The Proteus Effect 1 Running Head: THE PROTEUS EFFECT The Proteus Effect: The Effect of Transformed Self-Representation on Behavior

The prevalence of virtual environments, such as online games, chatrooms, and video conferences, increasingly allows us to alter our digital self-representations dramatically and effectively. But as we change our self-representations, do our self-representations change our behavior in turn? In two experimental studies, we explore the hypothesis that an individual’s behavior conforms to their self-representation independent of how others perceive them a process we term the Proteus Effect. In the first study, participants assigned more attractive avatars in immersive virtual environments were more intimate with confederates in a self-disclosure and interpersonal distance task than participants assigned to less attractive avatars. In our second study, participants assigned taller avatars behaved more confidently in a negotiation task than participants assigned shorter avatars. We discuss the implications of the Proteus Effect with regards to social interactions in online environments. The Proteus Effect 3 The Proteus Effect: The Effect of Transformed Self-Representation on Behavior The notion of transforming our appearances permeates our culture. On the one hand, minor alterations such as haircuts, make-up, and dressing up are seen as socially acceptable, if not socially desirable. On the other hand, the ability to truly transform oneself has been regarded in myths and legends as both dangerous and powerful. Consider for example werewolves and vampires from Europe, the kitsune (foxes that can take on human form) from Japan, the god Loki from Norse mythology, and the god Proteus from Greek mythology. The Greek god Proteus is notable for being the origin of the adjective “protean” the ability to take on many different selfrepresentations. And while extreme self-transformations are expensive (e.g., cosmetic surgery) or difficult to perform (e.g., gender reassignment surgery) in real life, nowhere is self-representation more flexible and simple to transform than in virtual environments where users can choose or customize their own avatars digital representations of themselves. For example, the documentation for the online social world Second Life notes that “using over 150 unique sliders, they can change everything from their foot size to their eye color to the cut of their shirt” (Labs, 2006). In other words, the mutability of our self-representations in online environments is a fundamental aspect of what it means to have a virtual identity (Turkle, 1995). Even though the plasticity of our self-representations is an important part of our online identities, the quantitative research in computer-mediated communication (CMC) has tended to focus instead on the impact of technical affordances on social interaction in online environments. For example, it has been argued that lack of social presence (Hiltz, Johnson, & Turoff, 1986; Short, Williams, & Christie, 1976) or the lack of social cues (Culnan & Markus, 1987; Kiesler, Siegel, & McGuire, 1984) creates an impoverished social environment, while others have shown that relationships develop slower in CMC but are not impoverished in the long term (Walther, The Proteus Effect 4 1996; Walther, Anderson, & Park, 1994). Other research has looked at how the narrow communication channels in CMC impacts impression formation (Hancock & Dunham, 2001; Jacobson, 1999; Trevino & Webster, 1992; Walther, Slovacek, & Tidwell, 2001). And while there has been research on self-representation in online environments, the focus has been on the impact of anonymity and authenticity (Anonymous, 1998; Flanagin, Tiyaamornwong, O'Connor, & Seibold, 2002; Jarvenpaa & Leidner, 1998; Postmes & Spears, 2002) in other words, the gap between the real and virtual self and how that difference changes social interactions. In the current work, we were instead interested in exploring how our avatars change how we behave online. As we change our self-representations, do our self-representations change our behaviors in turn? As we choose or create our avatars online and use them in a social context, how might our new self-representations change how we interact with others? Thus, we were interested in the impact of our actual self-representations on our behaviors in virtual environments rather than the effects of anonymity or authenticity. Behavioral Confirmation There is good reason to believe that our avatars change how we interact with others. Behavioral confirmation offers one potential pathway for this change. Behavioral confirmation is the process whereby the expectations of one person (typically referred to as the perceiver) cause another person (typically referred to as the target) to behave in ways that confirm the perceiver’s expectations (Snyder, Tanke, & Berscheid, 1977). In the seminal study by Snyder and colleagues (1977), male and female undergraduate students interacted over a telephone. Male perceivers who believed that a female target was attractive caused her to behave in a more charming and friendly manner regardless of how attractive the target actually was. Thus, in an online environment, a perceiver interacting with a target who is using an attractive avatar may cause the The Proteus Effect 5 target to behave in a more friendly and charming manner. In fact, the study by Snyder and colleagues itself occurred in a mediated context (i.e., over the telephone). It is important to note that the source of behavioral change from the effects of behavioral confirmation stem from the perceiver rather than the target. It is the perceiver’s behavior that in turn causes a change in the target’s behavior. Self-Perception Theory and Deindividuation Theory Behavioral confirmation provides one potential pathway for avatars to change how a person behaves online, but might our avatars change how we behave independent of how others perceive us? When given an attractive avatar, does a user become more friendly and sociable regardless of how others interact with them? Another line of research suggests a potential explanation for why this might occur. Bem (1972) has argued that people observe their own behaviors to understand what attitudes may have caused them (i.e., self-perception theory). For example, people given extrinsic rewards to do something they already enjoy doing are more likely to view the behavior as less intrinsically appealing (i.e., the overjustification effect) because this is what an impartial observer would have concluded as well. Other researchers have shown the far-reaching implications of this theory. In Valins’ study (1966), when participants were made to believe their heartbeat had increased while viewing a photograph of a person, they came to believe the person in the photograph was more attractive. In Frank and Gilovich’s study (1988), subjects that wore black uniforms behaved more aggressively than subjects in white uniforms. According to Frank and Gilovich, wearing a black uniform is a behavior that the subjects used to infer their own dispositions “Just as observers see those in black uniforms as tough, mean, and aggressive, so too does the person wearing that uniform” (pg. 83). The subjects then adhere to this new identity by behaving more aggressively. And finally, this effect has been The Proteus Effect 6 replicated more recently in a digital environment, where users given avatars in a black robe expressed a higher desire to commit anti-social behaviors than users given avatars in a white robe (Merola, Penas, & Hancock, 2006). Another line of research has shown that the impact of identity cues is particularly strong when people are deindividuated. Zimbardo (1969) originally used deindividuation theory to argue that urban or crowded areas cause deindividuation which leads to antisocial behavior, however it has also been shown that deindividuation can lead to affiliative behavior as well (Gergen, Gergen, & Barton, 1973). When dyads were placed in a darkened room for an hour, many deliberately touched or hugged the other person. On the other hand, dyads in the fully-lit room talked politely and did not engage in physical contact. Thus, the effects of deindividuation are not necessarily anti-social. The argument that deindividuation can lead to both pro-social and anti-social behavior has also been demonstrated in another well-known study. In a teacherlearner paradigm with electric shock as punishment, subjects in costumes that resembled Ku Klux Klan robes delivered significantly longer shocks than subjects in nurse uniforms (Johnson & Downing, 1979). It was also found that these effects were stronger when subjects were made anonymous in the study. Thus, deindividuation does not necessarily always lead to anti-social behavior as Zimbardo originally argued, but may in fact cause a greater reliance on identity cues whether they are anti-social or pro-social. In the computer-mediated communication literature, the Social Identity Model of Deindividuation Effects (Postmes, Spears, & Lea, 1998; Spears & Lea, 1994) argued that factors that lead to deindividuation, such as anonymity, might thus reinforce group salience and conformity to group norms. In this light, deindividuation does not, in and of itself, always lead to anti-normative behavior, but rather, behavioral changes depend on the local group norms The Proteus Effect 7 (Postmes, Spears, & Lea, 2000). More importantly, behavior that is typically seen as antinormative, such as flaming on message boards (Lea, O'Shea, & Spears, 1992), may in fact turn out to be normative and expected in particular contexts (Postmes et al., 1998). The Proteus Effect Online environments that afford anonymity are like digital versions of a darkened room where deindividuation might occur, and indeed, many researchers have suggested that deindividuation occurs online due to anonymity or reduced social cues (Kiesler et al., 1984; McKenna & Bargh, 2000). And in online environments, the avatar is not simply a uniform that is worn, the avatar is our entire self-representation. Whereas the uniform is one of many identity cues in the studies mentioned earlier, the avatar is the primary identity cue in online environments. Thus, we might expect that our avatars have a significant impact on how we behave online. Users who are deindividuated in online environments may adhere to a new identity that is inferred from their avatars. And in the same way that subjects in black uniforms conform to a more aggressive identity, users in online environments may conform to the expectations and stereotypes of the identity of their avatars. Or more precisely, in line with selfperception theory, they conform to the behavior that they believe others would expect them to have. We term this the Proteus Effect. While the Proteus Effect is similar to SIDE theory, there are several important theoretical differences. Most importantly, SIDE theory emphasizes conformity to local group norms (e.g., becoming more hostile on a hostile message board). On the other hand, the Proteus Effect emphasizes conformity to individual identity cues (e.g., becoming friendlier in an attractive avatar). Thus, theoretically, it would also be possible to pit one against the other i.e., having an attractive avatar on a hostile message board. We would also argue that having an attribute (e.g., The Proteus Effect 8 “being attractive”) is conceptually different from being amongst a group of individuals who have that attribute (e.g., “being in a group of attractive people”), while SIDE theory literature tends to conflate the two. Thus, in a situation where person A in a black uniform interacts with person B in a white uniform, SIDE theory might predict that the social identity of person A would default to the black uniform (i.e., become more aggressive) or the combined colors of the group in question – in other words, gray (i.e., remain neutral). The Proteus Effect would only predict the former. Another point of differentiation is that while the SIDE theory operates on the basis of an existing local group and its social norms, the Proteus Effect should operate even when the user is alone. This is because self-perception theory isn’t predicated on the actual presence of other people, but simply that a person evaluates him or herself from a third-person perspective (i.e., an imagined third party). Collaborative Virtual Environments and Transformed Social Interaction In designing of our studies, it was crucial that we isolate the impact of the Proteus Effect from that of behavioral confirmation. If participants were perceived to be attractive and believed themselves to be attractive at the same time, it would be impossible for us to claim that the Proteus Effect occurred independent of behavioral confirmation. To isolate the potential effect of the Proteus Effect, we employed a novel methodological paradigm. In the current set of studies, we utilized Collaborative Virtual Environments (CVEs, see Normand et al., 1999) to study the effects of the Proteus Effect. CVEs are communication systems in which multiple interactants share the same three-dimensional digital space despite occupying remote physical locations. In a CVE, immersive virtual environment technology monitors the movements and behaviors of individual interactants and renders those behaviors within the CVE via avatars. These digital representations are tracked naturalistically by optical sensors, mechanical devices, and cameras. The Proteus Effect 9 Because these avatars are constantly redrawn for each user during interaction, unique possibilities for social interaction emerge (Loomis, Blascovich, & Beall, 1999; Blascovich et al., 2002). Unlike telephone conversations and videoconferences, the physical appearance and behavioral actions of avatars can be systematically filtered in immersive CVEs idiosyncratically for other interactants, amplifying or suppressing features and nonverbal signals in real-time for strategic purposes. Theoretically, these transformations should impact interactants’ persuasive and instructional abilities. Previously, we outlined a theoretical framework for such strategic filtering of communicative behaviors called Transformed Social Interaction (Bailenson, Beall, Blascovich, Loomis, & Turk, 2005). In a CVE, every user perceives their own digital rendering of the world and each other and these renderings need not be congruent. In other words, the target may perceive his or her own avatar as being attractive while the perceiver sees the target as being unattractive. Previous work on transformed social interaction has demonstrated quite resoundingly that changing one’s representation has large implications on other’s in terms of social influence (Bailenson, 2006). In other words, transforming Avatar A strategically causes Avatar B to behave consistently with the representation of Avatar A (as opposed to the actual representation of Avatar A). In the current set of studies, this decoupling of representation allowed us to test a separate question relating to transforming a representation. Instead of seeing the strategic outcome of a transformation, we examined whether our changes in self-representations independent of how others perceive us cause the people behind the avatars to behave differently. Overview of Studies and Hypotheses In the current work, we conducted two experimental studies to explore the Proteus Effect. The Proteus Effect 10 Participants interacted with a confederate’s avatar in a virtual reality (VR) environment. In the first study, we manipulated the attractiveness of the participant’s avatar while the confederate was blind to condition. Studies have shown that attractive individuals are perceived to possess a constellation of positive traits (Dion, Berscheid, & Walster, 1972), and are evaluated more favorably by jurors in courtrooms (Friend & Vinson, 1974). Interpersonal Distance. According to nonverbal expectancy violations theory (Burgoon, 1978), when attractive individuals violate nonverbal expectancies, such as moving too close to someone, the positive valence that is created can be socially advantageous (Burgoon & Walther, 1990; Burgoon, Walther, & Baesler, 1992). Given that attractive individuals have higher confidence (Langlois et al., 2000), we hypothesized that: H1: Participants in the attractive condition walk closer to the confederate than the participants in the unattractive condition. Self-Disclosure. Friendliness was one of the measures used in Snyder, Tanke, and Berscheid’s original study (1977), and in this study we used self-disclosure as a behavioral operationalization. Because attractive individuals tend to be more extraverted and more friendly (Langlois et al., 2000), we hypothesized that: H2: Participants in the attractive condition would exhibit higher self-disclosure and present more pieces of information about themselves than participants in the unattractive condition. In the second study, we manipulated the height of the participant’s avatar again with the confederate blind to the condition. Similar to the attractiveness literature, taller people are perceived to be more competent (Young & French, 1996), more desirable as romantic partners (Freedman, 1979; Harrison & Saeed, 1977), and more likely to emerge as leaders (Stogdill, The Proteus Effect 11 1948). In this study, we implemented a negotiation task to best gauge confidence. H2: Participants in taller avatars would behave in a more confident manner and negotiate more aggressively than participants in shorter avatars. Experiment One Design In a between-subjects design, participants were randomly assigned to have an avatar with an attractive or unattractive face of his or her own gender and then interact with a confederate. We followed the paradigm in the study by Snyder and colleagues (1977) and always used a confederate of the opposite gender. The confederate was blind to the attractiveness condition such that the participant’s avatar appeared to the confederate with an untextured face – one which was structurally human but left uncolored. Participants Thirty-two undergraduate students (16 men and 16 women) participated in the study for course credit. Materials Facial Attractiveness Pretest. We ran a pretest to get subjective determinations of attractive and unattractive faces (for the participants), and also average-attractiveness faces (for the confederates). To minimize the chances that our findings would be driven by idiosyncrasies of a particular face, we chose two faces in each of these three attractiveness conditions. Thus, there were two attractive faces, two unattractive faces, and two average faces for each gender. In total, we used 12 faces in the study. To generate these 12 faces, digital photographs of 34 undergraduate students (17 male and 17 female) from a different academic institution from the main study were used in a pretest. The Proteus Effect 12 The chances of participant recognition of these faces were thus minimized. To reduce other variations in facial features, only Caucasians were used in the pretest1. Frontal and profile photographs of these 34 undergraduate students were converted into digital, three-dimensional head busts using 3DMeNow software. These three-dimensional head busts were then converted into Vizard 2.17 models, our CVE platform, and attached to generic male and female bodies. Finally, a frontal and three-quarter screenshot of every face was taken (see Figure 1). Thus, altogether, 68 screenshots were generated. Fourteen undergraduates from a separate subject population from the main study used a web-based survey to rate the attractiveness of every screenshot’s face on a unipolar 7-point fullylabeled construct-specific scale (from “Not Attractive At All” to “Extremely Attractive”). The frontal and three-quarter screenshot of every face were thus rated separately. Each screenshot was shown by itself and the order of faces was uniquely randomized for every rater. The ratings of the frontal and three-quarter image of every face were averaged. Then six faces were selected for each gender, where the two attractive faces were each rated as significantly more attractive than the two average faces, and the two average faces were each rated as significantly more attractive than the two unattractive faces. All pair-wise t-tests had a pvalue less than .05 (df’s = 26). The 12 faces used in the study are shown in Figure 1. The means and standard deviations of their attractiveness ratings are shown in Table 1. In the entire sample of faces we pretested, the mean attractiveness was 3.09 with a standard deviation of 1.30. The faces we chose for the high attractiveness condition had a mean of 4.63 and a standard deviation of 1.22, while the faces in the low attractiveness condition had a mean of 1.61 and a standard deviation of 0.83. Thus, our faces in the high and low attractiveness conditions varied from the average by about one standard deviation. The Proteus Effect 13 The Physical Lab Setting. The lab consisted of two rooms with an open doorway. In the room where the study took place, a black curtain divided the room. To ensure that confederates and participants were not biased by the attractiveness each other’s real faces, confederates stayed behind this black curtain until the virtual reality interaction began and thus never saw the participant’s real face and vice versa. The Virtual Setting. The virtual setting was a white room that had the same exact dimensions as the physical room participants were in (see Figure 2). Two meters behind the participant was a virtual mirror that reflected the head orientation (rotations along pitch, yaw, and roll) and body translation (translation on X,Y, and Z) of the participant with the designated face (See Figure 2). Thus, the mirror image tracked and reflected six degrees of freedom such that when the participant moved in physical space, his or her avatar moved in perfect synchrony in the mirror. The confederate’s avatar was located 5 meters in front of the participant, facing the participant, and remained invisible until the conversational portion of the experiment began. The confederate’s avatar also had an automated blink animation based on human blinking behavior and lip movement that matched the volume of the confederate’s speech. Apparatus Perspectively-correct stereoscopic images were rendered at an average frame rate of 60 Hz. The simulated viewpoint was continually updated as a function of the participants’ head movements, which were tracked by a three-axis orientation sensing system. The position of the participant along the x, y, and z planes were tracked via an optical tracking system. Participants wore an nVisor SX head-mounted display (HMD) that featured dual 1280 horizontal by 1024 vertical pixel resolution panels that refreshed at 60 Hz. See Figure 2 for equipment setup.