PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN Postmes, Spears / GENDER DIFFERENCES IN CMC
Behavior Online: Does Anonymous Computer Communication Reduce Gender Inequality? Tom Postmes University of Exeter, University of Amsterdam Russell Spears University of Amsterdam Two studies examined dominance and self-stereotyping in mixed-sex groups who had online discussions. Gender differences in dominance varied as a function of several contextual variables: individuation, the accessibility of gender stereotypes, and the fit between group task and stereotype. Results of the second study indicate that only when group members are depersonalized (anonymous and not individuated) does stereotype activation produce gender-stereotypic behavior. However, the nature of stereotypic behavior is moderated by the fit between group task and stereotypes, such that men dominate when the topic is masculine but not when it is feminine. These findings do not support suggestions that the anonymity offered by online communication would lead to equalization. Instead, results confirm predictions from a social identity model of deindividuation effects, that social effects of anonymity and identifiability in (online) groups depend on contextual factors such as stereotype accessibility and fit.
Gender differences have been documented in groups
as diverse as teams at work or self-help groups, in professional meetings or informal discussions. Differences have been observed in discursive dominance (Coates, 1984; Holmes, 1995; Meyers, Brashers, Winston, & Grob, 1997), productivity (Wood, 1987), nonverbal behavior (Hall, 1984), and socioemotional and task behavior (Anderson & Blanchard, 1982; Eagly, 1987). On the whole, men tend to take on more powerful and dominant roles in groups, whereas women perform more dependent and socioemotional roles. At the same time, however, there is evidence of contextual variability in gender differences, which suggests that these differences are certainly not universal (e.g., Deaux & Lafrance, 1998). Nonetheless, the multitude of contexts
in which gender inequality can be observed has prompted Geis (1993) to observe that “differential treatment is rampant in group tasks and discussions” (p. 18). In this article, we examine different contextual factors that help to explain variations in gender inequality and that also have relevance for the new communications technologies of e-mail and the Internet. Technological innovation has inspired the utopian vision that gender inequality would no longer be significant in the virtual world (Haraway, 1990; Poster, 1990; Sproull & Kiesler, 1991). On the information superhighway, we are told, intergroup divides of gender, and also of race, nationality, or class, become less relevant. In this new world, individuals derive their identity from knowledge and skill, not from social group memberships. This individualistic Internet ideology has been empirically backed by apparent support for what we shall call the equalization hypothesis: the idea that, for reasons elaborated below, traditional intergroup differences found in face-to-face interaction are equalized when people use Computer-Mediated Communication (CMC) systems (Dubrovsky, Kiesler, & Sethna, 1991). Social psychological research and theorizing has played a significant role in explaining why technology Authors’ Note: This research was made possible by a fellowship from the Royal Netherlands Academy of Arts and Sciences to the first author. The authors are grateful to Nyla Branscombe and three anonymous reviewers for their helpful comments. We also thank Djurre Kroon, Clemens Wenneker, and Wendy Zuidam for their help gathering the data. Correspondence concerning this article should be addressed to Tom Postmes, School of Psychology, University of Exeter, Exeter EX4 4QG, UK; e-mail:
[email protected]. PSPB, Vol. 28 No. 7, July 2002 1073-1083 © 2002 by the Society for Personality and Social Psychology, Inc.
1073
1074
PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
could reduce inequality. Researchers have argued that anonymity and isolation are responsible for online equalization (Kiesler, Siegel, & McGuire, 1984). CMC is characterized by a reduction in the amount of individuating information that we normally have in face-to-face interactions. Moreover, online identity is in some cases unknown, and in many other cases uncertain or unverifiable: People may not be who we think they are. Thus, even if people are themselves identified in online interactions, CMC filters out social cues, renders communicators relatively more anonymous, and may even deindividuate them (Dubrovsky et al., 1991). By using the capacities of the medium strategically, CMC may disguise power differences, cover up group memberships, and help portray “false” identities. Therefore, group memberships should no longer exert a significant influence in online interactions. Thus, the equalization hypothesis posits that because anonymity and isolation free individuals from the restraints imposed by group membership, interaction via CMC is more equal. This idea has had a major impact on the design of available systems for online interaction. Because equalization would improve the quality of decision making, for example, anonymity is a central feature of widely applied group-decision support systems (e.g., Nunamaker, Briggs, Mittleman, Vogel, & Balthazard, 1997; but see Postmes & Lea, 2000, for an empirical critique). However, there are several reasons, empirical and theoretical, why this hypothesis merits further investigation. EMPIRICAL INCONSISTENCIES IN STUDIES OF ONLINE EQUALIZATION
The equalization hypothesis posits that equalization should be greater in CMC, or more precisely that equalization should be greater in anonymous online interaction. Certain studies indeed provide strong and unambiguous support for the hypothesis in CMC (e.g., Bhappu, Griffith, & Northcraft, 1997; Dubrovsky et al., 1991) and in anonymous CMC (Kiesler & Sproull, 1992). For example, one empirical observation was that when groups of executives met face-to-face, the men in the groups were five times more likely than the women to make the first decision proposal. When those same groups met via computer, the women made the first proposal as often as the men did. (Sproull & Kiesler, 1991, p. 61)
Findings such as these substantiate the high hopes for online equalization. However, empirical support has been less forthcoming in other studies. Many studies have not found any evidence that online interaction equalizes participation (Adrianson & Hjelmquist, 1991; Berdahl & Craig, 1996;
Hollingshead, 1996; Matheson, 1991; Saunders, Robey, & Vaverek, 1994; Straus, 1996). Some studies even found reverse effects. For example, Weisband (1994; Weisband, Schneider, & Connolly, 1995) describes a series of studies demonstrating that status differences increase as a result of anonymous interaction via a CMC system (see also Scott & Easton, 1996). Because these are the conditions in which cues to status are most effectively filtered out (Dubrovsky et al., 1991), this is a surprising finding. The variability of results is reflected to some degree in sociological and anthropological studies of the Internet, which document that on one hand the “digital divide” has disappeared in terms of equality of access. On the other hand, however, the Internet is not exactly known for its equal accessibility to all groups (Hoffman & Novak, 1998) and it remains gender-segregated and sometimes blatantly female-unfriendly (e.g., sexist, hostile, or patronizing) in ways rarely found in real life (Herman, 1999). Moreover, even when factors such as proficiency and experience are corrected for, gender differences with regard to self-stereotyping, perceived self-efficacy, and attitudes toward computing and online activities remain quite considerable (Whitley, 1997). Finally, although these days women may be going online just as much as men, there is considerable evidence that what they do once online is considerably different. Whereas men spend more time surfing, women use the Internet more for e-mail and communication (Jackson, Ervin, Gardner, & Schmitt, 2001). Thus, the idea that virtual conditions would straightforwardly equalize gender differences through certain characteristics of the medium is not as solid as might appear. For social psychologists this makes the examination of online groups all the more theoretically interesting, because it suggests that even when gender is not immediately visible or unknown, traditional gender differences and blatant stereotyping still persist. This observation suggests that in the absence of social cues based on appearance, psychological factors may play a crucial role in reproducing certain social differences between men and women. ANONYMITY, IDENTIFIABILITY, AND GENDER STEREOTYPIC BEHAVIOR
The theoretical basis for the equalization hypothesis is the idea that anonymity and isolation reduce the restraints that we feel others may impose on us (and that others may actually impose on us) when we are more visible. This implies that individual identifiability is the basis for social influences exerted by group membership. Indeed, identifiability increases the control that powerful others may exert and makes one susceptible to the influence of their social expectations and social stereotypes, as proposed by self-presentation theory (see also
Postmes, Spears / GENDER DIFFERENCES IN CMC Barreto & Ellemers, 2000; Baumeister, 1982). For example, research has shown that women who were going to be meeting a person face-to-face for a job interview were responsive to the interviewer’s presumed gender stereotype by dressing in a stereotype-confirming way (von Baeyer, Sherk, & Zanna, 1981). Thus, identifiability leads to strategic behavior viz-à-viz a powerful (out)group, whereas anonymity and isolation set the individual free from these social restraints. However, stereotypical behavior is not merely externally imposed, but it may to some extent also be due to self-stereotyping. Self-stereotyping may occur when a social identity, those aspects of one’s identity that are derived from a certain group membership, is made contextually salient (e.g., Turner, 1987). This phenomenon also plays a role in the perpetuation of gender differences: In addition to existing social restraints, traditional gender stereotypes and roles are self-imposed to a certain extent (e.g., Eagly, 1987; Geis, 1993). Because the salience of social identity is not dependent on the presence of other group members alone, this means that even when one is anonymous or isolated, group membership may still exert a considerable influence (Turner, 1991). Indeed, some conceptualizations of power do not view stereotypes just as externally imposed constraints but also identity-based restraints (Spears & Lea, 1994). This might explain why equalization does not always happen when people interact anonymously via a CMC system, but it does not explain why in some studies status differences were accentuated under conditions of anonymity. In fact, the Internet does not present the only environment in which social phenomena are augmented under conditions that would seem least conducive to social influence. Anonymity is a key feature in some of the most powerful forms of collective behavior (Reicher, 1987). Social influence in the crowd is paradoxical because “normal” social influence would appear to be hindered by conditions such as high density, noise, or anonymity (Zimbardo, 1969). The paradox may be resolved when one considers that in situations in which a common social identity is salient or available, social identity may become accentuated when individuating information is scarce or absent, compared to when group members are individually identifiable (Reicher, Spears, & Postmes, 1995). The reason for this phenomenon is that when all members of a group are anonymous, any diversity among them is hidden from sight. Hence, the group appears even more homogeneous than it already is. The dress of the Ku Klux Klan illustrates the point well: The cloaks and hoods undermine individual distinctiveness while accentuating a collective identity (Reicher, 1987).
1075
According to the Social Identity model of Deindividuation Effects (SIDE), this phenomenon not only shapes outsiders’ perceptions of the group but also group members’ own perceptions of the group (Lea & Spears, 1991). The ability to individuate each member of a group undermines the perceptual unity of the group (Sassenberg & Postmes, in press; Spears & Lea, 1992; Spears, Lea, & Lee, 1990). Conversely, an absence of individuating information obscures personal features and interpersonal differences, thereby diminishing the relative importance of interpersonal concerns in favor of a focus on the (stereotypical) characteristics of the group as a whole. Provided a common identity is available, the inability to individuate members of a group thus increases the salience of group identity. Following self-categorization theory, this should then lead to enhanced social influence in line with group stereotypes (Turner, 1987). In sum, according to the SIDE model, anonymity should accentuate the effects of accessible stereotypes and the dominant behaviors associated with it (Reicher et al., 1995). For online interactions, this means that if individuation is hindered, one might find more stereotypical behavior rather than less, as suggested by the equalization hypothesis. This idea has received support in other domains, for example, in research showing that anonymous conditions promote normative behavior in offline (Postmes & Spears, 1998) and in online groups (Lea, Spears, & De Groot, 2001; Postmes, Spears, Sakhel, & De Groot, 2001). This shows that anonymity can enhance the salience of group membership. Likewise, research has shown that depersonalization of targets increases the influence of stereotypes (Locksley, Borgida, Brekke, & Hepburn, 1980; Locksley, Hepburn, & Ortiz, 1982; Wilder, 1978). However, no research known to us has yet shown that stereotypical behavior is increased when individuation is hindered. In sum, the SIDE model suggests that when individuation is difficult or impossible, as in anonymous online groups, behavior could be more gender-stereotypical. It is important to stress that this process is by no means automatic: Many of the conditions for the occurrence of these effects of social identity are specified by Self-Categorization Theory. To begin, what social identity is salient in a particular social context depends on various factors such as the cognitive accessibility of particular identities or stereotypes (Oakes, 1987). Thus, gender stereotypes would have to be accessible to find that anonymity would have any effect on this dimension at all. Furthermore, whether a particular identity is activated, and in what way this identity will exert its effects, also depends on the fit of a group’s identity to its current activity (Oakes, 1987; Oakes, Haslam, & Turner, 1994). Research has shown that men perform better on tasks
1076
PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
that are stereotypically consistent with their expertise, such as investing money or playing trucking games, whereas women perform equally well or better in relational and socioemotional tasks (Wood, 1987). Thus, the fit between task and stereotype (i.e., the relevance of a particular task or activity to the stereotype) is likely to determine whether a certain (gender) stereotype is expressed behaviorally. Overview of the Present Research The present article examines the equalization hypothesis, with the purpose of showing that gender differences can occur in online interaction. More particularly, we seek to demonstrate that, consistent with the SIDE model and contrary to the equalization hypothesis, behavioral gender differences may be accentuated when individuation is impossible, whereas gender differences are relatively smaller when people are individuated. We report two studies. In Study 1, the types of differences that may be expected between men and women in online settings are explored. In many studies of equalization, the quantity of contributions is used as the basis for inferring equality (McLeod, 1992). The purpose of this study is to demonstrate that one should also examine the content of group discussions. The second study then examines the hypothesis that gender differences are more marked in anonymous settings, but only when gender stereotypes are accessible, and in the context of tasks that fit the expression of particular gendered behaviors. STUDY 1
Method Participants were 56 volunteer undergraduates recruited in the Faculty of Social Sciences, 28 men and 28 women whose average age was 22. Eight participants at a time were individually placed in isolated cubicles equipped with networked PCs. After isolation, participants were allocated to groups comprising two of each gender, in which their identity was unknown to each other. Groups interacted via a CMC system for 10 mins about the preferred solution to a choice-dilemma (Kogan & Wallach, 1964, dilemma 4). This dilemma was altered so that it was stereotypically consistent with the expertise and interests of men. It was about “Anton, an IT specialist . . . who works in a large IT firm.” Anton had the opportunity to participate in an Internet startup, with high risks and large potential gains. During interactions, participants were identified by their gender and a unique identifier (for example, man2 or woman14) and were otherwise anonymous. Groups communicated via an easy-to-use synchronous computer-conferencing system, using a so-called IRC server. In conjunction with an IRC client program
(in this case MIRC) users can send messages within the group by entering text into a small window at the bottom of their screen. After pressing the return button the message was sent to all other members of their group within 0.4 secs. The message appeared on each screen in a large scrollable window in order of sending. Each message was identified with the sender’s nickname (i.e., “Woman6: . . . ”). The discussions using such computer software resemble, as it were, transcripts of a face-to-face discussion. Dependent variables were the content of communications. Codings were conducted by two independent raters who were blind to the gender of participants. The number of contributions and number of words contributed were counted as measures of dominance of the discussion. Coders also rated each message for the presence of questions. Questions are a very basic measure of gender-stereotypic behavior: Females tend to ask more questions during meetings than men do, it has been argued, because of the cautious and considerate feminine style (e.g., Fishman, 1983; Meyers et al., 1997). The interrater reliability for this simple task was very high (κ = .95). Differences of opinion were resolved. The number of questions was corrected for the total number of statements for each person. An additional dimension also was included that in our view better captures the nature of an unequal relation between men and women in their interaction style. We developed a coding scheme to identify the degree of dependence versus autonomy revealed in discussion statements. A statement was coded as autonomous when the statement was an explicit and unambiguous statement of the opinion of the sender, or when it was forceful, independent, directive, or explicitly reactive. An example of a statement coded as autonomous is as follows: “It’s obvious he should change jobs now.” By contrast, a statement was coded as dependent when it explicitly deferred to another discussant or to the group, when guidance was asked, when ignorance or incompetence was claimed, or when it was otherwise submissive. Examples of such statements are as follows: “I don’t know anything about this” and “I agree.” If none of these two categories could be applied, a statement was coded as neutral. Coders reached acceptable reliability on the autonomy categories (κ = .66). A composite autonomy scale was computed by recoding each statement such that an autonomous statement was given a score of 1, a dependent statement a score of –1, and a neutral statement a score of 0. Judgments were then averaged across raters and across statements and a composite score was thus obtained for each person’s level of autonomy on a scale ranging from –1 (consistently dependent) to 1 (consistently autonomous).
Postmes, Spears / GENDER DIFFERENCES IN CMC Results and Discussion Analyses were conducted at the group level. Gender was a variable that was nested within groups; therefore, we computed the difference between men and women within groups to examine between-subgroup differences (see Postmes et al., 2001, for a precise description of this type of analysis). The scores of women were subtracted from those of men: A higher difference score of, for example, the number of contributions made during the discussion would indicate that men contributed more than did women. On this particular measure of dominance, however, the results revealed no reliable differences between men and women (F < 1). Overall, participants contributed 18.54 statements, with 9.39 words on average. In research of online groups, the quantity contributed to the discussion is the conventional dependent measure for inferring equalization (McLeod, 1992). Thus, our results would appear to suggest that there are minimal gender differences in online interaction even when gender of participants was identified—a finding that would appear to be consistent with the equalization hypothesis. However, unlike face-to-face settings in which discursive dominance can be exercised by “holding the floor” more frequently and longer than others, most CMC systems provide the opportunity to type simultaneously. As a result of this “concurrency” (Valacich, Paranka, George, & Nunamaker, 1993), it is unlikely that male dominance in online settings would take the form of greater quantity because this would mean that men would have to type more or type faster than do women. Thus, quantity may not be the best indicator of dominance. Indeed, the content of communications portrayed a rather different picture, namely, that men did dominate the discussion in terms of the content of the interaction style. Men were more autonomous during the discussion, whereas women were relatively more dependent in their statements (Mdifference = .13, SD = .21), F(1, 13) = 5.19, p < .05. This was corroborated by the analysis of the number of questions asked during the discussion. Men asked 3.2% fewer questions than did women, and this was a reliable difference (Mdifference = –.03, SD = .05), F(1, 13) = 6.10, p < .05. Thus, men did appear to be more dominant when the discussion content was taken into account. We therefore suggest that quantity is not perhaps the best measure for the assessment of equalization, and we also suggest that some of the equalization phenomena that are so widely reported in the literature may be the result of an excessive reliance on this dependent measure. In sum, gender differences can be demonstrated in online discussions in which individuals’ gender is identified but they are otherwise anonymous to each other.
1077
The appropriate dependent measure for this is discussion content rather than quantity. STUDY 2
The purpose of the second study was to examine the effect of individuating cues on the types of gender differences found in Study 1. In addition, the effects of stereotype salience and discussion topic were considered. Participants were asked to participate in two ostensibly unrelated studies. In the first study, participants either unscrambled sentences designed to activate gender stereotypes or unscrambled neutral sentences (Srull & Wyer, 1979). Participants then discussed two topics in a second study. One topic was stereotypically masculine and one feminine. During discussions, participants were either individuated or depersonalized: They either exchanged autobiographical information or not. Special care was taken to ensure that participants’ gender was not identified in any condition: The purpose of the individuation manipulation was to vary the degree to which individuals would be distinctive while keeping the degree to which their gender was identified constant. We predicted that stereotype activation would affect participants’ perception of themselves (i.e., we expected a Gender × Stereotype Activation interaction on self-stereotyping). Moreover, we predicted that we would find most stereotype-consistent behavior when participants were depersonalized and when stereotypes were activated. Under these conditions, men should be more dominant if the topic was masculine, but not if it was feminine. Method PARTICIPANTS AND DESIGN
Participants were 64 male and 64 female undergraduates who were randomly assigned to 32 four-person groups consisting of two men and two women. Participants received 20 Guilders for their cooperation. The design was a 2 (individuation: individuated vs. not individuated) × 2 (stereotype accessibility: accessible vs. not accessible) × 2 (fit: masculine vs. feminine topic) factorial design. Accessibility and individuation were between-subjects factors; fit was a within-subjects factor. In addition, gender (male vs. female) was a nested within-groups factor. PROCEDURE
Participants were individually placed in an isolated cubicle equipped with a networked computer. Participants received instructions via the computer screen that they would participate in two separate studies. In the first study, they were asked to unscramble sentences for 6 mins. After this, they were thanked for their cooperation in the first task and welcomed to the second study, in
1078
PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
which they would discuss two topics in a group consisting, unknown to them, of two men and two women. After answering a series of questions about themselves, participants in the individuated condition were given quasi-biographical feedback describing the members of the group, whereas participants in the depersonalized condition were not given any individuating feedback. Participants then discussed two topics via a similar CMC system to Study 1. They discussed a masculine topic and a feminine topic for 12 mins each. Finally, participants filled in a postexperimental questionnaire and were thoroughly debriefed about the purpose of the study. During the discussion, groups members were identified by the nicknames green, blue, yellow, and red. This ensured that during the discussion, contributions could be attributed to one of the group members, without revealing identity or gender. The identification with a color meant that even in the depersonalized condition, anonymity was not absolute. This is important because complete anonymity confounds the ability to individuate people (the focus of our manipulation) with the ability to attribute contributions to sources. Not knowing the source hinders the comprehension of messages posted and the continuity of the discussion. INDEPENDENT VARIABLES
Stereotype accessibility was manipulated with a scrambled sentence test. For 6 mins, participants were presented with five words in random order, which they dragged across the computer screen to form four-word sentences. Sentences in the stereotype accessible condition contained seven traits and six behaviors that were part of the traditional gender stereotype of each gender group (26 in total, taken from Geis, 1993). Examples are “competitive” and “watches sports” and “emotional” and “plays with children.” Each of the traits and behaviors were featured in gender-stereotyped sentences that described person A (with a majority of female-stereotypic characteristics) and B (male-stereotypic). Seven neutral sentences were included to make conscious detection of the gender theme less likely. Participants in the accessibility control condition were presented with nongender-stereotypic traits and behaviors such as “reliable” and “cycling.” These formed sentences describing three persons—A, B, or C—so as to avoid the likelihood of the dichotomous categorization of gender. Individuation was manipulated by giving autobiographical feedback or not. Participants indicated their hometown, age, how many brothers and sisters they had, their major, how long they had been in university, topics of interest, and their favorite hobbies. Individuated participants were informed that “autobiographies” were shared within the group, so as to make each group mem-
ber personally identifiable and distinctive. Indeed, they ostensibly saw autobiographies of each group member. This was false feedback: All participants received identical biographical information without any cues to gender, such that individuating cues gave information about who other group members were but without conveying their gender. Depersonalized participants were informed that their autobiographical information was used only by the experimenter and that they would remain anonymous to the others in their group. Hence, individuation was more difficult in this condition. The fit of the discussion topic was varied to match the topic with male or female interests and capacities. Each group first discussed a more masculine topic and then a more feminine topic for 12 mins. Topics were selected according to Herring’s (1993) recommendations. The masculine topic was a discussion of the advantages and disadvantages of the city council’s policy to create a carfree zone. The feminine topic dealt with the importance of physical appearance: Participants were asked to discuss whether appearance was overvalued in our culture. DEPENDENT VARIABLES
Questionnaire data. The manipulation check of individuation consisted of three items: “I could form an impression of the people in my group,” “The others in my group were personally identifiable to me,” and “ I could not form a picture of the people I just discussed with” (recoded, Cronbach’s α = .84). Participants indicated agreement on 9-point scales with strongly disagree (1) and strongly agree (9) as endpoints. Self-stereotyping was assessed with the male subscale of the Dutch sex role inventory, an adaptation of the Bem sex role inventory for Dutch samples consisting of six male gender stereotypic traits and behaviors (Rojahn, 1996, with αs usually > .75). Responses were given on 6-point Likert-type scales with never (1) and often (6) as endpoints. Experience with computers was assessed with three questions (α = .63): “Do you have experience with computers?” “Did you like working with a computer in this study?” and “Were you hindered by your typing speed during the discussion?” Responses were given on the 9point scales described above. Finally, the questionnaire asked two open-ended questions: whether participants had any idea about the purpose of the study and a subsequent one informing participants that experimenters had tried to influence the nature of discussion and asking them if they knew in which way. Content analysis. The content of interactions was coded by three independent raters who were blind to the experimental conditions of the groups. As in Study 1, the number of questions was counted (κ = .92). Where differences of opinion existed, the judgment of the major-
Postmes, Spears / GENDER DIFFERENCES IN CMC ity of raters was followed. Coders also rated the degree of autonomy of each statement (κ = .61) and a scale from –1 to 1 was computed as in Study 1.
TABLE 1:
Effects of Individuation and Stereotype Accessibility on Gender Differences During Discussions With a Masculine and Feminine Topic, Study 2 Depersonalized
RESULTS
As in Study 1, we analyzed the differences between men and women within groups, with higher difference scores reflecting that scores of men were higher than those of women. Results were analyzed with 2 (individuation) × 2 (stereotype accessibility) × 2 (fit) analyses of variance, with repeated measures on the last factor. Because analyses are conducted at the group level they are statistically quite conservative, which means that reliable effects are also strong. In the 6 mins for unscrambling sentences, participants solved 18 sentences on average. None of the participants mentioned gender or gender stereotypes in the two open-ended control questions, indicating that participants were unaware of the stereotype accessibility manipulation. The individuation check indicated that this manipulation was successful. Participants in the individuated condition were better able to individuate their fellow group members (M = 3.92, SD = 0.94) than in the depersonalized condition (M = 3.18, SD = 0.95), F(1, 28) = 6.96, p = .01. Self-Stereotyping The manipulation of gender salience influenced selfstereotyping as predicted. There was a main effect of gender on the Dutch sex-role inventory such that selfdescriptions of men were more stereotypically masculine compared with those of women. However, this was qualified by the predicted Gender × Stereotype Accessibility interaction, F(1, 28) = 5.21, p < .05. When stereotypes were accessible, self-descriptions of men were more masculine than those of women (Mdifference = 0.43, SD = 0.52), F(1, 15) = 11.17, p < .01. In the accessibility control condition, gender groups did not differ from each other (Mdifference = 0.03, SD = 0.46), F(1, 15) = 0.07, ns. Men and women did not differ in their self-reported computing proficiency, F(1, 28) = 1.02, ns. Of interest, however, there was a reliable Gender × Stereotype Accessibility interaction, F(1, 28) = 8.22, p < .01, that was consistent with the self-stereotyping effect. When stereotypes were accessible, the self-reported computing experience of men was greater than that of women (Mdifference = 0.76, SD = 1.14), F(1, 15) = 7.09, p = .02. In the accessibility control condition, men and women reported equal levels of computing experience (Mdifference = –0.36, SD = 1.12), F(1, 15) = 1.68, ns. Thus, self-reported computing abilities were different between sexes when stereotypes were salient.
1079
Stereotype Accessibility Condition Masculine topic Autonomy Questions Feminine topic Autonomy Questions
Individuated
Accessible (n = 8)
Control (n = 8)
Accessible (n = 8)
.20 (.10) –.09 (.13)
.04 (.18) .01 (.10)
.04 (.23) .15 (.18) .04 (.16) –.03 (.13)
–.04 (.19) .16 (.20) .11 (.12) –.03 (.13)
.08 (.17) .04 (.11)
Control (n = 8)
.06 (.15) .04 (.13)
NOTE: Standard deviations are in parentheses.
Content Analyses The number of words and messages contributed were analyzed in three-way mixed-model analyses of variance. None of the main effects or interactions were reliable (Fs < 2.5). To minimize error variance between groups due to the quantity of interaction, the other content-coded variables were corrected for the average number of messages contributed per group. The differences between men and women in the degree of autonomy are summarized in Table 1. Only two effects were reliable. Overall, men made about 8.8% more autonomous statements than did women (Mdifference = 0.09, SD = 0.13), F(1, 28) = 13.69, p < .01. However, this main effect was qualified by the predicted three-way interaction, F(1, 28) = 8.52, p < .01. To test the prediction, a planned comparison analysis was carried out within each fit condition (Rosenthal, Rosnow, & Rubin, 2000; Wilkinson & Task Force on Statistical Inference, 1999). The contrast specified that gender differences would deviate in one condition only, namely, in the depersonalized condition in which stereotypes were made accessible (the contrast was 3 –1 –1 –1). In the masculine topic discussion, this planned contrast was reliable, F(1, 28) = 5.38, p < .05. As can be seen in Table 1, the results confirm the prediction that men dominate especially when participants were depersonalized and when stereotypes were accessible. For the feminine topic, the pattern of results was the reverse, F(1, 56) = 3.76, p = .06. In this fit condition, the gender differences were smallest when participants were depersonalized and when stereotypes were accessible. This was substantiated by the analyses of focused comparisons between fit conditions. Only in one condition was there a reliable difference between fit conditions, namely, in the depersonalized condition when stereotypes were activated. In this condition, men made 20.1% more autonomous statements than did women when the topic was masculine, whereas women made 3.8% more
1080
PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
autonomous statements than did men when the topic was feminine, F(1, 28) = 8.13, p < .01. The number of questions asked during interaction showed a similar pattern of results (Table 1). Women asked more questions when the topic was masculine (Mdifference = –0.02, SD = 0.14), whereas men asked more questions when the topic was a feminine one (Mdifference = 0.04, SD = 0.13). This effect was qualified by the predicted three-way interaction, F(1, 28) = 8.43, p < .01. The planned comparison analysis for the masculine topic was a trend, F(1, 28) = 3.18, p = .08, and the reverse planned comparison for the feminine topic was significant, F(1, 28) = 4.71, p < .05. Inspection of the means confirms that the pattern is as predicted. Differences between fit conditions were only reliable in the depersonalized condition when stereotypes were activated: Women asked 9.2% more questions than did men did when the topic was masculine, whereas men asked 11% more questions than did women when the topic was feminine, F(1, 28) = 13.52, p < .01. DISCUSSION
Results confirmed the predictions. The stereotype accessibility manipulation made self-stereotypes more gendered along traditional lines. When gender stereotypes were activated, self-reports of men were more traditionally masculine than those of women compared with the control condition. More surprisingly, the selfreported computing proficiency was affected in the same way by accessibility. This finding supports Whitley’s (1997) suggestion that self-stereotyping may be at the heart of the big gender differences that are generally observed in attitudes and behaviors related to computing. However, whether self-stereotypes were acted out depended on more than just stereotype accessibility. In line with predictions, stereotypical behavior was most accentuated when group members also were depersonalized. This is contrary to what the equalization hypothesis would predict, namely, that the anonymity of online communication fosters equality. This finding is consistent, however, with predictions derived from the SIDE model and demonstrates that the inability to individuate group members not only increases the stereotypicality of impression formation (Locksley et al., 1982) but also of behavior. However, there was one further important moderator. In the one condition in which gender differences were most accentuated, when stereotypes were accessible and group members were depersonalized, the kind of stereotypical behavior that was displayed depended on what the group task was. These results were obtained with the
two dependent measures that Study 1 had identified as being useful indicators of gender differences in online groups: a content-analysis of discursive dominance and a simple count of the number of questions asked. Moderation was evidenced by reliable three-way interactions and substantiated by planned comparisons with the predicted contrasts. Although two of the four contrasts were trends, it should be considered that the analyses were conducted at the group level; hence, statistical tests were extremely conservative (indeed the conventional alpha level for data analyses at the group level is sometimes adjusted). Overall, the pattern is clear and consistent: When stereotypes were accessible and group members were depersonalized, stereotypical differences were largest when the topic was masculine and smallest when the topic was feminine. Men in this condition did not simply dominate any discussions by adopting a more autonomous and overbearing discussion style. They dominated especially when there was fit between a traditional male stereotype and the topic of discussion. When there was fit between a traditional female stereotype and the topic, gender differences actually showed a tendency for reversal. Thus, when women discussed a topic that matched their stereotypical capacities and interests, they actually became slightly more dominant than men. GENERAL DISCUSSION
This research examined the question of whether disguising individual and gender identity during group interaction would lead to more equal participation of men and women and the disappearance of gender differences. Unlike in face-to-face groups, where gender is visible and signaled by various cues, gender can be concealed when groups interact using CMC systems. It has been hypothesized that this concealment of people’s identity would foster gender equalization. However, our findings do not support this equalization hypothesis. Although several factors contributed to produce them, gender differences in dominance were most accentuated when group members were actually unable to individuate each other. Thus, when unable to make clear individual distinctions between group members, and unable to identify them as men or women, gender differences were more accentuated than when such individual distinctions could be made. Of importance, however, depersonalization interacted with other factors, namely, with stereotype accessibility, and with the fit of the topic of discussion with that stereotype. When depersonalization was accompanied by accessibility and fit, behavior was most stereotypical. This finding is consistent with Self-Categorization Theory’s predictions that accessibility and fit are responsible
Postmes, Spears / GENDER DIFFERENCES IN CMC for social identity salience, which in turn is associated with self-stereotypical behavior (Oakes et al., 1994; Turner, 1987). The additional finding that the behavioral expression of self-stereotypes is most likely when group members are not individuated is consistent with the SIDE model. Showing that depersonalization in combination with stereotype accessibility of the gender categorization affects ones own behavior is important because it underlines the point that self-categorization and self-stereotyping, not just the depersonalized perception of others, are central to such “SIDE effects” (see also Lea et al., 2001). The gender differences found in these studies are unusual when compared with the traditional malefemale differences reported in small groups (Geis, 1993). In part, this is due to the nature of CMC. Because group members are able to type concurrently, male dominance is not expressed because one communicator holds the floor or types the most text. Indeed, none of the studies showed any evidence for male dominance with regard to the number of words or statements. Rather, dominance was expressed discursively through a linguistic style we have labeled “autonomy.” This was evidenced by forceful, independent, directive, or explicitly reactive statements that may not have been intended as intimidating or assertive but asserted that this was someone who knew what they were talking about and was not afraid to speak out. That this is a suitable indicator of dominance is corroborated by the opposite pattern of results that was obtained for the number of questions asked. The moderation of gender differences by the fit with the discussion topic showed that it is too simple to assume that gender salience automatically implies male dominance. When the group task was a more masculine one, the traditional pattern of male dominance was obtained, especially when stereotypes were accessible and group members were depersonalized. This pattern was somewhat different when the topic was feminine: In some cases, male dominance was reduced, whereas in other cases, we even found a reversal such that women became somewhat more dominant than men. It is noteworthy that this reversal occurred in the same conditions that evidenced the strongest male dominance (i.e., in the stereotype-accessible condition when participants were depersonalized). This suggests that in cases where gender is an issue, the expression is highly contextdependent: Dissimilar gender differences will resurface depending on the nature of the task that the group is facing (Deaux & Lafrance, 1998; Deaux & Major, 1987; Wood, 1987). However, that women dominated slightly only when they were “on their turf” paradoxically rein-
1081
forces the overall pattern of results by showing that gender dominance depends on the gender-specific domain. In sum, this research indicates that there are no easy remedies for gender differences that occur in groups. The literature reports many inequalities between men and women who meet face to face, ranging from discursive behavior through productivity to the acceptance of leadership. It has been suggested that providing groups with the anonymity and lack of identifiability offered by CMC systems would reduce many of these inequalities. The present results suggest the opposite: A lack of identifiability was actually associated with greater gender differences, although the stereotypic dominance of men was moderated by the gender-stereotypic domain. Within certain domains and tasks, therefore, it would appear that women risk being sidelined when they are online as much as offline. More generally, our findings support the idea that multiple contextual factors codetermine the extent to which gender identity will be activated in a particular context and illustrate how this translates into particular gender differences in behavior. REFERENCES Adrianson, L., & Hjelmquist, E. (1991). Group processes in face-toface and computer mediated communication. Behaviour and Information Technology, 10, 281-296. Anderson, L. R., & Blanchard, P. N. (1982). Sex differences in task and socio-emotional behavior. Basic and Applied Social Psychology, 3, 109-139. Barreto, M., & Ellemers, N. (2000). You can’t always do what you want: Social identity and self-presentational determinants of the choice to work for a low-status group. Personality and Social Psychology Bulletin, 26, 891-906. Baumeister, R. F. (1982). A self-presentational view of social phenomena. Psychological Bulletin, 91, 3-26. Berdahl, J. L., & Craig, K. M. (1996). Equality of participation and influence in groups: The effects of communication medium and sex composition. Computer-Supported Cooperative Work, 4, 179-201. Bhappu, A. D., Griffith, T. L., & Northcraft, G. B. (1997). Media effects and communication bias in diverse groups. Organizational Behavior and Human Decision Processes, 70, 199-205. Coates, J. (1984). Women, men, and language: A sociolinguistic account of sex differences in language. London: Longman. Deaux, K., & Lafrance, M. (1998). Gender. In D. T. Gilbert & S. T. Fiske (Eds.), The handbook of social psychology (4th ed., Vol. 2, pp. 788-827). Boston: McGraw-Hill. Deaux, K., & Major, B. (1987). Putting gender into context: An interactive model of gender-related behavior. Psychological Review, 94, 369-389. Dubrovsky, V. J., Kiesler, S., & Sethna, B. N. (1991). The equalization phenomenon: Status effects in computer-mediated and face-toface decision-making groups. Human Computer Interaction, 6, 119146. Eagly, A. H. (1987). Sex differences in social behavior: A social-role interpretation. Hillsdale, NJ: Lawrence Erlbaum. Fishman, P. M. (1983). Interaction: The work women do. In B. Thorne, C. Kramarae, & N. Henley (Eds.), Language, gender, and society (pp. 89-102). Rowley, MA: Newbury House. Geis, F. L. (1993). Self-fulfilling prophecies: A social psychological view of gender. In A. E. Beall & R. J. Sternberg (Eds.), The psychology of gender (pp. 9-54). New York: Guilford.
1082
PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
Hall, J. A. (1984). Nonverbal sex differences: Communication accuracy and expressive style. Baltimore: Johns Hopkins University Press. Haraway, D. (1990). A manifesto for cyborgs: Science, technology, and socialist feminism in the 1980s. In L. J. Nicholson (Ed.), Feminism/postmodernism (pp. 190-233). London: Routledge. Herman, C. (1999). Women and the Internet. In National Council for Civil Liberties (Ed.), Liberating cyberspace: Civil liberties, human rights, and the Internet. Chicago: Pluto Press. Herring, S. C. (1993). Gender and democracy in computer-mediated communication. Electronic Journal of Communication / Revue Electronique de Communication, 3. Retrieved from http:// www.cios.org/www/ejc/v3n293.htm Hoffman, D. L., & Novak, T. P. (1998). Bridging the racial divide on the Internet. Science, 280, 390-391. Hollingshead, A. B. (1996). Information suppression and status persistence in group decision making: The effects of communication media. Human Communication Research, 23, 193-219. Holmes, J. (1995). Women, men, and politeness. London: Longman. Jackson, L. A., Ervin, K. S., Gardner, P. D., & Schmitt, N. (2001). Gender and the Internet: Women communicating and men searching. Sex Roles, 44, 363-379. Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological aspects of computer-mediated communication. American Psychologist, 39, 1123-1134. Kiesler, S., & Sproull, L. (1992). Group decision making and communication technology. Special issue: Group decision making. Organizational Behavior and Human Decision Processes, 52, 96-123. Kogan, N., & Wallach, M. A. (1964). Risk taking: A study in cognition and personality. New York: Holt, Rinehart & Winston. Lea, M., & Spears, R. (1991). Computer-mediated communication, de-individuation and group decision-making. Special issue: Computer-supported cooperative work and groupware. International Journal of Man Machine Studies, 34, 283-301. Lea, M., Spears, R., & De Groot, D. (2001). Knowing me, knowing you: Anonymity effects on social identity processes within groups. Personality and Social Psychology Bulletin, 27, 526-537. Locksley, A., Borgida, E., Brekke, N., & Hepburn, C. (1980). Sex stereotypes and social judgment. Journal of Personality and Social Psychology, 39, 821-831. Locksley, A., Hepburn, C., & Ortiz, V. (1982). Social stereotypes and judgments of individuals. Journal of Experimental Social Psychology, 18, 23-42. Matheson, K. (1991). Social cues in computer-mediated negotiations: Gender makes a difference. Computers in Human Behavior, 7, 137145. McLeod, P. L. (1992). An assessment of the experimental literature on electronic support of group work: Results of a meta-analysis. Special issue: Computer-supported cooperative work. Human Computer Interaction, 7, 257-280. Meyers, R. A., Brashers, D. E., Winston, L., & Grob, L. (1997). Sex differences and group argument: A theoretical framework and empirical investigation. Communication Studies, 48, 19-41. Nunamaker, J. F., Jr., Briggs, R. O., Mittleman, D. D., Vogel, D. R., & Balthazard, P. A. (1997). Lessons from a dozen years of group support systems research: A discussion of lab and field findings. Journal of Management Information Systems, 13, 163-207. Oakes, P. J. (1987). The salience of social categories. In J. C. Turner, M. A. Hogg, P. J. Oakes, S. D. Reicher, & M. S. Wetherell (Eds.), Rediscovering the social group: A self-categorization theory (pp. 117141). Oxford, UK: Basil Blackwell. Oakes, P. J., Haslam, S. A., & Turner, J. C. (1994). Stereotyping and social reality. Oxford, UK: Blackwell. Poster, M. (1990). The mode of information: Poststructuralism and social context. Chicago: Polity. Postmes, T., & Lea, M. (2000). Social processes and group decision making: Anonymity in group decision support systems. Ergonomics, 43, 1252-1274. Postmes, T., & Spears, R. (1998). Deindividuation and anti-normative behavior: A meta-analysis. Psychological Bulletin, 123, 238-259.
Postmes, T., Spears, R., Sakhel, K., & De Groot, D. (2001). Social influence in computer-mediated communication: The effects of anonymity on group behavior. Personality and Social Psychology Bulletin, 27, 1243-1254. Reicher, S. D. (1987). Crowd behaviour as social action. In J. C. Turner, M. A. Hogg, P. J. Oakes, S. D. Reicher, & M. S. Wetherell (Eds.), Rediscovering the social group: A self-categorization theory (pp. 171-202). Oxford, UK: Basil Blackwell. Reicher, S. D., Spears, R., & Postmes, T. (1995). A social identity model of deindividuation phenomena. In W. Stroebe & M. Hewstone (Eds.), European review of social psychology (Vol. 6, pp. 161-198). Chichester, UK: Wiley. Rojahn, K. (1996). Gender in the context of leadership. Unpublished doctoral thesis, University of Amsterdam. Rosenthal, R., Rosnow, R. L., & Rubin, D. B. (2000). Contrasts and effect sizes in behavioral research: A correlational approach. New York: Cambridge University Press. Sassenberg, K., & Postmes, T. (in press). Cognitive and strategic processes in small groups: Effects of anonymity of the self and anonymity of the group on social influence. British Journal of Social Psychology. Saunders, C. S., Robey, D., & Vaverek, K. A. (1994). The persistence of status differentials in computer conferencing. Human Communication Research, 20, 443-472. Scott, C. R., & Easton, A. C. (1996). Examining equality of influence in group decision support system interaction. Small Group Research, 27, 360-382. Spears, R., & Lea, M. (1992). Social influence and the influence of the “social” in computer-mediated communication. In M. Lea (Ed.), Contexts of computer-mediated communication (pp. 30-65). Hemel Hempstead, UK: Harvester Wheatsheaf. Spears, R., & Lea, M. (1994). Panacea or panopticon: The hidden power in computer-mediated communication. Communication Research, 21, 427-459. Spears, R., Lea, M., & Lee, S. (1990). De-individuation and group polarization in computer-mediated communication. British Journal of Social Psychology, 29, 121-134. Sproull, L., & Kiesler, S. (1991). Connections: New ways of working in the networked organization. Cambridge, MA: MIT Press. Srull, T. K., & Wyer, R. S., Jr. (1979). The role of category accessibility in the interpretation of information about persons: Some determinants and implications. Journal of Personality and Social Psychology, 37, 1660-1672. Straus, S. G. (1996). Getting a clue: The effects of communication media and information distribution on participation and performance in computer-mediated and face-to-face groups. Small Group Research, 27, 115-142. Turner, J. C. (1987). A self-categorization theory. In J. C. Turner, M. A. Hogg, P. J. Oakes, S. D. Reicher, & M. S. Wetherell (Eds.), Rediscovering the social group: A self-categorization theory (pp. 42-67). Oxford, UK: Basil Blackwell. Turner, J. C. (1991). Social influence. Milton Keynes, UK: Open University Press. Valacich, J. S., Paranka, D., George, J. F., & Nunamaker, J. F., Jr. (1993). Communication concurrency and the new media: A new dimension for media richness. Communication Research, 20, 249276. von Baeyer, C. L., Sherk, D. L., & Zanna, M. P. (1981). Impression management in the job interview: When the female applicant meets the male (chauvinist) interviewer. Personality and Social Psychology Bulletin, 7, 45-51. Weisband, S. (1994). Overcoming social awareness in computer-supported groups: Does anonymity really help? Computer-Supported Cooperative Work, 2, 285-297. Weisband, S. P., Schneider, S. K., & Connolly, T. (1995). Computermediated communication and social information: Status salience and status differences. Academy of Management Journal, 38, 11241151.
Postmes, Spears / GENDER DIFFERENCES IN CMC Whitley, B. E. (1997). Gender differences in computer-related attitudes and behavior: A meta-analysis. Computers in Human Behavior, 13, 1-22. Wilder, D. A. (1978). Reduction of intergroup discrimination through individuation of the out-group. Journal of Personality and Social Psychology, 36, 1361-1374. Wilkinson, L., & Task Force on Statistical Inference. (1999). Statistical methods in psychology journals: Guidelines and explanations. American Psychologist, 54, 594-604.
1083
Wood, W. (1987). Meta-analytic review of sex differences in group performance. Psychological Bulletin, 102, 53-71. Zimbardo, P. G. (1969). The human choice: Individuation, reason, and order vs. deindividuation, impulse and chaos. In W. J. Arnold & D. Levine (Eds.), Nebraska symposium on motivation (Vol. 17, pp. 237-307). Lincoln: University of Nebraska Press. Received February 12, 1999 Revision accepted December 28, 2001