Nonverbal Norms in VEs 1 The Unbearable Likeness of Being Digital: The Persistence of Nonverbal Social Norms in Online Virtual Environments

Every day, millions of users interact in real-time via avatars in online environments, such as massively-multiplayer online role-playing games (MMORPGs). These online environments could potentially be unique research platforms for the social sciences and clinical therapy, but it is crucial to first establish that social behavior and norms in virtual environments are comparable to those in the physical world. In an observational study of Second Life, a virtual community, we collected data from avatars in order to explore whether social norms of gender, interpersonal distance (IPD), and eye gaze transfer into virtual environments even though the modality of movement is entirely different (i.e., via keyboard and mouse as opposed to eyes and legs). Our results showed that established findings of IPD and eye gaze transfer into virtual environments: 1) Malemale dyads have larger IPDs than female-female dyads, 2) male-male dyads maintain less eye contact than female-female dyads, and 3) decreases in IPD are compensated with gaze avoidance as predicted by the Equilibrium Theory (Argyle, 1988). We discuss implications for users of online games as well as for social scientists who seek to conduct research in virtual environments. Nonverbal Norms in VEs 3 A significant portion of the world’s population spends time online every day, using email, chat-rooms, instant messaging, and websites to interact with one another. In one type of these online environments, known as massively-multiplayer online roleplaying games (MMORPGs), millions of users spend on average 22 hours a week interacting with each other through avatars (Griffiths, Davies, & Chappel, 2003; Yee, 2006a; Yee, 2006b). In the current work, we were concerned with examining the subtle nonverbal and verbal behaviors of avatars in MMORPGs and other online forums, and comparing online behavior to typical face-to-face behavior. There are two main reasons for studying the similarities between user behavior in online environments and typical behavior in physical environments—to determine the positive and negative effects that online environments have on users in a media effects tradition, and to validate online games as a platform to study physical social interaction, both on the micro and macro level. The first concerns a “Media Effects” tradition within behavioral science. According to this paradigm, in order to understand the impact that a certain medium has upon an individual, it is important to understand how well he or she differentiates the media from actual reality. For example, early work by Bandura (1969) demonstrated that children would imitate violence that they watched on television. Demonstrating this oneto-one mapping between mediated behavior (i.e., behavior seen on television and video games) and actual behavior is one of the cornerstones of the media effects tradition. Similarly, a vast majority of behavioral science within virtual environments is predominantly concerned with the construct of presence. Presence (Blascovich, 2002; Heeter, 1992; Lee, 2004; Lombard & Ditton, 1997; Loomis, 1992) is a latent construct Nonverbal Norms in VEs 4 that roughly measures how ‘real’ one believes a mediated environment is in terms of nonverbal behaviors (Garau, Slater, Bee, & Sasse, 2001), physiological responses (Slater, 1994) and other measures. In sum, to understand how large the potential a medium has to change an individual, researchers have typically measured how realistically a user behaves while inside of that medium. The second main motivation for comparing online behavior in Second Life with the types of behaviors people exhibit in the physical world concerns the use of such online arenas as a testing platform. In other words, if it is the case that behavior online is largely similar to physical behavior, then it becomes possible to use these online worlds to test behavioral science theories that are predominantly concerned with physical behavior, both at the micro level and at the macro level. At the micro level, people have utilized this paradigm within the laboratory to study a number of types of social interaction behaviors. For example, by being able to take advantage of the experimental control, precise measurement abilities, and ease of replication that is intrinsic to virtual environments (Blascovich et al., 2002), researchers have learned more about learning (Hoyt, Blascovich, & Swinth, 2003) human conformity, therapeutic potential (Gaggioli, Mantovani, Castelnuovo, Wiederhold, & Riva, 2003; Rizzo et al., 2002), and nonverbal behavior (Bailenson, Blascovich, Beall, & Loomis, 2001; Beall, Bailenson, Loomis, Blascovich, & Rex, 2003). Alternatively, at the macro level, there is the potential to use online forums to explore economics (Castronova, 2006) and legal issues (Lastowka & Hunter, 2003; Noveck, 2005). For example, if one wants to test a theory about currency and inflation, these online environments can function as a much more generalizable simulation than a mathematical model. A researcher can simply change the value of an Nonverbal Norms in VEs 5 online currency and witness the effects it has on the online community. In another example, researchers interested in studying the effect of parental investment on gender and dating behavior can simply change the rules of the online forum such that males are forced to carry a child instead of females, or to change the gestation period from 9 months to a shorter period. Because online forums allow manipulations that are not possible to implement in the physical world, they allow us to examine unique research questions. However, before attempting such elegant manipulations, it is crucial for us to demonstrate that behavior in these online worlds has some similarity to typical behaviors in physical worlds. Second Life Second Life is the virtual world where we conducted our current study. Its maker, Linden Labs, describes it as “a 3D online persistent space totally created and evolved by its users” (Second Life, 2005). Users navigate, interact, and view the world through their own customized avatar a digital representation of themselves. Every object in the world, including the avatars, the trees, the buildings, and the roller coasters are 3D objects rendered in real-time (see Figure 1). Users communicate via typed chat and pre-recorded animations. More elaborate animations allow more complex behaviors, such as dancing. Users interact with objects via a graphical user interface that is largely mouse-driven. Relevant Nonverbal Theories Nonverbal communication has been recognized as a key component of interpersonal interaction (Argyle, 1988; Hall, 1959). One aspect of nonverbal communication that has received a great deal of attention is proxemics, also known as interpersonal distance (see Hayduk, 1983 for a review). This line of research began with Nonverbal Norms in VEs 6 Hall’s observation that people maintain personal buffer spaces around themselves and each other (Hall, 1959). Others have shown that the size of this personal space can be moderated by other factors, such as culture (Hall, 1966), race and gender (Rosengrant & McCroskey, 1975), age (Willis, 1966), and affiliation (Evans & Howard, 1973). Several theories and findings within the proxemics literature guided our predictions for the current study. Gender and Interpersonal Distance. Research on the effects of gender on interpersonal distance (IPD) has been mixed. Some researchers have argued that IPD is largest among male-male dyads, smallest among female-female dyads, and between those extremes for mixed dyads. And to a certain extent, this has been demonstrated (Adler & Iverson, 1974; Aiello, 1977), however, meta-analyses have shown how mixed the overall data is. In a review of 110 studies (Hayduk, 1983), only 27 were found to support this hypothesis. The other 83 found mixed results or no gender differences at all. While the literature on the effect of gender on IPD is inconclusive, we hypothesized, in accordance with the dominant theoretical proposition in the literature (see Hayduk, 1983), that malemale dyads in an online environment would have larger IPDs than female-female dyads. Equilibrium Theory. According to research (Argyle, 1988; Burgoon & Walther, 1990; Hayduk, 1981; Patterson, 1982; Rosenfeld, Breck, Smith, & Kehoe, 1984), the degree of intimacy within a dyadic interaction is maintained by compensatory changes in gaze or IPD. In other words, if we get too close to a person with whom we do not want to share high amounts of intimacy, we can avert our gaze to reduce that undesired intimacy and return to an equilibrium state (such as not standing face-to-face in an elevator). Thus, in our current study, we predicted that this equilibrium effect could be documented in an Nonverbal Norms in VEs 7 online environment. We hypothesized that mutual gaze and IPD would be inversely correlated in an online environment. Within the accepted social distance of 12 feet (Hall, 1959), the closer two avatars are, the less likely they will maintain mutual eye contact. Gaze and Turn-Taking. Eye gaze also plays an important role in regulating turntaking behavior in conversations (Argyle, 1988). In particular, eye gaze signifies attention (Breed, Christiansen, & Larson, 1972). Thus, we hypothesized that mutual gaze in a dyad in an online environment would be more likely to occur if one of the two interactants was talking. Previous research has also demonstrated gender differences in mutual gaze. In particular, female-female dyads are more likely to exhibit mutual gaze than male-male dyads and mixed dyads (Mulac, Studley, Wiemann, & Bradac, 1987). Thus, we hypothesized that this same pattern of gender difference would be observed in a virtual environment. Previous studies in immersive virtual environments. Another reason to suspect that these social norms do carry over into online environments is because previous research in immersive virtual environments (IVE, see Blascovich et al., 2002 for overview) has shown that participants given digital representations behave according to well-known IDP norms. In an IVE, a user moves through the virtual environment by moving in the physical world and having their movements tracked and rendered via a series of sensors and displays. Studies in IVEs have shown that the amount of IPD changes with situational aspects of the relationship such as familiarity (Bailenson, Beall, Blascovich, Weisbuch, & Raimmundo, 2001) and mutual gaze is inversely correlated with IDP as predicted by the Equilibrium Theory (Bailenson, Blascovich, Beall, & Loomis, 2003). In the same way that media interfaces are treated as social actors (Reeves Nonverbal Norms in VEs 8 & Nass, 1996), we hypothesized that the social norms of the physical world will be observed in virtual environments such as Second Life. Method Data Collection A triggered script was used to collect information from avatars in the world. When triggered by a designated key press, the script would collect the name, Cartesian coordinates (x, y), and yaw of the 16 avatars closest to the user within a 200 virtual meter radius. The script would also track whether the avatars were talking at that given moment. The script would then store the information as a text file. Six research assistants, paid at an hourly rate for 10 hours a week, collected data within Second Life over a period of seven weeks. There were 688 zones (discrete locations) in Second Life, and undergraduates were each assigned to 115 zones. These research assistants were instructed to systematically explore the zones and trigger the script near locations where a group of at least two people were interacting. Several types of locations were excluded from the data collection because of activity-specific positional configurations. These were: 1) dance clubs, 2) sex clubs, 3) classrooms, 4) casinos and parlor games, and 5) similar locations where physical architecture constrained position and orientation (such as movie theatres). Measures The following variables were captured or calculated for each snapshot. Because our analyses of interpersonal distance and mutual gaze were based on dyads, we parsed the data to extract all unique dyads. For example, in an interaction among A, B, and C, there are 3 unique dyads. Nonverbal Norms in VEs 9 Gender. After each triggered script snapshot, the research assistance noted down the gender of each avatar that had been captured. This was necessary because the gender of an avatar is not retrievable via the scripting system in Second Life. In many cases, however, the gender of avatars was unable to be determined, as users chose to be androgynous or non-human. Each dyad was then coded as male-male, female-female, or mixed. Interpersonal Distance. The distance between avatars in each unique dyad was calculated (in meters) from the positional data. Mutual Gaze. From the captured data, we calculated the gaze angles of avatars in each dyad. A gaze angle of 0 degrees means that the avatar was looking directly at the other avatar. A gaze angle of 180 degrees means that the avatar was looking directly away from the other avatar. And of course, the gaze angles of the two avatars in a dyad need not be congruent. We summed these two gaze angles for each dyad to create a new variable gaze sum that measured how much the two avatars were looking at each other, as Figure 1 demonstrates. Figure 1. Avatar A is gazing at avatar B at an angle of 8 degrees. Avatar B is gazing at avatar A at an angle of 40 degrees. Their gaze sum is 48 degrees. Nonverbal Norms in VEs 10 Talking. From the captured data, we were also able to determine whether the avatars were “talking” the moment the screenshot was captured. Second Life provides an animation while users are typing (and thus before their message itself is finished and seen). If users were in this “is typing” mode, they were coded as “talking”. We coded these as binary. We summed these two binary numbers to create a new variable talking sum that measured whether the amount of talking between two avatars. Location. The research assistants also noted whether the interaction occurred in an indoor or outdoor location. Because indoor spaces are necessarily smaller than unbound, outdoor spaces and thus would affect IPD, we included this factor in our analyses. Results Our sampling method produced 417 snapshots. From the snapshots, we extracted 8418 unique dyads. Because our snapshot script captured avatar data in a radius of up to 200 meters, we applied an exclusion criterion to filter out individual avatars that were too far away to be in a social interaction. Hall (1959) estimated that social distance extends up to 12 feet (or 3.66 meters). For the following analyses, we used a social distance of 3.66 meters as the cut-off for whether 2 avatars are considered to be a dyad. Altogether, our sampling method produced 835 unique dyads after applying this exclusion criterion. Interpersonal Distance Using this data composed of unique dyads, we ran an ANOVA with gender composition (male-male, female-female, mixed) and location (indoor and outdoor) as fixed factors, gaze sum and talking sum as linear covariates, and distance between the two avatars as the dependent variable. The effect of gender composition was significant, F[2,827] = 6.40, p = .002, η = .014. A comparison of the estimated marginal means Nonverbal Norms in VEs 11 revealed that avatars in the mixed-gender condition (M = 2.09, SD = .91) stood closer together than avatars in the male-male condition (M = 2.36, SD = .82) and female-female condition (M = 2.34, SD = .87), p’s < .005. The effect of gaze sum was also significant, F[2,827] = 8.32, p = .004, η = .01. The correlation coefficient between gaze sum and distance was -.10 (p = .001). In other words, within the range of 3.66 meters, the closer two avatars were, the less likely they were maintaining eye contact. The effect of location was not significant, F[2,827] = 1.36, p = .24, η = .002. The effect of talk sum was also not significant, F[2,827] = .16, p = .65, η < .001. There was also no significant interaction effect between gender composition and location, F[2,827] = .28, p = .75, η < .001. Mutual Gaze We also ran an ANOVA with gender composition (male-male, female-female, mixed) and location (indoor and outdoor) as fixed factors, distance and talking sum as linear covariates, and gaze sum as the dependent variable. The effect of gender composition was significant, F[2,827] = 4.37, p = .01, η = .01. Comparing the estimated marginal means, we found that avatars in male-male dyads (M = 140.30, SE = 74.98) were significantly less likely to look at each other than those in female-female dyads (M = 127.64, SD = 50.43) and mixed dyads (M = 127.64, SD = 96.85), p’s < .05. The effect of location was also significant, F[2,827] = 4.74, p = .03, η = .01. Comparing the estimated marginal means, we found that avatars in indoor locations (M = 135.48, SD = 65.30) were significantly more likely to be maintaining eye contact than avatars in outdoor locations (M = 123.80, SD = 63.54). Nonverbal Norms in VEs 12 The effect of the covariate talk sum was also significant, F[2,827] = 8.50, p = .004, η = .01. The correlation coefficient between talk sum and gaze sum was -.08 (p = .007). In other words, the more that two avatars were talking, the more likely they were looking at each other. The effect of the covariate of distance was also significant and has already been discussed above. Finally, there was a significant interaction effect of gender composition and location, F[2,827] = 4.26, p = .01, η = .01. The gaze sum of male-male dyads in indoor locations was significantly higher than all other dyads in both indoor and outdoor locations, p’s < .05. See Table 1 for means and standard deviations. In other words, male avatars are less likely to be maintaining eye contact with other male avatars in indoor locations as compared with all other gender composition and location combinations. Table 1. Means and standard deviations of gaze sum by gender composition and location