SOCIAL SCIENCE COMPUTER REVIEW Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
The Effects of Asynchronous Computer-Mediated Group Interaction on Group Processes JACQUI TAYLOR
Bournemouth University JOHN MACDONALD
Paisley University This article reports a study undertaken to investigate some of the social psychological processes underlying computer-supported group discussion in natural computer-mediated contexts. Based on the concept of deindividuation, it was hypothesized that personal identifiability and group identity would be important factors that affect the perceptions and behavior of members of computer-mediated groups. The degree of personal identifiability and the strength of group identity were manipulated across groups of geographically dispersed computer users who took part in e-mail discussions during a 2week period. The results do not support the association between deindividuation and uninhibited behavior cited in much previous research. Instead, the data provide some support for a social identity perspective of computer-mediated communication, which explains the higher levels uninhibited in identifiable computer-mediated groups. However, predictions based on social identity theory regarding group polarization and group cohesion were not supported. Possible explanations for this are discussed and further research is suggested to resolve these discrepancies.
Keywords: computer-mediated communication, e-mail, deindividuation, social identity
C
omputer-mediated communication (CMC) systems can include audio (e.g., voice mail), text (e.g., e-mail, computer conferencing, and bulletin boards), and visual (e.g., video conferencing) components. Text-based systems remain the most widely available type of CMC system (Holmes, 1995). These systems are often used to link geographically separate individuals so that they can conduct group-based project work at a distance. However, it is unclear how using these systems affects the different facets of group work. In 1990, Huber suggested that much of the theory for small group interaction needed to be reexamined to take account of computer-supported group activities and contexts. He concluded that the use of computer technology would profoundly affect group process, structure, and outcome. Since then, a number of researchers have investigated whether the social psychological literature on group interactions can be used to understand computer-mediated group interaction (Straus, 1997; Sudweeks, McLaughlin, & Rafaeli, 2000). However, systematic research has rarely been conducted within realistic contexts. Research has tended to study inexperienced participants in artificial situations. The limited research that has observed established groups indicates that group history and the expectation of future interaction are important factors in
Social Science Computer Review, Vol. 20 No. 3, Fall 2002 260-274 © 2002 Sage Publications
260
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
261
moderating the effects of technology on group process and performance (Walther, 1997). A body of research that emphasizes the importance of the social context in which CMC is used (e.g., Lea, 1992; Lyytinen, Maaranen, & Knuuttila, 1994) has been a strong influence in directing the design of the study reported in this article. A second criticism of much of the CMC research is that it frequently does not address the process of group work and the social interactions that occur within groups but rather concentrates on the outcome of group work (e.g., Sosik, Avollo, Kahai, & Jung, 1998). In particular, there has been very little research on the way that CMC affects interpersonal perception; for example, the development of group cohesion in computer-supported contexts has not been directly studied.
Effects of CMC GROUP COMMUNICATION AND CMC
Using e-mail for communication between group members has been shown to have a number of effects on the content and style of group discussion (e.g., Taylor & MacDonald, 1994). In particular, two effects have frequently been observed: the expression of more extreme opinions (flaming) and a more frank and open style of discussion (self-disclosure). Steele (1983) defined flaming as “to speak incessantly and/or rabidly on some relatively uninteresting subject or with a patently ridiculous attitude” (p. 65), whereas more recent definitions of flaming focus on the negative emotion involved; for example, Straus (1997) recorded the incidence of “blunt or destructive criticism.” Early CMC research (e.g., Sproull & Kiesler, 1991) consistently found more flaming in computer-mediated group discussions compared to face-to-face discussions. However, other researchers have questioned the extent and nature of flaming. For example, in an extensive review of CMC research, Lea, O’Shea, Fung, and Spears (1992) found that despite many references to flaming, only a few of these were supported with empirical research. Also, they suggested that flaming would only occur in specific CMC contexts; this suggestion has attracted much support recently (Douglas & McGarty, 2001; Walther, 1997). Sproull and Kiesler (1991) proposed that there is reduced audience awareness during computer-mediated discussion and that this results in reduced inhibition in group members and leads to increased levels of self-disclosure and openness. In an early study, they found that when they distributed a self-evaluation questionnaire by either paper or e-mail, those participants replying to the questionnaire by e-mail reported more types of undesirable behavior (Kiesler & Sproull, 1986). Since then, CMC has frequently been associated with more open and frank communication and recent work has shown that CMC can be used effectively to conduct sensitive surveys (Birnbaum, 2001) and counseling (Friedman, Kahn, & Howe, 2000). However, although many CMC studies have discussed the role of self-disclosure in computer-mediated exchanges (e.g., Olivero & Lunt, 2001), this has not been systematically investigated within the context of group processes. In contrast, self-disclosure has received much more attention in the study of small groups in face-to-face contexts. For example, Cathcart and Samovar (1992) suggested that “our impact and influence on a group is partially determined by what and how we disclose personal information to other members of the group” (p. 249). There needs to be investigation of the factors or contexts under which self-disclosure is shown during computer-mediated discussion (e.g., whether self-disclosure is related to the degree of individual anonymity), and second, whether self-disclosure is related to levels of intrapersonal perception (e.g., public and private self-awareness).
262
SOCIAL SCIENCE COMPUTER REVIEW
GROUP DECISION MAKING AND CMC
CMC has been shown to affect both the process of decision making and the final outcome (Sudweeks et al., 2000). During computer-mediated discussions there tends to be more balanced participation, compared to face-to-face discussions. For example, in a classic study, Huff and King (1988) explained the more balanced participation in CMC discussions between graduates and undergraduates in terms of the reduced status information, which enabled the less dominant members (undergraduates) an opportunity to be heard. Similarly, McGuire, Kiesler, and Siegal (1987) found more balanced participation between men and women in CMC discussions and attributed this to the lack of information regarding participants’ gender. Dubrovsky, Kiesler, and Sethna (1991) called this effect the equalization phenomenon; the effects of this phenomenon on group processes are many. Sproull and Kiesler (1991) found that, as a result of the increased democracy, CMC users gave more proposals for action compared to members of face-to-face groups. Also, Miranda (1994) found that there was less likelihood of groupthink in computer-mediated discussions as more alternatives in a decision-making task were put forward by group members, and McLeod (1997) found that there was more salience given to the minority influence. On the negative side, because of the more balanced participation, CMC groups have found it more difficult to arrive at a consensus, taking a longer time and reporting lower satisfaction with the consensus (Sudweeks et al., 2000). CMC also affects the final outcome of decision-making tasks, in particular increasing the likelihood of group polarization. This phenomenon of group polarization is defined as the movement of individual opinions, following group discussion, toward and beyond the initial group mean (Myers & Lamm, 1976). Sproull and Kiesler (1991) showed that the opinions of members in CMC groups shifted even more toward extreme positions following group discussion than did opinions of members in face-to-face groups. Spears and Lea (1994) interpreted this finding in terms of the social identity model of deindividuation effects (SIDE), proposing that under certain conditions computer-mediated group communication could be deindividuating. They hypothesized that when participants are deindividuated, adherence to group norms will be high and this will result in the normative phenomenon of group polarization. In an earlier study by Spears, Lea, and Lee (1990), participants were assigned to one of four conditions in which the CMC context was manipulated, with Identifiability (Low or High) and Group Salience (Individual or Group) as the two key variables. They predicted that the attitudes of participants in deindividuated groups (i.e., those receiving Low Identifiability and Group Salience) would become the most polarized. Participants were asked to discuss a number of controversial topics, and attitudes toward the topics were collected before and after discussion. The results confirmed their predictions. However, there are a number of methodological problems with this research. For example, the research was conducted in the laboratory with only three participants in each group, discussion was limited to 10 minutes, and not all participants were equally familiar with using computers as a method of communication. Therefore, it is not clear how much these results would generalize to group processes in real CMC systems. INTERPERSONAL PERCEPTION AND CMC
The absence of social cues in CMC has been hypothesized to affect interpersonal perception and result in the treatment of others in a depersonalized manner. For example, Sproull and Kiesler (1991) proposed that the anonymity in CMC creates the equivalent of “a tribe of masked and robed individuals.” However, they collected no data to substantiate this hypothe-
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
263
sis. In contrast to this, Matheson (1992) argued that rather than limited social information in CMC leading to reduced regard for status and gender, limited social information can lead to impression formation and behavior based on more extreme stereotypes. In support of this with regard to gender, she found that perceptions of others and differences in communication style between men and women were exaggerated when communicators possessed information regarding the gender category of the other participant. Thomson and Murachver (2001) have recently supported this finding. Perceptions of the group have received surprisingly little attention in computer-supported group research. In contrast, within organizational psychology and small group research, group cohesion (the feeling of unity among group members, of being closely knit) has been a major topic of study. Many of the studies conducted with face-to-face groups have shown group cohesion to be important for a group to be effective and for members to enjoy their experience together (Elias, Johnson, & Fortman, 1989). Without group cohesion, individual members are unlikely to commit themselves to the group, the task, or each other (Mudrack, 1989). Very little research has investigated how group cohesion develops and operates in computer-supported groups. One study by Straus (1997) that attempts to relate group cohesion to group outcome found that computer-supported groups expressed lower group cohesiveness than did face-to-face groups. Straus predicted that there would be a relationship between high levels of perceived group cohesion and more balanced group member participation but a negative correlation between group cohesion and levels of flaming. The former prediction was confirmed but levels of flaming were too low to justify further examination. In the discussion of her results, Straus suggested that group cohesion may be less variable in computer-supported groups because “group members will have less exposure to characteristics of others that they might find attractive or unattractive” (p. 237). Because in face-to-face groups members have greater access to information about other members’ characteristics, in computer-supported groups differences and similarities will not be revealed on which levels of attraction can be based. However, a major problem with this study is that it was conducted in a laboratory environment on very small groups of students; therefore, the results are of limited generalizability to real world contexts.
Explanations for the Effects of CMC on Group Interaction REDUCED SOCIAL CONTEXT CUES THEORY
Sproull and Kiesler (1991) proposed that the limited nonverbal and social context cues in CMC lead to a reduced regard for other users, which in turn lead to flaming and self-disclosure. However, there are a number of problems with this theory. First, studies have shown differences in degree of flaming across different CMC conditions (e.g., first shown by Smolensky, Carmody, & Halcomb, 1990), indicating that it cannot be the media per se that is influencing this aspect of communication behavior. This has stimulated research into examining the different conditions of CMC under which flaming is found (Douglas & McGarty, 2001). Also, there are many methodological weaknesses with the research that supports the theory. For example, often participants were unaccustomed to using e-mail; tasks were very narrow, and most studies were conducted in artificial laboratory environments. Finally, even though this theory emphasizes intrapersonal and interpersonal processes as mediating the effects of CMC, most research has failed to collect user perceptions of self, each other, or the group.
264
SOCIAL SCIENCE COMPUTER REVIEW
DEINDIVIDUATION AND SOCIAL IDENTITY THEORY
Deindividuation is a term used to describe the process of losing one’s distinctiveness and sense of personal identity (Reicher, 1984). Two important conditions known to influence deindividuation are the sense of group identity and degree of personal identifiability to other group members. Deindividuation is found in those groups where the group is emphasized more than the individual and where levels of personal identifiability are low (Turner, Hogg, Oakes, Reicher, & Wetherall, 1987). During the past 10 years, a program of research led by Spears, Lea, and Lee (1990) has used SIDE to guide the design of studies to investigate the effects of CMC (for a recent review of this research, see Postmes, Spears, Lea, & Reicher, 2000). Based on SIDE, it is proposed that an interaction of group immersion and reduced identifiability (i.e., conditions that restrict interpersonal cues) enhances the group context (rather than shift attention away from the context as proposed by Sproull and Kiesler [1991]) and leads to more inhibition in certain contexts of CMC. However, although the empirical work has found support for SIDE, in terms of group polarization, other factors involved in computer-mediated group interaction (e.g., group cohesion and self-disclosure) have not been measured.
Experimental Rationale and Hypotheses The study reported here aims to address some of the criticisms raised regarding previous research. A general problem is that researchers tend to focus their studies on either perception, communication, or the outcome of CMC but rarely set out to observe all three. In particular, previous research has neglected to collect measures of the way users perceive each other as individuals and the way they perceive the group. This study will record and examine the relationship between measures of interpersonal perception, group interaction, and group decision making. Second, it is not clear that the results from the laboratory studies can be generalized to real CMC networks where larger numbers of people would be involved and discussion would take place during longer periods of time. The effects of context, daily routine, and workload that occur in normal everyday use of e-mail need to be taken into account. The study reported here has been conducted in a more realistic context than previous research and specifically examines how group identity and e-mail structure affect aspects of interpersonal perception and group processes. In addition, as Huber (1990) recommended, we will investigate whether findings from social psychological research into group processes are applicable to the domain of electronic group communication: in particular, whether the social identity theory of deindividuation can explain computer-mediated group interaction effects. In summary, this study will manipulate the e-mail conditions under which communication will take place. The two manipulations will be: Identifiability (Low or High) and Salience (Group or Individual). It is predicted that these manipulations will lead to different degrees of deindividuation, which in turn will lead to differential effects on group processes and interpersonal perception. The deindividuated group is predicted to be that which receives Low Identifiability (i.e., very little identifying information about the other individuals in the groups) and Group Salience (i.e., instructions emphasizing that it is the group opinion and conclusion that is important). Five hypotheses will be tested. The first is that, in line with the predictions of Lea et al. (1992), there will be less uninhibited communication (in terms of flaming and self-disclosure) in deindividuated groups (Hypothesis 1). Hypothesis 2 will test whether there is more balanced participation (in terms of frequency and length of messages) in deindividuated groups, as found by Dubrovsky et al. (1991). In line with social
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
265
identity theory and the research of Postmes et al. (2000), it is predicted (Hypothesis 3) that attitudes of participants in deindividuated groups will become more polarized after group discussion, compared to groups receiving Low Identifiability and Individual Salience, whereas there will be no significant movement in the attitudes of those receiving High Identifiability. In line with the social identity model of group cohesion (Hogg & Abrams, 1992), Hypothesis 4 predicts that those participants in the Low Identifiability conditions will perceive more group cohesion, compared to those in the High Identifiability conditions. Hypothesis 5 predicts that those participants in the Group Salience conditions will perceive more group cohesion, compared to those in the Individual Salience conditions.
METHOD
Participants Forty-eight participants (37 men and 11 women) were recruited from a population of Internet users who responded to a request (posted on a number of bulletin boards) to participate in a study of CMC. Participants were randomly assigned to eight groups of six, with the constraint that to ensure equivalent gender balance between conditions each group contained at least one female participant. Participants were geographically dispersed and care was taken to ensure that participants obtained from the same bulletin boards were not placed in the same groups. Checks were made to ensure that no one had previously met or spoken to another member of their group. At the beginning of the study, participants completed a brief biographical questionnaire, which asked for information regarding education, employment, e-mail use, recreational activities and a personal description. The sample consisted mainly of university researchers and lecturers and the majority were employed in the computing and information technology sector. E-mail usage was high, with 70% of participants using e-mail more than once per day and 65% using e-mail for more than 3 hours per day.
Experimental Design A 2 × 2 between subjects factorial design was employed, with Identifiability (Low or High) and Salience (Individual or Group) as the independent variables. Two discussion groups were assigned to each condition. Personal identifiability was manipulated by providing those in the High Identifiability condition with an electronic biographic database containing details of each group member. Participants in the Low Identifiability condition were not provided with the biographic database and were given only the e-mail usernames of other group members. It is proposed that the provision of the biographic database will provide some of the social information about others normally gained in other methods of communication. Deindividuation is normally produced using anonymity or disguise in face-to-face studies; however, as complete anonymity in e-mail is not normally used, it is hypothesized that providing biographical information will be individuating, whereas the absence of this information will be deindividuating. Immersion in the group was manipulated by varying the emphasis (Salience) of instructions given to participants. In the Group Salience condition, group identity was emphasized by describing the objectives of the study as: “To observe communications between group members . . . and to see if there are any differences between various groups in the way that they communicate.” Participants were told that they were participating as members of this
266
SOCIAL SCIENCE COMPUTER REVIEW
group and not as individuals. The remainder of the instructions referred to the participant as a group member and the other participants as “the rest of the group.” In the Individual Salience condition, the briefing focused the aims of the study on individual factors. For example, the objectives of the study were described as: “To observe communications between individuals . . . and to examine personal styles in the way individuals communicate.” Participants were told that they were participating as an individual within this group and the remainder of the instructions referred to the other group members as “the other participants.” This has previously been shown to be sufficient to engender the importance of group membership (Turner, 1983).
Apparatus and Materials Participants used their own computer to send and receive e-mail. Although computers varied across the sample in terms of power, speed, and functionality, it was felt to be more important that participants should use familiar technology rather than try to standardize the systems being used. Participants sent messages to a group address, which were then automatically distributed to other group members and the experimenter. Each group discussed the same topic: “Issues Concerning the Definition of Rape.” This was chosen for topicality; at the time of the study there had been some discussion in the media concerning date rape, including the nature of consent and culpability on the part of the victim. This produced an active discussion and many differences of opinion among the participants. Participants were asked to produce a set of guidelines to help define the act of rape by the end of the discussion period. Changes in attitude toward the discussion topic and measures of interpersonal perception were obtained from two self-report questionnaires. Prior to, during, and at the end of the discussion period participants were asked to complete an “attitudes to rape” questionnaire, which consisted of seven statements about the discussion topic. At the end of the discussion period participants completed a questionnaire that contained measures of interpersonal perception (group cohesion and interpersonal attraction), and measures regarding participants’ experience of taking part in the study (the perceived success at completing the task, the perceived effectiveness of e-mail for discussion, and the perceived openness of email for group discussion). Attitudes and perceptions were measured on 7-point Likert-type scales ranging from –3 (disagreement) to +3 (agreement).
Procedure After participants had been allocated to one of the discussion groups, they were given background information to the study and instructions specific to the condition. At this point (T1) they completed an “attitudes to rape” questionnaire. The discussion period lasted for 2 weeks, during which participants were free to send as many or as few messages as they wished on as many days as they wished. During the discussion, approximately every 3 days (at T2, T3, and T4) and at the end of the discussion period (T5) attitudes were collected using the same seven-item “attitudes to rape” questionnaire. At T5, participants also completed a postdiscussion questionnaire. The transcripts were coded blind by one of the authors (headers and all information that could be used to identify conditions or individual participants were first removed) for instances of two behaviors: flaming and self-disclosure. Two types of flaming were differentiated: Type A consisted of abusive or impolite comments directed toward other participants, whereas Type B consisted of general comments containing uninhibited language. Self-disclosure was defined as any comments that revealed something private about the person. A
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
267
TABLE 1 Frequency of Flaming and Self-Disclosure and Mean Number and Length of Messages Sent per Person During the 2-Week Period
Communication Measure Frequency of flames: Type A Frequency of flames: Type B Frequency of self-disclosures Mean message frequency Standard deviation Mean message length (number of lines) Standard deviation
Low Identifiability, Individual Salience
Low Identifiability, Group Salience
High High Identifiability, Identifiability, Individual Group Salience Salience Total
1
4
13
8
26
4
2
13
11
30
4
1
17
31
53
3.2
6.9
9.7
8.4
1.2 51
4.9 71
5.3 160
4.0 206
29.2
46.2
103.3
131.2
7.0
122
Significant Effects χ2 = 12.45, p < .01 2 χ = 15.32, p < .01 2 χ = 42.62, p < .01 Identifiability p < .01 Identifiability p < .01
number of quantitative measures were extracted from the transcripts, including the number and length of messages sent by each individual.
RESULTS The analysis of the data will be considered for each type of measure: group communication, group decision making, and interpersonal perception. This will be followed by a consideration of the relationships between these measures.
Group Communication There were relatively few instances of self-disclosure and flaming per condition; therefore, Table 1 shows the total number of self-disclosure and flaming comments per condition. It can be seen that participants receiving a low level of Identifiability produced very few incidences of either kind of flaming—with only 5 out of a total of 26 Type A flames and 6 out of a total of 30 Type B flames. The observed frequencies were compared against expected values of equal distribution. The observed values were not equally distributed for either Type A flames, χ2(1) = 12.45, p < .0001; or Type B flames, χ2(1) = 15.32, p < .001. Therefore, flaming did not occur evenly across conditions, and it can be seen that Identifiability was the significant factor in both types of flaming. Contrary to previous research, it was the individuated, High Identifiability conditions that produced the most flaming, not the impersonal, Low Identifiability conditions. The Salience manipulation had little effect, with those in the Group and Individual conditions producing similar numbers of both types of flaming. The incidence of self-disclosure illustrated in Table 1 indicates that the more that is known about other members of the group (i.e., the High Identifiability groups), the more a person will disclose about themselves. Again, testing against an expected equal distribution, frequency of self-disclosure across conditions was significantly and strongly affected by the experimental manipulations, χ2(1) = 42.62, p < .001.
268
SOCIAL SCIENCE COMPUTER REVIEW
Two-way ANOVA tests were conducted for both measures of communication activity. Identifiability was shown to have a strong and statistically significant effect on messagesending activity: Participants in High Identifiability groups sent significantly more messages, F(1, 44) = 10.5, p < .01, and significantly longer messages, F(1, 44) = 21.93, p < .01, than those in Low Identifiability groups. The Salience manipulation had no significant effect on either the number or length of messages sent. When the standard deviations were examined, for both communication measures there was more variability in the High Identifiability groups, whereas standard deviations were lower for the Low Identifiability groups, indicating a more balanced distribution of messages of a similar length sent by each group member. To summarize, it appears that Identifiability is a significant factor affecting uninhibited communication, where the more that is known about other members of the group then the more that people are prepared to flame and self-disclose. Indeed, there were very few flames and self-disclosive comments from participants in groups receiving only limited interpersonal information. The results provide support for Hypothesis 1; in particular, the least number of Type B flames were produced in deindividuated groups. Also, limiting the amount of identifying information available produces a more balanced discussion, providing some support for Hypothesis 2, although there was less communication occurring in these groups.
Group Decision Making The data were analyzed to test the hypothesis (Hypothesis 3) that the attitudes of individuals in the deindividuated groups (Low Identifiability and Group Salience) would move further toward the prediscussion group mean, following group discussion, compared to the attitudes of those in groups receiving Low Identifiability and Individual Salience. To determine whether group polarization occurred during discussion as well as at the end of the discussion period (T5), measurements of attitude were collected on three occasions (T2, T3, and T4) during the discussion period for each of seven attitude items. In total, 28 measurements were compared with prediscussion attitude levels (at T1). The results revealed only six instances (Item 1 at T3; Item 2 at T4; Item 4 at T2, T3, and T4; and Item 7 at T2) where changes in attitude reflected the interaction predicted by Hypothesis 3, and none of these was statistically significant. Also, it was hypothesized that there would be no significant movement in attitude for the two High Identifiability conditions, but contrary to this some large (but not quite significant) differences were shown. Two-way ANOVA tests were conducted for the responses to postdiscussion questionnaire items; the significant findings and the mean responses are presented in Table 2. When participants were asked how successful they were in completing the task (to produce a set of guidelines to help define the act of rape), it can be seen that those receiving Group Salience and High Identifiability perceived the level of success to be high, whereas the other three conditions indicated negative mean responses. Although this interaction was not statistically significant, the Salience manipulation was significant, F(1, 44) = 6.68, p < .05. It can be seen from Table 2 that those receiving High Identifiability reported feeling significantly more open than those receiving limited Identifying information, F(1, 44) = 7.25, p < .01, supporting the data regarding self-disclosure presented in Table 1. Participants were also asked how effective e-mail had been for discussion and although there were no significant effects for this item, the responses of those participants receiving High Identifiability were more positive, with deindividuated groups indicating the lowest level. In summary, members of Low Identifiability groups indicated more negative perceptions of the group process, compared to the High Identifiability groups. Analysis of the data regarding the group outcome revealed no consistent changes in attitude during or following
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
269
TABLE 2 Mean Responses to Postdiscussion Questionnaire Items
Questionnaire Item Task success Openness Effective Group cohesion Attraction
Low Identifiability, Individual Salience
Low Identifiability, Group Salience
–1.33 –0.85 0.15 –1.31 0.19
–0.30 –0.36 –0.40 –0.73 0.36
High High Identifiability, Identifiability, Individual Group Salience Salience –0.58 0.42 0.33 0.42 0.42
0.91 0.45 0.81 0.36 1.20
Significant Effects Salience p < .05 Identifiability p < .01 ns Identifiability p < .01 ns
discussion and the strong interaction effects shown in research by Spears et al. (1990) were not found; therefore, Hypothesis 3 cannot be supported.
Interpersonal Perception The Identifiability manipulation strongly and significantly affected perceptions of group cohesion, F(1, 44) = 10.09, p < .01. However, contrary to Hypothesis 4, there was more perceived group cohesion in the High Identifiability groups than in the Low Identifiability groups. However, it is not clear whether Identifiability directly affected these perceptions or whether they are affected by the increased communication activity or increased uninhibited communication that occurred in these groups. The results appear to support the findings from face-to-face research that show a link between incidence of self-disclosure and high levels of perceived group cohesion. The Group Salience manipulation was not significant; therefore, Hypothesis 5 cannot be supported. Table 2 shows that perceptions of other group members were generally favorable: Participants receiving Group Salience and High Identifiability liked other group members the most, although this interaction was not statistically significant. In summary, the results showed that Identifiability rather than Group Salience was the important factor affecting interpersonal perception and that High rather than Low Identifiability was associated with the higher levels of cohesion. This is contrary to the social identity model of group cohesion, which predicts that group cohesion should be higher in the Group Salience groups (Hypothesis 5) and higher in the Low Identifiability groups (Hypothesis 4).
Relationships Between Communication and Interpersonal Perception Correlation tests were computed between the communication activity and the measures of interpersonal perception. There were two significant effects: Group cohesion and interpersonal attraction were positively and significantly correlated with the mean length of messages sent (respectively, r = 0.32, df = 44, p < .01; and r = 0.29, df = 44, p < .05). However, it is not clear from these tests whether high levels of cohesion within the group encouraged communication activity or whether the increased communication activity in these groups produced a higher level of perceived cohesion.
270
SOCIAL SCIENCE COMPUTER REVIEW
Previous research (Straus, 1997) has proposed that low levels of group cohesion are related to high levels of flaming. It can be seen from a comparison of the data in Tables 1 and 2 that this study finds the contrary relationship; that is, in those groups where flaming was the highest, levels of group cohesion were also high. Also, Straus (1997) proposed that high levels of group cohesion would be observed in those groups where participation was more balanced. When the standard deviations of message length are compared to levels of cohesion, the direct opposite is shown; that is, the lowest levels of attraction and the lowest standard deviation are shown in the Low Identifiability and Individual Salience condition, whereas the highest levels of attraction and the highest standard deviation are shown in the High Identifiability and Group Salience condition.
DISCUSSION This study had two main purposes. The first was to examine group interaction in realistic but controlled e-mail discussion groups where the contexts had been manipulated to provide more or less emphasis on a personal individual identity or an impersonal social identity. Previous CMC research has either compared electronic communication with face-to-face communication (e.g., Sproull & Kiesler, 1991) and confounded structural differences between the two (synchronicity, speed of communicating, etc.), or compared electronic groups under different conditions but generally in laboratory-based experiments (e.g., Lea & Spears, 1991) that are unlikely to generalize to everyday e-mail use. This study overcomes these problems by combining the power of experimental control and manipulation with real e-mail discussion groups. In this way, it attempted to provide a test of social identity theory in a more realistic setting than has so far been done to offer some generalization to the social identity research findings. The second aim was to consider the way that perceptions of group members and members’perceptions of the group as a whole relate to group interaction and in particular to investigate whether social identity theory could be used to predict the effect of the manipulations of identifiability and group identity on group cohesion. Much previous CMC research has ignored group cohesion, despite its apparent importance in face-to-face small group interaction. Also, despite the implication of changes in interpersonal perception as a mediating influence in many explanations of CMC effects, measurements of interpersonal perception have not generally been collected. The first conclusion to be drawn from the data presented here is that there is no support for the view that impersonal or deindividuated contexts encourage less inhibited communication either in quantity or quality. In fact, the evidence here directly contradicts that found by Sproull and Kiesler (1991). It was the conditions in which participants could be personally identified and in which they had provided information about themselves that produced higher levels of flaming and self-disclosure. Hence, the rather technologically deterministic arguments of Sproull and Kiesler (1991) regarding the effects of reduced social context cues are unfounded. More complex processes and factors are involved in influencing communication in electronic groups, particularly as regards the extent to which people obey the social norms that govern communication and interaction in face-to-face contexts. Whatever leads to antinormative behavior in CMC systems, anonymity does not appear to be the critical factor. If one takes the view that flaming and self-disclosure are more socioemotional rather than task-based discussion, then the provision of personal information has encouraged and facilitated the groups to interact at that level. Whether this can be seen as a positive release of socioemotional interaction or a negative release is difficult to consider here. Within the realistic context of this study, the prediction of more balanced participation in groups receiving limited identifying information (Dubrovsky et al., 1991) was confirmed,
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
271
although there was less communication occurring in these groups in terms of message length and frequency. The implications of this are important as regards the conclusions that can be drawn about the consequences of the use of CMC systems. It would appear that in circumstances where a high level of interactive discussion is desirable then this can be achieved by providing a means for the participants to obtain information about the other members of the group. However, if equality of participation is the critical goal, then the anonymity of group members should be preserved. It is difficult to give guidelines on how to realize a high level of discussion and equal participation. Further work needs to be conducted in this area. The results regarding group polarization are more difficult to explain. It appears that in this more realistic context, the normally highly consistent phenomenon of group polarization is not shown. This may be due to the longer time frame during which this study was conducted; most group polarization studies take place during minutes or hours rather than days. It may be that group polarization only occurs in the early stages of group discussion. The results provide some support for this argument; when the predicted interaction was shown, it occurred in the early or middle stages of discussion and was never shown in the final postdiscussion measurement. Further research needs to look at changes in attitudes over time and particularly those changes that may occur at the beginning of discussions, rather than just collecting measurements of attitude before and after discussion as is usually the case in group polarization research. McGrath and Hollingshead (1993) arrived at a similar conclusion in their work, which emphasized the importance of investigating the dynamics within groups and temporal issues in computer-supported groups. In a discussion of the differences between laboratory-based and field research, they discussed the predominance of CMC studies, which focus on the very early stages of group development. It is during the early stages that group members try to establish and understand the norms of the group, define and defend their position, and obtain a basis of influence over the decision process. CMC may have significant effects on groups at the early stages of development because it permits members to focus more rapidly and intensely on the task itself. Further discussion of the influence of temporal issues in CMC is considered elsewhere (Taylor & MacDonald, 2002). In summary, the results here suggest that group polarization in CMC may not be as widespread as has previously been suggested. Although the findings contribute to the recent body of research investigating the contexts of CMC, only two specific aspects of the context were investigated here. An aspect that has not been considered is the nature of the tasks that groups are asked to carry out. The task used here was intended to be realistic (to discuss the topic of rape and produce a set of guidelines outlining how rape should be defined at the end of the discussion period). There was no correct or incorrect answer, whereas previous group polarization research has used more structured tasks where a decision has been required within a short period of time or where a solution to the task is more definitive. It may well be the case that the findings of this study are partly a result of the rather open-ended task set to the group (i.e., to discuss the wider issues) and during a longer time period and may not apply more universally. How people approach the use of a communication medium will undoubtedly be affected by their conception of the nature of the activity required of them and their perception of how the medium will allow them to accomplish that requirement. Future research needs to examine the effects when different tasks are used. A number of researchers have started to address these issues; however, so far they have done so within laboratory contexts (e.g., Sosik et al., 1998). The second aspect of this study is the consideration of interpersonal perception. The importance of believing one is part of a group has been shown to significantly affect behavior (e.g., Tajfel, 1981). The study produced an unexpected set of significant group cohesion findings, contrary to the predictions based on the social identity model of group cohesion.
272
SOCIAL SCIENCE COMPUTER REVIEW
Rather than the manipulation of group immersion affecting group perception, it was the manipulation of personal identifiability, which significantly affected perceptions of group cohesion. Tajfel (1981) suggested that the only precondition for group cohesion to occur is the explicit categorization of individuals as group members; clearly, this was not the case in this study. Also contrary to social identity theory, the results showed that more group cohesion was perceived in groups receiving individuating information, whereas Hogg and Abrams (1992) proposed that there would be more cohesion in groups receiving limited individuating information. It may be that group cohesion operates differently in computermediated groups and that the lack of nonverbal cues qualitatively affects interpersonal perception. This would also support the comments of Straus (1997). In face-to-face contexts, group cohesion has been positively related to productivity. In this study also, perceptions of task success were higher in those groups perceiving high group cohesion (i.e., High Identifiability groups). However, an alternative explanation may be that members in Low Identifiability groups did not perceive the discussions as successful in achieving the task because the quality, as well as the quantity, of communication was inferior within these groups. The relationship between self-disclosure and group cohesion shown in previous face-to-face group research (e.g., Elias et al., 1989) was also shown in this study, although it is not clear whether the level of uninhibited communication (flaming and selfdisclosure) or quantity of communication encouraged group cohesion or whether the group cohesion encouraged increased communication. Further research needs to examine the concept of group cohesion in more detail, and specifically the relationship between selfdisclosure and group cohesion. This is particularly important considering the group research that emphasizes the important role that group cohesion plays in the performance and productivity of groups at work. Such results may have important implications for the implementation of groupware and CMC systems.
CONCLUSION The research has contributed to understanding how the contexts of CMC affect group interaction; in particular, the levels of social identity of group members were manipulated and investigated. Further research needs to focus on other aspects of the context, for example, task-type and temporal factors. Such work will help determine which features of CMC technology produce various outcomes and may also help clarify inconsistent results across studies. In the discussion of the findings from this study, it was suggested that perhaps some group processes (e.g., polarization and cohesion) do not operate in the same predictable ways in realistic computer-supported groups as they do in face-to-face groups. Lyytinen et al. (1994) came to a similar conclusion when they proposed that CMC research so far has been technology-focused, and they suggested that new, computer-mediated group processes need to be investigated. In the words of Lyytinen et al. (1994) “groups are not always the same” (p. 261).
REFERENCES Birnbaum, M. H. (2001). Introduction to behavioural research on the Internet. Upper Saddle River, NJ: Prentice Hall. Cathcart, R. S., & Samovar, L. A. (1992). Small group communication (6th ed.). Dubuque, IA: WCB. Douglas, K. M., & McGarty, C. (2001). Identifiability and self-presentation: CMC and intergroup interaction. British Journal of Social Psychology, 40, 399-416. Dubrovsky, V., Kiesler, S., & Sethna, B. N. (1991). The equalisation phenomenon. Human-Computer Interaction, 6, 119-146.
Taylor, MacDonald / ASYNCHRONOUS COMPUTER-MEDIATED GROUP INTERACTION
273
Elias, F. G., Johnson, M. E., & Fortman, J. B. (1989). Task-focused self-disclosure: Effects on group cohesiveness, commitment to task and productivity. Small Group Behaviour, 20, 87-96. Friedman, E. J., Kahn, P. H., & Howe, D. C. (2000). Trust online. Communications of the ACM, 43(12), 34-40. Hogg, M. A., & Abrams, D. (1992). Social identifications. London: Routledge. Holmes, M. E. (1995). Don’t blink or you’ll miss it: Issues in e-mail research. In B. R. Burleson (Ed.), Communication yearbook 18 (pp. 454-463). London: Sage. Huber, G. P. (1990). A theory of the effects of advanced information technology on organizational design, intelligence, and decision-making. Academy of Management Review, 15, 47-71. Huff, C. W., & King, R. (1988). An experiment in electronic collaboration. In J. D. Goodchilds (Chair), Interacting by computer: Effects on small group style and structure. Symposium conducted at the meeting of the American Psychological Association, Atlanta, GA. Kiesler, S., & Sproull, L. (1986). Response effects in the electronic survey. Public Opinion Quarterly, 50, 402-413. Lea, M. (1992). Contexts of computer-mediated communication. London: Harvester-Wheatsheaf. Lea, M., O’Shea, T., Fung, P., & Spears, R. (1992). “Flaming” in computer-mediated communication. In M. Lea (Ed.), Contexts of computer-mediated communication (pp. 89-112). London: Harvester-Wheatsheaf. Lea, M., & Spears, R. (1991). Computer-mediated communication, de-individuation and group decision-making. International Journal of Man-Machine Studies, 34, 283-301. Lyytinen, K., Maaranen, P., & Knuuttila, J. (1994). Groups are not always the same: An analysis of group behaviours in electronic meeting systems. CSCW, 2, 261-284. Matheson, K. (1992). Women and computer technology. In M. Lea (Ed.), Contexts of computer-mediated communication (pp. 66-87). London: Harvester-Wheatsheaf. McGrath, J. E., & Hollingshead, A. B. (1993). Putting the “group” back in group support systems: Some theoretical issues about dynamic processes in groups with technological enhancements. In L. M. Jessup & J. S. Valacich (Eds.), Group support systems (pp. 78-96). New York: Macmillan. McGuire, T. W., Kiesler, S., & Siegal, J. (1987). Group and computer-mediated discussion effects in risk decision making. Journal of Personality and Social Psychology, 52, 917-930. McLeod, P. (1997). The eyes have it: Minority influence in face-to-face and computer-mediated group discussion. Journal of Applied Psychology, 82, 706-718. Miranda, S. M. (1994). Avoidance of groupthink. Small Group Research, 25, 105-136. Mudrack, P. E. (1989). Defining group cohesiveness. Small Group Behaviour, 20, 37-49. Myers, D. G., & Lamm, H. (1976). The group polarization phenomenon. Psychological Bulletin, 83, 602-627. Olivero, N., & Lunt, P. (2001, November 7-9). Self-disclosure in e-commerce exchanges: Relationships among trust, reward and awareness. Paper presented at the Psychology and the Internet Conference, Farnborough, UK. Postmes, T., Spears, R., Lea, M., & Reicher, S. (Eds.). (2000). SIDE issues centre stage. Amsterdam: Royal Netherlands Academy of Arts and Sciences. Reicher, S. D. (1984). Social influence in the crowd: Attitudinal and behavioural effects of de-individuation in conditions of high and low group salience. British Journal of Social Psychology, 23, 341-350. Smolensky, M. W., Carmody, M. A., & Halcomb, C. (1990). The influence of task type, group structure and extraversion on uninhibited speech in CMC. Computers in Human Behaviour, 6, 261-272. Sosik, J. J., Avollo, B. J., Kahai, S. S., & Jung, D. I. (1998). Computer-supported work group potency and effectiveness: The role of transformational leadership, anonymity and task interdependence. Computers in Human Behavior, 14, 491-511. Spears, R., & Lea, M. (1994). Panacea or panopticon? Communication Research, 21, 427-459. Spears, R., Lea, M., & Lee, S. (1990). De-individuation and group polarisation 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. London: MIT Press. Steele, G. L. (1983). The hacker’s dictionary. New York: Harper & Row. Straus, S. G. (1997). Technology, group process and group outcomes: Testing the connections in computer-mediated and face-to-face groups. Human-Computer Interaction, 12, 227-266. Sudweeks, F., McLaughlin, M., & Rafaeli, S. (Eds.). (2000). Network and netplay: Virtual groups on the Internet. Cambridge, MA: MIT Press. Tajfel, H. (1981). Human groups and social categories. Cambridge, UK: Cambridge University Press. Taylor, J., & MacDonald, J. (1994). The effects of electronic interactions on group and individual communication processes. Journal of Intelligent Systems, 4, 113-132. Taylor, J., & MacDonald, J. (2002). A discussion of the changes that occur in computer-mediated groups over time. Manuscript in preparation.
274
SOCIAL SCIENCE COMPUTER REVIEW
Thomson, R., & Murachver, T. (2001). Predicting gender from electronic discourse. British Journal of Social Psychology, 40, 193-208. Turner, J. C. (1983). Some comments on . . . “the measurement of social orientations in the minimal group paradigm.” European Journal of Social Psychology, 13, 351-367. Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherall, M. S. (1987). Rediscovering the social group: A self-categorisation theory. Oxford, UK: Blackwell. Walther, J. B. (1997). Group and interpersonal effects in international computer-mediated collaboration. Human Communication Research, 23, 342-369.
Jacqui Taylor is a senior lecturer and course tutor of the BSc (Hons) applied psychology and computing degree at Bournemouth University. She has been researching psychological aspects of technology since 1989 and completed her Ph.D., “A Social Psychological Analysis of Computer-Mediated Interaction,” in 1995. Her research interests are CMC, CSCW, social identity theory, psychological research using the Internet, and e-learning. She may be reached via e-mail at
[email protected]. John MacDonald is a professor of psychology and head of the division of psychology at Paisley University. Having trained as a cognitive psychologist in the early 1970s, he developed an interest in human-computer interaction. His main research interests are in the use of new technologies as communication and interaction devices, in particular, the cognitive demands they place on people and the impact that has on communication processes. He may be reached via e-mail at
[email protected].
