Abstract
Although considerable research has identified patterns in online communication and interaction related to a range of individual characteristics, analyses of age have been limited, especially those that compare age groups. Research that does examine online communication by age largely focuses on linguistic elements. However, social identity approaches to group communication emphasize the importance of non-linguistic factors such as appearance and non-verbal behaviors. These factors are especially important to explore in online settings where traditional physical markers of age are largely unseen. To examine ways that users communicate age identity through both visual and textual means, we use multiple linear regression and qualitative methods to explore the behavior of 201 players of a custom game in the virtual world Second Life. Analyses of chat, avatar movement, and appearance suggest that although residents primarily used youthful-looking avatars, age differences emerged more strongly in visual factors than in language use.
Age acts as a significant organizing force in how we develop social relationships and in how we express ourselves. As such, it is a central aspect of identity (Logan et al., 1992). Scholarship suggests that people construct their identity in order to establish and maintain a positive self-concept and a particular status within a group. In particular, self-categorization theory suggests that perceptions of what specific identities are salient in a given context depend on both linguistic and non-linguistic cues – such as appearance and non-verbal behaviors – acquired from both individuals and the environment (LeBoeuf et al., 2010; Turner and Oakes, 1986).
In online social spaces, age is considered an important factor in interaction. Age identity allows for the bonding of people in social spaces through common characteristics. At the same time, intergenerational contexts are becoming more common online, especially in social media and multiplayer online games (MMOs). Indeed, in many MMOs, groups of players in their teens and in their 50s often interact as peers (Williams et al., 2006), rather than divide themselves into narrow age groups. Yet in many online social spaces, such as online games or chat rooms, age is relatively hidden as compared with face-to-face interaction. In such contexts, are there nevertheless ways that people communicate their age?
Prior scholarship suggests that linguistic cues such as swearing, emoticon use, and punctuation can be important markers that indicate age identity in online communication. Yet much of that research is focused on text-only communication (Rosenthal and McKeown, 2011; Tam and Martell, 2009). As online social spaces with an avatar or virtual representation become more common, it is important to consider the role of that visual representation in conveying age identity. Unfortunately, few studies have examined how the interplay of communication, avatar appearance, and behavior communicate age. Existing research tends to examine only a specific age group such as young adults (Leung, 2010; Schler et al., 2006) or seniors (Dell and Marinova, 2007; Lin et al., 2004; Reed and Fitzpatrick, 2008). Few studies of computer-mediated communication (CMC) systematically compare communication patterns across age groups.
This paper addresses those gaps by examining patterns in both visual and textual communication among 201 players of a custom game in the virtual world Second Life (SL). We ask, How do users in virtual worlds express their age identity? As social identity approaches would suggest, language is only a small part of overall identity performance. In avatar-based online spaces such as World of Warcraft or SL, avatar design and use are key aspects of self-presentation (Martey and Consalvo, 2011; Pearce, 2009). Indeed, research has found that individuals use avatars to express both bodily attributes (e.g. gender) and psychological ones (e.g. self-confidence) (Merola and Peña, 2010). Social identity approaches suggest that in virtual worlds, age could similarly be performed in a range of ways: through writing style and linguistic markers including slang, emoticons, punctuation, and sentence structure; through avatar movements and gestures; and through avatar attractiveness or visual cultural references.
We use multiple linear regression on logs of chat, avatar use, and avatar appearance to explore how players perform age identity in a specific context. We argue that avatar appearance adds a non-verbal dimension to ways that people interact and communicate, even when most players use avatars with youthful faces and bodies. Our analysis suggests that other visual factors provide important cues about age identity, and that linguistic differences in behavior are less powerfully associated with age than visual ones.
Age identity in online spaces
This study examines age identity, which McCann et al. argue ‘acts as a “pre-interactional” tendency whereby a strong sense of identification with a particular group (e.g. young adults) influences communication with outgroup members (e.g. older people)’ (2004: 89). As such, communication of age identity reflects how people perceive themselves in social contexts and in relation to others (Logan et al., 1992).
Scholarship suggests that age is culturally defined and is especially apparent in linguistic practices (Suslak, 2009). Cultural influences on age identity are often examined in studies of generational cohorts: groups of people born into particular socio-cultural environments, such as the Baby Boomers born after World War II who were shaped by the politics of the 1960s and 1970s, or Prensky’s (2001) digital natives, whom he describes as developing distinct ways of learning, speaking, and interacting based on their early exposure to computer technologies and games. Cohort research uses a cultural framework to examine the ways that such experiences define each age group, influencing behavior and attitudes towards social and political issues and events, as well as styles of dress, language use, and social interaction (Strauss and Howe, 1991; Winograd and Hais, 2008).
We examine the communication of age identity online as an aspect of social identity based on Tajfel and Turner’s (1986) work in this area. Research on the relationship between social identity and group behavior suggests that individuals associate particular behaviors to particular identities, including language (LeBoeuf et al., 2010) and material symbols such as clothing or product use (Shavitt and Nelson, 2000). In particular, self-categorization theory indicates that social information contributes to perceptions about what behaviors are perceived as salient and normative (Turner and Oakes, 1986). That information in turn influences which specific identities are activated and performed. Such theories align with Goffman’s (1959) role theory, in which social role performance is situated in a front that includes both intended and unintended expressions, and is ‘the expressive equipment of a standard kind intentionally or unwittingly employed by the individual during his [sic] performance.’ (p. 22). How individuals select such expressions might be different in online spaces where characteristics beyond the actor’s control offline such as height, weight, skin color – and to a lesser, but still important extent, age and gender – can be intentionally designed. Thus in virtual worlds, the salience of a broader range of identity markers can be influenced by the norms of the group, as self-categorization theory suggests.
As part of age performance, linguistic patterns can be used to signal membership in specific age categories. Use of slang, for example, is often considered a discourse marker of youth culture (Moore, 2004), and along with profanity, has been associated with younger users in online communication (Barbieri, 2008). In CMC, one study examined the ways that oral structures of communication practices, such as the use of ‘like’ by teenagers, have migrated into instant messaging (Jones and Schieffelin, 2009). In an analysis of age patterns in language used in blogs, Schler et al. (2006) found that older bloggers tended to use more prepositions and articles and fewer pronouns and assent/negation words. Tam and Martell (2009) used natural language processing to categorize online chat and successfully identified participant age groups through patterns in punctuation, capital letters, sentence length, and word types. Using automatic text categorization techniques on a large corpus of chat room text, Lin (2007) found that younger authors (largely teenagers) were more likely to use emoticons. Several other studies have argued that youth are more fluent in the use of online-specific language, including emoticons (Derks et al., 2007; Huffaker and Calvert, 2005). Although linguistic analyses have successfully predicted age of online speakers, little systematic research has addressed appearance or non-verbal behaviors that contribute to age performance online.
According to theoretical models of communication and aging (e.g. Hummert, 1994; McCann et al., 2004; Ryan et al., 1986), physical cues are key markers of age identity that trigger social stereotypes in encounters between generations. Correspondingly, appearance can play an important role in the ways avatars are used to communicate or express the self in a virtual world. Research on avatar design and appearance, although still limited, suggests that in virtual worlds such as SL, players customize their appearance to fit in with specific cultures or portray a specific persona (Merola and Peña, 2010). Messinger et al. (2008) found that although most people create avatars based on their offline appearance, many improve upon their looks, including appearing younger. Reed and Fitzpatrick (2008) asked 22 older people without experience in SL to design an avatar and found that the majority felt their avatar looked younger than them, in spite of the fact that they felt the avatar also looked extremely similar to them. Studies suggest that avatar design is related to what people wish to communicate about themselves when interacting with others in more idealized ways: creating a more attractive avatar for dating (Vasalou and Joinson, 2009) or a more frightening avatar to play ‘the bad guy’ (Salazar, 2009). Such research corresponds with ways that scholars of dress and fashion have identified age differences in offline appearance, especially among youth subcultures (Hebdige, 1981). In addition, research suggests that people use avatar movement, gestures, and facial expressions to communicate ideas and emotions (Antonijevic, 2008; Koda et al., 2006; Yee and Bailenson, 2007). The extent to which such choices differ by age, however, is unclear.
Methods
In order to examine the ways age is performed through verbal and non-verbal online communication behavior, we used mixed methods to analyze conversation and avatar use among 201 participants as they played a custom-built game in the virtual world SL. SL is a visual, digital environment where users or ‘residents’ can design and create avatars to move around and interact. Residents have significant control in avatar appearance customization. They can change the dress, complexion, hair, face, and body of their avatar, including height, breast and chest size. All default avatars appear youthful, however, and wrinkles cannot easily be added to face shapes. Residents also create and sell complex pre-made ‘skins’ that include realistic looking body features. Avatars do not have to be human. Residents can alter avatar physiques to be vampire-like, with long teeth and pale skin, they can be fairies with giant wings, or even objects such as a plane or car. They can be a ‘furry’, a part-animal, part-human avatar popular among some communities in SL (see Figure 1).
Different types of ‘furry’ avatars.
SL is a predominately social virtual world where users interact, chat, and play a wide range of games created by residents. For the study, we created an elaborate point-and-click mystery quest game set in a fantastical stylization of Victorian England known as steampunk. This style was selected because it is a rich and recognizable style within SL: several of SL’s most populated islands have steampunk themes. Developed as part of a larger study on interaction in virtual worlds, 1 the game was designed to evoke text chat conversation, movement in the environment, group cooperation, and problem-solving during the two to three hours it took to solve the mystery.
The game created for this study asked groups of three to five unacquainted participants to join the ‘Adamourne Detective Trainees’ in order to solve ‘The Case of the Missing Moonstone.’ Tasked by a non-player character (NPC), Chief Billingsly, to investigate the theft, players collected clues at the scene of the crime (an elaborate Victorian manor) and then explored dark town streets, a decrepit factory, and finally the apartments of the prime suspect (see Figure 2). Clues were displayed as text on the screen or in audio clips when players clicked objects in each zone. Sometimes players received items such as a small gray kitten that had to be delivered to a magical idol to comfort him. Other challenges required players to solve word puzzles to determine a secret code or to place their avatars in specific locations to activate security buttons.
The four main zones of the game (left to right): Blayfield Manor, Town, the Jeter Factory, and Morlock’s Apartments.
Study procedures
Approximately 400 participants were recruited through advertising in SL, Twitter, and Facebook. They were asked to answer an online questionnaire assessing demographics, experience with online games, and other characteristics including leadership and social conformity. Age was assessed three times to improve reliability: in the sign-up as current age, early in the pre-session questionnaire as current age, and at the end as date of birth. No discrepancies were found across these measures. Then, 260 qualified participants (18+ years old with at least 10 hours experience in SL) were assigned into 48 groups of three to five unacquainted people each to play the game for an average of 2.5 hours. After their session, participants filled out a 15-minute questionnaire assessing their experience. Due to cancelations, 225 actually participated, and 201 completed all stages of the study with no missing data for measures used in this analysis.
Upon arrival, participants attached to their avatars a data logger that recorded public chat, object clicks, and avatar movement. Two to three researchers, unseen by participants, observed and recorded the sessions on video and in field notes. One researcher, with the appearance of an automaton, accompanied the group through the game to answer questions or solve technical problems, but interacted with participants as little as possible, following ethnographic participant-observer techniques. Participants who completed all stages of the study received compensation in SL currency, $L5000 (~$20 USD).
Coding development and reliability
The data logged by the chat and movement logger were coded by computers and humans to develop the variables used in this study. These categories were developed to capture a variety of communication, appearance, and movement variables, not only to test predictors of age but also of other real world characteristics, such as gender and education. Approximately 40,000 chat lines and 45,000 object clicks from the 201 avatars were analyzed. Human coders identified dialog acts (function of messages), communication links (who is speaking to whom), avatar movement patterns, and avatar appearance. 2 Some features of text chat and avatar movements were auto-coded by computers, including punctuation, word length, and objects clicked. Lexicons were created to count emoticons, apologies, appreciation, profanity, praise, and laughter. Automatic language-processing algorithms identified and counted occurrences of text elements. For example, ‘lol’ was counted as an occurrence of laughter. Automatic coding was tested against human annotators on a subset of data to determine precision, recall, and ground truth with a .8 threshold set and met for all measures. 3 Automatic object click logging is a count of the times an avatar clicks objects. Avatar movement was automatically calculated, including distance the avatar traveled and proxemic measures including avatar’s average distance from the automaton participant-observer.
The chat codebook for this study was based on the theoretical discourse principles of Searle (1969) and Austin (1962), and on other projects from our prior research (Stromer-Galley, 2007). The chat codebook included over 20 categories. Here, we describe some of the factors that went into our model. Conventional Openings and Conventional Closings are statements such as ‘hello’ and ‘good bye.’ Other Conventional Phrases are those that fulfill conventional social functions, including appreciation and apology. Agree-Accept is a statement that agrees with or accepts others’ suggestion, request, or information. Corrected Misspellings are used to correct typos from a previous message, often indicated by the use of an asterisk (e.g. ‘*before’). Emotional Phrases are usually a single word or phrase that conveys the emotional state of a speaker, such as ‘ewww’ or ‘cool!’
A separate avatar appearance codebook was developed based on qualitative analyses performed for this study, as well as literature on avatar appearance (Merola and Peña, 2010; Vasalou and Joinson, 2009; Yee and Bailenson, 2007) and avatar use (Kafai et al., 2010). Measures used here include use of a gender-idealized avatar, which identified whether or not avatar shape reflects an idealized humanoid body (yes/no), including depicted weight, height, and musculature (but not clothing). For example, idealized female avatars are operationalized as having a narrow waist, slim legs, and large breasts; men were coded as idealized if they were tall with wide shoulders and strong facial features (see Figure 3). Avatar gender was identified as the gender players select in character creation (male, female, or unknown). Additional avatar coding identified costume use, clothing elaborateness, species (human, vampire, fairy, furry, etc.), and animations performed (see Martey and Consalvo, 2011, for a detailed analysis of avatar appearance). Although we developed a coding scheme for traditional age appearance, such as use of avatars with gray hair and wrinkles, no participants used avatars with such characteristics; all avatars in the study appeared as youthful adults.
Gender-idealized and non-idealized female and male human avatars.
For all human coding, researchers coded approximately 10% of the full dataset to establish intercoder reliability using a threshold of Krippendorff’s alpha of .8 for all measures. In addition to quantitative analyses presented here, we conducted extensive qualitative analyses based on session field notes, videos, and 30 interviews with participants selected in a purposive sample. These analyses contributed to refinement of categories and decision rules for the codebooks, survey item development, and model development and analysis.
Although there is a slight younger skew to the distribution of participant age, the average age of this sample is actually slightly older than other age estimates (Linden Labs reported a mean SL resident age of 32 in 2008). Overall, about 20% of our sample was 18–25, about 57% was 26–45, and 24% were over 45 years old. Participants were 51% female, 49% male, and 69% were either in college or had completed at least a Bachelor’s degree. The internet experience measure was a 10-item scale (Cronbach’s α = .728) assessing participants’ frequency of use from 1 (never) to 5 (several times a day) of different online activities. Experience in SL was assessed by asking participants how frequently they visited SL, from 1 (less than once a month) to 6 (several times a day). We also assessed how long ago they first started playing SL in years, from 1 (less than 1 month ago) to 5 (over two years ago) (see Table 1).
Participant characteristics.
SL: Second Life.
Table 2 shows descriptive statistics for the variables used in our two models. In addition, 11% of players used a furry avatar and 78% used a gender-idealized avatar. Steampunk costume style was used by 66%, and 13% used a costume that was coded as non-elaborate, 56% as moderately elaborate, and 31% as highly elaborate.
Descriptive statistics for continuous variables in the models.
Results and analysis
Survey data and coded logs of avatar chat and movement were analyzed using correlations and multiple linear regression in SPSS software. Models were developed to examine how observed communication behaviors, including non-verbal communication expressed in movement and appearance, are related to the age of participants. In order to explore if age is detectible in observations of online communication behavior alone, no self-reported measures were included in model development. For example, although communication behaviors are strongly related to factors such as education and national origin, these self-reported measures were not included in the models. This approach results in lower R2 levels than might otherwise be achieved, but emphasizes the ways in which observed online communication can indicate age without knowledge of non-observed characteristics. This approach allows analysis to focus on behaviors individuals actively perform in this space, rather than hidden or non-observable factors. All measures were calculated per 100 utterances to take differences in chat volume into account.
Seven chat measures and five movement and appearance measures are significantly correlated with age. Table 3 shows the simple Pearson correlations (two-tailed) between age and the main observed variables. Of particular note is that profanity, punctuation, and capital letters are not significantly associated with age (p > .05), contrary to previous research (Barbieri, 2008; Lin, 2007; Schler et al., 2006; Tam and Martell, 2009).
Correlation coefficients for coded variables and age.
NPC: non-player character.
p < .05; **p < .01 (two-tailed).
In order to test the relationships between age and factors identified as important in the literature, we created a regression model that includes only measures corresponding with previous research: punctuation, emoticons, profanity, capital letters, sentence length, Agree-Accept and Disagree-Reject statements (Barbieri, 2008; Lin, 2007; Schler et al., 2006; Tam and Martell, 2009). To avoid collinearity, we summed our four punctuation measures (commas, periods, exclamation points, question marks) into a single measure (correlation with Age: α = –.074, Sig. = .289). We included avatar appearance measures related to research on avatar design: use of a steampunk style as a measure of fitting avatar style to the game setting (Kafai et al., 2010; Pearce, 2009); use of a gender-idealized avatar as a measure of creating an attractive avatar (Messinger et al., 2008; Vasalou and Joinson, 2009); and costume elaborateness as a measure of using an avatar for cultural expression (Salazar, 2009) (see Figure 4).
Non-elaborately dressed avatar in a contemporary costume (a) and an elaborately dressed avatar in steampunk costume (b).
As Table 4 indicates, Model 1, which used variables indicated by the literature, explains approximately 15% of the variance in participant age (R2 = .154), with an F ratio of 3.50 (Sig. = .000; DF = 10). In this model, only the chat measure Agree-Accept and use of an avatar with a gender-idealized appearance were significant (p < .05).
Model 1: Summary of regressions analysis for variables predicting participant age (N = 201).
Note. R2 = .154.
p < .05; **p < .01.
To further examine factors predictive of participant age, a second model was developed using only variables with a significance of p < .05. Model 2 was developed to provide the leanest combination of significant predictors with the highest R2 value. To do so, various sets of variables were explored to identify a combination of predictors that, when considered together, best predict participant age. Non-significant variables were not included in Model 2. Some variables were highly collinear, such as total object clicks and button clicks; in such cases, variables were compared and the best predictor was selected.
As shown in Table 5, Model 2 demonstrates that younger players were more likely to use avatars that are taller, furries, and not gender-idealized in appearance. Avatar gender was a significant predictor, indicating that older participants were more likely to use a female avatar. Given that very few participants used an avatar gender different from their own (8%), this is close to a proxy for players’ actual gender. Older players used more conventional opening phrases, more other conventional phrases, and more agreement phrases; older players also used fewer emotional phrases, they corrected misspellings less, and used fewer sequential question marks. In addition, older players were less likely to click the vault buttons that solve a quest challenge. This model explains approximately 46% of the variance in participant age (R2 = .463), with an F ratio of 13.51 (Sig. = .000; DF = 12).
Model 2: Summary of regressions analysis for variables predicting participant age (N = 201).
Note. R2 = .463.
p < .05; **p < .01.
Discussion
Communication of social identity in online settings can incorporate a wide range of behaviors. As Goffman would suggest, the context, language, and non-verbal behavior are important aspects of the front players perform in social settings. In virtual worlds, the increased control over appearance, movement, and conversation can result in more strategically selected performances that may or may not express specific aspects of identity such as age. The setting and social context influence the extent to which age as an identity category is salient for participants, as self-categorization theory would suggest. Differences in how younger and older users behave in the three-dimensional virtual world of SL shed light on how players perform their age through not only linguistic but also visual communication in this context.
The appearance of participant avatars is a significant predictor of their age. Using a taller avatar that corresponds with traditional norms around men’s and women’s appearance and that is human is more common among older participants, corresponding with previous findings that people often use more idealized avatars in virtual worlds (Martey and Stromer-Galley, 2007; Messinger et al., 2008). Research has found that some older people create an ageless representation of themselves online to combat dissatisfaction with old age (Dell and Marinova, 2007). Our results confirm this. Our results also correspond to Reed and Fitzpatrick’s (2008) findings that most participants aged 50 and older designed avatars who looked younger than their actual age. The lack of older-appearing avatars might also be due to the relative scarcity of older shapes and skins in the SL marketplace. In addition, all the default avatars in SL appear youthful. Similarly, avatar customization options make adding certain age markers such as wrinkles difficult, although giving an avatar gray hair is simple. Our qualitative observations revealed that residents are largely youthful in appearance on many other SL islands as well, perhaps as a result of these constraints.
We also found furry avatars more common among younger players. In fact, not a single participant over age 35 used a furry avatar. These findings suggest that younger players are more likely to experiment with less traditional appearances, engaging in what Turkle (1995) calls identity play and corresponding with findings that young people use avatars for this purpose (Kafai et al., 2010). Alternatively, as Reed and Fitzpatrick (2008) suggest, older players could have been more focused on representing themselves as something they perceived as accurate to their offline appearance.
Another notable pattern was more frequent use among older players of language associated with politeness: conventional openings (e.g. ‘hello’ and ‘how are you?’), and other conventional phrases (e.g. ‘sorry’ or ‘thanks’). In addition, older players are more likely to use Agree-Accept statements (e.g. ‘yes, let’s go’). The increased use of these conventions may be due in part to the fact that group members did not previously know each other, and perhaps older players are more oriented toward being supportive and facilitating social interchange than younger players. Morand and Ocker (2003) suggest that the use of polite language that shows esteem and respect for others should be understood slightly differently in CMC. Although many shorthand forms of conventional phrases are used, such as ‘ty’ for ‘thank you’, the scarcity of non-verbal politeness cues as well as the need for lean communication (minimal typing) may change how politeness manifests in text chat. In particular, the authors argue that polite phrases may clash with clarity and efficiency in that medium, reinforcing a culture of fewer polite phrases used in conversation. Such a pattern is particularly likely in game settings where typing into chat necessarily stops other actions such as combating monsters or clicking objects. Our findings suggest that older players may have been more willing to sacrifice game efficiency in favor of politeness. The fact that younger players were more likely to click the buttons that solved the vault puzzle suggests that they were indeed more actively engaged than older players in certain game actions.
The tendency for older participants to use more polite language may also be a function of the text mode of communication in the game we created. SL provides a popular voice chat interface (used by 50% of residents at 1 billion minutes per month; Linden Labs, 2009). It is possible that our study’s requirement to use only text chat, which is associated with more formal language in both interpersonal (Chafe and Tannen, 1987) and mass media communication (Tolson, 2006), enhanced the use of more polite language. This association may be stronger among older players who did not use text communication in its current, casual instant messaging form during the formative years of their adolescence, as Prensky (2001) suggests. Although teen text chat is often characterized as informal (Tagliamonte and Denis, 2008), less is known about the instant message language of older adults. Some evidence suggests that for some, instant message chat has features of written rather than spoken communication (Baron, 2010), and that such conversation becomes more formal as speakers age (Ivy and Masterson, 2011).
Interestingly, younger players were more likely to use conventional closing phrases such as ‘bye’. Qualitative analyses of sessions indicate that this was because older players tended to say good-bye once, and then leave the island when the game was over. Younger players tended to stay in the ending area of the quest for longer, asking questions, adjusting their avatars, or chatting. As a result, these (younger) lingerers often said good-bye several times to other players as they left before they themselves left. This might reflect a greater interest among younger players in sociable interaction and just ‘hanging around’ – behavior that may be easier for younger players than for older ones with more pressing work/life demands. Indeed, 40% of those 35 and older were employed full-time and 24% had children under 18 living at home, as compared to 23% with full-time jobs and 19% with children under 18 at home among those under 35.
Another pattern was a tendency for younger players to correct their own misspelled chat more than older players. As age is strongly and significantly correlated with education (α = .392, Sig. = .000), it is possible that players in college at the time of the study were more focused on demonstrating their knowledge than older players. Alternatively, older people may be more focused on getting words out and paying attention to what else is going on, while young people who grew up with online communication might be able to process all of the information more efficiently, and hence devote additional time to checking what they themselves wrote. It should be noted that the exact count of misspelled words was not calculated for each age group. It is possible that younger players simply made more typing mistakes and both groups corrected themselves in equal proportion when errors were made. However, our qualitative analyses revealed that although older players tended to type more slowly, they did not make substantially fewer typing errors than younger players. Further research is needed on the relationship between committing errors and correcting misspelling to determine how such behavior relates to age.
The model also includes avatar gender as a significant predictor of age. In this dataset, avatar gender can almost entirely serve as a proxy for player gender, as only 8% of participants used an avatar with a gender different than what was reported in the survey. This finding suggests that age differences in CMC may be distinct for men and women, as argued by Schler et al. (2006), who found in their study of blogs that topics and word use associated with men were also more associated with older bloggers of both genders. It is also possible that women attempted to conceal their age in this space more than men did, as suggested by research on gender and ageing more generally (Harris, 1994). In our analysis, emoticons were not a significant predictor of age, but it is possible that their association with women more than men is influencing this result (Witmer and Katzman, 1996). Similarly, gender differences in the relationship between age and costume use may have caused this factor to drop out of the model as well. More research is needed to understand how age and gender work together to influence communication behaviors in this space. It is important to note, however, that there were no significant interaction effects of gender and the measures used.
Our findings are counter to some prior research on text chat features associated with age. For example, use of profanity, punctuation, capital letters, and sentence length, all found significant in the literature (Barbieri, 2008; Lin, 2007; Schler et al., 2006; Tam and Martell, 2009), were not significantly associated with age in the models. In fact, among these, only emoticons showed a significant correlation with participant age. Similarly, prior research found older populations used fewer assent/negation words in CMC (Schler et al., 2006); we found no influence of negation as measured by Disagree-Reject statements, and we found that assent, measured by Agree-Accept statements, was instead more common among older participants. These differences may be due to ways that players seek to coordinate with others in this space as Goffman (1959) would suggest, leading to a more consistent set of norms in language use than found in other settings. It may also be the case that the context of our study, one in which players knew they were being observed and recorded, inhibited their inclination to use profanity and led them to communicate more formally than they do elsewhere.
These differences also may be due in part to the specific population and social context within which data were collected. Scholars of communication note that media preferences (Dell and Marinova, 2007) and gratifications sought from media (Blumler, 1985) are related to age differences. Some scholars have noted that online communities, especially games, can serve as cultural unifiers in which participants tend toward similar uses of language and follow specific sets of social norms in behavior and appearance (Pearce, 2009). This would reduce age-related differences in how participants speak, move, and look, especially when it is more important to participants to express their involvement with the group rather than offline identities. As self-categorization theory would suggest, performing age identity might be secondary to establishing other markers of group involvement and coordination. As a result, players may have been selectively concealing their age based on to whom they were speaking, as has been found in other settings (Harris, 1994). In SL, this may influence age-related behaviors in particular due to its older population. In 2008, about 85% of SL’s users were over 25 and those over 44 were its heaviest users (Linden Labs, 2008). It is possible that younger users adjust their language to fit into this slightly older culture.
Conclusions
Identifying how people of different age groups communicate age in three-dimensional online spaces contributes to a more nuanced understanding of generational differences in communication technology use. Avatar appearance and use emerge as important factors in this study, emphasizing that text chat alone does not encompass all the ways age relates to online communication behaviors. In particular, the tendency for younger players to use avatars with a less traditional appearance emphasizes how non-verbal factors contribute to players’ self-expression in this space. In addition, the intersection of age and gender suggested by our analysis indicates that age may be related to communication behaviors in different ways for men and women, particularly in the use of polite and more formal forms of speech. Although our analysis suggests that politeness is more important for older players, additional research is needed to understand to what extent women of all ages are more polite than men in online social worlds, as has been suggested elsewhere (Martey and Stromer-Galley, 2007).
Our qualitative analyses revealed very little explicit representation of age as an identity in the sessions, although we did not track these systematically. Overall, we saw very few participants mention or refer to their age in conversation, perhaps because they sought to establish an ageless persona (Dell and Marinova, 2007). This relative lack of representation of age suggests that the indicators identified in this analysis might emerge differently in spaces where age identity is more salient – for example, an island designed specifically for one age group – or that communication patterns perceived as indicative of age are actively avoided. Further research on participants’ perceptions of their own behavior and age identity would help shed light on these issues.
Like any project, this research has limitations. Firstly, this analysis should be understood within a particular context that may influence how age relates to communication. The language and social interaction examined here are embedded in the overall culture of SL as well as within specific SL subcultures (Pearce, 2009). Secondly, the parameters of the game participants played for this study may have evoked different expressions of age and other identities than would be found elsewhere. The fact that players were strangers to one another may have enhanced some behaviors and reduced others. The use of text chat, rather than voice, may have also influenced how participants communicated during the sessions. Although our qualitative analyses and interviews suggest voice is rarely used with strangers in SL, the text-only requirement of our sessions may have been a more difficult adjustment for some than others, changing or reducing what they said. Future analyses of voice versus text chat are needed to examine the role of modality in the communication of different age groups. It should be noted that our sample was not a random sample of SL. We recruited broadly within the SL resident community, and oversampled men, people of color, and younger participants. However, at an average age of 36.8 our sample remained slightly older than Linden Labs’ last report of 32 as residents’ average age (Linden Labs, 2008). Thus our sample is likely more male, slightly less white, and slightly older than SL residents overall. As such, our ability to understand intergenerational communication is limited.
Some measures used here are unique to our quest in SL, such as clicking vault buttons to solve a puzzle. However, such measures might be comparable to object clicking in other spaces. In other words, can we generalize actions from our SL game to other virtual worlds? Such questions will be part of our ongoing research and can help provide insight about how a range of communication behaviors relate to player age.
Given that appearance turns out to be a rich source of information about age in SL even in avatars that look youthful in face and body, it is also worth considering how aspects of appearance that we did not categorize might provide additional cues to age identity. Given the great variety and nuance players can bring to their avatars, a more fine-grained analysis of hair style and color, facial shape and features, and body poses and walks could further illuminate age identity through avatar characteristics.
Overall, the ways in which individuals use chat, movement, and appearance to perform age identity in virtual worlds is complex and under-researched. Indeed, this is the first comprehensive, quantitative study of avatar communication and use in virtual worlds that focuses on differences in user age. Research in environments where players have less control over character customization, such as World of Warcraft, would be helpful for understanding the dynamics between avatar appearance and communication that convey age identity. In a context where appearance is more constrained by the system, do people express age more clearly or obviously through textual communication practices? Further research in additional settings is needed to more fully understand how individuals perform – or do not perform – age identity in virtual worlds.
Footnotes
Funding
This project was funded by the Air Force Research Labs.