Helen Harris* Jeremy N. Bailenson Alexia Nielsen Nick Yee Department of Communication Stanford University Stanford, CA 94305-2050 Presence, Vol. 18, No. 6, December 2009, 434���448 The Evolution of Social Behavior over Time in Second Life Abstract The current study tracked 80 participants who spent an average of six hours per week in Second Life over six consecutive weeks. Objective measures of movement and chat were automatically collected in real time when participants logged in to Second Life. Data regarding the number of groups and friends was self-reported through online questionnaires on a weekly basis. Results demonstrated that although the social net- works of users continued to broaden over the course of the study, users became less inclined to explore regions, decreased their use of high-energy actions such as flying or running, and chatted less. We discuss implications for theories of virtual social interac- tion as well as the use of Second Life as a social science research platform. 1 Introduction This paper describes a six week long study that tracked behavior in the virtual environment, Second Life. Blascovich (2002) proposed a model of how virtual human representations may influence users and result in realistic behav- iors, that is, behaviors that would be seen in the real world. Many studies based on this model have illustrated the ways in which the characteristics of virtual environments and interactions result in realistic user behavior. Although most research has focused on individual behavior within interpersonal or small group contexts, only a handful of studies have applied the social influence model to study individual or group behaviors within larger social contexts. The study presented here contributes to the current literature by providing addi- tional evidence for the social influence model in a large-scale context. Further- more, it is unique in its use of the Second Life platform as a methodological tool for an extended, controlled study within a naturalistic virtual setting. The following sections review Blascovich���s model and research of realistic individual behavior within small and large scale social contexts. 2 A Framework for Studying Realism in Virtual Environments Virtual environments are two- or three-dimensional representations of natural or imagined spaces (Biocca & Levy, 1995 Blascovich, 2002 Blas- *Correspondence to helenh@stanford.edu. 434 PRESENCE: VOLUME 18, NUMBER 6

covich et al., 2002 Kalawsky, 1993 Lanier, 2001) that often contain objects or representations of humans. Im- mersive virtual environments (IVEs) heighten the per- ceptive experience of individuals, typically the visual channel via a head mounted display or projection sys- tem. Immersion increases feelings of environmental and social presence. Individuals feel as though they are inter- acting with their virtual surroundings instead of the physical space they occupy, and there is a greater con- nection with other digital human representations in the same virtual space. According to Blascovich and col- leagues, the realism of immersive virtual environments can be leveraged as a methodological tool to study so- cial processes, in particular, the manner in which virtual agents and avatars influence real behavior. One goal of using virtual environments in research is to understand human social behavior. Blascovich, and colleagues have developed a model of social influence in virtual environments that delineates the manner in which realism and perceived agency of digital human representations moderate influence (Blascovich, 2002 Blascovich et al., 2002). They define realism, the degree to which a digital human representation looks and be- haves like a real human, and the three dimensions that contribute to the overall realism effect, communicative, anthropomorphic, and photographic realism (Blascovich Blascovich et al.). Communicative realism, the extent to which a digital representation physically and socially behaves like a real human, is the most consequential dimension (Blascovich Blascovich et al.). Anthropo- morphism is important in its contribution to communi- cative realism, as physical attributes (e.g., mouth, arms) are required for speaking, gesturing, and moving (Blas- covich Blascovich et al.). The greater the realism a digi- tal human representation achieves, the less perceived agency is needed for social influence, and vice versa. The model also predicts that individuals will respond differ- ently to varying degrees of perceived agency, with com- puter controlled representations (agents) being the least socially influential, and human-controlled representa- tions (avatars) being the most (Blascovich et al.). Thus, agents need to display more realism than avatars in or- der for the same amount of social influence to occur. As realism and social influence increase in virtual environ- ments, so does the likelihood that individuals will re- spond in realistic ways. The following sections detail previous research that illustrates how individuals respond realistically (i.e., ex- hibit social influence in response to virtual stimuli as they would to real, physical stimuli) in virtual environ- ments, with the goal of reviewing the literature and demonstrating that social influence does occur in virtual worlds. 3 Previous Research on Realistic Behavior within Virtual Environments 3.1 Social Influence in Immersive Virtual Environments 3.1.1 Personal Space. Personal space is broadly defined as the physical region around an individual. Hall (1966) defines personal space as the space between a 1.5 m radius and a 4 m radius from the individual. Al- though the typical amount of personal space varies with factors such as culture and population density, it still remains important in gauging how individuals feel about their surroundings, need for privacy, and attitudes about other people (Sommer, 1969). Using a virtual reality environment, Bailenson and colleagues (Bailen- son, Blascovich, Beall, & Loomis, 2001, 2003) studied personal space and found that female participants more closely approached and inspected agents than avatars, suggesting that they were respecting the human-con- trolled avatars��� personal space. 3.1.2 Persuasion, Conformity, and Obedi- ence. Research in virtual worlds has also replicated a number of interpersonal psychological processes. For example, Hoyt, Blascovich, and Swinth (2003), discov- ered that participants were more socially inhibited while performing a complex task when they believed they were performing in front of an audience of avatars than when performing in front of an audience of agents, or with no audience at all a replication of this study was provided by Zanbaka, Ulinski, Goolkasian, and Hodges (2007). Eastwick and Gardner (2008) used the virtual world There.com to study interpersonal persuasion. Ex- Harris et al. 435

perimenter-controlled avatars employed two common persuasion techniques in an attempt to get participants to have their snapshots taken, and both were successful in the virtual environment. Moreover, to study confor- mity, using a virtual blackjack game Swinth and Blascov- ich (2001) studied participants��� betting behavior. When participants were seated at a table with virtual human players, they conformed to the betting patterns of the other players more when they believed the player avatars were human-controlled instead of computer-controlled. 3.1.3 Realism and Influence. As predicted by the model of social influence, individuals react differ- ently to varying degrees of realism expressed by a virtual human. The amount of eye gaze directed toward an individual has real effects in virtual spaces. For example, in one study, when virtual presenters used augmented eye gaze directed toward participants compared to natu- ralistic eye gaze, the female participants expressed more agreement with the presented information, and both genders reported a greater feeling of social presence (Bailenson, Beall, Loomis, Blascovich, & Turk, 2003). Virtual eye gaze is most effective when it is relevant to the content of the conversation itself. Garau, Slater, Bee, and Sasse (2001) confirmed that avatars displaying conversation-relevant eye gaze behavior had higher eval- uations and were perceived to be more present and in- volved than avatars with random eye gaze. In another example, Slater and colleagues (2006) ma- nipulated the realism of agents when they used virtual reality to replicate Stanley Milgram���s obedience experi- ment (1963). Participants were asked to administer a virtual shock to a virtual human whenever there was an incorrect response to a word test. Even though the par- ticipants were aware than the virtual human was not real, the behavioral and physiological responses were greater when participants received visual and aural feed- back, instead of textual feedback. 3.2 Variability in Virtual Human Features Although virtual humans can be created with nearly endless types of features and abilities (Bailenson & Blascovich, 2004), individual differences in physical appearances play as important a role in determining atti- tudes and behaviors as they do in real life. For example, sex stereotypes extend into virtual environments. In a virtual reality study by Guadagno, Blascovich, Bailen- son, and McCall (2007) male and female participants were presented with arguments about campus security measures by either male or female virtual humans. The greatest attitude change between pre-tests and post-tests occurred when the participants and the virtual humans were the same sex. In that study, female virtual humans were thought to be more likeable, replicating the find- ing that women are warm whereas men are viewed as competent (Fiske, Cuddy, Glicke, & Xu, 2002). Fox and Bailenson (2009) presented individuals in virtual reality with virtual females who exemplified either high or low female stereotypes in terms of appearance and behavior. Individuals who saw the stereotypical virtual females demonstrated higher levels of sexism and rape myth acceptance than individuals who saw virtual fe- males with visual/behavioral inconsistency. Like sex, race is another characteristic that elicits ste- reotypes and in-group favoritism in both real and virtual contexts. Dotsch and Wigboldus (2006, 2008) found that when Dutch participants needed to inspect an ava- tar���s clothing at a virtual bus stop, participants more closely approached avatars with White features than ava- tars with Moroccan facial features. Implicit negative atti- tudes toward Moroccans were positively correlated with the distance kept between participants and the Moroccan- featured avatars. Not only do real world racial attitudes affect virtual behavior, but a virtual experience can also shape real world racial attitudes. Groom and colleagues (Groom, Bailenson, & Nass, 2009) showed that regard- less of actual race, participants who saw themselves transformed into Black avatars had stereotypes activated, showing greater racial bias than participants who saw themselves as White avatars. Although this instance of perspective taking resulted in greater prejudice, other uses of perspective taking have reduced negative stereo- types. In a study by Yee and Bailenson (2006), partici- pants who saw their avatars represented as elderly in virtual reality associated more positive words with older 436 PRESENCE: VOLUME 18, NUMBER 6