Choose your "buddy icon" carefully: The influence of avatar androgyny, anthropomorphism and credibility in online interactions

In both online and offline interactions, the visual representation of people influences how others perceive them. In contrast to the offline body, an online visual representation of a person is consciously chosen and not stable. This paper reports the results of a 2 step examination of the influence of avatars on the person perception process. Specifically, this project examines the reliance on visual characteristics during the online perception process, and the relative influence of androgyny, anthropomorphism and credibility. In the first step, 255 participants fill out a survey where they rated a set of 30 static avatars on their credibility, androgyny, and anthropomorphism. The second step is a between subjects experiment with 230 participants who interact with partners represented by one of 8 avatars (high and low androgyny, and anthropomorphism by high and low credibility). Results show that the characteristics of the avatar are used in the person perception process. Causal modeling techniques revealed that perceptions of avatar androgyny influence perceptions of anthropomorphism, which influences attributions of both avatar and partner credibility. Implications of these results for theory, future research, and users and designers of systems using avatars are discussed. People are selecting avatars to represent them during a variety of online interactions including work related projects, and relationship maintenance and development (Schroeder, 2002). The images that people select to represent them during online interactions have been called ‘buddy icons,’ avatars, or virtual images (see Suler, 1996; Damer, 1997) and can be created, borrowed, or even purchased. Many collaborative and communication applications, including instant messaging, have a buddy icon, or avatar. The Pew Internet and American Life project estimates that 53 million Americans are using instant messaging systems (IM), with 62% of those aged 18-27 using it and more than 46% of this age group using it more than email both for work related applications, and for keeping in touch with friends and family (Shiu & Lenhart, 2004). Companies have started using avatars in online advertising and customer support (Ikea, 2006). BBC even has a news segment anchored by avatars (Maseon, 2006). The types of avatars are quite diverse and may resemble other humans, animals, objects, or anything the user (or designer) can dream up and generally have little or no resemblance to the user’s offline appearance (Benford et al., 1995; Damer, 1997; Suler, 1996). Oravec (1996) argued that the avatar selected for an interaction is analogous to clothing selected for a meeting in that both are vehicles of expression. Some users have one avatar that they use most or all of the time (Becker & Mark, 2002) but most have a variety of avatars they use to represent them that they frequently change during interactions to represent their various moods and locations (Oravec, 1996; Soukup, 2004; Taylor, 1999, 2002). The use of avatars online influences people’s involvement in the interaction as well as their perceptions of others they encounter (Biocca & Nowak, 2002; Lee, 2004; Nowak, 2004; Schroeder, 2002; Taylor, 2002). Therefore, the increased use of avatars is likely to have implications for communication outcomes and relationship development. This project uses causal modeling techniques to examine the relative influence of the visible characteristics of the avatar (androgyny, credibility, and anthropomorphism), and the behavior of the person, on impressions of androgyny and credibility. The role of visual characteristics in person perception online and offline There are obvious differences and similarities between the avatar and the offline body. The effect of those differences on the perception process is not so obvious. They are similar in that both are the visible representation of another’s ‘presence’ during an interaction, and both are used to represent and identify a person. One important difference is that a person is born with an offline body, and it stays relatively consistent throughout his or her life, while the avatar is computer generated, chosen and can be easily modified or even completely changed (Benford, et al., 1995; Oravec, 1996; Taylor, 2002). This section considers the potential influence of these similarities and differences. Despite the obvious media differences, previous research suggests that the underlying process of perceiving people online is analogous to the process of perceiving people offline. Indeed, the process of reducing uncertainty has been shown to continue in online interactions where people adapt the uncertainty reduction process to utilize whatever information is provided in an interaction (Nowak, 2004; Ramerez, Walther, Burgoon, & Sunnafrank, 2002; Walther, In Press). Uncertainty reduction theory argues that people strive to understand the past and current behavior of those they encounter. People also work to interpret the motives of others and work to increase confidence in their ability to predict future behavior (Berger & Calabrese, 1975; Clatterbuck, 1979). People use all information available to reduce uncertainty and make interpersonal judgments in the online person perception process, including the characteristics of the avatar, usernames, and even images on a person’s webpage (Lea & Spears, 1992; Nowak, 2004; Sherman et al., 2001; Taylor, 2002; Wexelblat, 1997). Researchers have considered the distinction between conscious and unconscious signals in nonverbal behavior, and people seem to have more confidence in visually presented information that is not consciously controlled, such as appearance (Argyle, 1988; Burgoon, 1994). Impressions based on offline physical appearance are used to make attributions about other people’s personality with little conscious effort and people feel comfortable about the completeness and accuracy of these attributions (Ambady, Hallhan, & Rosenthal, 1995; Bull & Rumsey, 1988; Burgoon, 1994; Dion, Berscheid, & Walster, 1972; Ichheiser, 1970). Research suggests that “observers assume a person possesses personality characteristics that are consistent with his or her physical appearance” (Madison, 2000, p. 148). The fact that we can separate the avatar from the behavior allows an exploration of the extent to which this reliance on visible information in the perception process is due to the lack of conscious control and the relative stability of the body. Perceivers know that the avatar is consciously chosen, easy to change, and not stable. Therefore, if people rely more on a person’s behavior than on the visual information (avatar) when online, it is likely that people rely on characteristics of the offline body due to its stability and the fact that it is beyond conscious control. However, if the characteristics of the avatar have a stronger effect on the online person perception process than behavior, this implies that people rely on visual characteristics for some other reason. The effect of anthropomorphism on person perception Anthropomorphism can be defined in terms of either behavior or appearance. In behavioral terms, anthropomorphism implies the assignment of human qualities, such as mental abilities and behavior (DiSalvo & Gemperle, 2003), to objects that are not human. In appearance terms, anthropomorphism defines an object that has human morphology or visual characteristics (DiSalvo & Gemperle 2003; Nowak, 2003; Nowak & Rauh, 2005; Shapiro, 1997). Here we examine anthropomorphism only in terms of human morphology, or appearance of avatars and not to behavior and we extend this definition to apply to humans represented by avatars. In these terms, an anthropomorphic image has visible characteristics that make it appear more human than a less anthropomorphic image, such as an object (see Nowak, 2004). As with a person’s offline body, research has shown that the avatar influences how people categorize others, and has a large influence on the person perception process. Specifically, the characteristics of avatars, including anthropomorphism, strongly influence social liking and attributions of credibility (Nass, Steuer, Tauber, Reeder, 1993; Nowak, 2004). However, the direction of this effect is unclear. While some researchers found that more anthropomorphic avatars lead to more positive perceptions (more liking and credibility) than less anthropomorphic avatars (Koda & Maes, 1996; Wexelblat, 1997), others found that less anthropomorphic avatars led to more positive perceptions than more anthropomorphic avatars (Nowak, 2004; Nowak & Biocca, 2003). Understanding the influence of the avatar, much like the influence of offline physical appearance, is not trivial. These conflicting findings may be explained by the different contexts of the interactions, the different avatars used in the studies, or by other variables that were not considered in these studies. For example, in a study of virtual images realism, Garau et. al (2003) found a stronger effect for inconsistency of image type and behavior (low anthropomorphic image with very realistic behavior and vice versa) than for anthropomorphism alone. In some contexts, higher anthropomorphic images may set up higher expectations. These expectations are more difficult to meet, which results in disappointment and lower credibility and likeability ratings following interactions with those represented by anthropomorphic avatars (Slater & Steed, 2002; Nowak, 2004). Shneiderman (1988) argues that system designers arbitrarily apply certain anthropomorphic qualities to their design, which leads to user over-expectation and disappointment when the system fails to perform to these expectations. While researchers agree that the characteristics of avatars influence person perception, it is premature to conclude that the discerning factor is anthropomorphism. One limitation to the research discussed above is that researchers did not perform a manipulation check to ensure that participants perceived one character as more anthropomorphic than another as expected by the design and assumed in the conclusions (Koda & Maes, 1996; Nowak, 2004; Nowak & Biocca, 2003; Wexelblat, 1997). Nevertheless, there is enough research to suggest that anthropomorphism will influence attributions of credibility of both the avatar, and the person it represents. Further, it seems that more anthropomorphic avatars lead people to expect high degrees of sociability and credibility. Therefore, we predict H1: More anthropomorphic avatars will be rated as more credible than less anthropomorphic avatars. The effect of avatar androgyny on person perception In online interactions, the cues related to the biological sex of an interaction partner may not be obvious, in any way related to biological sex, or visible at all. Thus, it can be difficult to dichotomize a person, or image, as either masculine or feminine. It is possible to display both feminine and masculine characteristics, or neither (Bem, 1981). Thus, it makes sense to move away from what Murphy (1994) called the artificial dichotomization of “masculine’” and “feminine” (p. 22). Thus, gender should be considered as a continuum. The center of the continuum of masculinity and femininity is commonly called androgyny (Heilbrun & Pitman, 1979; Murphy, 1994; Bem, 1981). Many computer interfaces (especially text based interactions) are theoretically gender-blind, with no traditional gender cues, such as those related to the physical body, available. This led some CMC observers to predict that characteristics related to the physical body, such as androgyny and masculinity or femininity, would not be considered relevant in online interactions (Hert, 1997; Lea & Spears, 1992; Sproull & Kiesler, 1986). However, this prediction has been generally discredited. Users have continued to identify gender online (Balsamo, 1995; Cornetto & Nowak, 2006; Spender, 1996; Witmer & Katzman, 1997), and it has maintained both salience and meaning (Cherny, 1994; Clark, 1995; Herring, 1994; Herring, 2000; Yates, 1997). Research has shown that people rely on usernames (Cornetto & Nowak, 2006) and behavioral information such as speaking style, topics discussed, phrasing choices (Cherny, 1994; Herring, 1994; 2000; Jones, 1997), and self reports (Featherstone & Burrows, 1995; Spender, 1996; Yates, 1997) to make assignments of sex category (Witmer and Katzman 1997). Research has shown that in face to face interactions, the masculinity or femininity of physical features influences people’s perceptions of the personality traits and behaviors of others (Madison, 2000). Further, research has shown that people can and do make attributions of the gender of pictures, whether they are pictures of animals or humans (Karnoil, Reichman & Fund, 2000). However, the influence of the avatar or its characteristics on the gender attribution process is still unclear, though the gender of an avatar has been shown to influence the person perception process (Lee, 2004), and the likelihood that it will be selected for an interaction (Nowak & Rauh, 2005). Given the salience of sex category in offline interactions, it makes sense that it has maintained salience in online interactions and this may help to "ground or stabilize that which is new" (Clark, 1995, p. 114). Perceptions of gender have continued to be important and influential in online interactions, and androgyny, anthropomorphism and credibility are highly correlated (Nowak & Rauh, 2005). This suggests that level of androgyny is of primary importance in the perception process and that people will prefer less androgynous (more masculine or feminine) avatars. Therefore, we predict: H2: More androgynous avatars will be rated as less credible than less androgynous avatars. H3: People who are perceived to be more androgynous will be perceived as less credible than those who are perceived to be less androgynous. While research has shown that both anthropomorphism and androgyny influence the online perception process, and that these variables are highly correlated, the causal order of the relationship between these variables is still unknown. It is possible that people first discern the anthropomorphism of an object or entity they encounter online, and only once this distinction has been made, will people turn to assign their interaction partner to a sex category. However, it is likely that because all living things have sex categories, the classification of the masculinity or femininity of an entity will be determined before anthropomorphism is considered. Therefore, we predict: H4: More androgynous avatars will be perceived as less anthropomorphic than less androgynous avatars. What is certainly similar online and offline is the complexity of the person perception process. Factors beyond physical characteristics in the offline world, or the visible avatar in cyberspace, are likely to influence the process as well. For one, individual differences of the perceiver will likely influence perception in both contexts. Individual differences influence perception As discussed above, researchers have examined how different avatars influence person perception. There are likely to be individual differences that influence both how people perceive avatars and how they use that information in the perception process. In the process of examining how avatar characteristics influence person perception, we should consider which factors influence the perception of avatars. Computer self-efficacy refers to a person’s perceived ability to learn and use computers and computer programs (see also Compeau & Higgins, 1995). Thus, computer efficacy is relevant to ascertaining a person’s aptitude, and attitude, towards computers. Research has demonstrated a link between computer efficacy and a person’s adoption and use of computer technology (Compeau, Higgins & Huff, 1999; Compeau & Higgins, 1995; Eastin & LaRose, 2000). Therefore, we predict: H5: Those who spend more time using IM will feel more computer efficacy that those who use IM less. In addition, people will rely on their previous experiences with perceiving people in both online and offline contexts in the perception process. People who spend more time with a specific computer medium, such as Instant Messaging (IM), will have more online experiences with avatars to draw from during the perception process using these systems. Thus, they will likely perceive avatars, and the people they interact with differently from those with less computer efficacy and those who spend less time with IM. These experiences will influence how they interpret online information and their perceptions of their interaction partner. They are less likely to feel uncertainty, and will be more confident in their ability to get to know others they meet in these environments. These experiences will also likely lead to higher standards and expectations for anthropomorphism in avatars. Therefore, we predict: H6: Those with high computer efficacy will rate avatars as less anthropomorphic than those with low computer efficacy. Although we predict a strong relationship between androgyny and anthropomorphism, the higher expectations resulting from increased exposure to the technology may not have the same effect on people’s perceptions of androgyny. Avatars may be highly masculine, or feminine, and yet not anthropomorphic. Research presented above showed that people characterize gender of characters with very low anthropomorphism. The higher computer efficacy and more frequent usage of IM and consequent exposure to avatars that increases expectations for anthropomorphism may result in more tolerance for gender ambiguity. If this were true, there may be a tendency for those with greater computer efficacy and higher IM usage to see avatars as less androgynous, and have more confidence in their ability to categorize an avatar as more feminine or masculine, and less androgynous. Therefore, we predict: H7: Those with high computer efficacy will rate avatars as less androgynous than those with low computer efficacy. Considering the influence of behavior on perceptions of avatars and people In a social chat context, people make attributions of both the person and the avatar. Not only are each of these influenced by multiple factors but these attributions interact with, and influence, each other. People’s perceptions of an avatar are likely influenced by their perceptions of the person that the avatar is representing. However, as discussed in this section, in the offline perception process, people process visual information first. If the online process follows this pattern as research has suggested it will, people will process the avatar first, and perceptions of the avatar will influence perceptions of the represented partner. Information perceived early in an interaction influences the processing of information that is received later. As visual information is processed, an initial mental model of the other is created. As the interaction progresses, behavioral cues may be incorporated and the model adapted to reflect that information (Kunda, 1999). However, incorporating behaviors into the model requires more cognitive resources than processing visual information, which is thus processed earlier in the interaction. Also, creating a mental model requires less cognitive resources than the process of readjusting the model to fit new information. Because of that, people use their existing model to interpret new information and try to make the information conform to it. This means that the impressions based the physical appearance of others have a stronger influence on the perception process than impressions based on behavior (Kenny et. al, 1992). Thus, the visible characteristics of the avatar will likely influence the perceptions of the person represented. That is, if certain characteristics of the avatar cause certain inferences consistently across contexts, these same inferences will influence the perception of the person being represented by the avatar. This deduction leads us to predict that the characteristics of the avatar will influence the perceptions of those represented by the avatar. Therefore, we predict: H8: People represented by more androgynous avatars will be perceived as more androgynous than those represented by less androgynous avatars. H10: People represented by more credible avatars will be perceived as less credible than those represented by less credible avatars. Method This project consisted of two steps: a survey and an experiment. A survey was first conducted to determine how people perceive a group of 30 avatars in a static context and generate information in which to base the selection of stimulus materials for the second step, an experiment. The experiment consisted of a between subjects experiment with eight conditions (high and low avatar anthropomorphism plus high and low avatar androgyny by high and low avatar credibility). This study used a subset of 8 of the 30 avatars from step 1 as stimulus materials. The 8 avatars were selected based on the static context ratings from the first step in order to provide the above experimental conditions (see Table 2). Step 1 – Static Context Participants. Two hundred and fifty five participants were recruited from a large northeastern public American university where they received extra-credit for their participation. Of those, 115 were females, 136 were males and 4 of them did not report this information. Stimulus Materials. The avatar images used were created from 3D models using Poser 5 for the human characters and 3D Studio Max for the other characters. Thirty avatars were created with three classifications of avatars: 20 human-like characters, 5 animals/creatures and 5 objects that represent items that would not traditionally be animated, such as a bottle or an apple. The “human” avatars are composed of two versions of the same character, a version were the torso was present and another only with a floating head where the torso was not present. All avatars had identifiable eyes and mouth. See all images in Figure 1. The images in the figure are labeled as ‘m’ for male, ‘f’ for female, ‘a’ for animals/creatures and ‘o’ for objects. The male and female images had identical representations with only their heads and these had ‘h’ appended to their names. To illustrate our method, the 3 female image is labeled f3, and its head only counterpart is the f3h. Participants filled out a survey where they rated 8 randomly selected avatars from the set of 30. Measurement Instruments. Measurements were made of participant’s perceptions of the static avatars. All scales used semantic differential items. Avatar Static Anthropomorphism was measured with a 3-item perceived anthropomorphism scale including: ‘looks very human/does not look human’, ‘is very realistic/is not at all realistic’, and ‘looks very cartoon like/does not look like a cartoon’ from Nowak & Rauh (2005). Cronbach’s α = .83. Avatar Static Androgyny was measured with a 2 item 7-point scale included not masculine/masculine and not feminine/feminine reported in Nowak & Rauh (2005). The androgyny scale was computed by the absolute difference of the ratings of each scale (see also Heilbrun and Pitman, 1979; Lundy & Rosenberg, 1988) such that high androgyny indicates similar levels on both masculine and feminine items and low androgyny means a large difference between the ratings. Cronbach’s α=.79. Avatar Static Credibility was measured with a 7-point 3 item subset from the credibility scale developed by McCroskey, et. al (1974, 1981). The items consist of two adjectives from the competence dimension (“intelligent/unintelligent” and “informed/uninformed”) and one adjective from the character dimension (“reliable/unreliable”) of the original scale. Cronbach’s α = .92. Procedure: In first this step, each participant evaluated 8 randomly selected images from the total 30 images presented (see Figure 1). The avatars were not representing anyone but just statically displayed one at a time on a webpage. Participants indicated their impressions of the static avatar’s credibility, anthropomorphism and androgyny. These ratings were used to select the stimulus materials for step 2, see Table 1 for images, and Table 2 for means for each image used in step 2. See Nowak & Rauh (2005) for the values and ranking for all images for of each variable. Step 2 – Social Chat Context Participants. A different sample of two hundred and thirty participants (142 females and 87 males) was recruited for this step from a large northeastern American university where they received extra-credit for participation. Stimulus Materials. Participants were allowed to see all 30 images during image selection, but only 8 images were used to represent participants during the interaction. These 8 images compose 2 sets of four images for 2 2x2 matrices selected from the original set in order to maximize the variability of the stimulus set. One set maximized the variability of credibility and anthropomorphism and the other credibility and androgyny. Because these variables are correlated, completely orthogonal representatives for each cell is impossible. To address the challenge of filling the cells where the values are inverse of one another, we resorted to the following maximizing difference algorithm to fill the cells of each of the two experimental condition sets. First, the images were ranked according to credibility, androgyny and anthropomorphism. We then summed the rankings of each image for both variables and chose the one which had the lowest summed ranking. This algorithm is akin to making a scatter plot and locating the point that is closer to the diagonal lines towards each direction. For example, to find the image that should occupy the low credibility and high anthropomorphism cell, we ranked the images by less credible and more anthropomorphic. Image f4h ranks as the 8 more anthropomorphic and the 12 less credible which gives it a summed ranking of 20. This is the lowest summed rank of all images for these two variables. So while there were certainly images which were separately more anthropomorphic or less credible, there was none that was closest to that cell when these two variables were both taken into account. Table 2 lists the static context ratings from Step 1 of the avatars used in Step 2. Chat Program. Participants used chat software (Diamond, 2000) and connected via the internet to their partner where they were represented by avatars that were slightly larger than avatars used in most instant messenger programs. Participants interacted by typing text into a box and hitting ‘enter.’ Measurement Instruments. Participants rated the avatar that represented their partner on anthropomorphism (avatar chat anthropomorphism), androgyny (avatar chat androgyny) and credibility (avatar chat credibility). They also rated their partner’s credibility (partner chat credibility) and androgyny (partner chat androgyny) using the same instruments used in the first study and Cronbach’s reliabilities for the scales in this step were .88, .90, .91, .92 and .89, respectively. In addition, participants in Step 2 were assessed on their levels of computer efficacy and instant messaging use with the following instruments: Computer Efficacy. This variable indicates how well the participants perceive their ability to use, manage and generally understand a computer. The instrument was originally composed of 10 items measured on a 7-point Likert scale but factor analysis indicated that 4 of the items did not load properly into the primary factor and they were dropped. The final scale had alpha reliability of α = .93 and is composed of the following 6 items: “I understand terms describing computer hardware”, “I am good with computers, “I understand how computers work”, “I feel confident in my ability to trouble shoot computer problems”, “I can usually explain why a computer is not working”, “I know more than average about computers and the Internet”. Instant Messenger usage. This 1 item measure asked how much the participant used the computer for “Instant Messaging” purposes. The 7-point Likert scale had “very rarely” and “very frequently” as anchors. Procedure. Participants reported to rooms in separate buildings on campus. After both partners arrived, they signed their voluntary consent to participate and answered demographic questions, computer efficacy and instant messenger experience. All questionnaires were presented on computer screens. Once participants completed the pretest, they were taken to an online avatar selection screen. At the avatar selection screen, they were asked to indicate which of the 30 avatars (see Figure 1) they would like to represent them during this interaction with another person. At this point, participants were randomly assigned to a condition by the computer and that condition defined the avatar that would be shown to them as a representation of their partner. Thus, regardless of which image a participant selected to represent them, they were represented by the avatar assigned to them based on their partner’s condition and vice versa (see Table 1 for avatars). After the avatar selection screen, participants were taken to the online chat room for the interaction. Participants could see their partner’s avatar, but could not see the avatar that represented them to their partner given that this would reveal the manipulation. They then engaged in a chat with another person for 20 minutes with instructions to attempt to ‘get to know’ them, but they were asked to refrain from disclosing their names or information related to their physical characteristics. Following the interaction, participants filled out on line questionnaires. The post-test questionnaire included the avatar that had represented their partner during the interaction. Results In step 1, we examined how the participants would rate each individual image and 8 images were selected for step 2 based on these static ratings. The selected images are depicted in Table 1 and their ratings are reported in Table 2 (to view all 30 images and their ratings, see Nowak & Rauh, 2005). Many of the hypotheses were about the relationships between the user’s impression of their partner and that of the avatar that was representing him or her in a social context, as well as the causal order of the variables. The hypotheses were used to derive the hypothesized model depicted in Figure 2 (see also Table 4). We tested this model using the results from study 2 to conduct a path model, using Path, version 6.0. We used biological sex as an anchor, or exogenous variable for the model. First, male and female participants differed in terms of both computer efficacy and IM usage, with males reporting higher computer efficacy (r=.28), and less instant messaging (r=-.16) when compared to females. Further, males perceived somewhat lower partner credibility (r=-.11) than females. The remainder of the process was the same for males and females and there were no differences in how males and females rated the images in terms of androgyny, anthropomorphism or credibility. The hypothesized model depicted in Figure 2 was not a good fit with the data (RMSE=.13, χ(16) = 26.72, p=.04). We removed non-significant paths, and added missing links based on the difference between errors obtained and predicted by the model. The re-specified model is depicted in Figure 3, and it resulted in no errors over .10, and a good fit with the data (RMSE=.046, χ(15) = 5.96, p=.98). As shown in the summary table of hypotheses and results (Table 4), many of the hypotheses were supported, while others were not, and the data showed mediating variables and paths that were not hypothesized. As shown by the re-specified model in Figure 3, the individual differences did influence people’s perceptions of avatars as predicted by H6 and H7 and provide interesting insights into the process. Men report more computer efficacy, but less IM usage, and they also report their partners as less credible, than women. Participants reported extremely high usage of instant messaging (M = 6.62, SD = 1.00) and average computer efficacy (M = 4.10, SD = 1.39). While H5 was correct that IM usage predicted computer efficacy (r=.16), H7’s prediction that those with high computer efficacy would rate avatars as less androgynous was not supported. In fact, those with more computer efficacy had a slight increase in perceptions of an avatar’s androgyny (r=.19) and computer efficacy directly but negatively influenced avatar credibility (r=-.21), which was not predicted. H6 predicted that computer efficacy would reduce anthropomorphism, but this was not supported. However, frequent IM usage reduces perceptions of anthropomorphism (r=-.22) and slightly but directly increases partner credibility perceptions (r=.14), implying that those with more instant messenger use saw their partners as more credible regardless of the avatar that represented them. Computer efficacy predicted avatar androgyny, which (together with IM usage) predicted avatar anthropomorphism. As predicted by H1 and H4, the avatar sequence goes from androgyny to anthropomorphism (r=-.36) to credibility (r=.42). As predicted by H3, more androgynous partners were perceived as less credible than less androgynous partners (r=-.25). Although there was not a direct path from avatar androgyny to credibility as predicted by H2, there is a path from avatar androgyny to partner androgyny, which reduced partner credibility. As predicted by H8, H9, and H10, the characteristics of the avatar influenced people’s perceptions of their partner. Both androgyny (r=.30) and credibility (r=.29) flowed from avatar into partner perceptions. The characteristics of the avatar predicted the perceptions of the partner, meaning partner behavior did not influence perceptions of the avatar. To further examine the prediction the avatar characteristics would influence perceptions of the person, we first examined how the avatars were rated on androgyny, anthropomorphism, and credibility across the two steps. Avatars were rated similarly on anthropomorphism, androgyny and credibility, when they were static images on a webpage (step 1), and when they represented a person in a real time interaction (step 2). Next, we examined the correlations between the variables for the selected images in step 1 (see Table 3) and step 2 (see Table 4) and found a similar pattern in both steps, but the correlations, while still significant, are clearly weaker in step 2, suggesting the possibility that the behavior of the person represented by the avatar had some influence on the perceptions of the avatars. Overall, this suggests that behavior of the person in the chat context did not have a large influence on avatar ratings. Further, the causal model reveals that the causal direction is from representation to represented, from embodiment to embodied, and not the other way around. The similar effect sizes indicate that there is a consistent influence between avatar ratings and ratings of the partner at both levels. Though individual paths are moderate, the compounded effect across the entire interaction sequence results in a larger effect. Discussion These data support the prediction that people continue to generate impressions of others based on visible characteristics, whether those characteristics are computer generated (avatar) or natural (physical body). The visual characteristics of the avatar (anthropomorphism, credibility and androgyny) influenced perceptions of the partner represented by it. These results are consistent with previous research (Nowak & Rauh, 2005) that people reliably rate avatars in terms of anthropomorphism, credibility and androgyny and point to the importance of measuring and controlling for these factors in future research. The ratings of avatars across both static and chat context show that the influence of avatar characteristics on the partner is relatively consistent across contexts and individuals, though they were slightly influenced by biological sex, IM use and computer efficacy. People who use IM more see avatars as less anthropomorphic, but the small direct path from IM usage to partner credibility indicates that those with more IM experience rate partner as more credible regardless of the avatar they choose. This could suggest that more frequent IM users, who would likely be more comfortable using this chat interface, rely on avatar characteristics less, and the behavior of their partner more, than less frequent IM users. Those with more frequent IM usage rated the avatars as less anthropomorphic and felt their partners were more credible than less frequent users, regardless of the avatar that represented the partner. In this social chat context, anthropomorphism and credibility of the avatar increased perceptions of partner credibility. These data cannot address whether there is an intervening anthropomorphism variable between partner androgyny and partner credibility. There are probably other variables that can better explain the relationship between avatar androgyny and partner credibility, given that this relationship was not moderated by avatar credibility. Our results seem to suggest that this would be true. Specifically, the connection between avatar androgyny and partner androgyny and avatar credibility and partner credibility suggests that there would be a path from avatar anthropomorphism to partner anthropomorphism. However, uncovering how anthropomorphic a person’s unseen unmediated body is when they are represented by an avatar during the interaction presents a difficult methodological challenge. To manipulate the individual variables, and control for the possible influence of individual differences in behavior on avatar perception, participants were assigned to a condition that defined the avatar they saw and the avatar that represented them. Participants may have behaved in ways that were consistent with the avatar they selected, and not the avatar that their partner saw. While this allowed us to control the characteristics of the avatar and examine the effects of these characteristics, it limits the generalizability of these findings. The path model clearly shows that androgyny judgments come first, and that they directly influence anthropomorphism. This could mean that attributions of gender and anthropomorphism are made in parallel but that gender characteristics, when present, are more salient than anthropomorphism characteristics. Future research should continue to explore this process in preexisting interpersonal relationships, with different avatars, and in different contexts while controlling for anthropomorphism and androgyny. We note that this measure of anthropomorphism combined items that could represent realism and anthropomorphism. We are therefore unable to differentiate the relative effect of these distinct constructs, or their causal order. Future research should carefully measure and differentiate these constructs, and such research is currently underway. The causal model reveals that people first evaluate gender characteristics of a person or object in an environment and use that information to inform their perception of the object or person. Impressions derived from the avatar’s visual characteristics influence more abstract concepts further down the causal chain. Avatar androgyny influences partner androgyny and avatar credibility influences partner credibility. The fact that inferences of credibility were less strong in the social chat step as compared to the static step may provide some indication that the partner’s behavior influenced the process in that the links between avatar characteristics (anthropomorphism and androgyny). The relative influence of behavior is likely to increase during longer interactions outside a laboratory setting, and with pre-existing relationships. Future research should use repeated measure designs to examine these factors. As predicted, perceptions of avatar androgyny (the absence of masculinity or femininity) had a negative effect on partner credibility, mediated by the perceptions of partner androgyny. This implies that people prefer interacting with avatars that have clear indicators of gender (masculine or feminine) to androgynous avatars that are difficult to categorize as one or the other. One could conceivably argue that gender perceptions such as androgyny, as opposed to sex category perception, are more malleable and would also be similarly less grounded in physical characteristics than anthropomorphism. Conclusion Computer media enable users and researchers to control aspects of the interaction that cannot be modified or altered in offline interactions. The ability to alter things like the visual appearance of an interaction partner can provide insight into the perception process. Examining the influence of online visual information, such as the avatar, may help explain why people rely on the offline body in the perception process and how this information is used. These results suggest that as long as there is some representation, this representation will influence the perception of an interaction partner. It is not surprising that characteristics of an avatar influence perceptions of the person represented. The fact that the avatar characteristics influenced perceptions of the partner and not the other way around implies that people make initial mental models based on physical (visual) characteristics of people they encounter, whether they are online or offline. It is likely that this mental model is adapted with information from the partner’s behavior. Future research should continue to explore whether this reliance on visible information is based on the desire to reduce cognitive load, or a sense that visual information provides more accurate information, whether it is stable or not. Both IM usage and computer efficacy had some direct influence on perceptions of avatars. The small direct link from IM use to partner credibility suggests that those with more IM usage based at least part of their evaluation of their partner’s credibility on factors other than the characteristics of their avatar. At the same time, those with more IM usage reported avatars as less anthropomorphic overall, and those with more computer efficacy rated avatars as more androgynous (less masculine or feminine). Both of these may be interpreted to support the prediction that those with more computer efficacy and experience with online interactions and avatars have different expectations from those with less experience (Shneiderman, 1988; Slater &Steed, 2002; Nowak, 2004). Specifically, the data suggest that more frequent IM users have a higherstandard for what looks anthropomorphic and for what looks either masculine or feminine. As moreand more people use computers in general, and instant messaging in particular, we may seeincreasing user adaptation in the online socialization process. Future research should continue toexamine the distinction between computer efficacy and IM use and how those and other individualdifference factors may influence people’s perceptions of one another online.It is important for users and designers of these systems to understand that the characteristicsof the avatar they choose will influence how the avatars are perceived, and how that affectsperceptions of the users. These results reinforce the conclusion that perception of the avatar hasinfluence on perceptions of the partner. Users should choose their avatars as carefully as theymonitor their behavior. Simply put, these results suggest that wrong avatar can make you, literally,look bad, while using a more credible, more anthropomorphic, less androgynous avatar (whethervery masculine or very feminine), will make you appear more credible. ReferencesAmbady, N., Hallahan, M., Rosenthal, R. (1995). On judging and being judged acturatelly in zero-acquaintance situations. Journal of Personality and Social Psychology, 69(3), 518-529.Argyle, M. (1988). Clothes, Physique, and other aspects of appearance. In M. Argyle (Ed.), BodilyCommunication. (Second Ed., pp. 233-254). New York: International Universities Press, Inc.Balsamo, A. (1995). Forms of Technological Embodiment: Reading the Body in Contemporary Culture. InM. Featherstone & R. Burrows (Eds.), Cyberspace/Cyberbodies/Cyberpunk; Cultures of TechnologicalEmbodiment (pp. 215-237). Thousand Oaks, CA: Sage Publications.Bem, S. L. (1981). Gender Schema Theory: A Cognitive Account of Sex Typing. Psychology Review, 88(4), 354-364Benford, S., Greenhalg, C., Bowers, J., Snowdon, D., & Fahlén, L. (1995). User Embodiment inCollaborative Virtual Environments. Paper presented at the CHI '95 Mosaic of Creativity, Denver,CO.Berger, C., & Calabrese, R., (1975). Some explorations in initial interaction and beyond: Toward adevelopmental theory of interpersonal communication. Human Communication Research, 1, 99-112.Biocca, F. (1997). The Cyborg's Dilemma: Progressive Embodiment in Virtual Environments. Journal ofComputer Mediated Communication, 3(2), Retrieved fromhttp://www.ascusc.org/jcmc/vol3/issue2/biocca2.html.Biocca, F., & Nowak, K.L. (2002). Plugging your body into the telecommunication system: MediatedEmbodiment, Media Interfaces, and Social Virtual Environments. In D. Atkin & C. Lin (Eds.),Communication Technology and Society (pp. 407-447). Cresskill, NJ: Hampton Press.Bull, P., & Rumsey, N. (1988). The Social Psychology of Facial Appearance. Springer-Verlag.Burgoon, J. (1994). Nonverbal Signals. In M. L. Knapp & G. R. Miller (Eds.), Handbook of InterpersonalCommunication (pp. 229-285). Thousand Oaks, CA: Sage Publications. Clark, N. (1995). Rear-View Mirrorshades: The Recursive Generation of the Cyberbody,Cyberspace/Cyberbodies/Cyberpunk; Cultures of Technological Embodiment. (pp. 113-133). Thousand Oaks,CA: Sage Publications.Clatterbuck, G. (1979). Attributional Confidence and Uncertainty. Human Communication Research, 5(2), 147-157.Compeau, D. R., & Higgins, C. A. (1995). Computer self-efficacy: development of a measure and initial test.MIS Quarterly, 19(2), 189-212.Compeau, D., Higgins, C. A., & Huff, S. (1999). Social cognitive theory and individual reactions tocomputing technology: a longitudinal study. MIS Quarterly, 23(2), 145.Cornetto, K. & Nowak, K.L. (2006). Sex categorization and usernames in CMC; examining people’sperceptions and accuracy. Cyberpsychology and Behavior, 9(4), 377-387.Damer, B. (1997).Avatars!: Exploring and Building Virtual Worlds on the Internet. Berkley, CA: Peachpit Press.Diamond, G. (2000). Liquid Chat II. Downloaded from http://www.programmersresource.comDiSalvo, Carl & Gemperle, Francine (2003). From seduction to fulfillment: the use ofanthropomorphic form in design, Proceedings of the 2003 international conference on Designing pleasurableproducts and interfaces, June 23-26, 2003, Pittsburgh, PA, USA.Eastin, M. S., & LaRose, R. (2000). Internet self-efficacy and the psychology of the digital divide. 6(1). Journal ofComputer Mediated Communication. Available:http://www.ascusc.org/jcmc/vol6/issue1/eastin.html [2002, October 16].Garau, M., Slater, M., Vinayagamoorthy, V., Brogni, A., Steed, A., & Sasse, M. A. (2003). The impact of avatarrealism and eye gaze control on perceived quality of communication in a shared immersive virtual environment. Paperpresented at the CHI 2003, Ft. lauderdale, FL.Heilbrun, J., & Pitman, D. (1979). Testing Some Basic Assumption About Psychological Androgyny. Journalof Genetic Psychology, 135(2), 175. Heilman, M. (1995). Sex stereotypes and Their Effects in the Workplace: What we Know and what WeDon't Know. Journal of Social Behavior and Personality, special issue, Gender in the workplace., 10(6), 3-26.Hemp, P. (2006). Avatar-based marketing. Harvard Business Review, 84(6), p.48.Hert, P. (1997). Social Dynamics of an On-Line Scholarly Debate. The Information Society, 13, 329-360.Huang, P. (2006). Online Shopping. Forbes.com. Retrived on September 30,2006 fromhttp://www.forbes.com/technology/2006/06/14/american-apparel-retail_cz_ph_0614secondlife.html.Hu, Y., Wood, J., Smith, V., & Westbrook, N. (2004). Friendships through IM: Examining the relationshipbetween instant messaging and intimacy.Journal of Computer Mediated Communication, 10(1). Availableat .Ichheiser, G. (1970). Appearances and Realities; Misunderstanding in Human Relations. San Francisco: Jossey-Bass,Inc. Publishers.Ikea (2006). Company's website. Retrieved on September 30, 2006 from http://www.ikea.com/ms/en_US/.Karniol, R; Reichman, S; & Fund, L (2000). Children's gender orientation and perceptions of female, male,and gender-ambiguous animal characters. Sex Roles. 43 (5-6). 377-393.Kenny, D.A., Horner, C., Kashy, D.A., & Chu, L. (1992). Consensus at Zero Acquaintance: Replication,Behavioral Cues, and Stability. Journal of Personality and Social Psychology, 62, 88-97.Koda, T., & Maes, P. (1996). Agents with Faces: the Effects of Personification. Paper presented at the Human-Computer Interaction'96, London.Kunda, Z. (1999). Social cognition: Making sense of people. MIT Press: Cambridge, MA.Lea, M., & Spears, R. (1992). Paralanguage and Social Perception in Computer-Mediated Communication.Journal of Organizational Computing, 2(3&4), 321-341.Lee, E. (2004). Effects of gendered character representation on person perception and informational socialinfluence in computer-mediated communication, Computers in Human Behavior, 20(6), 779-799. Maseon, P. (2006). Do avatars dream of electric racoons?. BBC News. Retrieved September 30, 2006 fromhttp://news.bbc.co.uk/2/hi/programmes/newsnight/4583924.stm.McCroskey, J. (1971). Ethos, Credibility, and Communication in the Real World. North Carolina Journal ofSpeech, 4, 24-31.McCroskey, J. & Young, T. (1981). Ethos and credibility: the construct and its measurement after threedecades. The central state speech journal, 32, 24-34.Murphy, P. (1994). Social interest and psychological androgyny: Conceptualized and tested. IndividualPsychology: Journal of Adlerian Theory, Research & Practice, 50(1), 18-30. Retrieved Monday, April 16,2007 from the PsycINFO database.Nass, C., Steuer, J., Tauber, E., & Reeder, H. (1993, April 24-29). Anthropomorphism, Agency, & Ethopoea:Computers as Social Actors. Paper presented at the InterChi '93, Amsterdam, the Netherlands.Nowak, K. (2003). Sex categorization in CMC: Exploring the utopian promise.Media Psychology, 5(1). P. 83-104Nowak, K. L. (2004). The Influence of Anthropomorphism and Agency on Social Judgment in VirtualEnvironments. Journal of Computer Mediated Communication, 9(2), available athttp://jcmc.indiana.edu/vol9/issue2/nowak.html.Nowak, K.L., & Biocca, F. (2003). The effect of the agency and anthropomorphism on users' sense oftelepresence, copresence, and social presence in virtual environments. Presence: Teleoperators andVirtual Environments., 12(5), 481-494.Nowak, K.L. & Rauh, C. (2005). The Influence of the Avatar on Online Perceptions ofAnthropomorphism, Androgyny, Credibility, Homophily, and Attraction. Journal of Computer-MediatedCommunication, 11(1), article 8.Oravec, J. (1996). Virtual Individuals, Virtual Groups; Human Dimensions of Groupware and Computer Networking:Cambridge University Press. Palmer, M. (1995). Interpersonal Communication and Virtual Reality: Mediating Interpersonal Relationships.In F. Biocca & M. Levy (Eds.), Communication in the Age of Virtual Reality (pp. 277-299). Hillsdale,New Jersey: Lawrence Erlbaum Associates.Ramirez, A., Walther, J. B., Burgoon, J. K., Sunnafrank, M. (2002). Information-SeekingStrategies, uncertainty, and computer mediated communication Toward a conceptual model. HumanCommunication Research, 28(2), 213-228.Reeves, B., & Nass, C. (1996). The media equation: How people treat computers, television, and new media like realpeople and places. Stanford, CA: CSLI Publications.Schroeder, R. (2002). Social Interaction in Virtual Environments: Key issues, common themes, and aframework for research. In R. Schroeder (Ed.),The Social Life of Avatars; Presence and Interaction inShared Virtual Environments. London: Springer-Verlag.Shapiro, K. J. (1997). A phenomenological approach to the study of nonhuman animals. In R. W. Mitchell etal., eds. Anthropomorphism, Anecdotes, and Animals. New York: SUNY, 92-101.Sheehan, J. (1991). Introduction: Humans and Machines. In J. Sheehan & M. Sosna (Eds.), The Boundaries ofHumanity: Humans, Animals, Machines (pp. 135-141). Oxford, England: University of California Press.Sherman, R., End, C., Kraan, E., Cole, A., Campbell, J., Klausner, J., & Birchmeier, Z. (2001).Metaperception in Cyberspace. CyberPsychology & Behavior, 4(1), 123-129.Shiu, E.; Lenhart, A. (2004). Reports: Online Activities & Pursuits; How Americans Use Instant Messaging.Report from Pew Internet and American Life Project. Availablehttp://www.pewinternet.org/PPF/r/133/report_display.aspShneiderman, B. (1988). A nonanthropomorphic style guide: Overcoming the humpty-dumpty syndrome.The Computing Teacher, October, 9-10.Skitka, L., & Maslach, C. (1996). Gender as Schematic Category: A role Construct Approach. Social Behaviorand Personality, 24(1), 53-74. Slater, M., & Steed, A. (2002). Meeting people virtually: experiments in shared virtual environments. In R.Schroeder (Ed.), The social life of avatars: Presence and interaction in shared virtual environments, pp. 146-171.London: Springer-Verlag.Soukup, C. (2004). Multimedia Performance in a Computer-Mediated Community: Communication as aVirtual Drama. Journal of Computer Mediated Communication, 9(4), available athttp://jcmc.indiana.edu/vol9/issue4/soukup.html.Spender, D. (1996). Nattering on the Net: Women, Power and Cyberspace. Melbourne, Australia.: Spinifex Press.Sproull, L., & Kiesler, S. (1986). Reducing social context cues: Electronic mail in organizationalcommunication. Management Science, 32(11), 1492-1512.Suler, J. (1996). Life at the Palace; A cyberpsychology case study. Available:www1.rider.edu/~suler/psycyber/psycyber.html [ 2001, October 5].Taylor, T. L. (2002). Living Digitally: Embodiment in Virtual Worlds. In R. Schroeder (Ed.),The Social Life ofAvatars; Presence and Interaction in Shared Virtual Environments (pp. 40-62). London: Springer-Verlag.Turkle, S. (1995). Life on the screen: Identity in the age of the Internet. New York: Simon & Schuster.Walther, J.B. (In Press), Selective self-presentation in computer-mediated communication: Hyperpersonaldimensions of technology, language, and cognition, Computers in Human Behavior, CorrectedProof, Available online 30 June 2006. (http://www.sciencedirect.com/science/article/B6VDC-4K9C6JF-1/2/9e0fe5f957b0ea785fa9b3c4c94617bf)Waskul, D., & Douglass, M. (1997). Cyberself: The Emergence of Self in On-Line Chat. The Information Society,13(4), 375-397.West, C., & Zimmerman, D. (1991). Doing Gender. In J. Lorber & S. A. Farrell (Eds.), The Social Constructionof Gender (pp. 13-37). Newbury Park, CA: Sage Publications. Wexelblat, A. (1997). Don't Make That Face: a report on anthropomorphizing an interface, [Web Page].Available: http://wex.www.media.mit.edu/people/wex/anthro-expt-paper/Anthro-r.htm [1999,November].Witmer, D., & Katzman, S. (1997, May 24, 2001). On-Line Smiles: Does Gender Make a Difference in theUse of Graphic Accents?. Journal of Computer Mediated Communication, 2 (4). Retrieved October 18,2002 from: . Table 1 – Experimental Conditions of Step 2AnthropomorphismAndrogynyLowHighLowHigh