Evaluations of Ingroup and Outgroup Members: The ... - Science Direct

11 downloads 0 Views 245KB Size Report
Kelly Charlton, and Amy Mulholland. University of Missouri, Columbia. Received: January 30, 1996; revised: November 4, 1996; accepted: November 5, 1996.
JOURNAL OF EXPERIMENTAL SOCIAL PSYCHOLOGY ARTICLE NO.

33, 244–275 (1997)

JS961323

Evaluations of Ingroup and Outgroup Members: The Role of Category-Based Expectancy Violation B. Ann Bettencourt, Karen E. Dill, Scott A. Greathouse, Kelly Charlton, and Amy Mulholland University of Missouri, Columbia Received: January 30, 1996; revised: November 4, 1996; accepted: November 5, 1996 Examines the role of category-based expectancy violation in explaining extreme evaluations of ingroup and outgroup members. In three experiments, descriptions about ingroup and outgroup targets were varied to manipulate expectancy violation. Participants evaluated the global favorability and rated the trait characteristics of either an ingroup or an outgroup target. In addition, they indicated the extent to which the targets violated category-based expectancies. Experiment 1, which included two different scenarios that violated category-based expectancies, showed that expectancy violation affected global evaluations of target favorability and that measured expectancies mediated these evaluations. Experiment 2 replicated Experiment 1 and showed that ingroup and outgroup targets were evaluated similarly when category-based expectancies were not violated. Experiment 3 presented positive job resumes that violated or did not violate stereotyped expectancies for female and male job applicants. The results showed that female and male targets were evaluated more positively when they violated stereotyped expectations for their gender group and that these outcomes were not modified by the subject’s group membership. The results are discussed in terms of expectancy-violation theory as well as assumedcharacteristics and cognitive-complexity theories. r 1997 Academic Press

Imagine that an African-American student comes to your office seeking advice about graduate school. To get an idea of his qualifications, you inquire about his academic accomplishments. The student looks down and, casting an embarrassed eye at you, reveals that he has a G.P.A. of 3.96, is vice-president of an honor

This research was supported by National Science Foundation Grant SBR9409921 to B. Ann Bettencourt and by the Center for Research in Social Behavior. We thank Stephanie Black and Chris Schulze for their assistance in conducting the research. We also thank Doug Krull and two anonymous reviewers for their helpful comments on former drafts of this article. Finally, we thank Jeff Greenberg for his assistance in the design of Experiment 2. A preliminary report of this research was presented at the Nags Head Conference, June 1994. Address correspondence and reprint requests to B. Ann Bettencourt, Department of Psychology, 15 McAlester, University of Missouri, Columbia, MO 65211. E-mail: [email protected]. 244 0022-1031/97 $25.00 Copyright r 1997 by Academic Press All rights of reproduction in any form reserved.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05 jant No. of Pages—32 First page no.—244 Last page no.—275

245

INGROUP AND OUTGROUP EVALUATIONS

society, and studies under a prestigious scholarship. What would be your unguarded reaction? Perhaps, you would feel particularly impressed. In a similar vein, one might feel quite delighted to learn that an 80-year-old neighbor had just won a marathon in her division. On the other hand, one might feel particularly displeased to learn that a nun had been caught in the act of selling drugs to children. These scenarios are meant to exemplify the prediction that people are more affected when others’ behaviors violate category-based expectations for their respective ingroups than when they do not (Burgoon, 1986; Jussim, Coleman, & Lerch, 1987). Expectancy-violation theory (Burgoon, 1986; Jussim et al., 1987), in part, suggests that we evaluate persons more extremely when their behaviors violate stereotyped expectations for their salient ingroups and empirical research shows support for this prediction (e.g., Branscombe, Wann, Noel, & Coleman, 1993; Jackson, Sullivan, & Hodge, 1993; Jussim et al., 1987; Jussim, Fleming, Coleman, & Kohberger, 1996). Although this research provides considerable evidence for the effect of manipulated stereotype violations on evaluative extremity, to our knowledge, no study has explicitly tested the mediational role of stereotype violation on evaluative extremity. The primary purpose of the present series of experimental studies was to examine the mediational effects of category-based expectancy violations on perceivers’ evaluations of others. To do so, we not only manipulated the stereotypicality of the target’s described behavior but also measured the extent to which the behaviors violated stereotyped expectations for the target. In what follows, we review the literature that provides relevant theoretical predictions and raises important methodological issues concomitant to this area, and we report the outcomes of three experiments. Evaluative Extremity In a study of stereotyped expectancies, Taynor and Deaux (1973) showed that participants allotted greater reward to a woman described as having helped in a civic emergency than to a man described similarly. The authors reasoned that, because the rescue behavior was masculine, it was unexpected for the woman; as a result, evaluators were more impressed by the woman’s helpful behavior. Importantly, this finding was not moderated by the evaluator’s gender, and therefore ingroup membership of the evaluator appeared to be far less important in determining the evaluations than stereotyped expectancies. More recently, several studies (e.g., Jackson et al., 1993; Jussim et al., 1987) have explicitly tested the predictions of expectancy-violation theory. To test the premises of this theory as well as those of assumed-characteristics theory and complexity–extremity theory, Jussim and his colleagues (Jussim et al., 1987) manipulated both the quality and the ethnicity of a job applicant. The two critical conditions for testing expectancy-violation theory were those in which (a) a Black, upper-class, well-spoken applicant was depicted, and (b) a White, lowerclass applicant speaking broken, nonstandard English was depicted. The first

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

246

BETTENCOURT ET AL.

condition was meant to violate White participants’ expectancies for Blacks and the second was meant to violate their expectancies for Whites. In general, Jussim et al. interpreted their results as supportive of expectancy-violation theory but advised that future research should measure stereotyped expectations to determine their mediational effects. More recently, Jackson et al. (1993) and Jussim et al. (1996) have also reported findings consistent with expectancy-violation theory, showing that both positively and negatively depicted Black targets were evaluated more positively than similar White targets. Jussim et al.’s (1987) findings were also consistent with the predictions of assumed-characteristics theory (e.g., Locksley, Borgida, Brekke, & Hepburn, 1980). This theory predicts that a relatively small amount of individuating information about others leads perceivers to disregard group membership and category-based stereotypes and instead evaluate others on the basis of the background information. Jussim et al. (1987) concluded that, although target’s group membership did affect people’s evaluations, the relative valence (positive versus negative) of the individuating information has powerful effects on person evaluations. Finally, Jussim et al.’s (1987) results were also supportive of the complexity– extremity hypothesis (Linville, 1982; Linville & Jones, 1980), because they showed that the difference between the evaluations of the positive and the negative Black job applicants was of greater magnitude (i.e., more polarized) than that between the positive and negative White applicants. The complexity– extremity explanation for evaluative polarization is based on a proposed relation between one’s cognitive schema for a given group and one’s evaluations of specific members of that group. In explaining this relation, Linville (Linville, 1982; Linville & Jones, 1980) argues that an individual will be evaluated more extremely by an outgroup member than by an ingroup member because the outgroup evaluator’s cognitive schema for the individual’s respective ingroup lacks complexity. Thus, this hypothesis predicts an inverse relationship between cognitive complexity and evaluative extremity. It is important to note, however, that research has also shown some exceptions to the complexity–extremity hypothesis (Coleman, Jussim, & Kelly, 1995; Hass, Katz, Rizzo, Bailey, & Eisenstadt, 1991). For example, Coleman et al.’s (1995) replication of Jussim et al.’s (1987) study showed that, although Black participants had less complex schemas for the White outgroup, their evaluations for the outgroup targets were not polarized. Instead, the results showed more extremity in the range of the ratings for the ingroup. That is, like the White participants in Jussim et al.’s (1987) and Linville and Jones’s (1980) studies, the Black participants in Coleman et al.’s study showed more polarized evaluations of Black targets than of White targets. Thus, taken together, these studies may suggest that Black targets are generally rated with more range than are White targets, regardless of the evaluator’s group membership. Also in contrast with the complexity–extremity hypothesis, Marques and his colleagues (Marques & Yzerbyt, 1988; Marques, Yzerbyt, & Lyens, 1988;

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

247

INGROUP AND OUTGROUP EVALUATIONS

Marques, Robalo, & Rocha, 1992) have shown that unfavorable ingroup members are evaluated more negatively than similarly described outgroup members (i.e., black-sheep hypothesis), whereas favorable ingroup members are evaluated more positively than similar outgroup members. In one study (Marques & Yzerbyt, 1988), participants who were philosophy majors heard either a skillful or an unskillful speech that was attributed either to a philosophy major or a journalism major. The results showed that the most favorable evaluations were given to the skillful philosophy major and the most negative evaluations to the unskillful philosophy major. The apparent contradictions between the findings of Linville and Marques may be explained, in part, by expectancy-violation theory. For example, perhaps the excellent law-school applicant in Linville and Jones’s (1980) study violated category-based expectations for outgroup Blacks and thus was evaluated particularly favorably, whereas the unskillful speaker in Marques and Yzerbyt’s (1988) study violated expectations for ingroup philosophy majors and thus was judged particularly harshly. That violated expectations may produce the black sheep effect is indeed indicated in another of Marques’s studies (Marques et al., 1988). Participants judged a set of ingroup members more negatively than a set of outgroup members when each was described as violating norms exclusive to their respective ingroup, but participants did not judge these group members differently when each violated a general norm, unassociated with their ingroup. In general, expectancy-violation theory may provide an explanation for some instances of outgroup favoritism—cases of extremely positive evaluations of favorable outgroup members or extremely negative evaluations of unfavorable ingroup members. In general, expectancies for ingroup members should be relatively positive compared to those for outgroup members (although not always), and so the negative behaviors of ingroup members and the positive behaviors of outgroup members may be more likely to violate category-based expectancies. Nevertheless, we can also imagine instances in which expectancies might influence evaluations, regardless of whether the evaluator shares group membership with the target. For example, both nuns and those who are not nuns might be appalled by a nun who sold drugs to children. Whatever the categorybased expectancy, individuals who violate them may be more prone to be evaluated extremely. Within the specific framework of violated category expectations, the present studies not only allow a test of expectancy-violation theory, but they also allow examination of some of the premises of the assumedcharacteristics, complexity–extremity, and black-sheep hypotheses. Global Favorability and Trait Inference Ratings In their studies, Jackson et al. (1993) revealed that the expectancy-violation manipulations affected only participants’ ‘‘overall evaluations’’ of the target. In contrast, the target’s valence alone affected the ratings of target traits. This differential pattern of results was similar to what has been found in several other studies (e.g., Bettencourt, 1989; Bettencourt, 1990; Jussim et al., 1987; Glick,

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

248

BETTENCOURT ET AL.

Zion, & Nelson, 1988). The pattern of these data lead us to believe that a stereotype-expectancy violation might create a global affective reaction, detected by global evaluations of favorability or liking, but the same might have little effect on ratings specifically related to a target’s qualifications (i.e., traits). That violated expectancies are likely to cause affective reactions is well supported in the literature (e.g., Clary & Tesser, 1983; House & Perney, 1974). Moreover, in theories of emotion (Clore, Schwarz, & Conway, 1994; Roseman, Antoniou, & Jose, 1992, as cited in Clore et al., 1994), unexpectedness is considered to be a global intensity variable for all emotions. Thus, measures of global favorability (e.g., good versus bad) might be more likely to reflect the emotional intensity that unexpectedness elicits. As noted, the central purpose of the current investigation was to further examine the role of category-based expectancy violations in perceivers’ extreme evaluations of members of ingroups and outgroups. Because perceivers’ stereotyped expectancies had not been measured by other researchers, our highest priority was to specifically examine stereotyped expectancy as a mediator of the relationship between group membership of the targets and subsequent evaluations of the targets. We also wanted to further examine the effect of expectancy violations on global favorability toward the evaluative targets and a possible lack of this effect on ratings of traits that are closely linked to the individuating characteristics of the target. In each of three experiments, we manipulated stereotyped expectancies for evaluative targets and measured global-favorability, trait-specific, and expectancy ratings. Because the type of targets depicted and the manipulations vary across the three experiments, we state the hypotheses for each of the three studies just prior to the method sections. In doing so, we present our main predictions in terms of expectancy-violation theory, because these directly address our specific manipulations of violated expectancies and our primary purpose of testing the mediational role of category-based expectancies. However, we also state auxiliary predictions in terms of the assumed characteristics, complexity–extremity, and black-sheep hypotheses. EXPERIMENT 1 In the first study, participants read two brief scenarios: one about a male job applicant and another about a male college student giving a speech in a class. In the job-applicant scenario, the job applicant’s ethnicity was either Black or White and he was either presented as very favorable and well-spoken or as less favorable and using ‘‘broken’’ English. These scenarios were adapted from some of the conditions used in Jussim et al.’s (1987) study. The more favorable target description was meant to be a violation of stereotypes for Black job applicants whereas the unfavorable target description was meant to be a violation for White job applicants. In the speech scenario, the target was either a member of a football team or a speech team and either delivered a skillful or unskillful speech. The more skillful description was meant to be a violation of stereotypes for football-team members

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

249

INGROUP AND OUTGROUP EVALUATIONS

whereas the unskillful description was meant to be a violation for speech-team members. Participants themselves were outgroup members of both football- and speech-team members. Using these scenarios allowed us to accomplish two goals. First, the Black versus White job-applicant scenario allowed for methodological overlap with Jackson et al.’s (1993) and Jussim et al.’s studies (1987). Second, the footballversus speech-team scenario allowed us to test expectancy effects while eliminating the ingroup versus outgroup comparison. Because the ingroup–outgroup distinction was not present, cognitive complexity, ingroup favoritism, and the black-sheep effect should play less of a role or no role in differences in evaluations. In other words, comparing these two outgroup targets who are described as violating stereotypes for their salient group category allows for a test of expectancy violation theory while providing control for other variables (viz., cognitive complexity and black-sheep effects). Hypotheses For targets in both scenarios, we predicted that expectancy-violation manipulations would induce more extreme evaluations of the global favorability of the targets who violated stereotyped expectations and that these extreme evaluations would be in the direction of the target’s valence. Based on the results of Jackson et al. (1993), we also expected expectancy violations to have less of an effect on the trait ratings and we predicted that only target valence would affect trait-specific ratings of the target (which is also consistent with assumed-characteristics theory). It follows from these two predictions that prior category-based expectancies would only mediate the evaluations of the global favorability. Thus, we predicted that participants’ ratings of stereotyped expectancy would mediate the effects of group membership on the global-favorability evaluations. It is important to note that whereas expectancy is predicted to be a mediator of group membership, it is not a mediator of the valence of the target’s description. That is, we do not expect that the main effect difference of valence will be mediated by expectancy. Consistent with assumed-characteristics theory, we expected that negative descriptions would be evaluated more negatively than would positive ones. Because the cognitive complexity and the black-sheep hypotheses are specific to ingroup versus outgroup target comparisons, the following hypotheses only apply to the job-applicant scenario. The black-sheep hypothesis predicts that evaluation of favorable and unfavorable ingroup targets (White) will be more polarized than those of similarly described outgroup targets (Black). In contrast, the complexity– extremity hypothesis predicts that evaluations of favorable and unfavorable outgroup targets (Black) will be more polarized than those of similarly described ingroup targets (White). According to the complexity–extremity view, there should also be an inverse relation between evaluative extremity and measured cognitive complexity.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

250

BETTENCOURT ET AL.

Method Participants One-hundred and fifteen undergraduates at a large Midwestern university participated for partial credit in their Introductory Psychology course.

Target Descriptions Four separate packets were created so that each contained one job-applicant scenario and one speech scenario, with the restriction that no one packet had two scenarios that depicted stereotype violations. Each of these four scenario packets was presented in two possible orders: one order in which the job applicant was the first scenario and the speech scenario was the second and another order in which the opposite was true. The resulting eight packets were randomly assigned to experimental sessions (all participants in a given session received the same packet). Job applicant scenario. The design for the job application scenario was a 2 (Black or White target) 3 2 (skillful or unskillful applicant) between-subjects factorial. Participants were provided with the job applicant’s name (Edward Bower), age (30), sex (male), and ethnic description (Black or White). The vignette described Edward Bower’s job interview with a department manager. The quality of the job applicant was varied by his verbal communication style and by his style of dress; he either responded to the interviewer’s question in standard English (‘‘This is the kind of job . . .’’) and wore a formal suit (‘‘skillful’’), or he responded in nonstandard English (‘‘Dis is da kind a job . . .’’; see Jussim et al., 1987) and wore an older sport jacket and pants (‘‘unskillful’’). The results of a pretest (N 5 62) indicated that the skillful Black job applicant was rated as less stereotypical (M 5 1.94) than a similarly described White applicant (M 5 3.19), and that the unskillful White job applicant was rated as less stereotypical (M 5 1.60) than a similarly described Black applicant (M 5 3.47; Interaction F(1, 58) 5 62.79, p , .01). Speech scenario. The design for the speech scenario was a 2 (football-team or speech-team member) 3 2 (skillful or unskillful speech) between-subjects factorial. Participants were provided with the speaker’s name (Karl Henderson), age (20), sex (male), and ethnic description (White). In the vignette, participants were told that Karl was either ‘‘a member of the university speech and debate team’’ or ‘‘a member of the university football team.’’ The scenario stated that, in his speech class, Karl was assigned the topic of ‘‘Increased Government Spending on Education.’’ Participants read either a good or a poor speech on this topic. These speeches can be obtained from the authors. The results of the pretest indicated that, when paired with the skillful speech, the football player was rated as less stereotypical (M 5 2.00) than the speech major (M 5 3.33), but when paired with the unskillful speech, the speech major was rated as less stereotypical (M 5 1.38) than the football player (M 5 2.88; Interaction F(1, 58) 5 36.91, p , .01).

Dependent Measures Participants answered questions designed to assess their global favorability toward the target, specific judgments about the target’s capability, and category-based prior expectancies. Global favorability and trait-capability ratings. Participants were asked to indicate their beliefs and feelings about the target person on each of six semantic-differential scales. To measure global favorability, the following word pairs were used: likable/unlikable, favorable/unfavorable, and good/bad. To measure traits, the following word pairs were used: incapable/capable, motivated/ unmotivated, and resourceful/unresourceful. Each item had a 6-point Likert type response scale, and the directionality of the valence of responses and the order of the scales were randomly assigned. Typicality, expectancy, and stereotype ratings. After rating the scenarios, participants were asked to rate the expectedness, typicality, and stereotypic fit of each of the targets. For example, participants were asked, ‘‘Compared to other Black (White) job applicants, how unexpected (typical) is Edward’s behavior?’’ (the 1 endpoint was labeled ‘‘very unexpected’’ and the 8 endpoint, ‘‘very expected’’). The stereotype item asked participants to think about the common stereotype for Blacks (Whites) and answer the question: ‘‘To what extend does Edward Bower’s behavior fit common stereotypes for

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

251

INGROUP AND OUTGROUP EVALUATIONS

Blacks (Whites)?’’ (the 1 endpoint was labeled ‘‘does not fit stereotypes for Blacks (Whites)’’ and the 4 endpoint, ‘‘fits stereotypes for Blacks (Whites).’’ The questions for the speech scenario were rephrased in reference to football players or speech majors. Cognitive-complexity measure. Next, participants were asked to complete a trait-sorting task, modeled after those used by Linville (1982) and by Millar and Tesser (1986). In the present study, 40 trait adjectives were randomly selected from Anderson’s (1968) list of 555 trait adjectives in such a way as to select equal numbers of positive, neutral, and negative traits. Each subject was given a complete stack of cards and asked to think about the members of the group they had previously read about (Blacks or Whites). They were asked to sort the trait cards into stacks that represented meaningful categories for the group in question (see Linville & Jones, 1980, for a full description). Next they were asked to do the trait-sorting task again, this time thinking about the members of the other group they had read about (football players or speech majors). Participants completed the two trait-sorting tasks in the same order they had read about the two groups. Demographics. Participants indicated their ethnicity, major, and whether or not they belonged to a university athletic or speech team. Participants also indicated their gender and age.

Procedure Sessions included one to six participants. The instructions, read by one of two female experimenters, informed the participants that the study was about first impressions formed in common settings like school and the workplace. Participants were told that they would read scenarios about different people doing tasks relevant to school and jobs and that they would be asked to form impressions of these people. After these general instructions, each subject was placed in a private cubicle where he or she completed all tasks and measures. Finally, participants were fully debriefed.

Results The three items for each of these three indices (i.e., global favorability, trait-capability, and expectancy) were summed, averaged, and analyzed (a ranged from .74 to .97). Because two of the measures of expectancy had an eight-point scale and one had a four-point scale, the three measures of expectancy were transformed to Z-scores and these were then averaged. Each of the following subsections includes a report of (a) separate 2 3 2 ANOVAs for each of the three measures, (b) comparisons of cognitive complexity mean scores between the two target groups, and (c) regression analyses designed to test the effects of the proposed mediators.1 Because the two scenarios were presented in counterbalanced order, we tested the effects of order on the ratings of favorability, capability, and expectancy. Out of twenty possible main effects and interactions there was only one significant interaction. For the expectancy ratings of the job applicant, the interaction between group membership and job quality was moderated by order (3-way F(1, 102) 5 5.12, p , .05). However, the directions of differences between the cell means were identical for each of the two orders (i.e., there was not a reversal of the pattern of means). As such, we collapsed across the order variable for all subsequent analyses.

1 Two participants were excluded from the analysis because one did not follow the directions given for the measurements and the other left in the middle of the session to smoke a cigarette.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

252

BETTENCOURT ET AL.

Job Applicant To keep ingroup membership constant, only the data for participants who identified themselves as Caucasians were included in the following analysis.2 Expectancy ratings. The mean ratings based on the total of the three items in the expectancy index are depicted in the top panel of Fig. 1. The mean values for only the ratings of fit with common stereotypes are also presented in the lower panel of Fig. 1, because they more clearly reveal the stereotype-violation effect. As predicted, the interaction showed that participants rated the skillful Black job applicant as more unexpected than the skillful White job applicant but rated the unskillful White job applicant as more unexpected than the unskillful Black job applicant, F(1, 94) 5 48.11, p , .01. The main effect of job quality also revealed that, in general, skillful applicants were more expected than unskillful ones, F(1, 94) 5 76.53, p , .01. Global favorability ratings. We predicted that the expectancy-violation manipulation would affect global favorability evaluations but not those of trait-capability ratings.3 The mean values for global favorability are presented in the top row of 2 All data were tested for violations of normality and extreme outliers. According the recommendation of Tabachnick and Fidell (1989), Tukey (1990), and Wilcox (1995), outliers were identified examining boxplots of the data. A boxplot is a graphical display that indicates the range, the quartiles, the spread, and the median of the data. The length of the boxplot represents 50% of the cases, and thus the lower boundary of the box is the 25th percentile and the upper boundary is the 75th percentile. Data points with values more than 3 box-lengths from the upper or low boundary of the box are extreme outliers (Tabachnick & Fidell, 1989; Wilcox, 1995). Such analyses are commonly available in statistical packages such as SPSS and SAS programs. For the job-applicant scenario, 7 of 882 values were extreme outliers (less than 0.7%). As recommended by Tabachnick and Fidell (1989) and Wilcox (1995) these extreme outliers were modified to the value of the next closest nonoutlying value in the distribution. The pattern of the data for the White Unskilled (WU), Black Unskilled (BU), White Skilled (WS), and Black Unskilled (BU) targets before it was Winsorized was very similar to that for the Winsorized data. The means for the general favorability ratings before Winsorization were as follows: WU: M 5 3.47; BU: M 5 3.78; WS: M 5 5.02; BS: M 5 5.48 and the predicted main effect of ethnicity was significant, F(1, 94) 5 4.06, p , .05. The means for the trait-capability ratings before Winsorization were as follows: WU: M 5 3.25; BU: M 5 3.53; WS: M 5 5.14; BS: M 5 5.41, F(1, 94) 5 1.98, p . .10. Another approach for dealing with outlying values is to eliminate them from the data set (see Wilcox, 1995). Once again the outcomes of analyses that eliminated the outliers were also similar to the Winsorized data. The means for the general favorability ratings were as follows: WU: M 5 3.47; BU: M 5 3.78; WS: M 5 5.03; BS: M 5 5.65, and the predicted main effect of ethnicity was significant, F(1, 90) 5 6.07, p , .05. The means for the trait-capability ratings with the extreme outliers eliminated were as follows: WU: M 5 3.25; BU: M 5 3.53; WS: M 5 5.27; BS: M 5 5.45, F(1, 92) 5 1.50, p . .10. We adopted the Winsorization approach because it is often recommended (see Tabachnick & Fidell, 1989; Tukey, 1990; and Wilcox, 1995), and because this approach seemed more conservative than eliminating the extreme outliers. 3 The analysis treating type of measure (favorability, trait) as a within-subjects variable revealed a main effect of measure (F(1, 94) 5 3.94, p , .05) and a job quality by measure interaction (F(1, 94) 5 4.60, p , .05); however, the three-way interaction between race, job quality, and measure was not significant (F(1, 94) 5 .71, p . .20). These results seem to suggest that the manipulations might be affecting the favorability and the trait ratings somewhat differently. Given that these participants appeared to be responding to the measures differently and there was a theoretical basis for analyzing these indices separately, we did so.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

253

INGROUP AND OUTGROUP EVALUATIONS

FIG. 1. Mean values of the sum of the three ratings of expectancy and mean values of stereotypic fit for Black and White job applicants (Experiment 1).

Table 1. As predicted by expectancy-violation theory, the analysis revealed a main effect of group membership on participants’ global favorability toward the target, F(1, 94) 5 4.51, p , .05. Specifically, Black applicants were evaluated more favorably than White applicants. In addition, as predicted by assumedcharacteristics theory, skillful applicants were also rated more positively than

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

254

BETTENCOURT ET AL.

TABLE 1 MEAN VALUES FOR GLOBAL FAVORABILITY EVALUATIONS AND TRAIT-CAPABILITY RATINGS OF INGROUP AND OUTGROUP JOB APPLICANTS Unskillful applicant

Global favorability Trait-capability

Skillful applicant

White (19)

Black (30)

White (22)

Black (27)

3.47 3.25

3.78 3.53

5.03 5.18

5.52 5.43

Note. The numbers in parentheses indicate the number of participants in each cell. Higher mean values indicate more positive evaluations and more positive trait ratings.

unskillful applicants, F(1, 94) 5 78.03, p , .01. However, the Group Membership 3 Job Quality interaction was not significant, F(1, 94) 5 .24, p . .10. These findings suggest that neither the full predictions of the black-sheep hypothesis nor those of the complexity–extremity hypothesis were supported. Even so, it was the case that the more negative evaluations of the unskillful White (ingroup) job applicant were consistent with the black-sheep hypothesis and the positive evaluation of the skillful Black (outgroup) job applicant was supportive of the cognitive complexity hypothesis. The cognitive complexity hypothesis can also be tested by calculating the range of the difference between the evaluations of the positive and negative targets within each group category. These comparisons did show that the range of evaluations of the Black targets (difference 5 1.74) was somewhat greater than that of the White targets (difference 5 1.56), although this difference was not significant, t(94) 5 .048, p . .10. Trait-capability ratings. In contrast with the findings for the global favorability evaluations, but consistent with the predictions of assumed-characteristics theory and the findings of Jackson et al. (1993), only job quality affected participants’ ratings of the target’s trait-capability, F(1, 94) 5 100.63, p , .01; neither the main effect of group membership, F(1, 94) 5 1.98, p . .10, nor its interaction with job quality were reliable, F(1, 94) 5 .01, p . .10. A comparison of the range of the difference between the trait ratings of positive and negative targets showed the range of the trait ratings of the Black targets (difference in trait ratings 5 1.90) and the White targets (difference in trait ratings 5 1.92) to be similar. These outcomes may suggest that individuating information most likely affects inferences about traits that are conceptually linked to this information. Cognitive complexity. Based on the procedure reported by Linville and Jones (1980), we calculated Scott’s-H for the participants’ sortings of the 40 trait cards.4 As predicted, and consistent with Linville and Jones (1980), the results showed 4

In accordance with Linville and Jones (1980), Scott’s H was calculated for each subject as H 5 log2 n 2 (1/n) Sn i log2 ni,

where n is total number of traits (here 40) and ni is the number of traits that appear in a particular combination of groups.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

255

INGROUP AND OUTGROUP EVALUATIONS

that White participants’ cognitive representations for the category White job applicant (M 5 2.65) were more complex than that for the category Black job applicant (M 5 2.29), F(1, 96) 5 4.21, p , .05. Mediational role of expectancy violation. We hypothesized that the effects of group membership on favorability would be mediated by perceptions of categorybased expectancy. According to Baron and Kenny (1986), to test mediation, the results of three analyses must show that the proposed independent variable affects both the dependent variable and the mediator variable, and that when the proposed mediator and the independent variable are simultaneously entered into a third regression equation, the mediator must account for significant variation in the dependent variable (Baron & Kenny, 1986). Because group membership affected both the ratings of expectancy and the cognitive complexity scores, we tested both as mediators of global favorability evaluations. Assumptions of regression analyses (i.e., linearity, homogeneity, normality of variance) were first tested. The assumption of linearity and homogeneity were met and the expectancy ratings were log-transformed to better approximate normality of residuals. Furthermore, in the analyses of expectancy as a mediator, one observation was found to be an extreme outlier from the predicted line, and a test of distance from the line revealed that it was a significant outlier (studentized distance 5 23.34). Upon investigating why this point might be so far from the predicted line, we found that one participant’s responses seemed to be clearly inconsistent with the results from the pretest and the present study. Although the participants in the pretest and those in the current study indicated that the unskillful job applicant was relatively consistent with the stereotype for Blacks, this participant responded that this scenario was not consistent with common stereotypes for Blacks and instead indicate that the unskillful job applicant was a stereotype violation for Blacks. In the analyses that follow, we include this data point, but in one analysis, we report the results with this data-point removed. As reported in the previous subsections, group membership affected both global favorability and category-based expectancy. The results of the simultaneous regression showed that the effect of group membership on global favorability was minimized to nonsignificance, b 5 .02, t(95) 5 0.22, p . .20, but expectancy ratings remained as a predictor, b 5 .38, t(95) 5 3.97, p , .01.5 These results support the role of expectancy as a mediator of the relationship between group membership and global favorability. We also included expectancy in a simultaneous regression with group membership, job quality, and their interaction term. These results showed that global favorability was related to job quality, b 5 .54, t(93) 5 5.47, p , .01, and marginally related to expectancy, b 5 .19, t(93) 5 1.64, p 5 .052, but group 5 To assess the linear relationship between group membership and expectancy ratings, the expectancy ratings of the Black job-applicant condition were reversed coded for the regression analyses. In this way, expectancies would be positively related to evaluations.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

256

BETTENCOURT ET AL.

membership, b 5 10, t(93) 5 1.17, p . .20, and the interaction were not significant, b 5 .12, t(93) 5 1.13, p . .20. When we conducted this same analysis with the extreme outlier case removed, the results were similar but the correlation between expectancy and global favorability was reliable. Specifically, both job quality, b 5 .51, t(92) 5 5.11, p , .05, and expectancy, b 5 .25, t(92) 5 2.17, p , .05, predicted favorability, but group membership, b 5 .10, t(92) 5 1.27, p . .10, and the quality by group interaction did not, b 5 .13, t(92) 5 1.31, p 5 .10. Because the complexity–extremity hypothesis predicts that less complex schemas should be related to more polarized evaluations, we created a polarization score of the global favorability evaluations. To do so, we first calculated the Z-scores of the global favorability evaluations and then used the absolute value of the Z-scores as an index of polarization. This index ignores valence of the ratings and reveals only the degree of polarization from the mean of all scores. The results of the univariate regression analyses revealed no relationship between cognitive complexity and polarized evaluations (b 5 .01, t(96) 5 0.12, p . .20). In a second univariate regression analysis, expectancy ratings were related to polarized favorability evaluations (b 5 2.20, t(96) 5 22.03, p , .05), indicating that the more unexpected the target, the more polarized the evaluations. Speech Presenter No students indicated that they were either speech- or football-team members; therefore the speech and football targets were outgroup members for all participants.6 Expectancy ratings. The mean ratings based on the total of all three items in the expectancy index are depicted in the top panel of Fig. 2. The mean values for only the ratings of the targets’ fit with common stereotypes are presented in the lower panel of Fig. 2. Because all targets in this design were outgroup members, the predictors for complexity–extremity and the black-sheep hypotheses, which 6 As described in footnote 2, the data for the speech scenario were tested for violations of normality and extreme outliers. Sixteen of 882 values were extreme outliers (equivalent to 1.8%). The means for the general favorability ratings for the Speech Unskilled (SU), Football Unskilled (FU), Speech Skilled (SS), and Football Skilled (FS) targets before Winsorization were as follows: SU: M 5 3.47; FU: M 5 3.75; SS: M 5 5.01; FS: M 5 5.40, and the predicted main effect of ethnicity was significant, F(1, 106) 5 3.19, p , .05. The means for the trait-capability ratings before Winsorization were as follows: SU: M 5 3.19; FU: M 5 3.35; SS: M 5 5.05; FS: M 5 5.33, F(1, 106) 5 1.20, p . .10. The means for the total of the expectancy measures were as follows: SU: M 5 8.45; FU: M 5 15.74; SS: M 5 14.10; FS: M 5 6.92, F(1, 106) 5 182.26, p . .01. Another approach for dealing with outlying values is to eliminate them from the data set (see Wilcox, 1995). Once again, the outcomes of analyses that eliminated the outliers was also similar to the Winsorized data. The means for the general favorability ratings were as follows: SU: M 5 3.47; FU: M 5 3.89; SS: M 5 5.19; FS: M 5 5.40, and the predicted main effect of ethnicity was significant, F(1, 103) 5 3.39, p , .05. The means for the trait-capability ratings with the extreme outliers eliminated were as follows: SU: M 5 3.07; FU: M 5 3.53; SS: M 5 5.27; FS: M 5 5.46; F(1, 99) 5 1.77, p . .10. The means for the total of the expectancy measures were as follows: FU: M 5 15.79; SU: M 5 7.96; FS: M 5 6.92; SS: M 5 14.10, F(1, 102) 5 211.79, p . .01. We adopted the Winsorization approach because it is often recommended and because it seemed to be more conservative then eliminating the extreme outliers.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

257

INGROUP AND OUTGROUP EVALUATIONS

FIG. 2. Mean values of the sum of the three ratings of expectancy and mean values of stereotypic fit for football and speech team member (Experiment 1).

involve comparisons between ingroup and outgroups, were not tested. As predicted and commensurate with our pretest, for the unskillful speech, the speechteam member was rated as more unexpected than the football-team member, and for the skillful speech, the football-team member was rated as more unexpected than the speech-team member, F(1, 106) 5 190.14, p , .01. The main effect of

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

258

BETTENCOURT ET AL.

TABLE 2 MEAN VALUES FOR GLOBAL FAVORABILITY EVALUATIONS AND TRAIT-CAPABILITY RATINGS OF TWO OUTGROUP SPEAKERS Unskillful speaker

Global favorability Trait-capability

Skillful speaker

Speech (25)

Football (20)

Speech (34)

Football (31)

3.47 3.17

3.80 3.35

5.06 5.12

5.40 5.39

Note. The numbers in parentheses indicate the number of participants in each cell. Higher mean values indicate more positive evaluations and more positive trait ratings.

speech quality, F(1, 106) 5 6.48, p , .05, was significant, but that of group membership was not, F(1, 106) 5 0.04, p . .10. Global favorability and trait-capability ratings. As predicted by expectancy violation theory and shown in Table 2, participants rated the football-team member more positively than the speech-team member when the speech was skillfully delivered, and rated the speech-team member more negatively than the football-team member when the speech was poorly delivered. This main effect, F(1, 106) 5 3.67, p , .05, suggested that global favorability was affected by group membership. These evaluations were also affected by the quality of the speech, F(1, 106) 5 85.50, p , .01, such that skillful speakers were rated more favorably than unskillful speakers. The interaction was not significant, F(1, 106) , 1. Consistent with assumed-characteristics theory and the results for the jobapplicant scenario, group membership did not affect participants’ ratings of the targets’ capability, F(1, 106) 5 1.43, p . .20. Only the speech quality affected the specific trait-capability ratings, F(1, 106) 5 109.46, p , .01. As shown in Table 2, participants rated those who gave the skillful speech as more capable than those giving the poor speech. The interaction was not significant, F(1, 106) , 1.7 Cognitive complexity. Similar to the procedure for the job application scenario, we calculated Scott’s-H for the participants’ sortings of the 40 traits. Because both of these groups were outgroups for the participants, we did not expect the participants’ cognitive complexity to differ for the two groups. As expected, the cognitive complexity for these two groups was similar (football-team: M 5 1.83, speech-team: M 5 1.61), F(1, 108) 5 .56, p . .20. Mediational role of expectancy violation. We used the same series of regression analyses to test mediation as that described previously. Assumptions of regression analyses (i.e., linearity, homogeneity, normality of variance) were first tested. The assumption of linearity and homogeneity were met, and the expectancy ratings 7 Similar to the results for the job-applicant scenario, the ANOVA analysis revealed that participants responded differently to the two categories of measures (F(1, 106) 5 4.94, p , .05), and that type of measure interacted with the level of speech quality (F(1, 106) 5 6.09, p , .05), but the three-way interaction between group membership, quality, and measure was not significant, F(1, 106) 5 .10, p . .20.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

259

INGROUP AND OUTGROUP EVALUATIONS

were log-transformed to better approximate normality of residuals. One observation was found to be an outlier (studentized distance 5 23.28). Again, for this participant the ratings were clearly inconsistent with those of the present data set and the pretest data. In the analyses that follow, we include this data point, but in one analysis, we report the results with this data point removed. As reported previously, group membership affected both global favorability and category-based expectancy.8 The simultaneous regression results showed that this effect of group membership on global favorability was minimized to nonsignificance (b 5 .12, t(107) 5 1.59, p . .10), but expectancy ratings were reliably related to global favorability (b 5 .57, t(107) 5 7.37, p , .001). These results support the role of expectancy as a mediator of the relationship between group membership and evaluations. We also included expectancy in a simultaneous regression with group membership, speech quality, and their interaction to predict global favorability. These results showed that both the main effects of speech quality (b 5 .54, t(105) 5 4.58, p , .001) and group membership (b 5 13, t(105) 5 1.76, p , .05) were significantly related to favorability, but expectancy ratings were only marginally related to favorability (b 5 .17, t(105) 5 1.30, p , .10) and the interaction was insignificant (b 5 .03, t(105) 5 0.36, p . .20). When we conducted this same analysis with the extreme outlier case removed, the results were similar but the relationship between expectancy and global favorability was reliable. Specifically, both speech quality (b 5 .45, t(104) 5 3.67, p , .001) and expectancy (b 5 .25, t(104) 5 2.00, p , .05) predicted favorability; group membership (b 5 .11, t(104) 5 1.43, p , .10) was marginally related, and the quality-by-membership interaction (b 5 .06, t(104) 5 0.81, p . .20) was not related to favorability. Discussion The results for the global favorability evaluations support the predictions of expectancy-violation theory, which posits that category-based violations induce extreme evaluations. More importantly, the results of Experiment 1 extend our understanding of the mediational role of prior category-based expectations. The regression analyses showed that measured expectations mediated the relationship between the group membership of the target and the perceivers’ global favorability toward the target. The trait-rating results were affected only minimally by the expectancy manipulation and instead were affected primarily by the valence of the individuating information. These results of the specific trait-capability measures showed support for assumed-characteristics theory and are consistent with those of Jackson et al. (1993). The somewhat different results for the evaluations of global favorability and ratings of traits may suggest that category-based expectancy 8 To assess the linear relationship between group membership and expectancy ratings, the expectancy ratings of the football team member condition was reversed coded for the regression analyses. In this way, expectancies would be positively related to evaluations.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

260

BETTENCOURT ET AL.

violations influence affective reactions to a target but individuating characteristics influence cognitive inferences about the targets’ traits. The results of this study offered little support for either the complexity-extremity hypothesis or the black-sheep hypothesis, although there was a trend toward more polarized evaluations of outgroup members. However, a test of the relation between cognitive complexity and polarized evaluations did not reveal the predicted correlation. Although the findings are consistent with expectancy-violation theory, the results for the job-applicant scenario could possibly be interpreted in another way. Perhaps White participants judged Black applicants more favorably than White applicants because participants wanted to appear nonprejudiced and thus evaluated all Black targets more positively. However, several points cast doubt on this interpretation. For example, if social desirability were the causal mechanism, trait-capability ratings should also have been more favorable for Black targets. In addition, the mediational role of category-based expectancies, supported in the regression analyses, would not be consistent with a social-desirability explanation. Moreover, Jussim and his colleagues have shown that more positive evaluations of Black targets are obtained with White subjects (Jussim et al., 1987) and with Black subjects (Coleman et al., 1995). Nevertheless, we conducted a second experiment to help minimize the viability of the social-desirability explanation of the results for the job-applicant scenario. EXPERIMENT 2 According to expectancy violation theory, if a category-based expectancy is not violated, Black and White targets with similar neutral qualifications should be evaluated similarly. In contrast, the social-desirability explanation predicts more favorable ratings for Black applicants no matter what the target description. With these predictions in mind, we designed Experiment 2 to include the same skillful and unskillful conditions used in Experiment 1 as well as a more neutral condition.9 The procedures for Experiment 2 were largely the same as those used in Experiment 1, but the complexity task was omitted and a neutral job-applicant scenario was added. The predictions for the skillful and unskillful conditions were identical to those for the same scenario in Experiment 1. Method Participants One-hundred and seventy-eight undergraduates at a large Midwestern university participated for partial credit in their Introductory Psychology course.

9 We also attempted to create a neutral version of the speech condition for the football–speech scenario. Our pilot test suggested that even this ‘‘neutral’’ version was unexpected for the football player and thus was not of moderate quality. Because social desirability was only an alternative explanation for the comparison between the Black and White targets, we proceeded only with the job-applicant scenario.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

261

INGROUP AND OUTGROUP EVALUATIONS

Neutral Target Descriptions The skillful and unskillful job applicants described in the scenarios used in Experiment 1 were defined by their style of dress and verbal communication skills. For Experiment 2, we created a third version of this scenario by describing a job applicant the quality of whom would be more intermediate or neutral. For this neutral version, the target expressed his desire for the job by saying, ‘‘I am ready to move on to something new . . . the job sounds interesting to me.’’ The job applicant was described as wearing ‘‘a sport coat and brown pants.’’ A pretest showed that, for the Black target, the unskillful applicant was judged as fitting the respective stereotype more (M 5 3.32) than the intermediate/ neutral (M 5 2.30) or the skillful applicant (M 5 1.81), but for the White target, the unskillful applicant was judged as fitting the stereotype less (M 5 1.68) than the neutral (M 5 3.25) or the skillful applicant (M 5 3.73), interaction F(1, 117) 5 14.28, p , .001.

Dependent Measures The global favorability, trait-capability, and expectancy ratings were the same as those in Experiment 1 with the exception that participants answered the question about how well the target fit stereotypes for the target’s ingroup on a 1-to-5 scale, allowing a neutral midpoint.

Procedure Sessions included 8 to 20 participants who were seated at desks with one empty seat between them. The instructions, read by a male experimenter, informed the participants that the study was about first impressions formed in common settings. After the general instructions, the six versions of the scenarios were handed out in such a way that subjects were randomly assigned to conditions.

Results Only participants who identified themselves as Caucasian or White were included in the analyses. Because the reliabilities of the expectancy, globalfavorability, and trait-capability indexes were adequate (as ranging from .75 to .90), the three sets of three items each were summed, averaged, and analyzed. We report three series of analyses: (a) F tests using the contrast weights of the predicted interactions (as described by Kirk, 1982, p. 98) the expectancy ratings, and global favorability ratings, (b) an ANOVA for the trait-capability ratings, and (c) regression analyses designed to test the hypothesis of expectancy as a mediator of group membership effects on global-favorability.10 10 The data were tested for violations of normality and extreme outliers. For this data set, 17 of 1332 ratings (equivalent to 1.3%) were extreme outliers. As recommended by Tabachnick and Fidell (1989) and Wilcox (1995), these extreme outliers were modified to the value of the next closest nonoutlying value in the distribution. The pattern of the data for the results for the data White Unskilled (WU), Black Unskilled (BU), White Neutral (WN), Black Neutral (BN), White Skilled (WS), and Black Unskilled (BU) targets before it was Winsorized was very similar to that for the Winsorized data. The means for the general favorability ratings before Winsorization were as follows: WU: M 5 2.97; BU: M 5 3.28; WN: M 5 4.85; BN: M 5 4.79; WS: M 5 5.38; BS: M 5 5.56, and the weighted interaction was significant, F(1, 146) 5 26.00, p , .01. The means for the trait–capability ratings before Winsorization were as follows: WU: M 5 3.12; BU: M 5 3.36; WN: M 5 4.74; BN: M 5 4.59; WS: M 5 5.46; BS: M 5 5.48, F(1, 142) 5 .66, p . .10. The means for the mean total score for the expectancy ratings before Winsorization were as follows: WU: M 5 8.38; BU: M 5 10.41; WN: M 5 15.42; BN: M 5 11.88; WS: M 5 16.62; BS: M 5 12.86, and the weighted interaction was significant, F(1, 146) 5 21.78, p , .01. The outcomes of analyses that eliminated the outliers were also similar to the Winsorized data. The means for the global favorability ratings with outliers removed were as

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

262

BETTENCOURT ET AL.

Expectancy. We predicted that the skillful job applicant (S) and the unskillful job applicant (U) would be stereotype violations and that the neutral job applicant (N) would not be a violation; as such, we used the following contrast weights for the predicted interaction on expectancy ratings: BS 5 21, WS 5 11, WN 5 0, BN 5 0, BU 5 11, WU 5 21. As shown in the upper panel of Fig. 3, sums of the ratings for all expectancy items suggested that the skillful Black applicant and the unskillful White applicant were relatively unexpected and the skillful White applicant and the unskillful Black applicant were relatively more expected, F(1, 142) 5 29.85, p , .01. Because the mean values for only the ratings of the targets’ fit with common stereotypes are more consistent with the findings of our pretest, we present the means for this ratings separately in the lower panel of Fig. 3. Global favorability ratings. We predicted that White perceivers would be more favorable toward Black skillful applicants than they would be toward White skillful applicants and that these perceivers would be less favorable toward White unskillful applicants than Black unskillful applicants, and finally, favorability toward the neutral applicants would be equally moderate. To test this interaction, we used the following contrast weights: BS 5 12, WS 5 11, BN 5 0, WN 5 0, BU 5 21, WU 5 22. Supporting the predictions of expectancy-violation theory, the analysis of the global favorability evaluations revealed a reliable interaction between job quality and group membership, F(1, 142) 5 27.11, p , .01. As shown in Fig. 4, the unskillful job applicant was evaluated less favorably when described as White than when described as Black, and the skillful job applicant was evaluated more favorably when described as Black than as White, but the neutral job applicants were evaluated as similar. For this measure, we also calculated, separately for Black and White targets, the range between the positive and the neutral targets and the range between the neutral and the negative targets. The resulting difference scores revealed directional, but not statistically significant, support for both the complexity–extremity and the black-sheep hypotheses. Supporting the complexity–extremity explanation, the difference between the positive and neutral targets showed that the range in evaluations of the Black, outgroup targets (difference 5 0.74) tended to be greater than that for the White targets (difference 5 0.52). In contrast, but supporting the predictions of the black-sheep hypothesis, the difference between the neutral and negative targets showed that the range in evaluations of the White, ingroup targets (difference 5 1.89) tended to be greater than that for the Black targets (difference 5 1.56). The results are interesting because they imply that whether evaluations of ingroup targets are polarized or outgroup targets are follows: WU: M 5 2.97; BU: M 5 3.28; WN: M 5 4.87; BN: M 5 4.91; WS: M 5 5.38; BS: M 5 5.63, and the weighted interaction was significant, F(1, 142) 5 26.50, p , .01. The means for the trait–capability ratings with outliers removed were as follows: WU: M 5 3.13; BU: M 5 3.36; WN: M 5 4.80; BN: M 5 4.67; WS: M 5 5.56; BS: M 5 5.48, F(1, 143) 5 .72, p . .10. The means for the mean total score for the expectancy ratings with outliers removed were as follows: WU: M 5 8.00; BU: M 5 10.41; WN: M 5 16.09; BN: M 5 11.84; WS: M 5 17.67; BS: M 5 12.86, and the weighted interaction was significant, F(1, 127) 5 45.67, p , .01.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

263

INGROUP AND OUTGROUP EVALUATIONS

FIG. 3. Mean values of the sum of the three ratings of expectancy and mean values of stereotypic fit for Black and White job applicants (Experiment 2).

polarized may depend on situations in which one version of the target description involves a category-based violation for either the ingroup or the outgroup, but the other does not involve a violation. Trait capability ratings. Consistent with the results of Experiment 1, only the main effect of job quality affected the participants’ ratings of the trait-capability of

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

264

BETTENCOURT ET AL.

FIG. 4. Mean evaluations of global favorability for Black and White job applicants (Experiment 2). (For the neutral targets, the mean value for the Black target is 4.84 and that for the White target is 4.86.)

the targets, F(2, 142) 5 98.98, p , .001. Although the Black unskillful applicant (M 5 3.36) was rated somewhat more capable than the same White applicant (M 5 3.13), neither the moderate nor the skillful Black applicants were rated higher than the respective White applicants. (The means were: BN 5 4.63; BS 5 5.48; WN 5 4.78; WS 5 5.51). The main effect of group membership (F(1, 142) 5 .03, p . .20) and the interaction (F(1, 142) 5 .74, p . .20) were not significant. Mediational analyses. To test the mediational effect of category-based expectancy on the global favorability evaluations, we used the same series of regression analyses used for Experiment 1. As reported in the previous section, the interaction between group membership and job quality affected both global favorability evaluations and expectancy ratings. Supporting expectancy as a mediator, the expectancy rating remained a predictor of global favorability toward the applicant (b 5 .12, t(144) 5 1.94, p , .05). However, the interaction between group membership and job quality also remained significant (b 5 .69, t(144) 5 11.58, p , .001). With several

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

265

INGROUP AND OUTGROUP EVALUATIONS

outliers removed (studentized distance 5 23.66 & 23.24), the effect of expectancy was more reliable (b 5 .18, t(142) 5 3.07, p , .01) and the interaction between group membership and quality was also significant (b 5 .68, t(142) 5 11.57, p , .001). Discussion As in Experiment 1, the results of Experiment 2 indicated that the favorable Black target and the unfavorable White target were relatively unexpected compared with the unfavorable Black and favorable White targets. When target descriptions violated category-based expectancies, targets were evaluated more extremely on the ratings of global favorability. Moreover, these results again revealed support for category-based expectancies as mediators of the effect of group membership on evaluations of global favorability. Although the ratings of trait-capability for the unskillful applicant were affected somewhat by group membership, in general, the trait ratings were unaffected by category-based expectancies. Similar to Experiment 1, the results may suggest that expectancy violation might primarily affect global, affective reactions. Part of the purpose of Experiment 2 was to determine if social desirability concerns would also lead to more favorable evaluations of Black targets when stereotypes were not violated. This alternative explanation of social desirability had also been raised, but criticized, in the discussions offered by Jackson et al. (1993) and Jussim et al. (1987). The results of Experiment 2 cast some doubt on this alternative explanation. In fact, the global-favorability ratings showed that the Black applicant with neutral qualifications was evaluated slightly more negatively than the respective White applicant, which would not be explained by social desirability. Finally, by including the more neutral condition, these findings further contribute to our understanding of extremity in evaluation because the pattern of results simultaneously revealed outgroup and ingroup polarization. That is, if we focus on only the results for the very positive and the neutral job applicants, the pattern of results for the global-favorability evaluation reveals outgroup polarization. However, if we focus on only the results for the very negative and the neutral job applicants, the pattern of results reveals ingroup polarization. It may be that the research that finds outgroup polarization has tended to violate expectations for outgroups with positive behavior, whereas the research that finds ingroup polarization has tended to violate expectations for ingroups with negative behavior. As we discussed previously, at least one study revealed that the group membership of the evaluator had little effect on evaluations of a target who violated expectations. Taynor and Deaux (1973) reported that more favorable evaluation of a female target was not dependent on the gender of the evaluator. Moreover, as we have noted, the results of Coleman et al.’s (1995) study that included Black participants replicated those of Jussim et al.’s (1996) study that included White participants. Taken together, these studies suggest that ingroup target membership of the evaluator may play a less important role in evaluations

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

266

BETTENCOURT ET AL.

when targets violate stereotyped expectancies. With this possibility in mind, we designed a third study that included targets and participants who varied ingroup membership. EXPERIMENT 3 This experiment is similar to Experiments 1 and 2 except that it includes men and women as the two group categories. Using these groups allowed us to readily include participants who were ingroup and outgroup members of the targets (i.e., both female and male participants). Instead of a scenario, pictures of ostensible job applicants and their resumes were prepared, pretested, and presented. The female and male targets were depicted as either having a job objective as a fashion news-writer or as a sports news-writer. The female-sport and the male-fashion conditions were considered violations of gender-based stereotypes. All resumes depicted relatively favorable qualifications. Thus, in Experiment 3, valence of the job applicant was not varied. Participants responded to measures of global favorability, trait-capability, and stereotypicality. Using expectancy-violation theory as a guide, we predicted that participants would evaluate the female sports writer more favorably than the female fashion writer, but participants would evaluate the male fashion writer more favorably than the male sports writer. Thus, in Experiment 3 we predicted that target’s gender and job objective would interact in affecting the global favorability evaluations. However, in line with the findings of Taynor and Deaux (1973) and Jussim and his colleagues (1987; 1996), we predicted that participants’ gender would not interact with target’s gender and job objective. Moreover, we predicted that perceptions of category-based expectancies would mediate global favorability evaluations. Finally, consistent with Experiments 1 and 2, we predicted that trait-capability ratings would not be affected by the stereotype violations. The complexity–extremity hypothesis predicts a two-way interaction between gender of participant and gender of target. According to the extremity hypothesis, female participants should rate male targets more favorably than female targets, but male participants should rate female targets more favorably than male targets. That is, because all applicants have positive qualifications, evaluators should judge the outgroup target more favorably. In contrast, the black-sheep hypothesis predicts the mirror image of this later prediction such that participants should evaluate ingroup targets more favorably. Method Participants One hundred and thirty-one female and 111 male undergraduates at a large Midwestern university participated for partial credit in their Introductory Psychology course.

Target Description A template resume was prepared with details that presented relatively positive information associated with newswriting. The resume contained information about characteristics that were identical in all versions of the resumes. This information included the following: education was a B.A.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

267

INGROUP AND OUTGROUP EVALUATIONS

in journalism; work experiences were reporter and editor positions at one large newspaper and two local newspapers, as well as food server at a restaurant; qualifications were listed as excellent ability to communicate through writing and experience in many facets of newspaper production. A section on personal information described the applicant as married (with a female-named spouse for the male target and a male-named spouse for the female target), flexible, adventure-seeking, and interested in a variety of hobbies including nature hiking and reading mystery novels. The sex of the applicant was manipulated both by the names at the top of the resume and by pictures of the applicants, which had been previously matched on perceived attractiveness and liking as a news writer. The name of the applicant was either listed as Thomas A. Johnson or Tonya A. Johnson. In addition to the other information on the resume, the job objective was listed either as a ‘‘Fashion writer covering women’s apparel and jewelry’’ or as a ‘‘Sports writer covering men’s basketball and football.’’ Pilot testing showed that the fashion-writer resume was more stereotypic for women (M 5 3.20) than for men (M 5 2.00) and the sports writer resume was more stereotypic for men (M 5 3.20) than for women (M 5 1.67), F(1, 17) 5 33.59, p , .001.

Dependent Measures There were two global favorability items (good/bad and favorable/unfavorable) and two traitcapability items (creative/uncreative and capable/uncapable). These items were rated on six-point scales with the more positive end of the dimension represented by 6 and the more negative end of the dimension represented by 1. Participants were to indicate the extent to which Thomas’s (or Tonya’s) behavior was expected. Participants responded on a six-point Likert-type scale with endpoints labeled ‘‘very expected’’ and ‘‘very unexpected.’’ Participants were also asked to indicate ‘‘to what extent does Thomas’s (or Tonya’s) behavior fit the stereotype for males (females)?’’ They responded on a four-point Likert-type scale with endpoints labeled ‘‘does not fit stereotype for males (females)’’ and ‘‘fits stereotype for males (females).’’11 Finally, a manipulation check question was given to assess whether or not participants remembered the sex of the applicant. Participants were asked to indicate whether the applicant was male or female. Additionally, the participants were asked to indicate their sex, age, ethnicity, and country of citizenship.

Procedure Participants were randomly assigned to a computer and directed to begin reading the instructions on the computer screen. All instructions, materials, and measures were presented via a Macintosh computer using Super Lab (Cedrus Corporation, 1991) software.12 The cover story informed participants that with the increasing use of electronic mail on the Internet, the work force was anticipating its use in job applications. In the future, it was anticipated that jobs would be advertised via electronic mail and that individuals would be able to send their applications, along with a photo, in direct response to a specific job announcement. Participants were told that the experiment was being run in conjunction with the School of Journalism to determine the most effective resume styles for candidates seeking positions in the field of newswriting via the Internet. They were asked to assume the role of the employer when reviewing the candidate’s resume. After viewing the photograph and carefully reading the resume, they were asked to rate the candidate on several dimensions by pressing the number on the keyboard that corresponded to their ratings. Upon completion of the task, participants were thanked and sent a debriefing letter in the mail.

11 Other ratings were also made by the participants, including attributions and desire to hire the applicant. Because these variables are not of central interest to the hypotheses of the current work, we do not report the outcomes for these variables. 12 Analyses of the response time to answer the dependent variables showed no effect of the manipulated variables.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

268

BETTENCOURT ET AL.

Results The number of female and male participants were distributed relatively equally across the four conditions. If participants failed to answer the question in the time that was allotted for each computer screen, the computer recorded the response as missing.13 The global-favorability items and the trait-capability items were averaged into two indices (as were .79 and .63, respectively). The globalfavorability and the trait-capability indices as well as the stereotype measure and the expectancy measure were analyzed using 2 (Target Gender) 3 2 (Job Objective) 3 2 (Participant Gender) ANOVAs.14 Due to extremely low reliability between the measures of stereotype fit and expectancy (a 5 .46), these items were not averaged. These two items were analyzed using a 2 (Target Gender) 3 2 (Job Objective) 3 2 (Participant Gender) MANOVA with 2 repeated measures. Ratings of stereotype fit and expectancy. The multivariate analysis of the measures of stereotype fit and expectancy revealed a significant interaction between target gender and job objective, F(2, 184) 5 48.39, p , .01. The univariate analysis of the stereotype-fit item showed that participants indicated that the female fashion writer (M 5 2.72) was more stereotypical than the female sports writer (M 5 1.89), but the male fashion writer (M 5 2.20) was less stereotypical than the male sports writer (M 5 3.25), F(1, 202) 5 98.24, p , .001. It was also the case that male applicants (M 5 2.74) were more expected than female applicants (M 5 2.27), F(1, 202) 5 21.10, p , .001. There was no main effect of job objective on rating of expectancy, F(1, 202) 5 1.63, p . .20. Consistent with the stereotype measure, the female fashion writer (M 5 4.89) was more expected than the female sports writer (M 5 4.12), but the male fashion writer (M 5 4.35) was less expected than the male sports writer (M 5 5.35), F(1, 200) 5 27.88, p , .001. There was a main effect of target gender, F(1, 200) 5 4.49, p , .05, but none of job objective, F(1, 200) 5 0.42, p . .10.

13 Sixteen (distributed relatively equally across conditions) of the 231 participants failed to correctly identify the sex of the target; therefore, their data were dropped from all analyses. One participant’s gender was not identified and was also removed. 14 Seventy-five of the 1582 values were extreme outliers (equivalent to 4%). The means for the general favorability ratings of the Female Fashion (FF), Female Sport (FS), Male Fashion (MF), and the Male Sport (MS) conditions before Winsorization were as follows: FF: M 5 5.09; FS: M 5 5.39; MF: M 5 5.10; MS: M 5 4.99, and the predicted interaction was significant, F(1, 217) 5 4.78, p , .05. The means for the trait–capability ratings before Winsorization were as follows: FF: M 5 4.77; FS: M 5 5.02; MF: M 5 4.72; MS: M 5 4.71; F(1, 217) 5 1.06, p . .10. The means for the ratings of fit with stereotypes before Winsorization were as follows: FF: M 5 2.91; FS: M 5 1.89; MF: M 5 2.20; MS: M 5 3.02; F(1, 202) 5 67.92, p , .01. Once again, the outcomes of analyses that eliminated the outliers were also similar to the Winsorized data. The means for the global favorability ratings with the outliers eliminated were as follows: FF: M 5 5.24; FS: M 5 5.43; MF: M 5 5.26; MS: M 5 5.06, and the predicted interaction was significant, F(1, 215) 5 6.98, p , .05. The means for the trait–capability ratings with the outliers eliminated were as follows: FF: M 5 4.92; FS: M 5 5.13; MF: M 5 5.05; MS: M 5 4.90; F(1, 213) 5 3.18, p 5 .07. The means for the ratings of fit with stereotypes with the outliers eliminated were as follows: FF: M 5 2.65; FS: M 5 1.89; MF: M 5 2.20; MS: M 5 3.31; F(1, 181) 5 81.30, p , .01.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

269

INGROUP AND OUTGROUP EVALUATIONS

FIG. 5. ment 3).

Mean evaluations of global favorability for Female and Male job applicants (Experi-

Global favorability and trait-capability ratings. Consistent with the results found in Experiments 1 and 2, we predicted that expectancy violations would affect global favorability evaluations but not trait-capability ratings. The mean values for global favorability evaluations are presented in Fig. 5. As predicted by expectancy-violation theory, there was a significant interaction between target gender and job objective, F(1, 217) 5 6.08, p , .01. Specifically, the female sports writer was evaluated more favorably than the female fashion writer, but the male fashion writer was evaluated more favorably than the male sports writer. Additionally, the main effect of target gender revealed that, overall, female targets (M 5 5.30) were evaluated more favorably than male targets (M 5 5.11), F(1, 217) 5 4.02, p , .05. The cognitive-complexity and the black-sheep hypotheses would predict interactions with participants’ gender (i.e., group membership of the evaluator). The results did not reveal any 3-way, F(1, 217) 5 2.76, p . .10, or 2-way interactions with participant gender (Participant Gender 3 Target Gender: F(1, 217) 5 2.63, p . .10; Participant Gender 3 Job Objective: F(1, 217) 5 0.34, p . .10). However, there was a main effect of participant gender, F(1, 217) 5 10.38, p , .01, indicating that, overall, female participants rated all targets more positively than did male participants (Ms 5 5.31, 5.07, respectively).

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

270

BETTENCOURT ET AL.

As predicted, there were no category-based expectancy effects on the measure of trait-capability; the interaction between target gender and job objective was not significant, F(1, 217) 5 2.49, p . .10. The ratings of trait capability of the female fashion and sports writer were similar (Ms 5 4.86, 5.11, respectively) as were those for the male fashion and sports writer (Ms 5 4.88, 4.81, respectively). In addition, there were no main effects or interactions with participant gender. Mediational analyses. We used the same series of regression analyses to test mediation as that described in Experiments 1 and 2. As reported previously, the Target Gender 3 Job Objective interaction affected the measures of global favorability, stereotype fit, and expectancy. The simultaneous regression results did not support the mediation hypothesis. The analyses using the expectancy item revealed no significant effects. The analyses using the stereotype items showed that, when entered together, neither the interaction (b 5 .10, t(207) 5 1.24, p . .10) nor the stereotype-fit rating was significant (b 5 .09, t(207) 5 1.09, p . .10), but a supplementary analysis did show that, when entered alone, stereotypic ratings were related to global favorability (b 5 .15, t(208) 5 2.18, p , .05). This univariate regression suggested some relation, although not mediational, between greater perceived violation of stereotypes and greater global favorability. Discussion The results of Experiment 3 supported the predictions of expectancy-violation theory by showing that positively depicted job applicants who violated stereotyped expectations were rated more extremely than those who did not. These later results are consistent with those found in Experiments 1 and 2. However, the results of the regression analyses for Experiment 3 failed to show that stereotyped expectancies mediated the relation between target-group membership and global evaluations, as had been shown in Experiments 1 and 2. Several reasons may account for this outcome. First, our measures of prior expectations for the target group are actually stand-ins for participants’ individually held stereotyped expectancies. That is, to avoid social desirability in responding, our questions ask participants to indicate whether the target violated commonly held stereotypes. We use this measure in an attempt to reduce the likelihood that participants might be reluctant to indicate their own stereotyped expectancies for racial and gender groups. Because it is an indirect measure of violation, it may be less correlated with individual affective evaluations than would be a measure of individually held stereotypes. Second, by their nature, the stereotype violation measures are very highly correlated with the expectancy-violation manipulation. It may be that, although there was a correlation between measured expectancy and global evaluations, in the simultaneous regression the high degree of correlation between the manipulation of expectancy and measure of expectancy rendered the latter nonsignificant. We would expect less of a correlation between the manipulation and measured expectancy if we could tap each individual’s degree of experienced

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

271

INGROUP AND OUTGROUP EVALUATIONS

violation. In the future, researchers may attempt to use groups for which participants will be less apt to be reluctant to reveal previously held expectations. Moreover, also consistent with the other experiments, the results of Experiment 3 showed that ratings of trait-capability were much less affected by the stereotype violations. Finally, supporting prior findings reported by Taynor and Deaux (1973) as well as those reported in Jussim et al. (1987) and Coleman et al. (1995), the participants’ ingroup membership did not affect the global favorability evaluations. This latter outcome failed to show support for either the complexity– extremity hypothesis or the black-sheep hypothesis. GENERAL CONCLUSIONS Taken together, the results of the present set of studies provide support for the predictions of expectancy-violation theory, which states that category-based expectancy violations produce extreme evaluations in the direction of the valence of the violation. Experiments 1 and 2 clearly revealed mediating effects of stereotyped expectancies and as such these studies are the first to support their mediational role in extreme evaluations. In addition, in this research we included a variety of group memberships and types of target descriptions in an attempt to cast further doubt on the viability of social desirability as an alternative explanation (see also Coleman et al., 1995; Jackson et al., 1993; and Jussim et al., 1987, 1996, for their refutation of this alternate explanation). In these studies, we observed fairly consistent effects of manipulated category-based violations with groups that are less politically charged than are racial-group comparisons, and also we observed effects of stereotype violations among both ingroup and outgroup evaluators. The replication of category-based expectancy predictions across groups and target depictions suggests that stereotype expectancy effects provide a more straightforward account of the pattern of outcomes than does a social desirability account. The evidence also concurs with Jussim’s (1991) assertion that individuating information substantially affects social judgments. As such, the present findings support those of Locksley and her colleagues (e.g., Locksley et al., 1980). Moreover, although we failed to find any direct mediating effect of cognitive complexity, some of the patterns of means were consistent with those predicted by the cognitive-complexity hypothesis, which states that judgments of outgroup members should be more extreme than those of ingroup members. Other researchers (Taynor & Deaux, 1973; Jackson et al., 1993; Jussim et al., 1987) have suggested the importance of category-based expectancies in person evaluation, but the current work is the first to explicitly test the mediational role of category-based expectations. Although in Experiment 3 only a correlation, but not mediation, existed for stereotyped expectancies and evaluations, the results of Experiments 1 and 2 provided considerable support for stereotype expectancy as a mediator of the relationship between group membership and extreme evaluations. These results suggest that when a person’s behavior violates a relevant category-

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

272

BETTENCOURT ET AL.

based expectancy, the experience of unexpectedness for the perceiver impacts the perceivers’ global, perhaps affect-related, evaluations. That violations of expectations might influence affective reactions makes sense in light of emotion theorists’ claim that unexpectedness is an intensity variable of emotion (Clore et al., 1994; Roseman et al. as cited in Clore et al.). This potential effect of unexpectedness on affective responses highlights the probable requirement that, to observe effects of expectancy violations, a target’s behavior must be in stark violation of category-based expectations and not merely inconsistent with them. In this vein, it might be helpful to consider expectancy-violating behavior on one end of a continuum and expectancy-consistent behavior on the other end of the same continuum, with category-inconsistent behaviors lying somewhere between these two endpoints. The distinction between behaviors that are violations of expectations and those that are inconsistent is important, because behaviors that are merely inconsistent with expectations may fail to have enough impact to elicit emotional reactions. Thus, the degree of impact of stereotypes on extreme evaluations might depend on whether targets act in ways that are stereotype-consistent, stereotype-inconsistent, or stereotype violations. These studies also suggest that expectancy violations may more greatly affect these global assessments of favorability than influences about traits that are closely related to the targets’ behavior, and as such the results are consistent with those found by other researchers (Glick et al., 1988; Jackson et al., 1993; Jussim et al., 1987). Ratings of such traits may be more affected by cognitive inferences drawn from the individuating information than by violated expectations. A Further Integration of the Predictions for Evaluative Extremity Jussim et al. (1987; also see Jussim et al., 1996) discussed how assumedcharacteristics, complexity extremity, and expectancy-violation theories might be integrated. They proposed that the effect of individuating characteristics, cognitive representations, and category-based expectations might be additive, and supported this proposal in an analysis of their data. Specifically, they said, ‘‘The simplest way to integrate these theories is to assume that the combined influence of processes specified by each theory is an additive function of their separate influences . . .’’ (p. 542). We agree that in some instances these processes can combine additively in their effects on evaluations of others. However, in other instances some of these key variables may be so highly interrelated that their effects are not added to each other but are slightly different aspects of a single effect. It is likely that cognitive complexity for an outgroup and category-based expectancies for that same group, while not entirely synonymous, are interrelated. For example, if a person has a very limited cognitive representation of the elderly, the person may be more likely to hold easily violated stereotypes for people who are elderly. Moreover, because persons have less complex cognitive representations of their outgroups (Linville, Fisher, & Salovey, 1989), extreme evaluations for outgroup members may be more likely because there is a greater probability

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

273

INGROUP AND OUTGROUP EVALUATIONS

that any random behavior will violate the simple cognitive, and perhaps stereotyped, representations of the outgroups. To understand how these variables such as cognitive complexity and stereotyped expectancies affect extremity in evaluation, it might be useful to consider the distinction between the two types of category-based expectancies that Jones (1990) discussed: dispositional-based expectancies and norm-based expectancies. He explained that ‘‘dispositional expectancies reflect a belief that group members share similar dispositions . . . , [whereas] normative expectancies are more situation bound and introduce the additional consideration of social rules and social sanction’’ (pp. 79–80). It may be that dispositional expectancies for outgroup members are more likely to be violated and thus induce more extreme reactions, whereas for ingroup members it is norm-based expectancies that are more likely to be violated and induce extreme reactions. In line with this later proposition, Marques et al. (1988) showed more extremely negative evaluations of ingroup members who violated important ingroup norms. Even if correct, these propositions are merely probabilistic statements. So, although violation of dispositional-based expectations for outgroups may more likely cause extreme evaluations, this will not always be the case. Similarly, although violations of normbased expectations may more likely create extreme evaluations of ingroup members, this will not always be the case. In general, these propositions are consistent with cognitive complexity theory. That is, if people have limited cognitive structures for outgroups, then they may be more likely to hold dispositional-based expectations, whereas people’s complex cognitive structures for their ingroups may make it more likely that they will have more norm-based expectations than disposition-based expectations. In conclusion, stereotyped expectations about groups are likely to affect evaluations of others in highly complex ways (Jussim et al., 1987). The effects of group membership, individuating information, expectancy violation, and cognitive complexity might be additive, as Jussim et al.’s model predicts, or they may sometimes be one and the same, or the effects of one of these might outweigh the contribution of the others in affecting the evaluations of ingroup and outgroup members. REFERENCES Anderson, N. H. (1968). Likableness ratings of 555 personality-trait words. Journal of Personality and Social Psychology, 9, 272–279. Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182. Bettencourt, B. A. (1989, May). Evaluation extremity: The role of expectancy on intergroup evaluations. Paper presented at the meeting of the Western Psychological Association, Reno, NV. Bettencourt, B. A. (1990). (Effects of manipulation of expectancy violation on evaluations of different types). Unpublished raw data. Branscombe, N. R., Wann, D. L., Noel, J. G., & Coleman, J. (1993). In-group or out-group extremity: Importance of the threatened social identity. Personality and Social Psychology Bulletin, 19, 381–388.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

274

BETTENCOURT ET AL.

Burgoon, J. K. (1986, February). Expectancy violations: Theory, research, and critique. Paper presented at the annual meeting of the Western States Communication Association, Tucson, AZ. Brewer, M. B. (1979). Ingroup bias in the minimal intergroup situation: A cognitive–motivational analysis. Psychological Bulletin, 86, 307–324. Cedrus Corporation. (1991). SuperLab (version 1.4) [Computer software]. Phoenix, AZ: Cedrus Corporation. Clary, E. G., & Tesser, A. (1983). Reactions to unexpected events: The naive scientist and interpretive activity. Personality and Social Psychology Bulletin, 9, 609–620. Clore, G. L., Schwarz, N., & Conway, M. (1994). Affective causes and consequences of social information processing. In R. S. Wyer & T. K. Srull (Eds.), Handbook of Social Cognition: Vol. 1. Basic Processes (2nd ed.). Hillsdale, NJ: Erlbaum. Coleman, L. M., Jussim, L., & Kelly, S. H. (1995). A study of stereotyping: Testing three models with a sample of blacks. Journal of Black Psychology, 21, 332–356. Glick, P., Zion, C., & Nelson, C. (1988). What mediates sex discrimination in hiring decisions? Journal of Personality and Social Psychology, 55, 178–186. Hass, R. G., Katz, I., Rizzo, N., Bailey, J., & Eisenstadt, D. (1991). Cross-racial appraisals as related to attitude ambivalence and cognitive complexity. Personality and Social Psychology Bulletin, 17, 83–92. House, W. C., & Perney, V. (1974). Valence of expected and unexpected outcomes as a function of locus of goal and type of expectancy. Journal of Personality and Social Psychology, 29, 454–463. Jackson, L. A., Sullivan, L. A., & Hodge, C. N. (1993). Stereotype effects on attributions, predictions, and evaluations: No two social judgments are quite alike. Journal of Personality and Social Psychology, 65, 69–84. Jones, E. E. (1990). Expectancy effects. In R. C. Atkinson, G. Lindzey, & R. F. Thompson (Eds.), Interpersonal perception (pp. 77–111). New York: Freeman. Jussim, L. (1991). Social perception and social reality: A reflection–construction model. Psychological Review, 98, 54–73. Jussim, L., Coleman, L. M., & Lerch, L. (1987). The nature of stereotypes: A comparison and integration of three theories. Journal of Personality and Social Psychology, 52, 536–546. Jussim, L., Fleming, C. J., Coleman, L. M., & Kohberger, C. (1996). The nature of stereotypes II: A multiple-process model of evaluations. Journal of Applied Social Psychology, 26, 283–312. Kirk, R. E. (1982). Experimental design. (2nd ed.). Pacific Grove, CA: Brooks Cole. Linville, P. W. (1982). The complexity–extremity effect and age-based stereotyping. Journal of Personality and Social Psychology, 42, 193–210. Linville, P. W., & Jones, E. E. (1980). Polarized appraisals of outgroup members. Journal of Personality and Social Psychology, 38, 689–703. Linville, P. W., Fisher, G. W., & Salovey, P. (1989). Perceived distributions of the characteristics of in-group and out-group members: Empirical evidence and a computer simulation. Journal of Personality and Social Psychology, 57, 165–188. Locksley, A., Borgida, E., Brekke, N., & Hepburn, C. (1980). Sex stereotypes and social judgment. Journal of Personality and Social Psychology, 39, 821–831. Marques, J. M., Robalo, E. M., & Rocha, S. A. (1992). Ingroup bias and the ‘‘black sheep’’ effect: Assessing the impact of social identification and perceived variability on group judgements. European Journal of Social Psychology, 22, 331–352. Marques, J. M., & Yzerbyt, V. Y. (1988). The black sheep effect: Judgmental extremity towards ingroup members in inter- and intra-group situations. European Journal of Social Psychology, 18, 287–292. Marques, J. M., Yzerbyt, V. Y., & Lyens, J. (1988). The black sheep effect: Extremity of judgments towards ingroup members as a function of group identification. European Journal of Social Psychology, 18, 1–16. Millar, M. G., & Tesser, A. (1986). Thought-induced attitude change: The effects of schema structure and commitment. Journal of Personality and Social Psychology, 51, 1–11.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant

275

INGROUP AND OUTGROUP EVALUATIONS

Roseman, I. J., Antoniou, A. A., & Jose, P. E. (1992). Appraisal determinants of emotions: Revising current theories. Unpublished manuscript, University of California, Los Angeles. Tabachnick, B. G., & Fidell, L. S. (1989). Using multivariate statistics. New York: Harper & Row. Taynor, J., & Deaux, K. (1973). When women are more deserving than men: Equity, attribution and perceived sex difference. Journal of Personality and Social Psychology, 28, 360–367. Tukey, J. W. (1990). Data-based graphics: Visual display in the decades to come. Statistical Science, 5, 327–339. Wilcox, R. R. (1995). Statistics for the social sciences. San Diego, CA: Academic Press.

JESP 1323 @sp3/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_317z05

jant