Journal of Experimental Social Psychology Journal of Experimental Social Psychology 40 (2004) 239–246 www.elsevier.com/locate/jesp

Graded structure and the speed of category verification: On the moderating effects of anticipatory control for social vs. non-social categoriesq Kristina R. Olson, Alan J. Lambert,* and Jeffrey M. Zacks Department of Psychology, Washington University, 1 Brookings Drive, St. Louis, MO 63130, USA Received 26 August 2002; revised 12 June 2003

Abstract Speeded classification tasks have widely been used as leverage to discern whether exemplars are considered to be typical or atypical members of categories. However, little is known about the possibly different nature of the mechanisms that underlie such reactions for ‘‘social’’ vs. ‘‘non-social’’ categories. We presented participants with typical and atypical exemplars to classify and manipulated the opportunity that participants had to control their responses by varying whether stimulus onset asynchrony (SOA) was short (350 ms) or long (2000 ms). For non-social categories (e.g., BIRD), participants were faster to classify typical instances than atypical instances, regardless of SOA. In contrast, we predicted and found that manipulating SOA strongly moderated the pattern of response times when motives for control were greater, as was the case when participants were presented with occupations relevant to racial stereotypes (e.g., DOCTORS and JANITORS). The implications of these findings, and their relation to effects observed with gender-based occupational stereotypes, are discussed. Ó 2003 Elsevier Inc. All rights reserved.

Introduction One of the important conclusions to arise from the categorization literature is that most categories have ‘‘graded structure,’’ in which some exemplars are perceived as better fitting members than others (Rosch & Mervis, 1975). For example, although both robins and chickens are birds, the former is considered a better example of birds than the latter. Although perceptions of typicality are interesting in their own right, research in this area has additional value insofar as it carries potential to yield more general information regarding the nature of social, as well as non-social, categories (Medin & Smith, 1981). In the realm of typicality perception, however, there is at least one important difference in terms of how people classify members of social q

The research reported in this article was part of an undergraduate research project conducted by Kristina Olson under the supervision of Alan Lambert and Jeffrey Zacks. * Corresponding author. E-mail address: [email protected] (A.J. Lambert). 0022-1031/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0022-1031(03)00098-2

vs. non-social categories. Chickens do not (as far as we know) care whether people regard them as atypical. However, human beings may be more likely to be personally invested in whether other people view them as typical or atypical category members. Suppose, for example, that you have just completed your MD at a prestigious medical school, and are now interviewing for a position at a local hospital. However, suppose that the interviewer considers White doctors to be more typical of the occupation ‘‘doctors’’ than Black doctors. This possibility is important, because a person who is perceived to be a ‘‘better fit’’ to the occupation may be more likely to be hired for this reason alone. This scenario becomes more disturbing insofar as the color of the personÕs skin could drive perceptions of atypicality, independent of any other knowledge or ascriptions about the personÕs personal qualities. In light of these considerations, therefore, there would appear to be theoretical as well as practical reasons why it is important to gain greater insight into the cognitive processes underlying these kinds of perceptions (cf. Glick, Zion, & Nelson, 1988).

240

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246

Theoretical background When participants are asked to indicate whether certain objects (e.g., robins) are members of certain superordinate categories (e.g., BIRDS), response times are reliably faster when participants are classifying typical compared to atypical targets (e.g., Barsalou, 1992; Rosch & Mervis, 1975). Such effects are usually presumed to reflect the stronger associations of typical compared to atypical exemplars with the category. Much of this literature has focused on ‘‘non-social’’ categories, that is, concepts that do not contain or refer to people. Different mechanisms may well emerge for social categories. For example, indicating that chairs are better examples of the category FURNITURE than beanbags is one thing, but indicating that Whites are better examples of DOCTORS than Blacks is quite another. In the latter case, such appraisals could be interpreted as unfair, arbitrary or even racist. For this reason, people may be reluctant to make such judgments overtly. More generally, when people verify social categories, motivation for self-regulation may play a more important role compared to non-social categories. The present research To explore these matters, we employed a withinparticipant design in which participants made classification judgments with respect to three types of categories: (a) non-social categories, (b) occupational categories relevant to racial stereotypes, and (c) occupational categories relevant to gender stereotypes. In all cases, participants were first presented with the category, followed by a picture of an exemplar. Their job was simply to indicate whether or not the image was an example of the previously presented category. This design also included a within-subject manipulation of stimulus onset asynchrony (SOA) between presentation of the category label and of the picture to be verified (350 vs. 2000 ms). Our manipulation of SOA is predicated on the assumption that controlled processing is resource dependent and, for this reason, more likely to operate under long, rather than short, SOA. The notion that SOA moderates opportunity for control is now widely accepted among cognitive psychologists and was first articulated in a classic study by Neely (1977). In that study, participants were first presented with a category label (BUILDING or BODY) which was then followed by either a word or a non-word; their task was to verify whether the second stimulus was a word or not. In that aspect of the study most relevant to present concerns, participants were told that when the category BUILDING appeared, they should expect to see a body part (e.g., heart) and that when the category BODY appeared, they should expect to see a building-related

word (e.g., floor). Hence, successful completion of this task requires that participants carefully attend to both the category and the target word in a strategic manner. As Barsalou (1992) has noted, such strategic processing may be seen as an example of goal-directed control that is necessary to override the pre-existing semantic associations present in long-term memory. For example, participants must overcome their natural tendency to think of building parts after being presented with BUILDING in order to do well on this task. Importantly, long SOA trials permit more opportunity to implement these goals. For example, participants who were presented with BUILDING responded relatively more quickly to heart than to floor. Theoretically, such effects reflect the relatively greater accessibility of goal-appropriate (heart) compared to goal-inappropriate concepts (floor). (For further evidence bearing on the relation between goal-directed intentions and cognitive accessibility, see Gernsbacher & Faust, 1995.) Short SOAs, on the other hand, did not allow participants sufficient opportunity to implement these goals. Instead, such trials showed the opposite pattern (e.g., faster responses to floor than to heart after presented with BUILDING), reflecting the relative ease of retrieving typical exemplars from memory. In essence, then, long SOA trials allow controlled processing to predominate, more so than for short SOAs. (For additional empirical evidence supporting these assumptions, and a review of the relevant literature, see Balota, Black, & Cheney, 1992.) Our research is, however, different from that of Neely (1977) in at least one respect. In the case of the Neely paradigm, the decision to be made—lexical decision– does not require consideration of the prime in order to render a judgment. In contrast, classification judgments force the participant to consider both the initial category (e.g., DOCTOR) as well as the target exemplar in order to make a response. Also, in our paradigm, the assumption that long SOA allows relatively greater anticipatory control does not necessarily mean that responses under short SOA were guided by purely automatic processes. Our argument is merely that retrieval of atypical exemplars takes more time than the retrieval of typical exemplars and, for this reason, participants have less opportunity to recruit atypical exemplars from memory under short SOA. For reasons articulated earlier, it seems likely that peopleÕs motivation for control should generally be higher when people are making decisions about social compared to non-social categories. If motivation for control is greater for social categories, then it follows logically that varying opportunity for control through varying SOA should likewise have a greater effect as well (cf. Fazio, 1990). The specific empirical predictions that follow from these matters are discussed in more detail below.

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246

Overview of predictions Non-social categories First consider verification of exemplars belonging to non-social categories, such as confirming that ‘‘chicken’’ is a BIRD. Presumably, varying opportunity for control matters only if motivation for control is relatively high. Hence, we did not expect SOA to make an appreciable difference in this case. Indeed, the only effect we expected for these trials would be for the tendency for participants to make faster classification judgments for typical compared to atypical exemplars, regardless of SOA. Occupational categories primarily related to racial stereotypes Now consider trials in which participants make classification judgments for occupations considered stereotypically ‘‘White’’ (e.g., DOCTORS) or ‘‘Black’’ (e.g., JANITORS). As for the short SOA trials, the pattern of RTs should generally reflect the kinds of overlearned stereotypic associations linked with racial categories (cf. Devine, 1989), resulting in a variant on the classic typicality effect. This translates to faster responses to White compared to Black exemplars for stereotypically White occupations, with the reverse pattern for stereotypically Black occupations. As for the long SOAs, however, we expected a complete reversal of this pattern. According to the logic of the Neely (1977) paradigm, relatively long SOAs allow participants some degree of anticipatory control (i.e., after the category had been presented, but before the target is displayed on the screen), permitting them to consciously ‘‘bring to mind’’ material consistent with their underlying goals. Given the nature of our largely liberal sample and their underlying motivation to be relatively free of negative stereotypic bias we expected that most of our participants would be motivated to think about counterstereotypic, rather than stereotypic, constructs (e.g., try to think of Black, but not White, doctors). If so, then this would lead to faster responses toward counterstereotypic (atypical) compared to stereotypic (typical) images. Occupational categories primarily related to gender stereotypes The preceding predictions assume that people are strongly motivated to avoid thinking and acting in stereotypic ways in the realm of race. However, there are at least two interrelated reasons why such motivation might be less pronounced in the realm of gender. First, Blacks constitute a more salient stigmatized group given their distinctiveness in the population vis-a-vis females

241

(McGuire, McGuire, Child, & Fujioka, 1978). Second, issues of race might be more ‘‘touchy’’ given that overt claims about the differences between the races are less likely, and constitute more of a social taboo, than claims about gender. The debate as to whether Whites are more intelligent than Blacks, for example, and whether such differences are rooted in biological vs. sociological factors, represents one of the most inflammatory issues of our day (cf. Hernstein & Murray, 1994). In contrast, although debates about the differences between the sexes obviously exist (Maccoby & Jacklin, 1974), many differences between men and women (e.g., average size and strength) appear to be accepted by most people as ‘‘real,’’ that is, rooted in biological/genetic factors. These matters led us to consider two empirically testable propositions, both of which received support. First, we tested for, and found, differences in participantsÕ self-reported motivation to control their reactions in the realm of race vs. gender (cf. Dunton & Fazio, 1997; Plant & Devine, 1998), showing that such motivation was higher in the former case. Second, and more important, the relatively low motivation for control in this case led us to expect that the overall pattern of RTs would more closely resemble the pattern expected on the non-social trials. Summary of predictions We predicted a three-way Type of Category (nonsocial vs. race-based vs. gender-based categories)
Typicality (Typical vs. Atypical)
SOA (short vs. long) interaction. For non-social categories, RTs should be faster for typical compared to atypical exemplars, regardless of SOA. For race-based categories, the pattern of RTs should resemble those of the non-social categories, but only under short SOA; under long SOA, the reverse pattern should obtain. As for gender, we expected that motivation for control would be lower compared to racial categories (but perhaps higher than for the non-social categories). Hence, we generally expected a pattern similar to that of the non-social categories. In other words, it seemed more likely that we expected in this case that it would be more likely that RTs would be faster for typical compared to atypical exemplars, regardless of SOA.

Method Overview of design Participants made classification judgments about typical and atypical instantiations of non-social categories (e.g., BIRDS) as well as social occupations (e.g., DOCTORS). Regardless of whether participants were presented with typical or atypical targets, the correct

242

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246

answer on the trials of interest was always yes (e.g., robins and chickens are both birds, and White and Black doctors are both doctors). If participants were always presented with such (congruent) trials, they could easily anticipate their intended response, even before the target is presented. To avoid this problem, we included equal number of incongruent trials, in which the correct answer was no (e.g., the word ‘‘DOCTOR’’ paired with an image of a robin). We used a completely crossed design in which each given category label was paired with four types of targets: congruent typical (CT), congruent atypical (CA), incongruent typical (IT), and incongruent atypical (IA). For example, for each participant the category cue DOCTOR would be paired with a picture of a White doctor (CT), a Black doctor (CA), a robin (IT), and a chicken (IA) over the course of the experiment. Each participant was presented with a total of 16 unique categories, including four relevant to race (two stereotypically ‘‘White’’ and two stereotypically ‘‘Black’’ occupations) four relevant to gender (two stereotypically ‘‘male’’ and two stereotypically ‘‘female’’ occupations), and eight control non-social categories (e.g., birds, vehicles, weapons, fruits, vegetables, toys, tools, and sports). This yielded a total of 64 unique category–target pairings for each participant (16 categories
four target types) arising from a complete crossing of the categories and targets involved in the design. As for SOA, we used a counterbalanced within-subject design such that, for half of the participants, they initially completed a block of 32 trials at short SOA, followed by a block of 32 trials at long SOA; this order was reversed for the other participants. Finally, to maximize generalizability of our findings, we constructed two parallel sets of category cues and target items. In the end, all subjects saw the same 64 targets but they were matched with one of the two sets of categories. Preliminary analyses revealed a nearly identical pattern of results across these two sets (and across the counterbalancing of SOA blocks) and hence the main analyses are collapsed over these factors.

based on normative data reported by Rosch (1975). The occupational categories were selected from a pilot test conducted on a separate group of participants (n ¼ 10) who were asked to name the 10 jobs they most often associated with Blacks, Whites, males, and females. In almost all cases, the four most common responses were used.1 This yielded a total of 16 occupational categories, of which eight related to gender stereotypes (4 males and 4 females) and eight related to racial stereotypes (4 White and 4 Black). As with the non-social categories, the specific images of targets were colored line drawings of people in various occupations. Special care was taken to make sure that, for each type of occupation (e.g., DOCTOR), the corresponding typical and atypical exemplar (here, White vs. Black doctors) each contained a similar visual cue (e.g., a stethoscope) indicating the personÕs probable membership in the category in question. A manipulation check employed at the end of the study (see ahead for details) verified the viability of our assumptions in this regard.2 Procedure Classification task Participants were informed that: (a) they would be presented with a series of categories followed by a particular image, and that (b) their job was to indicate whether the stimulus was an example of the previously presented category by indicating ‘‘yes’’ or ‘‘no’’ on two marked keys on the keyboard. In all cases, category cues were presented for 300 ms. The delay prior to presentation of the target varied (50 and 1700 ms inter-stimulus intervals) yielding a short and long SOA of 350 and 2000 ms, respectively. Motivation for control Motivation for control in the realm of race was assessed on the basis of two distinct but conceptually related instruments, developed by Plant and Devine (1998) and Dunton and Fazio (1997). In all cases, participants

Participants The participants were 99 Washington University undergraduates (70 females and 29 males) who received course credit for their participation. We excluded four Black participants (all female), which left us with 95 participants (66 females, 29 males). None of the effects involving RTs were contingent on participant gender (all ps > :20) and hence the analyses are collapsed over this factor. Materials The non-social categories in our design and our selection of typical vs. atypical members of them, were

1 The one exception to this rule applied to the stereotypically male occupations. In this case, some of the most common ‘‘male’’ occupations listed were also listed among the ‘‘White’’ occupations, reflecting the fact that racial and gender stereotypes are not completely orthogonal. Although, we could not completely eliminate this confounding factor, for the stereotypically ‘‘male’’ occupations we picked the four most common responses that were not also listed in the White category. 2 On the race trials, all the pictures depicted males. As for the gender trials, 94% of the pictures depicted Whites. In both cases, therefore, the ‘‘non-focal’’ category membership of the target predominantly represented the ‘‘cultural default’’ (i.e., White and male). Although, it would be of interest to explore the potentially different effects arising from the various subcategories crosscutting race and gender (e.g., Black males vs. Black females), our design does not permit this kind of fine-grained analysis.

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246

responded to these items along a scale ranging from )3 (strongly disagree) to +3 (strongly agree). The Plant and Devine (1998) scale consists of 10 items, all of which refer to participantsÕ motivation to avoid acting or appearing in a prejudicial manner toward Blacks (e.g., ‘‘I try to hide any negative thoughts about Black people in order to avoid negative reactions from others’’). We constructed a parallel version of this scale to assess analogous motivation in the realm of gender stereotypes by simply rewording each of the 10 items to refer to women, rather than Blacks. The race vs. gender version of this scale appeared on separate pages. The Dunton and Fazio (1997) scale consists of 17 items that also assess motivation to avoid prejudice. However, only three items refer specifically to Blacks (e.g., ‘‘I feel guilty when I have a negative thought or feeling about a Black person’’). Hence, although participants completed all 17 items from the original Dunton and Fazio (1997) scale, we were mainly interested in the three items that pertain to Blacks. In addition to these items, we included three additional statements that represented modification of the three Black-related items to pertain to gender. Manipulation checks Subjective base rates Next, in order to verify our assumptions about the occupations, participants were provided with a list of all 16 occupations from the main task along with the following instructions: ‘‘In this task, we are interested in your personal experience with persons belonging to various occupations. Specifically, we are interested in the percentage of persons within any given category that you have personally met or seen that are female vs. male.’’ On the next page, participants completed the same sort of task, this time with respect to the categories of ‘‘Black’’ vs. ‘‘White’’ vs. ‘‘Other race.’’ For this task, we constructed four composite indices designed to capture the average perception of occupations deemed to be (a) stereotypically White (DOCTOR, LAWYER, POLITICIAN, and PROFESSOR), (b) stereotypically Black (ATHLETE, JANITOR, PREACHER, and COOK), (c) stereotypically male (CONSTRUCTION WORKER, POLICE OFFICER, FIREFIGHTER, and MECHANIC), and (d) stereotypically female (TEACHER, HOMEMAKER, SECRETARY, and NURSE). Verification of assumptions about social exemplars The last task simply presented participants with the same targets as presented in the main task and were given unlimited time to label each one as male or female or as Black or White. This task was presented in two blocks, one soliciting judgments of race for the targets used on race trials, and another soliciting judgments of gender for images used in the gender trials. Following

243

these two manipulation checks, participants were debriefed, thanked for their participation, and dismissed.3 Results Manipulation checks Perceived base rates Participants estimated that 88% of the stereotypically male occupations were held by males and 12% by females. This pattern was reversed for the stereotypically female occupations (17% vs. 82%). We expected a similar pattern for race, except that this pattern should be qualified by the large difference in baserate for Whites vs. Blacks in the population. For the stereotypically White jobs, participants indicated that 76% of the persons holding these jobs were White, and 12% were Black. In contrast, participants indicated that 48% of persons holding stereotypically Black jobs were White and 37% were Black. Note that the estimated percentage of Blacks in stereotypically Black occupations (37%) is nearly three times higher than those in stereotypically White occupations (12%). Interestingly, then, participants showed an apparent sensitivity to population baserates but, above and beyond this fact, confirmed our expectations about the differential familiarity with Whites vs. Blacks in these professions. Perception of typical vs. atypical social targets Recall that after the main task, participants were presented with each of the person targets and, for each one, were asked to indicate their racial or gender membership. The overall error rate in identification of the 32 social images was quite low (M ¼ 2:0 errors) and this was true for both the pictures relevant to race (i.e., those that varied the race of the person), (M ¼ 0:9), as well as gender, (M ¼ 1:1). Hence, the pictures we chose were clear and appeared to convey their intended category membership. 3 Following the classification task, participants completed the Modern Racism (McConahay, 1986) and Attitudes Toward Women scales (Spence & Helmreich, 1978). We included these measures for exploratory purposes, to investigate whether these attitudes, along with control motivation, would moderate the pattern of RTs. Although space does not permit formal consideration of these analyses, supplemental regression analyses on the race related trials revealed that both factors (attitudes and control) jointly moderated responses. Interestingly, there appeared to be one ‘‘type’’ of participant who showed especially strong typicality effects under short SOA, coupled with strong reversal under long SOA: those who scored high in control motivation, and had relatively negative views toward Blacks. This is consistent with our framework insofar as the classic typicality effect should be especially strong among participants whose personal attitudes are negative (Lepore & Brown, 1997) and the reversal under long SOA should be especially strong when motivation for control is also strong. A similar although weaker trend emerged on gender trials. (A full description of these analyses may be obtained from the first author.)

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246 Reaction times in milliseconds

244

Race-related categories

Gender-related categories

Non-social categories 1000

1000

1000

900

900

900

800

800

800

700

700 short

long

short

long

700

SOA

SOA

Typical Exemplars

short

long SOA

Atypical Exemplars

Fig. 1. Response times (in ms) for congruent RTs only as a function of SOA and Typicality for non-social categories (left panel) gender-related categories (middle), and race-related categories (right).

Category verification task Preliminary analyses Given that our main focus was on the speed of making correct (yes) responses on congruent trials, we initially coded such trials for whether the RT in question was derived from a correct or incorrect response. (Overall, participants made relatively few errors [i.e., ‘‘no’’ responses], M ¼ 2:24, or 7%.) (We excluded one participant because 55% of her responses were errors.) We then performed within-subject analyses on the correct responses only, in which we excluded all RTs that were faster than 200 ms, and more than three standard deviations from the mean for that participant. This trimming procedure led, on average, to the exclusion of 2.4% of participantsÕ congruent RTs. Very similar findings were obtained when we used a more stringent [2SD] cutoff for slow responses.4 Analyses of RTs We obtained strong support for the predicted Category Type
Target Type
SOA interaction, F ð2; 84Þ ¼ 4:60; p < :01; see Fig. 1. Additional analyses further confirmed our expectations. As for the non-social categories, we predicted and found that participants would respond faster to typical than to atypical targets, regardless of SOA, F ð1; 85Þ ¼ 69:99; p < :001. A similar pattern obtained for the gender-related categories, F ð1; 85Þ ¼ 26:84; p < :001. In contrast to the preceding two sets of data, and consistent with predictions, participants responded more quickly to typical targets than to atypical targets at a short SOA for race-related categories but the reverse was true for long SOAs, 4 There are tradeoffs involved in coding trimmed RTs as missing (as we have done here) vs. replacing these values (e.g., with the mean of each participantsÕ responses). Although the former avoids possible distortion of the results, it can lead to a small number of participants being dropped in multivariate analyses due to missing data; the analyses to follow were restricted to 86 participants (out of the original 94) for this reason. However, both approaches yielded very similar effects and both yielded the predicted three-way interaction (both ps < :01).

F ð1; 85Þ ¼ 6:85; p < :01 for the SOA
typicality interaction. (Additional analyses revealed that the SOA
Typicality interaction was not qualified by whether the occupation in question was actually a stereotypically Black vs. White occupation (p > :20) indicating that these results were a function of category related to race, rather than something idiosyncratic about either the White/Black occupational categories per se.) Comparison of mean levels of motivation for control for race vs. gender One important implication of Fig. 1 is that SOA mattered only for occupations primarily relevant to race, but not gender, stereotypes. This suggests, in turn, that motivation for control was higher in the former compared to the latter case. As seen in Table 1, this was in fact the case. Moreover, this was true for both the Plant and Devine (1998) instrument, F ð1; 93Þ ¼ 41:07; p < :001, as well as for the items derived from the Dunton and Fazio (1997) instrument, F ð1; 93Þ ¼ 66:08; p < :001. (Because we found nearly identical effects across the indices of internal vs. external control for the Plant and Devine scale, results are collapsed over this factor.) These findings are important, because it provides converging support from two very different types of data (RTs vs. self-reported motivation) regarding our interpretation of why SOA moderated RTs on race-related, but not gender-related, trials.5 5 It is reasonable to assume that motivation for control was driven by the fairly transparent nature of our task in terms of its relevance to stereotyping. Indeed, if our participants had been completely obvious of this fact, they presumably would not have initiated the kinds of anticipatory control that drove the pattern of responses under long SOA. We assume that, on average, most participants recognized the nature of the task fairly early in the experiment. However, in order to explore the possibility that such insight might have increased as the experiment progressed, we conducted supplementary analyses to discern whether evidence for anticipatory control might have been even greater among those participants for whom the long SOA trials occurred in the second, as opposed to the first, block of trials. However, there was no evidence of this pattern in the data, p > :20 for the relevant comparison of effects across blocks.

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246

245

Table 1 Descriptive statistics for and intercorrelations between motivation for control across race and gender domains Variables

Descriptive statistics M

1. Motivation to control-race (from Plant & Devine; 10 items) 2. Motivation to control-gender (modification of Plant & Devine; 10 items) 3. Motivation for control-race (from Fazio & Dutton; 3 items) 4. Motivation for control-gender (modification of Fazio & Dutton; 3 items)

SD

Intercorrelations a

1

2

3

1.26

0.78

0.72

—

0.73

0.91

0.76

.56

—

0.18

1.15

0.51

.46

.49

—

)0.63

1.03

0.41

.19

.51

.62

4

—

Note. Higher numbers on motivation scales indicate greater motivation to control prejudice; scales range from )3 to +3. * p < :01.

General discussion Despite the large literature on the ‘‘structural’’ properties of categories, relatively few studies have focused on the controlled mechanisms that might be involved in such judgments. For good reason, we think. Beliefs that chickens represent poor examples of BIRDS, for example, do not carry any connotation beyond their perceived typicality. Expressing these beliefs (either directly, or indirectly, such as through the classification task used here) normally would not engender a need to monitor certain responses over others except, of course, to make sure that they responded with the correct answer. Hence, one reason why cognitive psychologists have not focused on SOA in this domain is simply that varying the opportunity for control should be largely irrelevant. Our results confirmed this, showing that SOA was indeed irrelevant for non-social categories. As expected, a different picture emerged for racerelated trials. In this case, the usual typicality effect was replicated under short SOA, but this pattern was reversed under long SOA. Theoretical contribution of the present research to the social cognition literature

First, and most important, our design included three distinct kinds of categories (non-social, race, and gender) within the same experimental design. Hence, instead of focusing on only one type of category—as is usually the case in this area—this aspect of our design allowed us leverage in exploring the different mechanisms that might be arise across different category structures. Aside from a few notable exceptions (e.g., Fazio & Dunton, 1997) we are aware of few if any, studies in the literature to do so. Second, we varied—rather than held constant— opportunity for control by employing SOA as a within-subject factor. This is contrast to most studies in this area, which deliberately attenuates participantsÕ efforts to exert control over their responses in all conditions (e.g., by always keeping SOA short). This arguably reflects a more general zeitgeist in social cognition, in which substantially more attention has been accorded toward an understanding of automaticity, rather than control (cf. Payne, Jacoby, & Lambert, in press). Indeed, the differential role of controlled mechanisms across the three types of categories investigated here represents one of the key findings of our research. A final word

Similar to other studies (e.g., Banaji & Hardin, 1996; Blair & Banaji, 1996), we assume that activation of social categories has the potential to activate stereotypic constructs which can, in turn, facilitate responses to information consistent with those constructs, and inhibit responses toward information inconsistent with them (Wittenbrink, Judd, & Park, 1997). In addition, we assume, as do many researchers in this area, that both motivation and opportunity jointly moderate the likelihood that controlled processes actually occur in any given context (Fazio, 1990; Wilson & Brekke, 1994). However, the contribution of the present research to the literature derives from the fact that our paradigm differs from this earlier work in at least two ways.

Future work is obviously needed to further explicate not only the differences between race and genderrelated occupations, but also the relationship between occupational bias (as seen in experimental research) and real world hiring decisions. Nevertheless, if we were to offer any tentative advice to people as to how they might want to eliminate occupational bias in their own hiring decisions, our findings suggest that faster is not always better. ‘‘Take it slow’’ might not seem to be very inspiring advice on how to reduce prejudice, but given the inherent difficulties in counteracting the effects of automatically activated stereotypes (Bargh, 1999), such simple strategies might justify a longer look.

246

K.R. Olson et al. / Journal of Experimental Social Psychology 40 (2004) 239–246

References Balota, D. A., Black, S. R., & Cheney, M. (1992). Automatic and attentional priming in young and older adults: Reevaluation of the two-process model. Journal of Experimental Psychology: Human Perception and Performance, 18, 485–502. Bargh, J. A. (1999). The cognitive monster: The case against the controllability of automatic stereotype effects. In S. Chaiken, & Y. Trope (Eds.), Dual Process Theories in Social Psychology (pp. 361– 382). New York: Guilford. Banaji, M. R., & Hardin, C. D. (1996). Automatic stereotyping. Psychological Science, 7, 136–141. Barsalou, L. W. (1992). Cognitive psychology: An overview for cognitive scientists. Hillsdale, NJ: Erlbaum. Blair, I. V., & Banaji, M. R. (1996). Automatic and controlled processes in stereotype priming. Journal of Personality and Social Psychology, 70, 1142–1163. Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56, 5–18. Dunton, B. C., & Fazio, R. H. (1997). An individual difference measure of motivation to control prejudiced reactions. Personality and Social Psychology Bulletin, 23, 316–326. Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model as an integrative framework. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 23, pp. 75–110). New York: Academic Press. Fazio, R. H., & Dunton, B. C. (1997). Categorization by race: The impact of automatic and controlled components of racial prejudice. Journal of Experimental Social Psychology, 33, 451– 470. Gernsbacher, M. A., & Faust, M. (1995). Skilled suppression. In F. N. Dempster, & C. N. Brainerd (Eds.), Interference and inhibition in cognition (pp. 295–327). San Diego, CA: Academic Press. Glick, P., Zion, C., & Nelson, C. (1988). What mediates sex discrimination in hiring decisions? Journal of Personality and Social Psychology, 55, 178–186. Hernstein, R. J., & Murray, C. A. (1994). The bell curve: Intelligence and class structure in American life. New York: Free Press.

Lepore, L., & Brown, R. (1997). Category and stereotype activation: Is prejudice inevitable? Journal of Personality and Social Psychology, 78, 275–287. Maccoby, E. E., & Jacklin, C. N. (1974). The psychology of sex differences. Palo Alto, CA: Stanford University Press. McConahay, J. B. (1986). Modern racism, ambivalence, and the modern racism scale. In J. Dovidio, & S. L. Gaertner (Eds.), Prejudice, discrimination, and racism: Theory and research. New York: Academic Press. McGuire, M. J., McGuire, C. V., Child, P., & Fujioka, T. (1978). Salience of ethnicity in the spontaneous self-concept as a function of oneÕs ethnic distinctiveness in the social environment. Journal of Personality and Social Psychology, 36, 511–520. Medin, D. L., & Smith, E. E. (1981). Strategies and classification learning. Journal of Experimental Psychology: Human Learning and Memory, 7, 241–253. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limitedcapacity attention. Journal of Experimental Psychology: General, 106, 226–254. Payne, B. K., Jacoby, L. L., & Lambert, A. J. (in press). Attitudes as accessibility bias: Dissociating automatic and controlled processes. In R. Hassan, J. A. Bargh, & J. Uleman (Eds.), The New Unconscious. Oxford. Plant, E. A., & Devine, P. G. (1998). Internal and external motivation to respond without prejudice. Journal of Personality and Social Psychology, 75, 811–832. Rosch, E. (1975). Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 104, 192–233. Rosch, E., & Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573–605. Spence, J. T., & Helmreich, R. L. (1978). Masculinity and femininity: Their psychological dimensions, correlates and antecedents. Austin: University of Texas Press. Wilson, T. D., & Brekke, N. (1994). Mental contamination and mental correction: Unwanted influences on judgments and evaluations. Psychological Bulletin, 116, 117–142. Wittenbrink, B., Judd, C. M., & Park, B. (1997). Evidence for racial prejudice at the implicit level and its relationship with questionnaire measures. Journal of Personality and Social Psychology, 72, 262–274.