In a seminal study by Devine (1989), the processes involved in activating stereotypic information were investigated. Devine firstly endeavoured to tap knowledge of a racial stereotype. To do this, she employed a free response measure that required participants to generate a list of features that they associated with the target stereotype of Blacks. On the basis of previous stereotype assessment literature, participants were found to list features typical of the Black stereotype. As the features listed were similar across participants Devine was able to establish that the content of this stereotype was commonly known to participants. In order to ascertain participants' explicit beliefs about the racial category, Devine employed an attitudinal scale which contained stereotype-related statements associated with the category of Blacks (e.g., Black people are generally not as smart as whites; McConahay, 1986). In administering such a measure, Devine found that participants varied in the extent to which they endorsed features of the racial category.
Subsequently, Devine set out to determine whether stereotypic information was available at the level of automatic activation. Participants were parafoveally primed at a level below conscious awareness with a set of features pertaining to the Black stereotype. Priming below the level of awareness was employed in order to negate the possibility that controlled processes would be involved in participants' responses. Each feature was presented for a period of 80 ms, and participants were required to indicate whether the prime word appeared on the left or right-hand side of a screen. Some of the features employed as priming words were: nigger, afro, poor, lazy, musical, and unemployed. Recognition of primes was subsequently tested, and indicated that participants had no conscious awareness of the content of the primes. In one condition participants were primed with a set of traits in which 20% were stereotype-related, and in another condition, the frequency of priming increased to a level in which 80% of the items were stereotype-related. This was done to determine whether temporarily increasing the accessibility of stereotypic information by increasing the frequency of priming, had any effect upon subsequent responses. To examine the effect upon subsequent responses, a measure known as the Donald paradigm was employed. In this, participants were required to evaluate a set of ambiguously hostile behaviours displayed by a person in a vignette. It was found that participants' ratings of the hostility of Donald's behaviour was higher in the condition that contained 80% stereotype-related primes, than in that which contained 20% of these primes.
This study suggested that for all participants priming with features of the stereotype activated a link between the Blacks category and hostility. It also suggested that as participants' responses were found to be a function of priming frequency, the priming had activated certain knowledge. As the priming material was stereotype-related, this activated knowledge was assumed to be knowledge of the racial stereotype (Higgins, 1998). Moreover, this effect was obtained across participants whose attitudes were measured to be both high and low in prejudice toward the racial stereotype.
In order to further explain the overall findings, Devine drew upon a model developed by Dovidio et al. (1986). This model holds that when people encounter a member of another social group, well-learned associations facilitate an automatic activation of the relevant stereotype. The model further contends that this activation occurs irrespective of whether personal attitudes are high or low in prejudice toward the social group (Dovidio et al., 1986). Hence, while stereotypic information may be automatically activated uniformly across people, it is also the case that attitudes towards social group members may show individual differences (Dovidio et al., 1986). In this way, the model advocates a dissociation of automatic and controlled cognitive processes, wherein implicit knowledge that is activated in an automatic or spontaneous manner is seen to operate independently from explicit beliefs which result from intentional or controlled thought processes (Wittenbrink, Judd & Park, 1997).
In addition to providing support for this model, Devine (1989) also found that participants who possessed stereotypic knowledge at an automatic level were able to express nonprejudiced attitudes in a condition allowing for controlled processing. As such, it suggested that implicit stereotypic knowledge could be inhibited in favour of nonprejudiced attitudes. Devine was thus able to argue against the inevitability of prejudice perspective which views the holding of prejudiced attitudes as an inescapable consequence of stereotyping (Lepore & Brown, 1997). However, it must be noted that in conditions where time pressure, high cognitive load, and little motivation prevails, the tendency for those who possess implicit stereotypic knowledge to express prejudiced attitudes is increased (Fiske, 1998). In order to illustrate these findings further, a model based upon Devine's assumptions has been developed (Table 1).
|The Nature of Knowledge/Attitudes|
|Automatic Processing||Stereotypic Knowledge||Stereotypic Knowledge|
|Controlled Processing||Prejudiced||Non- prejudiced*|
|Predicted Attitudes||Prejudiced||Non- prejudiced*|
* If motivated, and time pressure and cognitive load are minimal.
This model serves to differentiate people's attitudes on the basis of the alignment between their implicit knowledge of and their explicit beliefs about social groups. Moreover, based on the alignment of these constructs, the model provides predictions of the type of attitudes likely to be exhibited.
Devine's study (1989) did, however, invite criticism with respect to her methodology (Lepore & Brown, 1997). It was noted that many of the primes employed to activate stereotype-related information, were negatively valenced stereotypic traits. As such, Lepore and Brown indicated that the priming procedure did not enable an unbiased spontaneous activation of the participant's implicit stereotypic knowledge. They further noted that Devine's set of primes included both traits (e.g., lazy), which serve to prime the stereotype, and category labels, (e.g., African-American) which serve to prime the racial category. Thus, it was unclear as to whether her findings may have resulted from priming the stereotype directly, or from priming the racial category.
In order to address the methodological problems evident in Devine's study (1989), Lepore and Brown (1997) conducted contrasting studies: one which employed category labels and stereotype-related traits as primes and the other, category labels and neutral associates of the category. Prior to this, however, they measured attitudes toward the category of Black people. In using a scale similar to that employed by Devine, they found that participants varied in the extent to which they endorsed the stereotype content as being representative of their explicit beliefs about the stereotype. They then conceptually replicated Experiment 2 of Devine's study (1989) by parafoveally priming participants with category labels (e.g., Afro-Caribbean), negative trait words (e.g., dirty), and words evoking the category (e.g., reggae). In this, the components of the priming set were similar to those in Devine's study.
In the priming procedure, each prime was presented for 100ms and was followed by a 100ms presentation of a letter string mask. As individual prime words were repeatedly exposed, this mask was used to reduce the residual effects of repeated exposure to the same prime. After exposure to the primes, participants were required to read a vignette which described a set of behaviours displayed by a target person. They were then required to rate the target person on four dimensions (athletic, fun-loving, unreliable, aggressive).
Support was found for Devine's results in that the stereotypic primes increased the negativity of ratings of the target person in the subsequent task for those with high and low prejudiced attitudes. More importantly, however, when participants were parafoveally primed only with category labels (e.g., West Indian), differential responses were obtained in line with participants' attitudes. In this condition, participants low in prejudice rated the target person more positively, and those high in prejudice increased the negativity of their ratings, when compared to a condition in which no primes were used. In essence, Lepore and Brown found that when the content of the stereotype was primed directly, similar responses were produced in both high and low prejudiced people. Priming with category labels, however, resulted in differential ratings between these two groups. In an attempt to explain these findings, Lepore and Brown argued that as the stereotype of the target group was predominantly negative in content, priming with stereotype-related words served to automatically activate negative associations and affect across participants. As priming the category resulted in more negative ratings for participants with high as opposed to low prejudiced attitudes, this suggests that at the automatic level, links between the stereotype and the category are stronger for those whose attitudes are higher in prejudice toward the target group. Thus, as Lepore and Brown found that participants' implicit knowledge differed according to whether the prime used was evocative of the category, or of the stereotype, it suggests that there is more variation in the stereotyping process than is credited by Devine's model.
Devine's model has further been challenged by other studies that have found implicit knowledge to be correlated with explicit attitudes (Augoustinos et al., 1994; Fazio & Dunton, 1997; Locke et al., 1994). In attempting to tap implicit knowledge of the Australian Aborigine stereotype, Augostinous et al. (1994) employed a response latency paradigm, in which the speed at which participants indicated endorsement of positive and negative stereotype-related words was measured. The theory underpinning response latency measurement is cognitive psychological in nature, and essentially posits that well-learned sets of associations are activated more readily or are more accessible when interpreting incoming information than newer sets of associations (Hilton & von Hippel, 1996; Wittenbrink et al., 1997). As implicit knowledge associated with a given social category is argued to be well-learned (Allport, 1954; Devine, 1989), faster responses to stereotype-related information than to -unrelated information, suggests that implicit knowledge of the stereotype has been activated (Banaji & Hardin, 1996). This type of activational process differs to that employed by both Devine (1989), and Lepore and Brown (1997), as it aims to access chronically accessible information rather than access information that has been temporarily activated by frequent priming (Higgins, 1998). Using the theory underlying response latency measurement, Augostinous et al. inferred that as participants with attitudes high in prejudice were faster to respond to predominantly (negative) features of the Aboriginal stereotype than participants with low prejudiced attitudes, implicit knowledge of the cultural stereotype was more available in those with high prejudiced attitudes. Furthermore, those low in prejudiced attitudes responded faster to positive descriptions of the stereotype than did those whose attitudes were high in prejudice. These results suggested that the two participant groups varied in their endorsement of stereotypic knowledge at an automatic level, in the direction of their explicit attitudes. While these findings clearly differed to those obtained by Devine (1989), it is important to note that the methodologies employed to access implicit stereotypic knowledge, also differed between studies.
In addition, a study by Locke et al. (1994) found that participants with highly prejudiced attitudes towards the Australian Aborigine showed greater interference in a Stroop task, to stereotype-related trait words than to unrelated trait words, whereas those low in explicit prejudice showed no difference in interference between stereotype-related and -unrelated trait words. Contrary to Devine's assertion that those with non-prejudiced attitudes tend to inhibit their implicit stereotypic knowledge, Locke et al. found that stereotypic knowledge seemed to be less available at an automatic activation level for participants whose attitudes were low as opposed to high in prejudice. This finding was further bolstered by results obtained from participants who had little knowledge of the Australian Aborigine stereotype (Singaporeans recently arrived in Australia). In this sample, there were participants whose attitudes were low as opposed to high in prejudice, and who further failed to demonstrate implicit knowledge of the stereotype at the automatic level. Hence, for these participants the expression of non-prejudiced attitudes did not require an inhibition of implicit stereotype information (Locke et al., 1994). This finding can be seen to add another dimension to Devine's model. In addition to there being those with implicit knowledge of the stereotype who have attitudes which are either high or low in prejudice, there appear to be those who posses non-prejudiced attitudes that are congruent with their implicit knowledge of the stereotype (Table 2). The importance of this is that prejudice is not inevitable, even under conditions thought to allow only automatic processing.
|The Nature of the Attitudes/Behaviour|
|Automatic Processing||Stereotypic Knowledge||Stereotypic Knowledge||Stereotypic Knowledge|
|Controlled Processing||Prejudiced||Non- prejudiced*||Non- prejudiced|
|Predicted Attitudes||Prejudiced||Non- prejudiced*||Non- prejudiced|
* If motivated, and time pressure and cognitive load are minimal.
Research suggests that the degree of individual variability found in the implicit knowledge and explicit attitudes of racial stereotypes extends to those of gender stereotypes (Blair & Banaji, 1996; Banaji & Hardin, 1996; Foddy, 1997; Hadjion; 1997; Stroupe & White, 1996). In a study by Banaji and Hardin (1996) it was found that while participants differed in their explicit endorsement of gender stereotypes, differences were also apparent in the measurement of implicit knowledge. To ascertain implicit knowledge, they employed a response latency paradigm wherein the primes male or female were presented below the level of awareness (for 200ms) and were then followed by masculine or feminine pronouns. It was found that female participants were faster to determine that she was a pronoun than they were to determine that he was a pronoun, and male participants were similarly faster in their responses to he than she irrespective of the nature of the preceding prime. In this study, attempts were also made to deduce the boundaries within which stereotypic priming would elicit activation of implicit stereotype knowledge. It was found that primes denoting gender (e.g., mother, father) elicited faster response latencies than primes that merely connoted gender (e.g., doctor, nurse). Faster response latencies to consistent gender label-pronoun pairs (e.g., mother-she) were also produced when the judgement task that succeeded presentation of the prime was altered from a gender irrelevant lexical decision (pronoun or not pronoun?) to a gender relevant one (male or female pronoun?). Overall, these findings suggest that implicit stereotypic knowledge may be differentially available across people. They further suggest that this knowledge is activated as a function both of the types primes employed, and of the nature of the task following presentation of the prime. In the context of developments in stereotype research, these factors can be seen to add further complexity to the debate concerning what, in fact, constitutes an adequate measure of implicit knowledge.