One of the crucial issues concerning stereotypes is whether it should be assumed that they are individual beliefs or whether they should be seen as inherently collective. As Schneider (1996) puts it: “the issue is, of course, not whether some beliefs about other groups are held widely – clearly many are – but rather whether this should be a defining feature of stereotypes” (p. 430). In other words, does it make sense to claim that an individual, qua individual and independently from any group membership, holds a stereotype of a social group? Or should the term “stereotype” be applied only to those beliefs about social groups that are widely shared?

The social cognition approach (Ashmore & Del Boca, 1981; Fiske & Taylor, 1991) tends to account for stereotypes as products of individual cognitive processes. This assumption has been explicitly endorsed by Mackie, Hamilton, Susskind and Rosselli (1997, p. 41) who state that a stereotype is a cognitive structure, “meaning that it resides in the head of the individual perceiver. In contrast to other definitions that emphasize the shared nature of stereotypic beliefs systems, in our view the degree of consensus of any given belief is an empirical issue, rather than a definitional property” (p. 42). These authors justify their opinion by arguing that each individual’s experience is unique, therefore each individual’s social stereotype might also be different. They contend that the reason that the content of many social stereotypes tends to become widely shared among members of various groups is that stereotype formation is massively influenced by a common social context.

This view has been criticized by proponents of self-categorization theory (SCT; Haslam et al. 1998; Oakes, Haslam, & Turner, 1994). For instance, Haslam (1997) defines the social cognition approach as a “common informational input model of stereotype sharedness”, since it suggests that “shared stereotypes arise from shared experiences that provide the common informational input upon which generic processes of information-processing and social learning set to work” (p. 125). He rejects this model and argues that the idea that two people who are exposed to identical experience will have identical reactions is not only theoretically but also empirically questionable. For instance, the same racist propaganda has different effects on racists and non-racists.

The alternative proposal of Haslam and colleagues is that, rather than deriving from exposure to similar information, shared stereotypes result from common group membership that structures information-processing. More specifically, it is argued that when a collection of individuals share the same subjective identification with a social group, they will be depersonalized (they will feel similar and interchangeable on identity-relevant dimensions), and therefore will expect to agree on issues concerned with group identity, including their representation of the ingroup and relevant outgroups. This expectation to agree will turn into a motivation to reach agreement, and so the group members will engage in a process of discussion and argumentation which will tend to eliminate disagreement and to lead to a common perspective.

These ideas imply that stereotype consensus should be enhanced (1) by factors which increase the salience of the shared social identity of a group of perceivers, and (2) by group discussion premised upon the shared social identity and relevant to the stereotype, as this gives rise to mutual social influence (Haslam et al. 1998).

To test these hypotheses, Haslam et al. (1998) asked Australian participants to stereotype Australians (the ingroup), either compared to Americans (intergroup context) or without comparing them with any other group (intragroup context). Moreover, participants were asked to perform this task either individually or via group discussion. Consistent with predictions, it was found that (a) Australian participants were more consensual when stereotyping Australians compared to American people (intergroup context) than when stereotyping only Australians (intragroup context), and that (b) when stereotypes were formed via group discussion, rather than individually, both the stereotype of Australians alone and the stereotype of Australians compared to Americans were more consensual than when they were formed individually.

The present authors believe that the view held by SCT is an enlightening one, but they also contend that it remains to be established whether those findings can be generalized to other social groups and contexts. Our aim in this paper was to test SCT’s view of stereotype consensus using less inclusive social groups (i.e., “students” as the ingroup and “managers” as the outgroup) and exploring stereotypes based on a more specific and perceptual dimension (i.e., “dressing style” rather than “psychological traits”). The assumption was made that, because the way people dress is often related to their collective identity (Davis, 1992; Joseph, 1986) and because dressing styles may be used as a potent symbol of shared beliefs and values, as well as a device to clarify and perceptually accentuate differences between members of different groups (Hatch, 1993; Pratt & Rafaeli, 1997), the stereotypes of student and managerial dressing style should become more consensual when such stereotypes are formed within an intergroup context, and when they are created via interaction premised upon shared social identity.

At the start of each experimental session participants were given a checklist modeled on that compiled by Katz and Braly (1933). This checklist consisted of 52 adjectives related with the way in which people may dress (e.g., trendy, fashionable, funky, glamorous, gothic, biker). Depending on the condition to which they had been assigned, participants were asked to describe the way in which either “students” or “students in contrast to managers” or “managers” typically dress. They did so by underlining the 16 adjectives that described the target group and then marking with an “X” the five words most typical of people in that group. Subsequently, participants in each condition were divided up randomly into small groups of three individuals, and each group was given a single checklist. They were asked to discuss their target group for 10 minutes, and to consensually select five adjectives to be assigned to the target group by following the same procedure as in the individual phase.

The adjectives chosen to describe students in the intragroup context were less consensually endorsed than were both those selected to describe managers (implicit intergroup context) and those selected to describe students in contrast to managers (explicit intergroup context). Furthermore, this difference was exceptionally marked at the group phase. On average, 35% of individuals and 46% of groups endorsed the five most commonly selected adjectives for students in the intragroup context, whilst the first five adjectives for managers (implicit intergroup context) were assigned by 63% of individuals and 74% of groups, and the first five adjectives for students in the explicit intergroup context were assigned by 43% of individuals and by 62% of groups.

In order to provide appropriate analysis of these results, a measure of the extent to which the adjectives selected by each participant in the individual phase, or by each group of three participants in the group phase of the experiment, contributed to intersubject (or intergroup) consensus in the stereotype that arose in each condition was calculated. In accordance with the procedure used by Haslam et al. (1998), the consensus coefficient was obtained by counting the number of times each adjective selected by a participant (or by a group) was also selected by other participants (or other groups) in the same condition, and by dividing this number by the maximum number of times that those adjectives could have been selected by other participants (or other groups) (i.e., N - 1). Thus, the consensus coefficient has a minimum value of 0.00 and a maximum value of 1.00. However, in order to allow comparison between the two phases of the study, consensus coefficients of participants in the individual phase of the study were averaged. See Table 1 for adjectives assigned to groups and for the mean consensus in each condition.

Scores were subjected to a 3 (stereotyped group) × 2 (experimental phase) analysis of variance with repeated measures on the second factor (where N = 10 in both independent conditions). This analysis revealed main effects for both stereotyped group (F(2, 27) = 88.53, p < 0.01) and experimental phase (F(2, 27) = 58.36, p < 0.01), as well as a two-way interaction between stereotyped group and experimental phase (F(2, 27) = 4.56, p < 0.01). Analysis of simple effects showed that in all three conditions stereotypes were more consensual at the group phase than at the individual phase (students: Ms = 0.19, 0.10 respectively; t(9) = 2.41, p < 0.05; managers: Ms = 0.27, 0.13 respectively; t(9) = 4.58, p < 0.001; students-not-managers: Ms = 0.32, 0.13 respectively; t(9) = 6.09, p < 0.001.