Computer-mediated communication, deindividuation, and group decision-making

This paper discusses social psychological processes in computer-mediated communication and group decision-making in relation to findings that groups communicating via computer produce more polarized decisions than face-to-face groups. A wide range of possible explanations for such differences have been advanced, in which a lack of social cues, disinhibition, ‘deindividuation' and a consequent tendency to antinormative behavior are central themes (Kiesler et al.,1984; Kiesler, 1986; Siegel et al., 1986). In these explanations, both disinhibition and greater equality of participation are thought to facilitate the exchange of extreme persuasive arguments, resulting in polarization. These accounts are briefly reviewed and attention is drawn to various problematic issues.

Video Demonstration of the Experiment

The BBC filmed a simulation of our experiment for a programme called “Computers and Conversation” which was first broadcast on BBC2 in 1995. Here's an extract.

Introduction

Computer support for group decision-making has become a central topic of concern in CSCW research and in this context the effects of computer-mediated communication (CMC) on the process and outcomes of group interaction are of particular interest. Numerous studies have attempted to evaluate the efficiency and quality of CMC in comparison with the face-to-face “standard” (see Hiltz and Turoff, 1978; Rice, 1984 for reviews). Early work in this field was mainly descriptive and atheoretical, but more recently social psychologists in particular have begun to investigate the psychological processes that are involved. Research conducted by the Committee on Social Science Research in Computing at Carnegie-Mellon University has been especially influential in this regard (Kiesler Siegel and McGuire, 1984; Kiesler, 1986; Siegel, Dubrovsky, Kiesler and McGuire, 1986; Sproull, and Kiesler, 1986; McGuire, Kiesler, and Siegel, 1987). Their account of the social psychological processes in CMC as applied to group decision-making was the impetus for the research reported here. We shall therefore begin by outlining their descriptions of these processes and address some of the important questions raised by this work. We then introduce an alternative approach, based on social identity theory and a recent re-conceptualization of deindividuation. The empirical section of the paper reports on a group-polarization experiment designed to address the relevant conceptual issues. Finally, we relate the results of the experiment to other recently reported work and offer a reinterpretation of the earlier studies in terms of our new approach.

The central thesis advanced by the Carnegie-Mellon team is that various technological features of electronic communication trigger psychological states and processes that result in less normative influences on individuals and groups and more deregulated and extreme behavior. These tendencies are manifested in uninhibited behavior such as “flaming” by individuals and in more extreme, or polarized decisions by groups. Kiesler et al. draw on a major account of group polarization, namely persuasive arguments theory, to support their line of reasoning. Persuasive arguments theory posits that polarization results from the exposure to a greater pool of arguments favouring the preferred pole, exchanged during discussion (Vinokur and Burnstein, 1974; Burnstein and Vinokur, 1977). Kiesler et al. (1984) suggest, for example, that if the exchange of information is greater and more evenly spread when communicating via computer, participants will be exposed to more extreme persuasive arguments.

The explanations identified in the Carnegie-Mellon account of CMC can be considered under five headings: (i) lack of social cues, (ii) deindividuation (iii) difficulties of co-ordination and feedback, (iv) depersonalization and/or attentional focus, and (v) conformity to a particular norm or etiquette associated with the computing subculture (Kiesler et al., 1984).

A central and recurring theme is the absence of social influence and contextual cues, and thus the reduced impact of social norms and constraints. For example, Siegel et al. (1986) suggest that “…the absence of social context cues in computer mediation will reduce normative influence, relative to informational influence.” (p. 182). The lack of such cues also results in more equal participation, by masking cues to status and power, which allegedly leads to a greater exchange of extreme arguments. Furthermore, the absence of cues undermines leadership so that there are fewer constraints on uninhibited behavior and hence the exchange of extreme arguments (Kiesler et al. 1984, p. 1125).

A second explanation invokes the concept of “deindividuation” which, classically defined, is the process whereby submergence in a group produces anonymity and a loss of identity, and a consequent weakening of social norms and constraints (e.g., Festinger, Pepitone and Newcomb, 1952; Zimbardo, 1969). Kiesler et al. (1984) observe that “Computer-mediated communication seems to comprise some of the same conditions that are important for deindividuation—anonymity, reduced self-regulation and reduced self-awareness.” ( p. 1126).

Because uninhibited behavior is traditionally linked with deindividuation (e.g., Festinger et al., 1952; Zimbardo, 1969), this may once again explain the communication of more extreme views. Siegel et al. (1986) concur in identifying deindividuation as a possible explanation for uninhibited and antinormative behavior. Kiesler et al. (1984) argue that disinhibition so derived may also contribute to the normative (social comparison) explanation of group-shifts in decision-making by supporting a more extreme group norm. Briefly, social comparison theory argues that polarization shifts result from conformity to a socially desirable but relatively extreme norm which only becomes apparent during group communication (Sanders and Baron, 1977).

A third explanation is that difficulties of communication arising from the lack of feedback from participants may cause uninhibited behavior and polarization. Reflecting on the results of one experiment, Kiesler et al. (1984) suggest, “Perhaps it was frustrating for people to be discussing a problem inefficiently; they might have become angry and, hence, more extreme in decision-making and more uninhibited” (p. 1130). Absence of feedback may also help to explain the equality of participation thought to facilitate the persuasive arguments route to polarization (Kiesler et al. 1984).

The fourth explanation of polarization shifts is in terms of depersonalization or a redirection of attention away from the audience. Kiesler et al. (1984) assert that participants are “…less responsive to immediate textual cues… and less bound by precedents set by societal norms” (p. 1130). Siegel et al. (1986) also maintain that the nature of CMC encourages participants to focus on the content and context of the message, rather than on the social context. They argue that a heightened self-consciousness or self-absorption in the message, may result in reduced sociability and relatively unrestrained and antinormative behavior (p.182). According to these authors, equality of participation (and hence uninhibited behavior) may also depend on this focus of attention.

The fifth class of explanation advanced both by Kiesler et al. (1984) and Siegel et al. (1986) is based on a particular etiquette or norm apparently associated with the computing subculture “…which rejects conventionality and social restrictions” (Siegel et al., 1986, p. 183).

To summarize then, certain themes receive particular weight in the explanations that have been advanced, namely the absence of social cues, the breakdown of social constraint and regulation (“antinormative behavior”), and the emphasis on informational influence as the agent of polarization. However, these explanations at times appear problematic and contradictory when taken both as a whole and in their elements. We will briefly consider the theoretical consistency between the ideas before going on to assess the support for specific explanations and the assumptions underlying them.

To begin at a general level, there is a potential conflict between the characterization of CMC as a potentially fast and efficient means of communication on the one hand, and the citing of the relative inefficiency of the system as a cause of frustration and uninhibited behavior on the other. Clearly the concept of efficiency is a complex and multidimensional one which needs greater elaboration in the CMC context. The theme of “rationality” is similarly double edged in this regard; on the one hand CMC seems to be more rational, more equalized, and focussing on hard information rather than “biasing” peripheral cues (cf. Petty and Cacioppo, 1986). Conversely, CMC is also characterized by a lack of rationality (disinhibition, de-regulation) as invoked in the traditional concept of deindividuation (namely a loss of identity leading to an antisocial state). These two themes—the enhanced exchange and processing of information, and uninhibited and impulsive behavior—sit rather uneasily together.

In terms of the relationship between efficiency and persuasive arguments, it seems at times that the greater efficiency of electronic communication should lead to a fast and free flow of information, feeding persuasive arguments, unhindered by irrelevant social cues. Meanwhile, the communication inefficiencies associated with CMC (caused by lack of feedback and cues) also seem to kindle the generation of persuasive arguments by stimulating frustration and disinhibition, and by equalizing information exchange. This apparent flexibility clearly requires some attention to prevent the underlying concepts from losing their heuristic value.

The traditional conception of deindividuation used in this work also raises certain questions. First, employing the concept of deindividuation (submergence in a group leading to the weakening of social norms) to explain a classic group influence effect (group polarization) is not theoretically straightforward. Secondly, there is a serious problem in relation to the attentional focus argument. Although deindividuation is associated with a lack of self-awareness, Siegel et al. (1986) argue for heightened self-consciousness in CMC. An elevated state of “private self-awareness” in CMC—precisely the opposite of the state predicted by deindividuation theory (Diener, 1980)—is also confirmed in other CMC research (Matheson and Zanna, 1988, 1989). Indeed, the increase in self-attention processes for people isolated at computer terminals could plausibly be argued to be “individuating” rather than deindividuating (Spears, Lea and Lee, 1990). In addition, the idea that people are somehow “submerged in the machine” (Kiesler et al. 1984), or absorbed by the message (Siegel et al., 1986) appears to be contradictory to the earlier observation; it is not clear how people can be both more self-aware and more absorbed by the information at the same time.

However, perhaps the greatest difficulties rest with the central argument concerning the lack of social cues and consequent weakening of social norms and standards in CMC. It may be noted that it is undermined by another suggestion, that uninhibited behavior and thus extreme decisions may be due to a particular computing norm or etiquette. If people in CMC are impervious to social norms in general, it is not entirely clear how this particular norm penetrates though the system. Meanwhile, if it does penetrate successfully, why not other more pervasive norms too?

More important is the problem of defining normative versus “antinormative” behavior. The idea of a weakening of norms is clearly manifested in the concept of disinhibition associated with “flaming” in CMC. However, it could be argued that “antinormative” behavior, where it has a clear and directional form precisely describes a “norm” (albeit an extreme or negative one). In group discussions, if behavior was really socially deregulated as implied by the deindividuation principle, then a haphazard and random distribution of decision responses (or those favoring both extremes) should be expected, rather than the consistent favouring of a single decision extreme. Such deregulated behavior should, almost by definition, result in group responses closer to the mid-point than to one polar extreme of a scale.

Again, many explanations of the group-polarization effect within social psychology are based on the role of social values or norms (Stoner, 1961; Sanders and Baron, 1977; Mackie, 1986; Turner, Hogg, Oakes, Reicher, and Wetherell, 1987). It seems therefore that there is an important distinction to be made between uninhibited behavior or a state of disinhibition on the one hand and polarization in group decision-making on the other. The group-polarization literature has long since shown that decisions in the group can become more polarized in the direction of caution, as a result of group discussion—that is, both more extreme and less risky or “uninhibited” at the same time (e.g., Fraser, 1971).

Empirically, the assumption that CMC is characterized by a weakening of social norms seems to have little direct or independent support. In fact it could be argued that an absence of social cues from other interacting individuals, together with the resulting uncertainty, forces people to resort to default social norms to guide their behavior. Both our own and other research in this field indicates that people may be particularly susceptible to social norms in CMC, particularly when their group identity is salient (Hiltz, Johnson, and Turoff, 1989; Spears et al., 1990).

One report by the Carnegie-Mellon group would seem to undermine the generality of their explanation of group polarization in CMC (McGuire et al., 1987). In this study, as in Siegel et al. (1986), CMC participants actually produced a smaller volume of argumentation (due to the efficiency constraints of the system) compared to a face-to-face condition. This is then used to support a persuasive arguments explanation of polarization in face-to-face groups, whereas the same principle is cited in the previous work as mediating greater polarization in CMC groups. As we have seen, the earlier research invokes equalization of participation and thus argumentation to explain the generation of more extreme arguments, so there appears to be a shift away from this explanation to the importance of the quantity of arguments in McGuire et al. (1987).

Advocates of persuasive arguments theory do not refer to equality of participation so much as to the volume of new arguments exchanged (e.g., Burnstein and Vinokur, 1974). In fact, it could be argued that greater equality of participation would underlie a wider spread of advocated positions (reflected in arguments) leading to diffusion rather than polarization. Conversely, greater “floor-taking” by a particular participant could actually be argued to bias the decision in that person’s preferred direction, leading to imbalance and greater polarization. Finally, while on the subject of persuasive arguments, it should also be noted that the exchange of arguments or information, is not necessary for group polarization to occur (see Wetherell, 1987). In sum, the link between equality of participation and polarization is not clearly established, and appears to be contradicted by the McGuire et al. (1987) study.

Thus far, we have identified some important questions raised by the explanations that have been proposed for CMC decision-making phenomena. In the computer-mediated group decision-making experiment to be reported below, we attempted to explore some of these explanations and their underlying assumptions while advancing our own alternative position. Our theoretical approach is described in detail elsewhere along with a preliminary report of some of our findings (Spears, et al., 1990). In the present context our aim is to add some empirical support for the conceptual issues we have highlighted above. We will therefore present a brief recapitulation here and in addition present new data (derived from the message traffic log, content analysis of the group discussions and post-discussion responses) which are directly relevant to an evaluation of the explanations advanced by the Carnegie-Mellon group.

In essence our view is that, rather than being of decreased importance, the social and normative contexts are of central relevance to decision-making in CMC. In making our case, we draw on a recent re-conceptualization of deindividuation which avoids many of the problems which seem to bedevil not only the work on CMC, but also the more general social psychological research on collective behavior with which deindividuation is classically associated (for critiques see Reicher, 1987; Hogg and Abrams, 1988). Adopting an approach developed from social identity theory (Tajfel and Turner, 1986), Reicher (1984, 1987) argues that deindividuation or anonymity associated with immersion in a group does not weaken social norms, but can act to enhance the salience of the group and thus the relevant norms. He asserts that if people identify with a group, and that group membership is made salient to them, then they will be more likely to be influenced by the group under deindividuating conditions. This is because (visual) anonymity leads to a perceived reduction of intragroup differences, thereby further increasing the salience of the group. However, if one’s group identity is not already salient, deindividuation can enhance one’s sense of individuality, reducing the salience of group norms (Reicher, 1984).

Applying this point to the CMC context, Spears et al. (1990) predicted and found greater norm-directed social influence and polarization in deindividuating conditions (where the participants were in separate rooms and could not see each other) when group salience was high, but least (actually “depolarization”) when participants individuality was made salient. This pattern of findings is not explicable by any of the accounts proposed earlier. The finding that polarized group decision-making within CMC can be associated with an increased salience and influence of group norms has also been confirmed in other contemporaneous research. Hiltz et al. (1989) found greatest polarization under conditions of anonymity (a “pen-name” condition) which is consistent with our own results and explanation. Moreover, Hiltz et al. arrive at a comparable re-conceptualization of deindividuation to account for their findings.

Another related explanation for these results which should be considered turns on the role of self-attention processes (e.g., Carver and Scheier, 1981). As proposed by Siegel et al. (1986) and Matheson and Zanna (1989), social isolation in CMC is likely to be associated with high (private) self-awareness. However, again we argue that the effects of this will be to heighten the salience of particular norms given by the social context, rather than reduce them. If a participant’s group identity is made salient, increased self-attention should enhance the influence of social norms associated with the group, whereas a more individual standard will be influential when the participant’s individuality is already salient. This account draws on the idea in social identity theory that people have different identities, both as unique individuals or “personalities” on the one hand, and as members of social groups on the other. These various identities become salient as a function of the social context. In other words, self-attention and private-self awareness do not necessarily have to be “individualizing” in the sense of reducing adherence to social standards. Instead, we see this account as a complementary explanation based on the “socialized” conceptualization of deindividuation outlined above (for a more detailed discussion of this relation see Spears et al., 1990; Lea and Spears, forthcoming). Matheson and Zanna (1989) also come to similar conclusions relating to self-awareness and the role of social factors.

To summarise then, a unifying theme for these two explanations which distinguishes them from those discussed earlier is the provision made for the role of the social context, and the influence of social norms specified therein. This allows for a relatively parsimonious theoretical structure, which is nevertheless able to account for considerable variation in CMC behavior. We argue that there has been a tendency to confound the form and content of behavior in CMC in earlier explanations of group polarization whereby the form (disinhibition, “flaming”) has been confounded with the content of group norms which underlie polarization. It seems to us that the content of these norms will depend more on the nature of the pre-existing group of CMC participants (i.e., degree of cohesiveness, ingroup identification, group salience, and so on) than on features of the technology, per se, although the technology may introduce certain conditions which crucially mediate its influence (e.g., anonymity, isolation).

In the following report we do not attempt to provide a definitive test of all the explanations outlined above, as this would be impossible within a single study. Also, given the number of explanations considered, the listing of competing hypotheses would not be an efficient use of space. However, the data reported here do speak to many of the explanations proposed, and this is followed up in the discussion.

In our design we manipulate what we see as two critically important social psychological variables independently of each other, namely whether the group or individual identity of participants is made salient, and whether participants are deindividuated by virtue of being isolated and anonymous as opposed to being co-present. This last point is important and requires some further elaboration. Previous relevant research in this area has, to our knowledge typically employed a conventional face-to-face discussion group as the comparison or control condition for the CMC group. However, because CMC by its very nature is not usually “face-to-face”, this has in our view confounded the effects of the technology with an important socio-structural variable (namely whether or not participants are visible or anonymous during the discussion). So while this may have been an interesting and relevant comparison in practical terms, as can be surmised from the foregoing analysis this point is of crucial theoretical importance and remains unresolved. In the present study therefore, our face-to-face condition also communicated via computer, in order to control for the constraints and effects of the communication system.