In this section we first describe theoretical perspectives on the contemporary racial atti- tudes of Whites toward Blacks, which suggests the existence and operation of dual atti- tudes – one attitude explicit and the other implicit. We then review the evidence on implicit racial attitudes and the reliability and validity of these measures. After that we explore, conceptually and empirically (through meta-analysis), the relationship between implicit and explicit racial attitudes. Next, we examine the relationship between implicit and explicit racial attitudes and interracial behavior. Finally, we consider the malleability of implicit attitudes. Although our focus is on the racial attitudes of Whites toward Blacks, we also discuss related work and findings in other domains.
Theoretical perspectives
Whereas the traditional, “red-neck” form of racial prejudice is considered to be direct and univalently negative, the contemporary racial attitudes of Whites, particularly White Americans, are hypothesized to be more complex – reflecting both negative and positive reactions. Two approaches exemplifying this complexity are the aversive racism frame- work and the symbolic racism framework.
According to the aversive racism perspective (Gaertner & Dovidio, 1986; see also Dovidio & Gaertner, 1998), many people who consciously, explicitly, and sincerely support egalitarian principles and believe themselves to be nonprejudiced also develop, through normal cognitive (e.g., social categorization), motivational (e.g., needs for group status), and sociocultural (e.g., social learning) processes, unconscious negative feelings and beliefs about Blacks and other historically disadvantaged groups (e.g., Latinos and women). The attitudes of aversive racists are consciously egalitarian but unconsciously negative. The aversive racism framework further suggests that contemporary bias is
expressed in indirect ways that do not threaten the aversive racist’s nonprejudiced self- image, such as when inappropriate behavior is not obvious or when a negative response can be justified on the basis of some factor other than race.
Symbolic racism theory (Sears, Van Laar, Carillo, & Kosterman, 1997) also proposes that racism is now more subtle, indirect, and less conscious than in the past. According to this theory, negative feelings toward Blacks, which Whites acquire early in life, are rel- atively stable across the life span and persist into adulthood. When primed, these implicit predispositions influence responses to racially associated attitudinal objects, such as race- related policies. However, because explicit beliefs change more fully and rapidly than these racial feelings and are more likely to conform to prevailing egalitarian norms, these neg- ative attitudes are expressed indirectly, symbolically, or in rationalizable ways (as in terms of opposition to busing or resistance to preferential treatment) rather than directly or overtly (as in support for segregation).
In general, both aversive and symbolic racism theories propose that Whites may simul- taneously hold egalitarian attitudes about Blacks while also harboring negative racial feel- ings. Although the aversive racism framework has its historical roots in psychodynamic principles (see Kovel, 1970), recent treatments (e.g., Dovidio, Kawakami, Johnson, Johnson, & Howard, 1997; Dovidio & Gaertner, 1998) have reconceptualized it in terms of dual attitudes, one explicit and egalitarian and the other implicit and negative. Simi- larly, symbolic racism posits that many Whites develop nonprejudiced explicit attitudes as a consequence of prevailing norms and may genuinely believe that discrimination is no longer a problem, but they continue to have implicit negative attitudes that are ves- tiges of earlier socialization and experience. Thus, both frameworks are compatible with Wilson et al.’s (2000) recent general model of dual attitudes. Also consistent with the dual attitudes model, aversive racism and symbolic racism perspectives propose that whereas explicit, nonprejudiced attitudes may govern overt and deliberative forms of interracial behavior, implicit negative attitudes are related to indirect, subtle, and less obvious racial biases. In the next section, we examine evidence for the activation of implicit negative racial attitudes of Whites toward Blacks.
Implicit attitudes
Evidence of implicit racial attitudes has been generally consistent and strong. Response latency procedures, in particular, have demonstrated that racial attitudes and stereotypes may operate like other stimuli to facilitate responses and decision making about related concepts (e.g., doctor–nurse). In general, the greater the associative strength between two stimuli, the faster people can make decisions about them (e.g., Gaertner & McLaughlin, 1983).
Convergent evidence has been obtained with a variety of different priming procedures.
For example, a representation of Black and White racial categories is often presented for a short period of time before participants are asked to make a decision about a positive or negative word that follows. Faster response times are assumed to reflect stronger associations between the category and the evaluation. Dovidio, Evans, and Tyler (1986) and Judd, Park, Ryan, Brauer, and Kraus (1995) using semantic categories as primes;
Fazio, Jackson, Dunton, and Williams (1995) employing photographs of Blacks and Whites as primes; Dovidio et al. (1997) using subliminally presented schematic faces of Blacks and Whites as primes, and Wittenbrink, Judd, and Park (1997) using subliminal category labels as primes have found faster response latencies to negative traits after Black than White primes and faster response latencies to positive traits after White than Black primes.
The Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998) represents another technique for assessing implicit attitudes.1The IAT is based on the assumption that it is more difficult, and thus takes longer, to produce evaluatively incom- patible than compatible responses. Specifically, participants in the IAT are instructed to classify targets and attributes by pressing the appropriate key. In an IAT task, for example, participants’ first two steps may involve practice classifying Black and White photographs and then positive and negative evaluative words using specified response keys on the left or right. In subsequent steps, participants respond to combinations of photographs and words using the same response key. It is assumed that when the target (e.g., a social group) and an attribute (e.g., pleasant) share the same response key, people will make the response more quickly when the evaluation of the target group and the attribute are congruent than when they are incongruent. Studies using the IAT (Dasgupta, McGhee, Greenwald, & Banaji, 2000; Greenwald, McGhee, & Schwartz, 1998, Experiment 3; Ottaway, Hayden, & Oakes, in press; Rudman, Ashmore, & Gary, 1999) have also reliably provided evidence of relatively negative implicit attitudes of Whites toward Blacks.
Other studies assessing implicit racial prejudice have employed indirect techniques for measuring intergroup bias more generally. For instance, research on the linguistic inter- group bias (Maass, Salvi, Arcuri, & Semin, 1989) has shown that undesirable actions of outgroup members are encoded at more abstract levels (e.g., she is hostile) whereas desir- able behaviors are encoded at more concrete levels (e.g., she walked across the street holding the old man’s hand) relative to the same behaviors of ingroup members. These biases produce a persistence of more positive cognitions and evaluations about ingroup than outgroup measures. Because behaviors encoded at concrete levels are individual instances that can be discounted, they do not necessarily affect more general beliefs – rep- resented by abstract schematas. Thus, for example, positive stereotype-disconfirming behaviors of outgroup members, which are encoded at a concrete level, would leave the more abstract, dispositionally based, stereotypic schema undisturbed. Von Hippel, Sekaquaptewa, and Vargas (1997) found evidence of implicit prejudice of Whites toward Blacks on a measure of linguistic intergroup bias, as well as other measures of biased attributions.
Evidence for implicit social attitudes has not only been obtained for Whites toward Blacks, but also for ingroups and outgroups generally (Perdue, Dovidio, Gurtman, &
Tyler, 1990, with subliminal priming), and for Blacks toward Whites (Judd et al., 1995;
Fazio et al., 1995, with supraliminal priming), men and women toward women (Banaji
& Greenwald, 1995, in judgments of fame), between Japanese and Korean students (Greenwald et al., 1998, using the IAT), between northern and southern Italians (Maass, Ceccarelli, & Rudin, 1996, with the linguistic intergroup bias), and between northern Germans and Bavarians (Newman et al., 1998, on the IAT).
Reliability and validity of implicit attitudes
If implicit measures are assumed to reflect actual attitudes, then it is also important to examine their psychometric properties. For instance, in their edited book, Measures of Personality and Social Psychological Attitudes, Robinson, Shaver, and Wrightsman (1991) outline basic evaluative criteria for these measures that include “reliability (both test-retest reliability and internal consistency), and validity (both convergent and discriminant)” (p.
2). These criteria have only rarely been applied to response latency measures of attitudes or stereotypes.
With respect to test-retest reliability, we were unable to locate any published studies for implicit racial attitudes. Unpublished data by Rudman et al. (1999), however, did reveal a test-retest correlation of .50 over a nine-week period for an IAT prejudice measure.
Data for implicit stereotypes also suggest some stability over time. Using a variation of Banaji and Hardin’s (1996) task, in which stereotype-related traits were used as primes that preceded trials on which participants were instructed to categorize photographs of people as either White or Black, Kawakami and Dovidio (in press, Study 2) found evi- dence of stereotypic biases. Participants responded faster to stereotype-consistent trait- photograph trials than to stereotype-inconsistent ones. The test-retest correlation of the extent to which each individual exhibited this stereotypic bias across two administrations of the task in a one-hour session was .42 (p<.001). Across a 5- to 15-day period, the correlation for an independent sample of White participants was comparable, and in fact somewhat higher, .60. Test-retest reliability over a three-week period using a subliminal priming task (see Dovidio et al., 1997) was .50 (Kawakami & Dovidio, in press, Study 3). Using an IAT procedure, Rudman et al. (1999) found test-retest correlations of .48 for negative racial stereotypes and .54 for positive racial stereotypes over nine weeks. For each of these four samples, participants showed significant evidence of implicit racial stereotyping for both the first and second administrations of the task, but individual dif- ferences in implicit stereotyping reflected only modest reliability. The reliability coeffi- cients for implicit stereotyping were substantially lower than those for explicit prejudice.
Over a three-week period Kawakami and Dovidio (in press, Study 3) found the reliabil- ity coefficient for the Modern Racism Scale (McConahay, 1986) was .64 and for the Atti- tudes Toward Blacks Scale (Brigham, 1993) was .98.
The test-retest reliability for implicit racial stereotyping obtained in these studies is similar to the level for other implicit measures. Kawakami and Dovidio (in press, Study 3) reported a test-retest correlation of .51 for two administrations of the Banaji and Hardin (1996) priming task for gender stereotypes within the same session. Pelham and Hetts (1999) found a test-retest correlation of .47 for a measure of implicit self-regard (based on a word-completion task) over a five-week period. A measure of implicit group regard was less stable, r =.22. Taken together, this line of research suggests that at an aggregate level evidence of implicit stereotypes and negative racial attitudes can be reli- ably obtained, but the stability of individual differences in these areas is modest – and sub- stantially less than one would expect for established explicit measures.
Evidence for a correlation among different measures of implicit attitudes and stereo- typing is also sparse. Dovidio and Kawakami (in press) obtained a moderate correlation,
.28, between responses on one measure of implicit racial stereotyping based on a Banaji and Hardin-type (1996) task and another measure from a subliminal priming procedure (Dovidio et al., 1997). This relationship is somewhat weaker than the correlation, .39, between implicit racial attitudes on the IAT when White and Black names were used as target stimuli and responses of the same participants when White and Black pictures were the target stimuli (Dasgupta et al., 2000). With respect to a different type of attitude object, attitudes toward cigarettes, Sherman, Presson, Chassin, Rose, and Koch (1999) found an overall correlation of .22 between a priming measure (based on Fazio et al., 1995) and the IAT in one study (Study 2) but a lower correlation, .06, in another (Study 3).
The modest relationship for implicit measures assessed with different techniques may initially be unsettling, but it is not altogether surprising given work in cognitive psy- chology. Methodologically, De Houwer (1999) has argued that techniques, such as Stroop and priming procedures, may share some superficial similarities in how participants are asked to make decisions but they differ in fundamental structural ways. As a consequence, they tap into somewhat different cognitive processes. Conceptually, research in cognitive science (Squire & Kandel, 1999) indicates that various forms of implicit memory can be largely dissociated from one another. Thus the relatively weak relationship among implicit measures of social attitudes and stereotypes is consistent with empirical and theoretical work in cognition more generally. In the next section, we explore the relationship between implicit and explicit measures of racial attitudes.
Relationship between implicit and explicit measures
Intuitively, one might expect that implicit attitudes and explicit measures would be directly related because they are likely to be rooted in the same socialization experiences.
Moreover, at first glance one might consider this to represent a test of convergent validity. However, researchers in this area have argued that implicit and explicit measures involve different processes and thus are not necessarily expected to be highly corre- lated (see Dovidio et al., 1997; Greenwald & Banaji, 1995; Wilson et al., 2000). Theo- retically, response latency measures and self-report measures may reflect the distinc- tion between activation and application (Bargh, 1999; Devine, 1989; Gilbert & Hixon, 1991). The presentation of an attitude object may automatically activate an associ- ated evaluation from memory (Fazio et al., 1995) which may influence its subsequent use in judgments. However, as Gilbert and Hixon (1991) argue, automatic activation
“does not mandate such use, nor does it determine the precise nature of its use. It is possible for activated information to exert no effect on subsequent judgments or to have a variety of different effects” (p. 512; see also Fiske, 1989). Thus, it is quite possible that implicit attitude activation and explicit expressions of prejudice could be empirically unrelated.
As we noted earlier, a dissociation between implicit attitudes and self-reported atti- tudes may be likely to be observed generally for socially sensitive issues (Dovidio & Fazio, 1992; Fazio et al., 1990), and particularly for racial attitudes. Devine (1989; see also Devine, Plant, & Blair, this volume, chapter 10), for example, proposed that high- and
low-prejudiced people are equally knowledgeable about cultural stereotypes about minor- ity groups and similarly activate these stereotypes automatically with the real or symbolic presence of a member of that group. Low- and high-prejudiced individuals differ, however, in their personal beliefs and their motivations to control the potential effects of the automatically activated cultural stereotypes. Lower prejudiced people are more moti- vated to control, suppress, and counteract their initial, automatic, biased reactions. Thus unconscious associations, which are culturally shared and automatically activated, may be disassociated from expressions of personal beliefs that are expressed on self-report mea- sures of prejudice and systematically vary.
Other theoretical perspectives on contemporary racism also imply a relatively weak relationship between implicit and explicit attitudes. The aversive racism framework, for example, focuses on three theoretically relevant cells representing the combinations of implicit and explicit attitudes: People who are low in self-reported prejudice and nonbi- ased on implicit measures (nonprejudiced); people who are high on both (traditional racists); and people who are low in self-reported prejudice but biased on implicit mea- sures (aversive racists). This taxonomy of racial attitudes associated with the aversive racism framework – three of the four cells in an Explicit x Implicit Attitudes matrix – suggests a positive but weak empirical relationship. For instance, Dovidio, Gaertner, and Kawakami (1998) conducted a series of simulations based on the assumptions of the aver- sive racism framework and on the data from implicit and explicit measures to determine correlations between implicit and explicit measures. The resulting average correlation was .30, which is positive but modest. Thus weak correlations between explicit and implicit attitudes may not reflect weak measures, but may instead represent the nature of con- temporary prejudice.
To summarize the empirical findings on the relationship between explicit and implicit racial attitudes, we conducted a meta-analytic review of studies using a range of different implicit measures of racial prejudice, including the physiological measure of GSR as well as response latency measures, by Whites toward Blacks. The results of these 27 tests involving 1562 participants are presented in Table 9.1. Overall, there was a significant, Z=7.44,p<.001, but modest positive relationship, mean Fisher Z =.249, mean r = .244 (fail-safe number =524). The magnitude of this effect approximated Dovidio et al.’s (1998) simulations. Although there was some variability in magnitude, the relationships between implicit and explicit attitudes were significant across 14 tests involving priming measures, Z=3.87,p<.001, mean Fisher Z=.156, mean r=.155 (fail-safe number = 63); three tests using other latency measures (e.g., times for looking at pictures), Z=3.84, p<.001, mean Fisher Z =.500, mean r =.463 (fail-safe number =13); and four tests with physiological measures, Z=3.98,p<.001, mean Fisher Z=.388, mean r=.4370 (fail-safe number =19).
Blair (in press) also conducted a comprehensive, narrative review of the relationship between implicit and explicit attitudes, not only of Whites toward Blacks but of attitudes toward other groups as well. Blair’s review indicates that the relationship between explicit and implicit attitudes toward women is variable and, on average, even lower than the rela- tionship between implicit and explicit attitudes toward Blacks: Banaji and Greenwald (1995, Study 3), r = -.10; Moskowitz, Wasel, Gollwitzer, and Schaal (1998, Study 1), r= +.31; von Hippel et al. (1997), r =.01.
Study Implicit prejudice measure Explicit prejudice Statistic (df, n, Effect
measure direction) size (r)
Best et al. (1976) Color Meaning Test II PRAM II r(54)=.62 0.62
{56}{+}
Blascovich et al. Latency for identification Modern Racism Scale F(1,13)=4.60 0.51
(1997, study 1) of Black and White faces {15}{+}
Blascovich et al. Latency for identification Modern Racism Scale F(1,36)=4.40 0.33
(1997, study 2) of Black and White faces {38}{+}
Bray (1950) Conformity in Attitudes Toward r(48)=.108 0.11
autokinetic judgments Negroes Scale {50}{+}
Dasgupta et al. IAT using names and Overall effect r(73)=.145 0.15
(2000) pictures {75}{+}
Feelings Thermometer r(73)=.22 Semantic Differential r(73)=.31 Modern Racism Scale r(73)= -.055 Diversity Scale r(73)=.06 Discrimination Scale r(73)=.045
Dovidio et al. Evaluative race priming Overall effect r(22)=.215 0.22
(1997, study 1) (Black vs. White faces; {24}{+}
positive vs. negative Attitudes Toward r(22)=.28
adjectives) Blacks Scale
Modern Racism Scale r(22)=.15
Dovidio et al. Evaluative race priming Overall effect r(31)=.55 0.55
(1997, study 2) (Black vs. White faces; {33}{+}
positive vs. negative Modern Racism Scale r(31)=.60
adjective) Old-Fashioned r(31)=.49
Racism Scale
Dovidio et al. Evaluative race priming Overall effect r(31)= -.03 -0.03
(1997, study 3) (Black vs. White faces; {33}{-}
positive vs. negative Modern Racism Scale r(31)=.01
adjectives) Old-Fashioned r(31)= -.07
Racism Scale
Fazio et al. Evaluative race priming Modern Racism Scale r(51)=.15 -0.15
(1995, study 1) (Black vs. White faces; (53}{-}
positive vs. negative adjectives)
Fazio et al. Evaluative race priming Modern Racism Scale r(47)=.28 -0.28
(1995, study 2) (Black vs. White faces; {49}{-}
positive vs. negative adjectives)
Fazio et al. (1995, Evaluative race priming Modern Racism Scale NS 0
study 4) (Black vs. White faces; {117}{0}
positive vs. negative adjectives)
Study Implicit prejudice measure Explicit prejudice Statistic (df, n, Effect
measure direction) size (r)
Greenwald et al. Implicit Race Overall effect r(24)=.144 0.144
(1998) Association Test {26}{+}
(Black vs. White Semantic Differential r(24)=.21 names; positive vs. Modern Racism Scale r(24)=.07 negative words) Feelings Thermometer r(24)=.13 Diversity Index r(24)=.24 Discrimination Index r(24)=.07
Porier & Lott GSR Overall effect r(58)=.25 0.25
(1967) {60}{+}
E-scale r(58)=.38
Opinionation Scale r(58)=.13
Rudman, Ashmore, Implicit Race Overall effect r(62)=.30 0.30
& Gary (study 1) Association Test {64}{+}
(Black vs. White Modern Racism Scale r(62)=.37 names; positive vs. Feelings Thermometer r(62)=.23 negative words)
Rudman, Ashmore, Implicit Race Overall effect r(45)=.295 0.30
& Gary (study 2) Association Test (Black {47}{+}
vs. White names; Modern Racism Scale r(45)=.36 positive vs. negative Feelings Thermometer r(45)=.23 words)
Schnake & Ruscher Linguistic Intergroup Modern Racism Scale F(1,62)=3.97 0.25
(1998) Bias {65}{+}
Sensening et al. Examination times of Semantic Differential F(1,22)=8.6 0.53
(1973) pictures of Black and {24}{+}
White students
Thomsen (1991) Evaluative race priming Overall effect r(45)=.191 0.19
(Black vs. White faces; {47}{+}
positive vs. negative Modern Racism Scale r(45)=.16
adjectives) Multifactor Racial r(45)=.221
Attitude Inventory
Tognacci & Cook GSR Multifactor Racial F(1,20)=7.71 0.53
(1975) Attitude Inventory {24}{+}
Vanman et al. Facial EMG in Modern Racism Scale F(1,23)=7.485 0.49
(1997, study 3) response to photos {25}{+}
of Black vs. White students
Vidulich & GSR Attitudes Toward F(1,36)=.98 0.16
Krevanick (1966) Negroes Scale {40}{+}
von Hippel et al. Linguistic Intergroup Modern Racism Scale r(188)= -.05 -0.05
(1997, study 1) Bias {190}{-}
von Hippel et al. Linguistic Intergroup Modern Racism Scale r(116)=.03 0.03
(1997, study 2) Bias {118}{+}