2), 109-125. Walther, J. (2006). Nonverbal dynamics in computer-mediated communication, or :( and the net :('s with you, :) and you :) alone. In V. Manusov & M. Patterson (Eds.), The sage handbook of nonverbal communication (pp. 461-480). Thousand Oaks, CA: Sage. Walther, J. B. (1992). Interpersonal effects in computer-mediated interaction: A relational perspective. Communication Research, 19(1), 52-90. Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23(1), 3-43. Walther, J. B., Van Der Heide, B., Hamel, L. M., & Shulman, H. C. (2009). Selfgenerated versus other-generated statements and impressions in computermediated communication: A test of warranting theory using facebook. Communication Research, 36(2), 229-253. Wang, P., Robins, G., & Pattison, P. (2005). Pnet: A program for the simulation and estimation of p* exponential random graph models: University of Melbourne. Wartella, E., & Reeves, B. (1985). Historical trends in research on children and the media: 1900-960. Journal of Communication, 35(2), 118-133. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge ; New York: Cambridge University Press. Wasserman, S., & Pattison, P. E. (1996). Logic models and logistic regressions for social networks: I. An introduction to markov graphs and p*. Psychometrika, 61, 401425. Webb, E., Campbell, D., Schwartz, R., & Sechrest, L. (1966). Unobtrusive measures: Non-reactive research in the social sciences. Chicago: Rand McNally and Company. Wellman, B. (2001). Computer networks as social networks. Science, 293(5537), 20312034.

209 Wellman, B., & Gulia, M. (1999). Virtual communities as communities: Net-surfers don't ride alone. In M. Smith & P. Kollock (Eds.), Communities in cyberspace (pp. 167-194). London: Routledge. Wellman, B., & Haythornthwaite, C. (2002). The Internet in everyday life. Malden, Massachusetts: Blackwell Publishing. Wellman, B., Wong, R. Y.-l., Tindall, D., & Nazer, N. (1997). A decade of network change: Turnover, persistence and stability in personal communities. Social Networks, 19(1), 27-50. Wenger, E. (1998). Communities of practice: Learning, meaning, and identity. Cambridge: Cambridge University Press. White, P. (2009). MMOGData: Charts. Gloucester, United Kingdom. Williams, D. (2003). The video game lightning rod. Information, Communication & Society, 6(4), 523-550. Williams, D. (2006a). On and off the 'net: Scales for social capital in an online era. Journal of Computer-Mediated Communication, 11(2), 593-628. Williams, D. (2006b). Why game studies now? Gamers don't bowl alone. Games and Culture, 1(1), 13-16. Williams, D. (2007). The impact of time online: Social capital and cyberbalkanization. CyberPsychology & Behavior, 10(3), 398-406. Williams, D. (in press). The mapping principle, and a research framework for virtual worlds. Communication Theory. Williams, D., Caplan, S., & Xiong, L. (2007). Can you hear me now? The social impact of voice on internet communities. Human Communication Research. Williams, D., Consalvo, M., Caplan, S., & Yee, N. (2009). Looking for gender: Gender roles and behaviors among online gamers. Journal of Communication, 59(4), 700725. 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.

210 Williams, D., Kennedy, T. L. M., & Moore, R. J. (2010). Behind the avatar: The patterns, practices, and functions of role playing in MMOs. Games and Culture. Williams, D., Yee, N., & Caplan, S. E. (2008). Who plays, how much, and why? Debunking the stereotypical gamer profile. Journal of Computer-Mediated Communication, 13(4), 993-1018. Woodcock, B. (2008). MMOG subscriptions market share - April 2008. Retrieved June 10, 2010, from http://www.mmogchart.com/Chart7.html Yee, N. (2001, May). The norrathian scrolls (version 2.5). Retrieved June 10, 2010, from http://www.nickyee.com/eqt/report.html Yee, N. (2006a). The demographics, motivations and derived experiences of users of massively-multiuser online graphical environments. Presence: Teleoperators and Virtual Environments, 15, 309-329. Yee, N. (2006b). The labor of fun: How video games blur the boundaries of work and play. Games and Culture, 1(1), 68-71. Yee, N. (2006c). Motivations for play in online games. CyberPsychology & Behavior, 9(6), 772-775. Yee, N. (2009). Befriending ogres and wood-elves: Relationship formation and the social architecture of norrath. Game Studies, 9(1). Yuan, Y. C., & Gay, G. (2006). Homophily of network ties and bonding and bridging social capital in computer-mediated distributed teams. Journal of Computer Mediated Communication, 11(4), available from: http://jcmc.indiana.edu/vol11/issue14/yuan.html. Zaheer, A., & Soda, G. (2009). Network evolution: The origins of structural holes. Administrative Science Quarterly, 54(1), 1-31.