In this chapter we explore the social psychological dimensions of computer-mediated communication (CMC). We shall not be directly concerned here with the processes by which people choose CMC in preference to other communications media (see e.g., Fulk & Boyd, 1991, for a recent review), although it will become obvious that our argument has implications for media choice models. Rather we are interested here in the social psychological factors operating once “inside” CMC and particularly the role played by social influence processes in relation to outcomes produced in this medium, such as group attitudes and decisions. Although CMC may also produce differences in the organization, style and content of communication (for example), it is these outcomes that in a real sense will “make a difference” in the many applied settings in which CMC is used.
Download full text
Given our interest in the processes of social influence operating within CMC, and the use of CMC to support collaborative group work, we shall focus particularly on group processes and decisions. A central tool in this regard, and by no means an uncontroversial yardstick for the measurement of social influence effects, is the “group polarization paradigm”, of which more later. Finally we do not have room to consider the process and outcomes associated with other forms of cooperative activity, such as group problem-solving, and the technologies that have been designed to support these activities (see Kraemer & Pinsonneault, 1990, for a recent review). Such issues take a backseat to a more basic theoretical analysis of CMC processes, although our arguments have a range of implications and applications. Our main aim, then, is to examine the social psychological processes underlying the products of group communication and decision-making using electronic mail; and computer conferencing, as compared to face-to-face communication.
In fact a cursory review of the literature might convince the reader quickly to assume that the social dimension is of very little relevance to CMC. Broadly speaking, the prevailing analyses—which we collectively term the “social cues perspective”—indicate that communicating via computers is bereft of social cues. This is quickly followed by the inference that CMC is devoid of social or normative context and (thus) forms a particularly inefficient medium for social influence. If all this were true there would presumably be little need for any detailed social psychological analysis of CMC at all. We consider this received wisdom to be profoundly flawed. Our chapter aims to challenge this orthodox view and to restore the social dimension to an analysis of CMC behaviour and effects. Indeed, we shall argue that paradoxically CMC may represent a more intrinsically “social” medium of communication than the apparently “richer” context of face-to-face interaction, and one that gives fuller reign to fundamentally social psychological factors.
We begin by reviewing research in the “social cues” tradition which forms and informs the orthodox view. Much of this work has been concerned with earlier telecommunications technologies, although there have been no shortage of attempts to apply this theory to the CMC context. We shall focus here on three influential approaches: the Social Presence Model (Short, Williams & Christie, 1976), the Cuelessness Model (Rutter, 1987) and the Reduced Social Cues approach (Kiesler, Siegel & McGuire, 1984). These models are discussed in increasing detail commensurate with their recency, sophistication and relevance to the CMC literature. Although not the only models in this field, we consider them to be broadly representative of the social cues approach generally, and they serve to illustrate thematic problems to which it gives rise. In this respect, our general critique can also be applied to similar approaches not elaborated here, such as the media richness model (Daft & Lengel, 1984; Daft, Lengel & Trevino, 1987; Daft & Macintosh, 1981). After critically evaluating these approaches, we present our own model based on social identity theory and de-individuation processes (the SIDE model). Empirical data supporting the model are reviewed, after which we briefly return to the earlier work and reconsider it within this new framework.
Models of communication and social influence ;within CMC:
The “social cues” perspective
The Social Presence Model
One of the most influential theoretical frameworks for analyzing mediated communication is the Social Presence Model developed by the Communication Studies Group at University College, London (Short, Williams & Christie, 1976). Funded by the British Post Office (now British Telecom), this research program was concerned with communication via telephone, audio and video links. However the influence of this approach has been more widely felt and it has been repeatedly applied in more recent analyses of CMC (Daft & Lengel, 1986; Johansen, 1977;; Rice 1984; Rice & Love, 1987).
According to Short, Williams and Christie (1976), the critical factor in the communication medium is its “social presence“. social presence was originally defined as an objective “quality of the medium itself” (p.66), although on the following page greater weight is given to social presence as “a subjective quality of the medium” (p.67). In this latter respect social presence corresponds to the psychological feelings or “mental set” associated with a given medium.
social presence is conceived as a single factor that comprises a number of dimensions relating to degree of interpersonal contact. It is closely related to the notions of “intimacy” (Argyle & Dean, 1965) and “immediacy” (Wiener & Mehrebian, 1968). In rating studies using semantic differentials, social presence is typically characterised by dimensions such as unsociable-sociable, insensitive-sensitive, cold-warm and impersonal-personal (Short et al., 1976), although it does not always appear as a unique factor. The personal-impersonal dimension also correlates significantly with items designed to tap social presence (Champness, 1973;). A series of these rating studies suggested that communications media could effectively be ranked according to social presence operationalized in this way (namely from business letters, telephone/ single speaker audio, multiple speaker audio, television, up to face-to-face communication). electronic mail then presumably occupies a relatively low position level in this ranking, somewhere between business letters and the telephone. Subsequent comparative media analyses including some which have encompassed electronic mail or teleconferencing have produced very mixed results. Some of these have supported the ranking implied by the social presence model while others have identified a number of conditions under which the relative positions that media occupy give way to more important considerations or else can be overturned (e.g., Daft & Lengel, 1986; Hiemstra, 1982; Lea, 1991 Reid, 1977; Rice, 1987 Rice & Love, 1987 Rice & Williams, 1984; Sproull & Kiesler, 1986; Sumner, 1988; Steinfield, 1986; see also Walther, 1992).
Given the importance of such dimensions as sociable-unsociable and personal-impersonal to the social presence model, it is not hard to see how media low in social presence have been seen as less “social” per se. Indeed this seems to be entailed in the very construct. We shall return to this point in detail later. Our immediate concern, however, is with the predictions this model makes for the social influence of others in CMC, or analogues thereof. A more-or-less explicit corollary of the social presence Model is that the greater the social presence the more likely that communicators will be able to bring all their powers of persuasion to bear and exert influence, whereas such powers should be progressively eroded as “presence” decreases. Short tested this hypothesis in a series of studies comparing a face-to-face condition (high social presence) with a closed circuit TV condition (intermediate), and an intercom condition (low) (Short, 1972a; 1972b; 1972c; 1973). For our purposes CMC might be considered at least as low in terms of social presence as the sound only or intercom condition.
Perhaps surprisingly, the result of these studies was that influence and attitude change followed a pattern directly contrary to that predicted by the model. When attitude change occurred, it was consistently more in the direction of the source in the sound only condition than for face-to-face, with the CCTV condition in between. Puzzled by these findings Short offered certain post hoc explanations, none of which however were particularly encouraging for the equanimity of the model or its predictive utility (see also Rutter, 1987). For our part, the problem arises from construing “presence”—to the extent that it plays any role in mediating such effects—in largely physical and informational terms. Suffice to say here that this failure to predict and satisfactorily explain this basic media difference in influence effects is a serious blow to the model’s viability.
Rutter (1987) has also questioned the phenomenological emphasis of the social presence model and criticized its lack of theoretical specification (e.g., the conflation of informational cues with “mental set”) as well as raising other theoretical and methodological problems (e.g., that social presence is usually determined circularly and post hoc). Despite these criticisms Rutter acknowledges that there is much overlap between the social presence Model and his own “Cuelessness Model”. Given that the latter model builds on the social presence approach and attempts to account for its shortcomings within a more sophisticated theoretical framework, we devote some attention to it here.
The Cuelessness Model
The concept of “cuelessness” and its accompanying model reflects the culmination of an impressive body of experimentally-based work by Rutter and his associates (see e.g., Rutter, 1984; 1987). Like the Communication Studies Group much of this research was funded by British Telecom, although it had a broader theoretical scope than earlier work, the aim being to integrate a wide variety of apparently contradictory and disparate findings (many generated by the research program itself). Again, much of this research was conducted at a time when CMC was a less prominent technology, so that many of the explicit or implied comparisons were between face-to-face communication, audio and closed circuit television (CCTV). Nevertheless, much of the research does reproduce certain conditions common to CMC so this work is of theoretical interest here.
Cuelessness clearly has conceptual similarities with information richness and social presence. In contrast to social presence however, cuelessness was originally defined as an information-based rather than a phenomenological concept, being simply the aggregate number of social cues available to subjects: ” …the fewer the social cues, the greater the cuelessness” (Rutter & Stephenson, 1979). Nevertheless, in purely operational terms cuelessness is virtually indistinguishable from these other concepts, with face-to-face interaction rating lowest and letters, bulletins and so forth rating high on cuelessness. In this framework then, CMC must be regarded as relatively cueless. The most “cueless” condition used in the experiments of Rutter and his colleagues was a voice-only condition, which arguably still provides more social cues than the remote text-based communication afforded by CMC.
The cuelessness concept and the model have undergone a number of refinements in order to accommodate awkward and contradictory results (charted in Rutter, 1987). Thus, although cuelessness was originally predicted to directly affect behaviour (Rutter, Stephenson & Dewey, 1981), its effect was subsequently argued to be indirect, mediated by “psychological distance” and dependent not simply on the number of available cues, but on the number of “usable” cues (Rutter, 1987). In this respect psychological distance seems to be very close conceptually to the more phenomenological notion of social presence. The effects of psychological distance are then further mediated by the ” content” of the communication which putatively results in particular communication styles and outcomes:
“cuelessness leads to psychological distance, psychological distance leads to task-oriented and depersonalized content, and task-oriented depersonalized content leads in turn to a deliberate, unspontaneous style and particular types of outcome” (Rutter, 1987;, p. 74).
Although the cuelessness model has not been applied in its entirety to CMC research, nevertheless the key concepts of cuelessness and psychological distance have resurfaced in CMC research within the social presence tradition (e.g., DeScanctis, & Gallupe, 1987; Rice & Love, 1987).
Before considering problems with the cuelessness model it is important to examine some of the empirical work with which it has been associated. Given the range and heterogeneity of findings, our selection must at best be considered illustrative, and once again we focus on outcomes rather than other aspects of the model (see Rutter, 1987 ;for a review). A series of formative studies in this tradition involved simulations of industrial bargaining in which subjects (usually students) role-played negotiators for the union or management side in an industrial dispute (e.g., Morley & Stephenson, 1969, 1970; Stephenson, Ayling and Rutter, 1976;. See also Short et al., 1976). The consistent finding from this line of research was that the side which was given the stronger case, either union or management, tended to persist through to victory, but only in the intercom condition. In the face-to-face condition the most common result was for the sides to compromise. Morley and Stephenson originally interpreted these findings in terms of their concept of “formality“, arguing that face-to-face communication is less formal and increases the salience of the interpersonal dimension at the expense of “interparty” concerns. Thus communication conducted face-to-face may be more concerned with maintaining satisfactory personal relationships than that in more formal communication systems where interparty considerations become more salient, such that so that settlements in favour of the side with the stronger case tend to be associated with the formality of the system used (Morley and Stephenson, 1977). This distinction between the interpersonal and intergroup dimensions is an important feature of the work of Morley and Stephenson (e.g., Morley & Stephenson, 1977; Stephenson, 1984) and plays a central role in our own model to be articulated shortly.
Rutter (1987), however, re-interprets these findings in terms of the cuelessness Model: the psychological distance associated with cuelessness leads to a task-oriented and depersonalized content which results in outcomes which lack compromise. Differences between media seem to turn on a ‘task/person’ distinction echoing the division between task-oriented and socio-emotional functions introduced by Bales (1950) and subsequently applied to the CMC context (e.g., Hiltz & Turoff, 1978; Rice, 1984; Hiltz, Johnson & Turoff, 1986; Sumner, 1988) despite substantial problems (Lea, 1991; Walther, 1992). face-to-face communication is rich in interpersonal cues, more personalized and thus socially rich. By associating more cueless media with task-orientation on the other hand, the implication is that these are somehow less social so that here the intergroup dimension to the ‘social’ is marginalized. In sum the ‘social’ element in the cuelessness Model is defined in largely quantitative and ‘informational’ terms as the number of social cues, and is optimized in the interpersonal exchange of face-to-face communication. Short et al., (1976) also cite these bargaining studies, and like Rutter, use them to underscore the idea that face-to-face communication is more social (and sociable) than communication via media less rich in interpersonal cues. More recently Rice & Williams (1984) cite these studies in applying the social presence model to CMC.
Part of the impetus for the rejection of “formality” and reformulation in terms of cuelessness can be related to the contradictory findings that soon began to emerge in this field. Under certain circumstances it appeared that face-to-face conditions could lead to less compromise (e.g., Carnevale, Pruitt & Seilheimer, 1981; Pruitt, Kimmel, Britton, Carnevale, Magenau, Peragallo & Engram, 1978; Short, 1974)—for example when people had to argue their own position instead of on behalf of the group (Short, 1974). In a more recent attempt to reconcile the contradictory findings of Morley and Stephenson (1969, 1970) and Carnevale et al. (1981), Schruijer (1990) pointed to the competitive or co-operative nature of the pre-existing normative context, and argued that the salience of the interpersonal dimension in face-to-face communication would enhance these behavioural responses (cf. Zajonc, 1965). Her results were not clear-cut, but the evidence tended to go against her hypothesis, such that in the neutral and competitive conditions the party with the stronger case was more likely to win under telephone than face-to-face conditions (Schruijer, 1990). In fact these findings are more in line with the original approach of Morley and Stephenson, and our own model developed below.
The immediate point however is that the cuelessness dimension was incorporated into the model in an attempt to explain these contradictory findings. However, it is not at all clear that this refinement is any more successful in explaining why people sometimes exercise more influence (and show less compromise) in sound only conditions, and sometimes in more “personalized” face-to-face conditions (see Rutter, 1987). In sum the model does not sufficiently specify how social influence might operate in both these contexts, nor the quality or direction of outcomes that might be predicted. To be fair Rutter (1987) acknowledges some conceptual shortcomings in the cuelessness approach (e.g., the criteria for the usability of cues are not specified so the advantage of cuelessness as an objective and informational concept seems to be somewhat lost). Also the mechanisms mediating the relationship between cuelessness and its (albeit indirect) influence on process and outcomes are imprecise. For example although the effect of cuelessness is normally to increase psychological distance, Rutter notes that some media that are high in cuelessness can be very close psychologically in certain contexts, such as ‘telephone hotlines’. Here it is precisely the lack of social cues that seems to guarantee a degree of intimacy. In theoretical terms then the truly psychological concepts in the model like the ‘usability’ of cues and ‘psychological distance‘ seem to protect cuelessness and the associated informational model from empirical exceptions, as much as serving a priori to explain processes and outcomes.
Finally a remark concerning communication style is in order. One of the clear predictions of the cuelessness model is that cueless communication media will lead to task-oriented and depersonalized content which should result in an unspontaneous communication style. It is not entirely clear how ‘unspontaneous’ should be interpreted here, but in any event evidence from CMC suggests quite the opposite regarding this prediction. The phenomenon of paralanguage; produced by keyboard tricks, the lack of etiquette for salutations, the playfulness of much CMC and the variety of persona or “voices” that are adopted in computer conferences all suggest stylistic spontaneity in CMC that, if anything, is promoted because of the relatively cueless environment (Finholt & Sproull, 1990; Lea & Spears, 1992; Myers, 1987; Ord, 1989; Spitzer, 1986; Van Gelder, 1985; Walther, 1992; Wilkins, 1991).
In sum, although the cuelessness model has been successively refined to account for disparate findings it is not at all clear that cuelessness, at least in any such unidimensional informational sense can adequately accommodate them. In this respect we would argue that a retreat from an analysis of the distinctive features of the social and normative context towards a purely informational and quantitative analysis of social cues is a retrograde theoretical step which tends to exclude the broader structure and social meaning of situations and their influence on behaviour. The question of extrapolation to text-based electronic media only exacerbates these problems and a direct test of the model’s predictions in the CMC context still awaits. We now move on to consider an approach which has tackled the social psychological aspects of CMC directly.
The Reduced Social Cues approach
In this section we consider the research program of another very influential group: the Committee on Social Science Research in Computing at Carnegie-Mellon University (e.g., Kiesler, Siegel & McGuire, 1984; Kiesler, 1986; McGuire, Kiesler & Siegel, 1987; Siegel, Dubrovsky, Kiesler & McGuire, 1986; Sproull & Kiesler, 1986). Their extensive work in this field covers a cluster of interrelated processes and explanations which we refer to collectively as the “Reduced Social Cues” approach (RSC).
From our point of view, this work has a number of clear advantages over the research previously discussed. First and foremost it has been directly concerned with CMC, so many of the problems of extrapolation from more general experimental studies designed to manipulate social cues simply do not arise. Second this work directly addresses social influence and decision making processes using established paradigms (notably the group polarization paradigm) so comparisons across research is much less ad hoc or hazardous. It also offers a sophisticated theoretical account of influence processes which is specifically tuned to effects and outcomes, making it easier to generate and evaluate predictions. At the same time, the theoretical underpinnings of this research share many similarities with the two models already discussed. Once again the absence of social cues occupies a central role in explaining the social psychological effects associated with CMC.
What are these effects? Kiesler and her associates present and review an impressive body of empirical work that suggests that behaviour within CMC tends to be relatively uninhibited (as reflected in the phenomenon of “flaming“) and that, relatedly, group decisions made in CMC tend to be more polarised, extreme, and risky than in face-to-face interaction (e.g., Kiesler, Siegel & McGuire, 1984; Kiesler, 1986; Siegel, Dubrovsky, Kiesler & McGuire, 1986; Sproull & Kiesler, 1986).
It is perhaps important at this point to clarify and define ” group polarization” because this phenomenon and the processes underlying it will form a recurring issue in the remainder of the chapter. group polarization is simply the tendency for the mean attitudes or decisions of individuals to become more extreme in the direction of the already preferred pole of a given scale, as a result of discussion within the group. (This is usually combined with some convergence of positions towards each other, although this aspect of group polarization is far less commonly noted.) Ever since early work on the “risky shift” (Stoner, 1961) this phenomenon has been established as one of the most robust and well-researched findings within social psychology (see e.g., Isenberg, 1986; Lamm & Myers, 1978; Wetherell, 1987 ;for reviews). The research of Kiesler and her colleagues was particularly important in showing that this effect was often even stronger when group discussion took place via CMC (e.g., Kiesler et al., 1984; Siegel et al., 1986).
The basic thesis of the RSC approach is that certain features associated with CMC encourage psychological states which undermine the social and normative influences on individuals or groups, leading to more deregulated and extreme (“antinormative”) behaviour. A central factor underlying this approach is the absence of social and contextual cues, leading to the reduced impact of social norms and constraints. Thus Siegel et al. (1986) indicate that: “the absence of social context cues in computer mediation will reduce normative influence relative to informational influence” (p. 182).
The model of informational influence used to explain greater polarization in CMC is “persuasive arguments theory” (e.g., Burnstein & Vinokur, 1977; Vinokur & Burnstein, 1974). This theory posits that group polarization is a function of the number of (novel )persuasive arguments favouring the already preferred pole to which one is exposed during group discussion. It is argued that both the uninhibited behaviour and the more equal participation characteristic of CMC facilitate the exchange of more persuasive arguments favouring the preferred pole.
In fact this brief summary does not do justice to the full complexity of the RSC approach. As suggested above, the approach implicates a range of theories and concepts which also seem to have been developed and refined across time (cf. Kiesler et al., 1984; McGuire et al., 1987; Siegel et al., 1986). We have tried to summarise the essential aspects of this eclectic model and the relation between its elements in Figure 1. We should stress that this is not the reproduction of a model presented anywhere else in this diagrammatic form; it is our attempt to formalise and integrate the various elements from publications (notably Kiesler et al., 1984 ;and Siegel et al., 1986). The aim is to facilitate comprehension of the elements and their interrelations, and, no less importantly, to afford later comparisons and contrasts with our own approach.
[Insert figure 1 about here]
Following our earlier summary (Lea & Spears, 1991) we shall briefly consider under five headings the Carnegie-Mellon team’s range of explanations for the distinctive behaviour and products associated with CMC. These are (1) lack of social cues; (2) difficulties of co-ordination and feedback; (3) de-individuation; (4) depersonalization and/or attentional focus; (5) conformity to a computing subculture norm. We shall summarise these in turn (see also Figure 1).
The absence of social cues associated with CMC is a central theme which in many respects also underlies the other explanations. Lack of social cues is argued to have two basic effects both of which putatively lead to the exchange of more extreme arguments—the basis of the persuasive arguments explanation of group polarization. In general then, removing such cues arguably removes those social inhibitions or normative constraints which could act as a brake on the generation or articulation of extreme arguments, particularly by removing or reducing indicators of status, power and leadership (path b, Figure 1). Related to this, the absence of social cues results in more equal participation of group members putatively facilitating the greater exchange of extreme arguments (path a). uninhibited behaviour is possibly also provoked by the lack of feedback and general communication difficulties encountered in CMC (path c), while the relative absence of feedback could also help to explain more equal participation (path d)—again, both routes to more extreme arguments (Kiesler et al., 1984).
The de-individuation explanation is also founded on the absence of social cues. de-individuation has been classically defined as the loss of identity and weakening of social norms and constraints associated with submergence in a group or crowd (e.g., Festinger, Pepitone & Newcomb, 1952; Zimbardo, 1969). Kiesler et al. (1984) note that: “Computer-mediated communication seems to comprise some of the same conditions that are important for de-individuation—anonymity, reduced self-regulation and reduced self-awareness” (p. 1126).
Given the traditional link between de-individuation and uninhibited behaviour then, this should provide another route to the articulation of extreme arguments (Kiesler et al., 1984; Siegel et al., 1986; path e, Figure 1). Interestingly, Kiesler et al. also use this disinhibition element on which to premise a normative route to group polarization (path f, Figure 1). The traditional normative explanation of group polarization, social comparison theory (e.g., Sanders & Baron, 1977), posits that polarization results from conformity to a socially desirable but extreme norm that only becomes publicly revealed during group discussion. The disinhibition associated with de-individuation presumably helps reveal more of this hidden norm, or pushes it further to the extreme.
Depersonalization or a shift of attention away from one’s audience is a further possible explanation considered by both Kiesler et al. (1984), and Siegel et al. (1986) (path g, Figure 1). Thus Kiesler et al. suggest that CMC participants are “…less responsive to immediate textual cues… and less bound by precedents set by social norms” (p. 1130) whereas Siegel et al. concur in asserting that CMC reflects a shift of attentional focus from the social context to the content and context of the message, sui generis. According to Siegel et al., a heightened self-consciousness or self absorption in the message may produce less sociable and more uninhibited, or anti-normative behaviour (p. 182).
Finally both Kiesler et al. (1984) and Siegel et al. (1986) suggest that there may be a particular local etiquette or norm associated with computing subculture which rejects convention and social restraint, helping to explain the relatively uninhibited behaviour and extreme outcomes (path h, Figure 1). This norm also feeds into the depersonalization of the communication context (path i).
In summary, a common feature of these explanations is their stress on various effects of the absence of social cues, notably the reduction of inhibition, social constraint and regulation resulting in “antinormative behaviour”, together with the more equal participation of group members. Both these channels feed into the informational explanation of group polarization based on persuasive arguments theory (see Figure 1). In our earlier paper we have provide a conceptual and empirical critique of this eclectic model (Lea & Spears, 1991). Nevertheless, it is necessary to reiterate the main problems as we see them again here. We focus first on conceptual issues and problems of theoretical integration before turning to empirical anomalies.
First of all the RSC approach combines aspects of what Rice (1984) has termed the “cool” and “warm” models of CMC within the same framework. Thus, there seems to be some degree of contradiction between the characterization of CMC as an fast, efficient medium on the one hand whilst invoking inefficiency (communication difficulties, delayed feedback and so forth) as a cause of frustration and disinhibition on the other. Although CMC may in certain respects be less information-rich or efficient than face-to-face communication, other media that are even less so (e.g., letter writing) do not appear to evoke disinhibited behaviour. It may be that expectations of efficiency are not fulfilled in CMC, or that the on-line nature of communication is particularly frustrating. In any event the relationship of CMC to inefficiency and uninhibited behaviour is not yet unequivocally established and would seem to contradict the more generally proclaimed advantages of electronic mail as a communication medium (see chapter by Lea et al., this volume). Moreover, it could be argued that media such as face-to-face interaction which promote a less hampered flow of information are more likely to result in the greater exchange of novel arguments required by informational influence (a position these researchers were in fact to adopt later). Further empirical work on this issue, specifying the features of the medium which may underlie such effects would therefore seem to be warranted.
“Rationality” seems to occupy a similarly contradictory position in this approach: both rational (e.g., informational influence) and irrational elements (deregulated and antinormative behaviour) are combined somewhat promiscuously within the same model. At times it is difficult to conceive how the deregulated and antisocial psychological state of classic de-individuation theory (for example) is compatible with the relatively reasoned deliberation of the pros and cons of a choice dilemma, as in persuasive arguments theory. At the very least then, these irrational and impulsive beginnings sit rather uneasily with the essentially cognitive and information processing explanation of group polarization, to which they apparently give rise.
The employment of the concept of de-individuation within this account is problematic in other respects also. It should be recalled that de-individuation is being used here to explain a classic social influence effect (group polarization) whereas it would seem to imply the very rejection of social norms and standards (i.e. deregulated and “antinormative” behaviour). Its relationship to self-awareness is also contradictory within the wider context of the model. de-individuation typically implies a lack of self-awareness or self-consciousness (Diener, 1980), where as Siegel et al. (1986) argue precisely for a heightened self-consciousness within CMC—an assumption supported in other empirical work (Matheson & Zanna, 1988, 1989; see also chapter 4 this volume). In this respect at least, isolation at the computer terminal could be regarded as individuating rather than de-individuating, and this could help to explain evidence of greater (private) self-awareness in CMC (cf. Carver & Scheier, 1981; Matheson & Zanna, 1988, 1989; Spears, Lea & Lee, 1990).
Nevertheless a perhaps more fundamental problem concerns the central thesis that group polarization follows from a weakening of social norms due to the absence of social cues. It is worth remarking that this general position would immediately seem to be undermined by the inclusion in the same framework of an explanation based on a specific etiquette or norm associated with the computing environment, while the latter explanation begs the question how one particular norm can penetrate an otherwise “normless” context, and if it does, why other norms should be excluded.
More critical still for the “anti-normative” domain of the model is the thorny issue of defining normative versus “anti-normative” behaviour. It appears that both behaviour (flaming) and outcomes (group polarization) have been described as anti normative to the extent that they are more negative or extreme than some general social standard. However, if antinormative behaviour has a clear directional form, this would seem to define some sort of “norm”, albeit an extreme and negative one. If behaviour was really socially deregulated as implied by the de-individuation concept, then presumably we should sooner predict a random distribution of decision responses (or positions favouring both extremes), than polarization in the already preferred direction. A random distribution should then result in outcomes closer to the midpoint (depolarization). Similarly the distinctive language and behaviour hitherto associated with CMC— flaming—can also be interpreted as the expression of a particular group norm rather than reflecting deregulated behaviour per se (see Lea et al., chapter 5 this volume).
persuasive arguments theory notwithstanding, the notion of group norm and normative influence remain central explanatory concepts in accounting for group polarization generally (Isenberg, 1986; Mackie, 1986; Stoner, 1968; Sanders & Baron, 1977; Wetherell, 1987). It seems that part of the problem for the RSC model derives from the translation of the state of disinhibition, as reflected in uninhibited behaviour, into group polarization and more risky or extreme decisions. We would argue that disinhibition on the one hand and the riskiness or extremity of decisions are quite separate phenomena, or at least that no direct link between the two has been demonstrated. This becomes more apparent when one considers that it has long been shown within the group polarization paradigm that certain choice dilemmas result in extremity shifts to caution, and not risk, following group discussion (Fraser, Gouge & Billig,1970; Pruitt, 1971). Indeed Hiltz, Turoff and Johnson (1989) obtained just such a cautious shift in a recent CMC field study. These findings would seem conceptually hard to square with the idea that they reflect “disinhibition” or uninhibited behaviour. Indeed there seems to be surprisingly little if any direct empirical support in the literature that CMC is characterised by an absence or weakening of social norms.
Further empirical problems for RSC approach emerge from a subsequent study. McGuire, Kiesler and Siegel (1987) found that participants actually produced a smaller volume of message traffic (“argumentation”) in the CMC condition than in face-to-face interaction—supporting this aspect of the earlier studies by the same team (Siegel et al., 1986). However, in the Maguire et al. study there was actually more evidence of group polarization in the face-to-face condition (although the experiment diverged in certain respects from the standard group polarization paradigm). This difference was then used to argue for a persuasive arguments explanation of greater polarization in the face-to-face condition. Recall however, that in the earlier studies persuasive arguments theory was used to explain the greater polarization in CMC. A second difference in the later study was to shift from the equality of participation to the actual amount or number of arguments exchanged as the operational definition of persuasive arguments. However persuasive arguments theory hinges, strictly speaking, on the number of valid and relevant novel arguments that are brought to bear on the issue (Burnstein & Vinokur, 1977) and data relating to this dimension of the arguments exchanged were not reported in these studies. In sum it is not clear whether equality of participation, or the sheer amount of exchange is the m ore valid indicator, although based on persuasive arguments theory we would argue that volume is the more likely candidate. Indeed, it could be plausibly argued that greater equality of participation should undermine polarization by giving expression to a broader spectrum of views. To the extent that one person or only a few people dominate the discussion, they are more likely to sway decisions in their preferred direction. Following the “law of small numbers” small samples are likely to have more extreme means than large samples (Tversky & Kahneman, 1971) implying that inequality of participation is more likely to lead to polarization. A further telling point is evidence that exchange of arguments does not even seem to be necessary for polarization to occur, nor is it obvious why relatively more arguments should be generated beyond the mean group position (see e.g., Wetherell, 1987 ;for a recent review).
In summary, these difficulties leave the status of persuasive arguments theory decidedly shaky within the context of the RSC model. The use of equality of participation to underpin the greater generation of extreme arguments is particularly dubious and one that the Carnegie-Mellon team themselves later appear to have rejected. In general, we would argue that problems with the RSC model stem from both of its basic elements— namely a commitment to the idea that behaviour within CMC is socially deregulated or “antinormative” and (consequently) a reliance on an informational explanation of social influence. In the following section, after drawing together and critiquing the various theoretical approaches to CMC considered thus far, we propose an alternative formulation which addresses the distinctive features of CMC within a more parsimonious theoretical framework whilst avoiding many of the pitfalls of the present eclectic model.
A summary critique of the “social cues” approach
In this section we have critically reviewed three influential theoretical approaches to CMC behaviour and effects, and examined a range of specific theoretical and empirical problems with each. Here we draw these points together and focus on a recurring theoretical blind spot. A unifying theme in all these approaches, as well as others, is to stress the lack of social cues afforded by CMC. This starting point would indeed seem indubitable and we do not dispute the relative absence of direct visual cues in CMC. However, the problem for us arises with the referents of the term “social” in this formulation.
Part of the difficulty in our view is that what counts as social has largely been over-extended and misconstrued. Specifically, the social has come to be equated with the interpersonal and the informational in these models. To make the underlying assumptions explicit then, what is social about being and behaviour is interpersonal interaction and literally being with others. Thus in a statement reminiscent of social presence theory Rutter (1987) defines psychological distance as someone’s ‘feeling that their partner is “there” or “not there”‘ (p. 137). The social is thus defined in individualistic, physicalist, and “interdependence” terms so that social information constitutes the verbal and nonverbal information conveyed in situ, and most powerfully in face-to-face interaction. It is then perfectly logical to expect that social influence and one’s social grip on the other will be greatest in the face-to-face interaction, as predicted by social presence theory. Social interdependence implies dependence on others for social information and influence (Hogg, 1987; Turner, 1991). The foundations for social deregulation and antinormative behaviour when direct interpersonal bonds are cut are then already laid.
Kiesler and her colleagues are not alone in taking this view that CMC is less socially regulated for these reasons. For example Rice (1984) considers a degree of control or authority, maintaining the stability of the social structure to be a prerequisite f or organizational efficiency. In the social cues tradition he sees this social fabric as being kept intact by social cues and barriers; their elimination implies a reduction of certainty and control. This theorizing is heavily influenced by the work of Bales (1950), where the social elements of group behaviour are defined in terms of functional social roles occupied by individuals, and the interaction between them. Once again, this locates social influence and cohesion in interpersonal relationsand in the transmission of social cues; once these disappear, social deregulation and disorder are presumably more likely.
Another recurring theme, which we argue reflects the interpersonal conception of the social, is the distinction between the task-oriented and the social or socio-emotional dimensions of group activity and communication This division can again be traced back to Bales (1950), and resurfaces in the work of many researchers in CMC (e.g., Kiesler et al., 1984; Rice, 1984; Rice & Love, 1987; Siegel et al., 1986; Sproull & Kiesler, 1986; Steinfield, 1986; Sumner, 1988). The interpersonal basis of the social in this formulation is made most explicit in Rutter’s opposition between the “task” and the “person”; task-orientation is associated with cuelessness while face-to-face communication is more personal and thus more “social” (Rutter, 1987). This idea is directly echoed by CMC theorists who couple “immersion in the task” with the less social mode of CMC (e.g., Siegel et al., 1986). Again the social is equated here with interpersonal interaction, the implication being that task related activity is not itself social activity or socially regulated.
We argue that this equation of interpersonal with social provides a very circumscribed, and misleading notion of the social that theoretically excludes important aspects of this dimension. This is no accident, but flows directly from the methodological individualism characteristic of the social cues approach, and its heritage in experimental research on nonverbal behaviour and interpersonal communication and on small ad hoc groups. By equating the social with the interpersonal nexus of communication, and with the transfer of information in situ, this approach inevitably tends to neglect the important pre-existing social categories, norms and identifications which position communicators and define their relations to each other. The experimental focus of much social psychological research on CMC also tends to occlude the historical dimension to these relationships (cf. Rice, 1984), and how people may develop new social identities within CMC, often precisely because of the anonymity afforded by the lack of “social” cues (Myers, 1987; Van Gelder, 1985). In sum we argue that a major distinction necessarily excluded by this formulation is that between the personal and the social, where the social is in certain crucial respects the opposite of the personal or interpersonal. We now develop this argument in t he context of our own alternative approach to understanding the social psychological processes in CMC.
An alternative approach: The SIDE Model
In order to avoid the shortcomings of previous theories outlined above we wish to make a clear distinction between social cues qua interpersonal cues conveyed in situ (e.g., nonverbal signals, verbal intonation and so forth), and cues to the social, namely information about the participants, the context, and particularly relevant social category information. We take as our theoretical point of departure social identity theory (Tajfel & Turner, 1986) and its sister theory, self-categorization theory (Turner, Hogg, Oakes, Reicher & Wetherell, 1987). It is necessary first to outline a few basic principles in order to proceed with our argument.
In delineating the self-concept these theories draw a distinction between personal identity; on the one hand and social identity or identities on the other. Whereas personal identity corresponds to one’s identity as a unique individual (one’s “personality” if you like) social identities are those aspects of the self corresponding to valued groups or social categories to which one belongs and with which one identifies. According to this theoretical tradition, either personal identity or particular social identities may become salient and affect behaviour depending on the situation or context; when personal identity is salient we act in terms of our identity as unique individuals, but when a particular social identity becomes salient we are more likely to act in accord with the norms and standards associated with the corresponding group or social category. In fact this distinction between personal and social identity is conceptualized as a continuum rather than a simple dichotomy so that social behaviour can be conceived as varying on a dimension ranging from individual and interpersonal relations at the one pole, to intergroup relations at the other. In the former individual identities and interpersonal psychological processes will be central, whereas in the latter intra and intergroup identities and processes will come to the fore. Here it is easy to see the parallels with the interpersonal/interparty distinction of Morley and Stephenson. This distinction is crucial because as should be clear from the foregoing discussion, the social cues perspective has tended to define the social both theoretically and operationally in terms of the interpersonal domain only.
Certain points follow from this theoretical framework. First of all we would argue that an absence of social cues as interpersonal contact does not necessarily imply the absence of social cues per se. Whereas the communication of cues at the interpersonal level would seem to be quite sensitive to the information richness of the medium it is not clear that conveying categorical information, or the salience of such categories, depends at all upon an information rich environment, or at least not in the same way. The nuances of personal style, charisma, and appearance which are so important in the interpersonal context are indeed likely to be impaired in a relatively cueless medium. In contrast, by its very nature social categorical information is very sparse and unelaborated, discrete rather than continuous (e.g., Fiske & Neuberg, 1990; Tajfel & Forgas, 1981), and it is often likely to be given or already inferred in the CMC context (Culnan & Markus, 1987). In fact categorical information is the type generally provided by message headers (e.g. name, giving cues to gender and ethnicity, organizational affiliation, distribution list/conference, and so forth). Such information might be summarised in a single social category label and could be given in answers to questions such as “are the other participants part of a social category to which I belong?”. In fact a feature of CMC networks frequently remarked-on is the ease with which they bring together people with similar interests through membership of specific distribution list or computer conferences (e.g., Compuserve, Usenet). Indeed some have argued that because CMC is often used from within a network of supportive relationships, merely to use the medium may in itself signify membership of a particular group (cf. Fulk, Schmitz & Steinfield, 1990; Trevino, Daft & Lengel, 1990). In most organizational contexts in which CMC is used then (e.g., managers communicating with other managers or their staff) these social categories, roles and relationships are likely to be well defined quite independently of the presence of interpersonal cues or mystificatory notions of “social presence” or “psychological distance“.
In sum the salience of category cues underlying social identification and group processes is likely to be to some extent independent of the interpersonal context cues that are lacking in CMC. In fact it is possible to argue more strongly that these two dimensions are to some extent inversely related. Following the interpersonal— intergroup continuum of social identity theory, the salience of personal identity in the context of interpersonal relationscan in certain respects be conceived as quite the reverse of one’s identity as a member of a group or social category. Indeed Turner (in Turner et al., 1987) argues that there is a “functional antagonism” between these two levels of identity. Thus a context that is informationally rich in terms of interpersonal and personalized cues might undermine or counteract the group and social categorical bases of interaction. Although an oversimplification (we are not saying that personal identity can never be salient within CMC, as we shall see), this point forms an important theoretical base from which to explain a number of empirical anomalies that have bedevilled extant models of CMC effects and provides a foundation for our own analysis. We shall now develop this argument further in the context of a reconceptualization of de-individuation within social identity theory. Like previous attempts to explain CMC effects, our approach is grounded in the distinctive features of CMC use—namely the visual anonymity of the participants, and relatedly, their physical isolation or separation. We focus first on the effects of visual anonymity in relation to de-individuation.
Our approach draws on a social identity-based reinterpretation of the concept of de-individuation articulated by Reicher (1984, 1987) in the context of explaining social influence processes in crowds. (To indicate the social identity and de-individuation elements we refer to it as the SIDE Model.) Contra classic de-individuation theory which posits a loss of identity in the mass, and a corresponding deregulation of normative behaviour (cf. Festinger et al., 1952; Zimbardo, 1969; Diener, 1980), Reicher argued that anonymity accompanying immersion in a social group can actually serve to enhance social identity, with a consequent strengthening of group response and normative behaviour. Rather than a loss of identity, de-individuation may often reflect a switch to social identity (Reicher, 1984, 1987). Assuming people identify with a salient group, they are more likely to be influenced by it under de-individuating conditions because the visual anonymity will further reduce perceived intragroup differences, thereby increasing the salience of the group. However, Reicher adds that if the reverse situation pertains—where personal identity is already salient— then de-individuating conditions are likely to enhance one’s sense of individuality, reducing the salience of group norms (see Reicher, 1984;, and also Abrams, 1990; Hogg & Abrams, 1988). In other words, de-individuation as anonymity should serve to accentuate the effects of the salient identity (social or personal), and the dominant normative response associated with it. In applying this idea to CMC we argue that the context in which CMC takes place corresponds to de-individuating conditions (i.e. with visual anonymity), and that the effects of this should crucially depend on whether group or individual identity is already salient. social influence by the group should only result if one identifies with the group, and this identity is salient.
It is important to consider another possible route to the social influence effects within CMC that draws on self-attention processes and which leads to quite similar predictions (see Matheson, chapter 4 this volume). As noted earlier, both Siegel et al. (1986) and Matheson and Zanna (1988, 1989) argue that the social isolation associated with CMC produces increased (private) self awareness (cf. Fenigstein, Scheier & Buss, 1975). Following self-attention theory (e.g., Carver & Scheier, 1981) we argue that the effect of this will be to heighten the person’s awareness of already salient identities and their accompanying norms or standards. Thus once more, if a participant’s group or social identity is salient, increased private self-awareness should increase conformity to group relevant norms, whereas a more individual standard will be influential if personal or individual identity is salient.
In sum the effects of visual anonymity and isolation are predicted in our model to have similar effects, albeit by slightly different routes. These are unified by the overarching framework of social identity and self-categorization theories with their stress on the distinction between the salience of social or group identity on the one hand and personal or individual identity on the other. These effects, due to a re-socialized concept of de-individuation and self-attention processes, can thus be seen as complementary rather than contradictory (cf. Spears et al., 1990), each invoking a different aspect of the CMC context.
Before considering the empirical support for the model it is necessary to say something further about the mechanisms underlying group polarization in order to be clear about how effects in this paradigm relate to our predictions concerning social influence. It may be apparent to the reader by now that we advocate a more normative explanation of group polarization in favour of the purely informational account in terms of persuasive arguments theory offered by the RSC approach. However, this is a slight simplification because the theoretical account we follow, again grounded in social identity theory, attempts to transcend the dualism of informational and normative influence that has plagued the social influence literature (cf. Turner, 1991). In criticising purely informational and purely normative explanations of group polarization, Turner and his colleagues propose a theory of “Referent Informational Influence” (e.g., Hogg & Turner, 1987; Turner et al., 1987; Turner, 1991; Wetherell, 1987) which combines both normative and informational elements. According to this model, social influence reflects conformity to the norm of the relevant group with which one identifies.
The central conceptual problem with traditional normative accounts of group polarization has been to explain how apparently going beyond the mean group response can be normative. Because it is premised on a specific group norm, and not some general societal norm (as in classic normative influence), referent informational influence resolves this problem by positing an intergroup dimension to social influence. According to this view the group norm is defined not only by the positions of the own group, but also in contrast to positions taken up by other (out)groups. This implicit contrast category or “outgroup” is represented by positions contrary to those actually taken up by the group discussants. To the extent that initial positions are usually displaced to one end or the other of the judgemental continuum, the normative or “prototypical” position for the group will then actually be more extreme than the mean group response (see e.g., Wetherell, 1987;, for a fuller account). In this way group polarization can be conceived as convergence on an extremitized group norm.
In an empirical test of the model we manipulated what we regarded as the two crucial factors independently of each other, namely whether the group or the individual identity of the participants was salient, and whether participants were isolated and visually anonymous (de-individuated) or visibly present in the same room (individuated). The study therefore formed a 2×2 factorial design. Group and individual identity were rendered salient in the introduction to the experiment and by consistently referring to subjects as either group members or individual participants throughout the course of the experiment (see Spears et al., 1990 ;for further details). Subjects were required to discuss a series of attitude topics which had been pretested to establish a distinct (left-wing) norm for the student population that took part and participants were also made aware of the direction of this norm via feedback about the views of the student population. In all four conditions communication was by means of an asynchronous CMC system. An important feature of the design then was that, unlike previous research, CMC was used to communicate in face-to-face as well as isolated conditions in order to control for confounding factors associated with verbal vs. written communication (cf. Siegel et al., 1986). This aspect of the design lent additional ecological validity to the study of CMC conditions by relating the face-to-face conditions to electronic meeting room environments (e.g., Nunamaker, Vogel, Heminger, Martz, Grohowski & McGoff, 1989; Vogel & Nunamaker, 1990).
Following Reicher (1984), we predicted an interaction such that de-individuation (visual anonymity) should increase normative behaviour, and thus group polarization under conditions of high group salience (by reducing perceived intragroup differences) and decrease normative behaviour when group salience was low (individual identity salient).
Overall our predictions received strong support. Limitations of space prevent our presenting them in full here, but the reader is referred to Spears et al. (1990) and Lea and Spears (1991) for m ore detailed and comprehensive descriptions of our analyses and findings. First and foremost the predicted interaction between group vs. individual identity and de-individuation factors was confirmed (see Figure 2). De-individuated subjects whose group identity was made salient exhibited most norm-directed group polarization and de-individuated subjects whose individual identity was made salient displayed the least (actually “depolarization;”— a shift in the opposite direction to the directional norm). The means for subjects in the “individuated” or face-to-face conditions occupied intermediate positions.
[Insert figure 2 about here]
A number of other measures taken in this study are also relevant to the evaluation of our model, particularly in comparison to that of the RSC approach. In figures 3 and 4 we present the data pertaining to the number of discussion or ‘task-oriented’ remark s, and the number of ‘social’ remarks exchanged in the four conditions as coded and content analysed from the message traffic log (see Lea & Spears, 1991). Both of these analyses resulted in significant 2-way interactions, with, again, greatest divergence occurring between the two ‘de-individuation‘ conditions. Focusing on these, it turns out that significantly more discussion oriented remarks were exchanged in the individual identity condition than in the group identity condition whereas this pattern is reversed for social remarks. In terms of discussion oriented remarks then, given that it was the de-individuated individual identity condition which actually produced least group polarization, this would seem to present considerable problems for the persuasive arguments explanation of group polarization, premised on the volume of arguments exchanged. If anything there seems to be an inverse relation between the amount of issue-relevant information exchanged and group polarization. By contrast, the fact that the number of social remarks is greatest in the de-individuated group identity condition is quite consistent with a more normative explanation of group polarization (and problematic for an exclusively informational one).
[Insert figures 3 & 4 about here]
Nor does it seem possible to salvage the RSC approach’s earlier articulation of the persuasive arguments explanation couched in terms of equality of participation. Two measures of inequality of participation resulted in significant 2-way interactions between the factors such that the individuated individual identity condition, and the de-individuated group identity condition manifested greater inequality of participation than the remaining conditions (Lea & Spears, 1991). These results suggest, if anything, that inequality of participation rather than equality of participation mirrors the pattern of group polarization across the four conditions. Once again this would undermine an exclusively informational explanation of group polarization based on persuasive arguments. This pattern is however interpretable in normative terms. Indeed the more normative explanation based on Referent Informational Influence (Hogg & Turner, 1987; Turner, 1991; Turner et al., 1987) suggests that certain “prototypical” group members will be more influential than others, so that differential floor-taking during discussion is quite consistent with such a normative explanation of group polarization.
A final feature of these results worth noting is the marked absence of flaming behaviour supposedly characteristic of CMC. The use of exaggerated or uninhibited paralanguage in communication was meagre. More importantly in theoretical terms, it did not vary across conditions— that is to say it did not covary with the significant differences in group polarization (cf. Lea et al., chapter 5 this volume). Again, this would seem to undermine the conceptual link between the uninhibited behaviour associated with CMC and the causes of group polarization as suggested by the Carnegie-Mellon group.
To summarise then, our results would appear to provide support for our SIDE model, but problems for the RSC approach, and the cuelessness and social presence models. In particular it is not clear how any of the latter approaches would have predicted the interaction between de-individuation and group/individual identity obtained here. All three approaches are ill-placed to explain the discrepancy in group polarization between the group and individual identity de-individuation conditions. Both these conditions were identical in terms of their basic CMC features and presumably exhibit the same absence of context or “social cues”, and the same degree of ” social presence” or “psychological distance“. The individuated or face-to-face conditions by contrast provided a relatively information-rich or “cueful” environment presumably maximising social presence and minimizing psychological distance— conditions assumed to provide a normative context and enhance social influence. However, the degree of norm directed group-polarization here lay in between those of the de-individuated conditions.
It is also impossible to explain our results in terms of the normative explanation proposed in the RSC approach, namely that there is a specific norm or etiquette associated with computer culture. Our subjects were using CMC for the first time so were unlikely to be aware of such a generic norm should it exist. Moreover as with t he other explanations this could not account for the pattern of differences between the conditions—because all participants were equally exposed to CMC, they were presumably equally subject to this etiquette.
On the other hand, this pattern was predicted from the SIDE model and is explicable in terms of the distinction drawn between group and individual identity. The model proposes that the norms associated with these two levels of identity become particularly salient in typical CMC conditions characterised by visual anonymity and individual isolation. The visual anonymity associated with CMC attenuates the perception of intragroup differences which “individuate” group members and undermine the salience of the group in its purest form (the prototype or “ideal” of the group if you like). anonymity may also enhance one’s sense of individuality if individual identity is already salient (Reicher, 1984). Thus the fact that subjects displayed de-polarization in the de-individuated individual identity condition may have been because this was the only way participants could express their individuality as distinct from the group within the terms of our experiment (Lea & Spears, 1991; Spears et al., 1990; cf. Turner, Wetherell & Hogg, 1989).
Isolation from others may also increase (private) self-awareness in ways predicted to increase adherence to salient norms or standards (Carver & Scheier, 1981). Once again the effects of self-attention depend crucially on whether a social standard (group norm) or an individual standard (personal norm) are contextually salient (cf. Diener, 1980). In both these explanations then the theoretical distinction between the personal and the social level of identity is central. Both the de-individuation and the self-attention route to the norm directed group polarization predicted here can be considered within the overarching framework of social identity theory, and the theory of Referent Informational Influence. The present experiment, where we actually made no measure of self-awareness doe s not allow us to arbitrate between these two mechanisms, but as we have argued elsewhere these are likely to be complementary and conceptually related (Spears et al., 1990). Establishing their respective contribution must be the subject of future research which attempts the by no means straight-forward task of manipulating the anonymity and isolation components of CMC independently from each other (Spears, 1990). Our model incorporating both these two routes to the social influence effects in group polarization is summarised in Figure 5.
[Insert figure 5 about here]
In sum, it is arguably precisely because of the lack of social cues (qua personal information or interpersonal signals) that the respective norms or standards may become so salient in CMC. Ironically, people may only have their social or personal standards to go by in the absence of other cues. As suggested earlier, the notion that CMC is somehow less social or antinormative is no accident but follows directly from the prevailing “interdependence” notion of the group (cf. Hogg, 1987; Hogg & Turner, 1987; Turner, 1982). According to this traditional view the group is defined as a structural entity comprising interdependent individuals, bonded by ties of cohesiveness or mutual attraction (see e.g., Hogg, 1987 ;for a review and critique), fulfilling functional roles (cf. Bales, 1950). From an interdependence perspective then it is quite logical to assume that physically removing people from the group somehow undermines or even eliminates the group an d its associated norms. The fact that traditional theory of normative influence holds that conformity to a group norm depends upon surveillance by physically present others— the so-called dependency theory of social influence (Moscovici, 1976; Turner, 1991)—is a direct consequence of this interdependence conceptualization. The social identity approach rejects the individualistic metatheory of interdependence, viewing social categorization and social identification as the critical mediators of group processes (Hogg, 1987; Turner, 1982, 1991). In these terms the group is conceptualized in socio-cognitive, and not first and foremost in structural or relational terms. This then allows for the impact of group influence independently of the co-presence of others, and also for the influence of more general social categories defined independently of individuals per se.
It is important to distinguish this view of the social founded in salient group or social identity from some notion of general societal norms (implicit in the “antinormative” accounts of CMC behaviour) or from the generic norm associated with the computing context. These latter accounts are descriptive and are ill-placed to account for variations in outcomes or behaviour. Locating the norm in the group itself however offers a more more flexible and heuristic approach; particular groups may differ widely in their local or contextual norms and still produce “normative” behaviour in their members. In these terms, flaming behaviour may be much more amenable to a normative explanation than hitherto supposed (see Lea et al., chapter 5 this volume). Such norms do not have to be exclusively or even most powerfully communicated by physically present individuals, nor are they simply part of the context. Rather they are specified in the content of the group stereotype or social identity.
A reappraisal of previous research findings
Although our research supports the SIDE Model and provides problems for established approaches, it still forms a somewhat narrow empirical base and there remains much to be done to test and develop this line of inquiry (Lea, 1989, 1990; Lea & Spears, 1992, in prep;.; Spears, 1990). It is nevertheless possible to reconsider previous research to see if the present model can account for findings that have proved problematic for these other approaches. Restrictions of space prevents a detailed review in this respect, so what follows is selective and focuses on findings discussed earlier. We address these models in the same order as before.
In terms of the social presence Model it was assumed that media with the highest social presence (i.e. face-to-face communication) would provide the greatest facility for social influence. In fact in the series of studies conducted by Short (1972a,b,c, 1973) quite the reverse turned out to be the case—a telecommunication condition consistently produced more influence than face-to-face. As should now be apparent, this pattern of findings provides no problems for our own approach, and more strongly, is quite consistent with it. In our model, the impact of the social is not related in any simplistic or mechanical way to “presence”, and may often be inversely related to the tangible presence of other individuals. Indeed the intercom condition models in many respects the de-individuating conditions of our own experiment. It need only be assumed that the target shared some common category identification with the communicator for us to predict more effective social influence than in a face-to-face context.
The results of much research assimilated by Rutter under the umbrella of his cuelessness model are ripe for a similar reappraisal. Recall the simulated industrial bargaining studies in which the relatively cueless sound-only communication medium gave the advantage to the side with the stronger case, compared to face-to-face. In terms of our model, the social identity associated with the negotiators’ “side” may have been particularly salient, and also uncomplicated by interpersonal considerations under relatively cueless conditions (as Morley and Stephenson originally suggested). The fact that the advantaged party pressed home their advantage to victory is then perfectly consistent with the demands of their role. Thus while Rutter may be correct in supposing that the cueless environment lends itself particularly well to “interparty” considerations, his interpretation that this reflects an immersion in the “task” unnecessarily attempts to relocate the cause somewhere beyond the manifest social level of this phenomenon. The fact that the cueless environment is more open to interparty considerations may more parsimoniously reflect that the participants simply perceive the context more in interparty terms (cf. Stephenson, Ayling and Rutter, 1976).
Further studies confirm that a recurring effect of cuelessness is to accentuate conformity to salient social norms, roles or standards (Rutter, 1984; Rutter & Robinson, 1981; Stephenson, Ayling & Rutter, 1976). For example in a ‘teaching by telephone’ study Rutter (1984) found the same 2-way interaction between medium (face-to- face vs. telephone) and role (student vs. tutor) on no less than eight dependent measures such that in the telephone condition ‘tutors became “more like” tutors, students “more like” students’ (Rutter, 1987;, p. 160) compared to face-to-face. This is very reminiscent of our own findings where it is no exaggeration to say that individuals became “more like” individuals, and group members “more like” their group, under de-individuating conditions. Other research in the tradition of de-individuation theory also shows that isolation and anonymity can produce more normative behaviour (Zimbardo, 1969; see Spears & Lea, 1991). These findings correspond well with the social identity interpretation and nicely illustrate the paradoxically “social” and normative nature of the apparently cueless environment.
The finding that face-to-face communication became more effective in bargaining when participants had to argue their own position (e.g., Short, 1974) may reflect the fact that such conditions tend to render personal identity salient; visibility may enhance the sense of personal accountability in defending one’s own position. However we suggested earlier that the de-individuating conditions associated with anonymity can also enhance the salience of individual as well as group identity. Since participants’ individual identity was not explicitly manipulated in the studies reviewed by Rutter (1987), we can only speculate that face-to-face interaction which involves arguing one’s own case may be particularly likely to render personal identity salient, but that if personal identity has already been made salient then de-individuating conditions may accentuate its effects. Nevertheless, the impact of accountability and self-presentational factors in visible communication should clearly not be underestimated. The same interpersonal impression management factors which may encourage compromise in bargaining face-to-face on behalf of a group, may also underlie a more competitive stance when one’s own position and credibility as an individual is on the line. Thus taking into account the nature of the normative context (competitive vs. co-operative), its specific relation to salient identity level (personal vs. group), and the functions of visibility and anonymity associated with the communications medium may help to explain the apparently anomalous outcomes in this research tradition. Like Morley and Stephenson before her, Schruijer‘s (1990) failure to distinguish adequately between the salience of personal vs. group identity and tie norms to these identities may help to explain the inconclusiveness of her attempted integration. Unfortunately a full theoretical integration is beyond the scope of the present chapter.
Reduced Social Cues
We have already provided an alternative explanation to the RSC account of enhanced group polarization in CMC as well as detailing its difficulties in explaining our own results. Here we attempt more explicitly to accommodate their results within the framework of the SIDE Model (see also Lea & Spears, 1991).
First of all we would liken the CMC conditions of the studies reported by Kiesler et al. (1984) and Siegel et al. (1986) to our own de-individuated group identity condition. Subjects in these conditions were visually anonymous and isolated from each other, and they were drawn from a single homogeneous student pool, which would seem to guarantee a relatively cohesive group context and a basis for shared social identity. The face-to-face conditions with people in full view of each other were by contrast more comparable to the face-to-face or ‘individuated’ conditions of our own experiment which also resulted in less group polarization. In sum these findings are perfectly consistent with our own analysis.
However, it should be recalled that in a subsequent study by McGuire et al., (1987) less rather than more group polarization was obtained in a CMC condition compared to a face-to-face condition. Passing over certain differences from the standard group polarization paradigm, the obtained effect may have been due to a number of features of the experiment which could have weakened the salience of the group and accentuated the individuality of participants. It is impossible to tell from the details of the study whether there was an established group identity or norm for the sample of managers and administrators that participated in the experiment. However, the fact that the most of the participants were familiar to each other, that their names were displayed throughout, and that they could see each other during the session, would all be predicted to have further individuating effects. The fact that participants were drawn from two samples, namely managers from the same organization and university administrators, may have compounded the problem. It meant that mo re than half of the experimentally created groups comprised subjects from two different backgrounds with the result that one level of group categorization (occupation) was cross-cutting another (the experimental discussion group). In this context it is possible to argue that competing group loyalties and group norms may have been circulating such that one faction polarized away from the other producing “bipolarization” (Paicheler, 1977) rather than group polarization proper. These and other problems (Lea & Spears, 1991) all suggest that the lack of group polarization in the CMC condition may have been caused by undermining the participants sense of being part of a cohesive group with a common identity and goals. As we have tried to argue in our own research, it is not the use of CMC per se that results in group polarization, but the favourable conditions for the salience of group identity that CMC conditions afford (anonymity, isolation and reduced interpersonal cues).
Taking our own research together with the reappraisal of previous research findings, we believe there is now mounting support for the explanation of social influence effects and for social psychological processes more generally within CMC, in terms of the SIDE Model. Of course much of the re-interpretation of previous work is post hoc and provides no substitute for empirical inquiry designed to test the limits of our own approach. Nevertheless, the success of this framework in accounting for the distinctive and often apparently contradictory results in this field provides an encouraging start. In particular the model’s capability to predict the incidence or absence of polarization effects, based on the contents of salient norms or standards, gives it far greater flexibility and explanatory power than its current competitors which lapse into ad hoc and eclectic explanation in an effort to account for the range of effects and outcomes associated with CMC.
In this chapter we have attempted to show the that there is a very important social dimension to CMC which has been neglected by previous approaches in what we have called the “social cues” tradition. These approaches assume that because many social cues a re excluded in CMC, the social dimension is itself partialed out. We have argued that the tendency to neglect the social dimension has arisen from a theoretical blindness deriving from individualistic interdependence conceptions of social behaviour in which the interpersonal dimension of interaction is confused with the social per se. Once this equation is made it then becomes logical to assume that the social context impinges less on behaviour within CMC and there is then no reason to consider the distinctive social and normative contexts which we argue shape group behaviour in much experimental work in this field.
Our analysis based on social identity theory posits that the interpersonal and the social (qua intra and intergroup relations) can be conceived as opposing poles, evoking different levels of identity and their associated norms. Moreover we argue that the salience of the social as opposed to the interpersonal is far less sensitive to ‘information richness‘, ‘bandwidth’ or the number of cues afforded by the communication channel. Indeed the salience of interpersonal cues may actually undermine the salience of social categorical information by individuating participants. However, just as the salience of social or group identities may be strengthened under typical conditions of CMC (anonymity, isolation), our research suggests that under different circumstances personal identity can be also strengthened by these conditions (Although we would quickly add that one’s personal identity may be more chronically salient in much face-to-face interaction.) We argue that the effects of both social and individual aspects of behaviour can be accentuated by CMC and that, within these terms, the critical social psychological differences between CMC and face-to-face communication are of degree rather than of kind. To paraphrase a famous quote of Floyd Allport (1924) referring to the effects of immersion in the crowd, it is possible to think of behaviour within CMC as being like ordinary behaviour only “more so”. Unlike Allport’s original reference however, we do not take this to imply some regression to a more natural antisocial and uninhibited state as some have implied. We assume that the social is not something simply achieved or maintained in social interaction, but forms the very constitution of being. Viewed in these terms, interpersonal interaction can be seen as an overlay on the default social bases of behaviour which come more to the fore in the absence of these “social” (interpersonal) cues. If people behave in an uninhibited fashion within CMC, we suggest that this too may be normatively produced and directed (cf. Lea et al., chapter 5 this volume); apparently extreme or idiosyncratic behaviour can also reflect group norms as much as individually motivated or antinormative tendencies.
We should stress that we are not trying to deny the importance of social information within CMC, or rather its absence. We are however criticising simplistic linear and quantitative models of information exchange which concentrate on the number of cues or their “richness” and fail to consider the meaning of information for the communicators. Meaning does not depend purely on information still less on the amount of information exchanged. Given our limited attentional capacity, the salience of certain information tends to rule out the salience and use of other sorts of information. Information conspicuous by its absence (e.g., individuating interpersonal cues) can equally have informational significance and influence. The traditional transmitter-decoder model of communication implies that only information actually exchanged in the medium counts, but communication does not take place in a social vacuum (Tajfel, 1972). A more semiotically grounded approach allows for the availability of information without any active exchange in the medium at all. In these terms, the question is not the quantity or lack of information, but what sort of information is contextually salient (or absent), and its meaning for those involved. The idea that CMC is somehow informationally impoverished and that this cons trains effective communication is then open to reappraisal. Formalistic and objectivist models which take a narrow empirical definition of social information and ignore the complex relation between such information and what it signifies, are doomed to flounder in a morass of complex and apparently contradictory effects.
Our approach also challenges attempts to reify the effects of CMC and fix them in some general socio-psychological syndrome or “nature” associated with this technology. We argue that behaviour in CMC, as elsewhere, cannot be understood abstracted from its constituent social context. For this reason we regard as seriously misleading the distinction between the “task-oriented” nature of CMC compared to the more social nature of face-to-face communication that has been a recurring theme in the social cues perspective. Our own work apart, recent research suggests that CMC may be just as much a medium of social as of task-related communication. More fundamentally we would argue that the very opposition between task-related and social aspects of CMC is itself a misleading one that once more derives largely from interpersonal and interdependence conceptions of the social dimension. Defining task-oriented behaviour in opposition to the social, appears to rule out socially regulated task-oriented behaviour tout court, so that the social becomes trivialised in comparison with more serious task-related matters, though providing some functional or therapeutic value. According to our approach task-related uses are just as much infused by the social and normative context, and these may even be more social in the group sense. We would argue that a neglected distinction is not so much between task and person or task and social, but between the interpersonal and the social. Viewed in this way effects and outcomes do not have to be reified in terms of effects or attributes of CMC, or immersion in the task (which in any case may have a strong social dimension to it), but can be seen in terms of their constituent inter personal and social relations. CMC is then quite literally no more than a medium for these, although one which can effectively focus socially determined effects.