How Dissociated Are Implicit and Explicit Racial Attitudes? a Bogus Pipeline Approach Jason A

How Dissociated Are Implicit and Explicit Racial Attitudes? A Bogus Pipeline Approach Jason A. Nier

To cite this version:

Jason A. Nier. How Dissociated Are Implicit and Explicit Racial Attitudes? A Bogus Pipeline Approach. Group Processes and Intergroup Relations, SAGE Publications, 2005, 8 (1), pp.39-52. 10.1177/1368430205048615. hal-00571590

HAL Id: hal-00571590 https://hal.archives-ouvertes.fr/hal-00571590 Submitted on 1 Mar 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 39

Group Processes & G Intergroup Relations P 2005 Vol 8(1) 39–52 I R How Dissociated Are Implicit and Explicit Racial Attitudes? A Bogus Pipeline Approach

Jason A. Nier Connecticut College

The current study examined the implicit and explicit attitudes of White Americans toward African-Americans. A variation of the Bogus Pipeline procedure was employed to determine if the apparent dissociation between implicit and explicit measures of racial attitudes that is reported in previous research might be exaggerated. The results indicated that the relationship between implicit and explicit attitudes was only signiﬁcant under Bogus Pipeline conditions, while implicit and explicit attitudes were largely dissociated when they were measured under normal circumstances. Thus, it appeared that as the motivation to accurately report explicit attitudes increased, the implicit–explicit relationship strengthened and the dissociation between implicit and explicit racial attitudes was substantially reduced. The results indicate that Whites’ implicit and explicit attitudes toward African-Americans may not be as greatly dissociated as some theories of racial attitudes have presumed.

keywords implicit attitudes, racial attitudes

FOR DECADES social psychologists have observed participants’ self-reports (Greenwald et al., that White attitudes toward African-Americans 1998). Furthermore, implicit measures of White have become increasingly positive (Jones, 1997). attitudes toward African-Americans generally Despite the increased positivity of self-reported yield attitude assessments that are more negative attitudes, subtle measures of prejudice suggest than explicit measures, suggesting that they may that negative attitudes persist (Dovidio & relate to more subtle forms of negative inter- Gaertner, 1998), calling into question the racial behavior (Dovidio, Kawakami, Johnson, validity of the increasingly positive self-reports. Johnson, & Howard, 1997). Recently, however, the introduction of implicit measures of racial attitudes, such as the Implicit Author’s note Association Test (Greenwald, McGhee, & Address correspondence to Jason A. Nier, Schwartz, 1998), have promised to enhance the Department of Psychology, Box 5305, accuracy of racial attitude assessments. These Connecticut College, 270 Mohegan Avenue, measures are believed to be less reactive than New London, CT 06320, USA explicit measures since they do not rely upon [email: [email protected]]

Group Processes & Intergroup Relations 8(1)

The relationship between implicit and explicit attitudes are independent cognitive explicit measures of racial attitudes representations that are the result of different psychological processes. While there are differ- While implicit measures seem to yield attitude ing theoretical variations on this general theme assessments that are more negative than self- (Devine, 1989; Wilson, Lindsey, & Schooler, reported attitudes, interpreting the precise 2000), the basic idea is that the poor meaning of these measures has proven prob- implicit–explicit relationship is due primarily to lematic. Perhaps the most distinctive aspect of the fact that implicit and explicit measures implicit measures is their apparently inconsis- assess different evaluations of the same attitude tent relationship to traditional, explicit object. In those circumstances where implicit measures of racial attitudes (for reviews see and explicit evaluations differ, as is often Blair, 2001; Dovidio, Kawakami, & Beach, 2001; presumed to be the case for White attitudes Fazio & Olson, 2003). Although there have toward African-Americans, attitudes are said to been numerous exceptions (e.g. Cunningham, be ‘dissociated’ (Greenwald et al., 1998) or Preacher, & Banaji, 2001; Wittenbrink, Judd, & ‘independent’ (Wilson et al., 2000). Park, 1997), many studies examining the Furthermore, some theories of racial relationship between the two measures have attitudes are based, in part, on the notion of found that implicit attitudes are often not ‘dissociation’ that is associated with the represen- correlated with explicit attitudes (e.g. Boniecki tational perspective. For example, Devine’s & Jacks, 2002, r = .03; Dasgupta, McGhee, (1989) influential model of prejudice holds Greenwald, & Banaji, 2000, r = .02; Greenwald that all Whites hold automatically activated et al., 1998, r = .07). This has led researchers to negative beliefs, yet individuals vary in the neg- disagree about what implicit measures are ativity of explicit beliefs toward African- actually assessing and has even led some to Americans. As evidence for this model, Devine question the utility of implicit measures pointed out that responses on the Modern (Cameron, Alvarez, & Bargh, 2000). Racism Scale, an explicit measure of racial atti- While it remains clear that there is a con- tudes, were unrelated to the degree to which siderable degree of inconsistency between racial stereotypes were automatically activated implicit and explicit measures of White atti- using a subliminal priming procedure. tudes toward African-Americans, there have Further evidence for the dissociation been several different explanations offered to between implicit and explicit attitudes comes explain this poor correspondence. One expla- from the differential predictive validity of these nation focuses on methodological concerns, measures. Implicit measures tend to predict particularly the psychometric properties of spontaneous behaviors that are difficult for implicit measures. The reliability of implicit individuals to consciously monitor and control, measures is generally lower than the reliability while explicit measures tend to predict deliber- of explicit measures (Kawakami & Dovidio, ative responses that are easily monitored and 2001), and as a result the error variance that is controlled (Dovidio et al., 1997; Dovidio, present in the measurement of implicit atti- Kawakami, & Gaertner, 2002; Wilson et al., tudes may suppress the implicit–explicit 2000). Thus the notion that implicit and relationship. Consistent with this interpre- explicit attitudes are largely dissociated seems tation, procedures that have been introduced to be consistent with prevailing theories of atti- to improve the reliability of implicit measures tudes and the available empirical evidence. strengthen the magnitude of the implicit– However, the empirical evidence upon which explicit relationship (Cunningham et al., 2001; the representational perspective is based does Greenwald, Nosek, & Banaji, 2003). not account for the impact of social desirability Another explanation for the weak implicit– concerns, a motivational variable that can inﬂu- explicit link, which Blair (2001) called the rep- ence the expression of explicit racial attitudes. resentational perspective, holds that implicit and Specifically, since most individuals are

40 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 41

Nier implicit and explicit racial attitudes

motivated to present a positive self-image strengthen under conditions designed to (Schlenker & Weingold, 1992), they may minimize socially desirable responding. This manipulate their explicit responses so as to question has important theoretical implications appear unprejudiced. Supporting this view, since the representational explanation holds Sigall and Page’s (1971) classic ‘bogus pipeline’ that the dissociation between implicit and study indicated that Whites expressed more explicit attitudes is due primarily to the negative attitudes toward African-Americans different psychological processes that produce when they believed that a device was able to these two types of attitudes, rather than social accurately assess their true attitudes. Thus when desirability forces. participants were led to believe that a machine was a ‘pipeline’ to their true inner states and Current study attitudes, social desirability forces presumably exerted less influence on the self-reported racial In the current study, the implicit and explicit attitudes and as a result they reported attitudes attitudes of Whites toward African-Americans that more closely reflected their true beliefs. were assessed under different circumstances in More recent research indicates that social desir- order determine the degree to which implicit ability forces continue to influence Whites’ and explicit attitudes are dissociated when explicit attitude toward African-Americans (e.g. social desirability forces are minimized. A vari- Fazio, Jackson, Dunton, & Williams, 1995). ation of the bogus pipeline procedure was Whereas the social desirability perspective has employed to examine whether the implicit– focused exclusively on the malleability of explicit relationship would be stronger under explicit measures under different conditions bogus pipeline conditions, relative to the (Roese & Jamieson, 1993) and there seems to be circumstances under which implicit and little doubt that social desirability forces influ- explicit measures have traditionally been ence the expression of explicit attitudes, less administered. Participants in all conditions attention has been directed toward the influ- were asked to complete the Implicit Association ence of social desirability on implicit measures. Test (IAT), an implicit measure (Greenwald et This is not entirely surprising, since implicit at., 1998), and the Modern Racism Scale measures are perhaps less likely to be influ- (MRS), an explicit measure of racial attitudes enced by social desirability forces than explicit (McConahay, 1986). The order in which measures. Nevertheless, the strength of the participants completed the IAT and MRS was relationship between implicit and explicit measures counterbalanced so that the presence of order may be moderated by the influence of social effects could be determined. desirability on explicit measures (Dunton & In order to manipulate participants’ moti- Fazio, 1997), since explicit attitudes may shift vation to provide accurate self-reports, the under conditions that minimize the impact of information that participants received about social desirability motivations. Specifically, if the nature of the IAT was systematically varied. Whites report more negative explicit attitudes In one condition (the Accurate condition) toward African-Americans due to a bogus participants were told that the IAT was an pipeline procedure and these explicit responses accurate measure of racial attitudes, not unlike are less contaminated by social desirability a lie detector. Thus the motivation to accurately concerns, it is possible these attitudes may be report one’s explicit attitude was strong since more highly correlated with implicit measures, participants believed that the ostensibly since implicit measures also tend to yield fairly accurate implicit measure would be used to cor- negative attitudes assessments and are also less roborate their self-reported attitude. In a likely to be influenced by social desirability second condition, participants received infor- forces. To date, there is no published research mation that stressed that the implicit measure that has systematically examined the degree to was not an accurate measure of their racial atti- which the implicit–explicit relationship may tudes. Thus in this condition (the Inaccurate

41 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 42

Group Processes & Intergroup Relations 8(1)

condition), participants were aware that their counterbalancing of the order in which the responses on the explicit measure would be MRS and IAT were administered allowed this compared to their performance on an implicit alternative explanation to be empirically tested. measure, but they did not have to be concerned Specifically, if a stronger implicit–explicit with the ability of the implicit measure to relationship is observed only when the IAT is accurately assess their true attitude. In the final completed first and the MRS second, thereby condition (the No Information condition) allowing participants to adjust their responses participants did not receive any information on the MRS to fit with their perceived perform- about the nature of the IAT. Thus this con- ance on the IAT, then a strengthened dition reflects the context in which the implicit–explicit relationship could indeed be implicit–explicit relationship is generally due to the fact that participants adjusted their assessed; the motivation to accurately report an MRS responses to fit with their IAT perform- explicit attitude was weak and social desirability ance. However, if the magnitude of the motivations were presumably high. implicit–explicit correlation is similar regard- It was predicted that participants in the less of the order in which the measures are Accurate condition would express more completed, then the results cannot be due to negative explicit racial attitudes compared to the adjustment of MRS responses to fit with per- participants in the No Information and Inaccu- ceived IAT performance, since participants rate conditions. Consequently, the relationship who completed the MRS before the IAT could between implicit and explicit racial attitudes not be aware of their IAT performance at the was predicted to be stronger in the Accurate time they completed the MRS. Such a pattern condition, relative to the Inaccurate condition of results would strongly suggest that the mini- and the No Information condition. Such a mization of social desirability forces due to the pattern of results would indicate that the bogus pipeline manipulation would be the increased motivation to accurately report more likely explanation for a strong explicit racial attitudes, which is strong in the implicit–explicit relationship. Accurate condition and weaker in the Inaccu- rate and No Information conditions, reduces Method the dissociation between implicit and explicit attitudes, indicating that the true degree of dis- Participants sociation between implicit and explicit racial Participants were 112 White undergraduates attitude may be smaller than some theories of enrolled at a small college in New England. racial attitudes have presumed. Participants received course credit for their Additionally, the order in which the MRS and participation. the IAT were administered was counterbalanced in order to examine the plausibility of Design a potential alternative interpretation of the A 3 (Information about IAT: Accurate, Inaccu- results. As Whites complete the IAT they often rate, No Information) 2 (Order of Measures: realize that the results of the test will indicate IAT First, IAT Last) between-subjects factorial that they are biased against African-Americans design. (Monteith, Voils, & Ashburn-Nardo, 2001). Consequently, when participants then com- Measures plete an explicit measure they may adjust their The IAT was used to assess implicit racial atti- explicit responses to ‘fit’ with their IAT tudes toward Blacks and Whites (Greenwald et performance. Thus a participant’s perception al., 1998). The IAT requires individuals to of their performance on the IAT, rather than categorize target concepts (in this case, Black changes in social desirability forces, might and White faces) and attributes (good and bad cause a shift in explicit attitudes and a strength- words). In the version of the IAT used in the ening of the implicit–explicit relationship. The current study, 10 faces (5 male and 5 female; 5

42 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 43

Nier implicit and explicit racial attitudes

Black and 5 White) were employed as stimuli to 1 to 7 scale, where 1 indicated strong disagree- represent Black and White social groups; these ment and 7 indicated strong agreement. The faces were taken from the group of 12 faces reliability of the scale was acceptable at pre-test used by Cunningham et al. (2001). The follow- (alpha = .74) and post-test (alpha = .86). The ing evaluative words (drawn primarily from average of the six MRS items constituted the Greenwald et al., 1998) were utilized as ‘good’ explicit measure of participants’ racial attitudes words: lucky, rainbow, love, peace, heaven, at each time period. pleasure, cheer, and happy. The following words were used as ‘bad’ words: evil, death, Procedure pain, disaster, ugly, vomit, stink, and rotten. The study consisted of two separate sessions; These stimuli were employed in a procedure one session was a pre-test session in which the very similar to previous IAT research where the MRS was administered and the second session IAT task was composed of five blocks; blocks 1, was the laboratory portion of the study.1 Upon 2, and 4 were practice blocks and the data were arrival at the laboratory, participants were collected during blocks 3 and 5, with data being informed that they would be participating in a collected on 40 trials from each block. For one study of word associations and attitudes. Partici- of these blocks, participants classified White pants were then randomly assigned to one of faces and good words on one key and Black three conditions. faces and bad words on another key In the Accurate condition, participants com- (white+good/black+bad block). For the other pleted the IAT and the MRS. Also, the experi- block, participants classified White faces and menter gave the participants information that bad words on one key and Black faces and good ostensibly established the accuracy of the IAT as words on another key (white+bad/black+good a measure of racial attitudes. Specifically, block). The order of these two data collection immediately prior to completing the MRS blocks was determined randomly for each participants were told that the IAT was participant. The accuracy and latency of partici- designed to assess White attitudes toward Black pants’ responses were recorded on each trial. Americans. Participants then watched a five- As suggested by Greenwald et al. (2003) if a minute videotape of a segment from Dateline, participant gave an incorrect response on a a television news program, which attested to the trial, the trial continued until the correct accuracy of the IAT (Banaji & Greenwald, response was given and these latencies were 2000). After the clip was over, the experimenter included in subsequent IAT analyses. read aloud the following information about the A six-item version of the MRS was admin- accuracy of the IAT: istered to assess explicit attitudes toward African-Americans (McConahay, 1986). This Previous research by social psychologists has version of the MRS contained the following demonstrated that these computer tasks are very accurate measures of people’s true racial attitudes. items: ‘Blacks are getting too demanding in For example, social psychologists at Yale Uni- their push for equal rights’, ‘Discrimination versity, University of Washington, University of against Blacks is no longer a significant Colorado, University of Delaware, and Colgate problem in the United States’, ‘It is easy to University have all successfully used these understand the anger of Black people in computer measures of racial attitudes. Although America’ (reverse scored), ‘Over the past few these computer tasks may not be a perfect measure of your racial attitudes, it’s the closest thing to a lie years the government and news media have detector that social psychologists can use to determine your shown more respect for Blacks than they true beliefs about race. deserve’, ‘Blacks should not push themselves where they are not wanted’, ‘Over the past few Similar to most research employing the Bogus years Blacks have gotten more economically Pipeline procedure (Roese & Jamieson, 1993), than they deserve’. Participants expressed their where participants are asked to predict the degree of agreement with each statement on a output of the pipeline device, when completing

43 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 44

Group Processes & Intergroup Relations 8(1)

the MRS participants were instructed that their (rather than predicting the results of the response should reflect ‘what the computer computer’s analysis) when completing the would predict’ their response to be for each of MRS. Thus the No Information condition the six statements, rather than indicating their reflected the procedure through which implicit personal attitude toward African-Americans.2 In and explicit measures of racial attitudes are order to test for the presence of order effects, often administered (e.g. Dasgupta et al., 2000; the order of completion of the IAT and the Kawakami & Dovidio, 2001). MRS was counterbalanced so that half the participants first completed the IAT, were then Results given the information about the IAT, and then completed the MRS. The remaining partici- Implicit attitudes pants first were given information about the Extreme IAT reaction time outliers were IAT, completed the MRS, and then completed recoded, such that responses quicker than the IAT. 300 ms were recoded as 300 ms and responses The Inaccurate condition was identical to the greater than 1200 ms were recoded as 1200 ms Accurate condition, except that the experi- (similar to Dasgupta et al., 2000). Mean menter stated that the IAT was a poor measure response latencies and error rates were calcu- of racial attitudes. Specifically, the experi- lated for each data collection block of trials menter read aloud the following information (the white+good/black+bad block and the just before participants completed the MRS. white+bad/black+good block).3 Replicating [The IAT was] originally proposed to assess indi- previous IAT research, response latencies were vidual’s attitudes toward Black Americans. Psychol- significantly longer for the white+bad/black+good ogists hoped that an individual’s performance on trials (M = 708.0 ms), relative to the these computer tasks would predict whether white+good/black+bad trials (M = 586.1 ms), people hold negative attitudes about members of t(111) = 11.3, p < .001, indicating that partici- different racial groups. Unfortunately, it was dis- pants displayed an implicit evaluative prefer- covered that people’s performance on these ence for White over Black. computer tasks is virtually unrelated to their racial attitudes. For example, Dr. Russell Fazio, a social A similar analysis was also conducted on psychologist at Indiana University, has found that error rates. The proportion of trials in which a people’s performance on these computer tasks is participant gave an incorrect response was unrelated to their actual racial attitudes and other calculated for both of the data collection beliefs about race. In other words, he found that blocks. A t test revealed that the proportion of the computer tasks don’t tell us anything about your incorrect trials was significantly higher for racial attitude. white+bad/black+good trials (M = 6.88 %), relative Participants then completed the MRS to white+good/black+bad trials (M = 5.20 %), immediately afterward. Similar to the Accurate t(111) = 3.35, p < .001, which again indicated condition, as participants completed the MRS that participants displayed an implicit prefer- they were instructed to indicate ‘what the ence for White over Black. computer would predict’ their response to be Next, a 3 (Information: Accurate, Inaccurate, for each of the six statements and the order of No Information) 2 (Order: IAT First, IAT the completion of the MRS and IAT was Last) 2 (Block: white+good/black+bad, counterbalanced. white+bad/black+good) mixed multivariate In the No Information condition participants analysis of variance (MANOVA) was conducted completed the MRS and IAT without receiving to examine whether implicit attitudes varied any information about the accuracy of the IAT across experimental condition, where the IAT as a measure of racial attitudes, and the order effects for reaction time and error rates were of the completion of the IAT and MRS was treated as the dependent variables, Block was again counterbalanced. Also, participants were treated as the within-subjects factor, while asked to indicate their true beliefs about race Information and Order were treated as

44 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 45

Nier implicit and explicit racial attitudes

Table 1. Changes in Modern Racism Scores as a function of Information condition

Pre-test Post-test ————————————— ——————————————— Experimental condition MSDMSD

Accurate 2.30a 0.92 3.55b 1.17 Inaccurate 2.13a 0.77 2.96c 1.08 No Information 2.02a 0.63 1.82a 0.62

Note: The means that do not share the same superscripts are signiﬁcantly different, Tukey, p < .05.

between-subjects factors. As expected, the inter- (t (33) = 1.5, p = ns). Participants in the Inac- action between Information condition and curate condition displayed a significant Block was not significant, Wilks’ Lambda = .99, increase in their MRS scores from pre-test (M = p = .66, indicating that the implicit attitudes 2.13) to post-test (M = 2.96) (t (36) = 5.2, p < toward African-Americans were similar across .05).5 Additionally, the MRS scores for partici- experimental condition.4 In sum, the IAT pants in the Accurate condition were signifi- results were consistent with previous research. cantly higher at post-test (M = 3.55), relative to The IAT data indicated that participants, on their pre-test scores (M = 2.30) (t (40) = 7.3, p average, held an implicit preference for White < .001). Planned comparisons also revealed that over Black. Furthermore these implicit atti- participants in the Accurate condition dis- tudes did not vary as a function of experimental played significantly higher post-test MRS scores, condition. relative to participants in the Inaccurate condition (t (76) = 2.4, p < .05) and the No Infor- Explicit attitudes mation condition (t (73) = 8.0, p < .001).6 Thus A 3 (Information: Accurate, Inaccurate, No the hypothesis, that the degree of change in Information) 2 (Order: IAT First, IAT Last) MRS scores would vary according to experi- 2 (Time: Pre-test, Post-test) mixed ANOVA mental condition, was supported; participants was conducted to examine whether participants in the Accurate condition displayed the greatest in the Accurate condition would display greater increase in their MRS scores, participants in the changes in MRS scores, relative to participants Inaccurate condition displayed a significant but in the Inaccurate and No Information con- smaller change in their MRS scores, and partici- ditions. There was a main effect for order (F(1, pants in the No Information condition did not 106) = 4.8, p < .05), with participants who com- display any change in their MRS scores. Also, pleted the IAT before the MRS reporting this crucial interaction effect was not qualified higher MRS scores (M = 2.62 ) than those who by the Order variable. completed the MRS before the IAT (M = 2.31). However, there were no other significant Relationship between implicit and explicit effects involving order. measures More importantly, the expected Time In order to examine the relationship between Information interaction was significant (F(2, implicit and explicit attitudes, the IAT effect was 106) = 21.6, p < .001), indicating that the collapsed into a single composite score. To degree of change in MRS scores from pre-test accomplish this goal for response latencies, the to post-test varied as a function of the Infor- mean white+good/black+bad latency was sub- mation condition (see Table 1). Planned tracted from the mean white+bad/black+good follow-up tests revealed that the pre-test MRS latency. Similarly, the mean white+good/black+bad scores for participants in the No Information error rate was subtracted from the white+bad/ condition (M = 2.02) were not significantly black+good error rate. Next, the IAT effects for different than their post-test scores (M = 1.82 ) both latency and error rates were standardized

45 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 46

Group Processes & Intergroup Relations 8(1)

and these standardized scores were then and IAT. The three contrast coded variables, averaged together. This average captured the the composite IAT effect, and their corre- IAT effect into a single composite score, where sponding interaction terms were then used to higher scores indicated more negative implicit predict post-test MRS scores. It was predicted attitudes toward Blacks.7 that the interaction between the IAT effect and This IAT effect was then used as the depen- the first contrast coded variable (that con- dent variable in a multiple regression analysis trasted the Accurate versus the Inaccurate and that was designed to test the hypothesis that the No Information conditions) would be a signifi- relationship between implicit and explicit cant predictor of MRS scores. Such a pattern of measures would be stronger in the Accurate results would signify that the relationship condition than in the Inaccurate and No Infor- between the IAT effect and MRS scores was mation conditions. To test this hypothesis, two significantly stronger in the Accurate condition contrast codes representing the Information than in the Inaccurate and No Information condition were employed (see Aiken & West, conditions. 1991; Cohen & Cohen, 1983). The first contrast Overall the regression model accounted for code compared the Accurate condition versus 49% of the variance in post-test MRS scores (see the Inaccurate and No Information conditions Table 2). In terms of individual variables, the and the second contrast code compared the composite IAT effect (beta = –.21, p < .01), the Accurate and Inaccurate conditions to the No contrast comparing the Accurate condition to Information condition. We also included a final the Inaccurate and No Information conditions contrast coded variable that represented the (beta = –.47, p < .01), and the contrast com- order in which participants completed the MRS paring the No Information condition to the

Table 2. Multiple regression analysis of the implicit–explicit relationship

Predictor variable Beta tpvalue

Main effects IAT effect –.21 2.91 <.01* Order condition .13 1.80 .08 Information condition Contrast One (Accurate vs. Inaccurate/No Info) –.47 6.48 <.01* Contrast Two (No Info vs. Pipeline/Inaccurate) .37 5.17 <.01*

Two-Way Interactions IAT effect Order condition –.03 0.36 .72 IAT effect Information condition IAT effect Contrast One (Accurate vs. Inaccurate/No Info) –.16 2.21 .03* IAT effect Contrast Two (No Info vs. Pipeline/Inaccurate) .04 0.51 .62 Order condition Information condition Order condition Contrast One (Accurate vs. Inaccurate/No Info) –.08 1.04 .30 Order condition Contrast Two (No Info vs. Pipeline/Inaccurate) .14 1.91 .06

Three-Way Interaction IAT effect Order condition Information condition IAT effect Order condition Contrast One .05 0.63 .53 IAT effect Order condition Contrast Two –.02 0.23 .82

R2 = .49 Notes: Regression model predicting post-test MRS scores from the composite IAT effect, Order condition and Information condition. Asterisks indicate statistically signiﬁcant predictors, p < .05.

46 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 47

Nier implicit and explicit racial attitudes

6 Information Condition 5 Pipeline Inaccurate Control 4

MRS Score (1-7) 2

Ð2 Ð1 0 +1 +2 IAT Effect (Z-Scores)

Figure 1. IAT effect Information condition interaction. Lines represent simple regression lines predicting MRS scores from the composite IAT effect in each condition.

Accurate and Inaccurate conditions (beta = .37, scores and IAT scores. However, for partici- p < .01), were all significant predictors of post- pants in the Accurate condition there is a fairly test MRS scores. The contrast that corre- strong positive relationship between MRS sponded to the order in which participants scores and IAT scores. This pattern of results completed the MRS and IAT was not signifi- suggests that as the motivation to accurately cant, nor were any of the interactions involving report one’s racial attitudes increases, the this variable.8 relationship between MRS scores and IAT Most importantly, the only significant inter- scores strengthens.9 action term was the interaction between the A supplementary analysis was also performed IAT effect and the contrast comparing the in which the implicit–explicit relationship was Accurate condition to the Inaccurate and No examined using a correlational approach (see Information conditions (beta = –.16, p < .05), Table 3). Supportive of the regression analysis, indicating that the relationship between the the pattern of implicit–explicit correlations in IAT effect and MRS scores was moderated by each condition yielded a similar interpretation. the information that participants received The correlation between the composite IAT about the IAT. The exact nature of this inter- effect and post-test MRS scores was significant action is depicted in Figure 1, which presents in the Accurate condition (r = .51, p < .001) and the simple regression lines predicting post-test was not significant in either the Inaccurate con- MRS scores from the composite IAT effect in dition (r = .18, p = ns) or the No Information each condition. As the figure indicates, among condition (r = .14, p = ns). Similar to the results participants in the Inaccurate and No Infor- of the regression analysis, the implicit–explicit mation conditions there is a very weak (and correlation in the Accurate condition was nonsignificant) relationship between MRS significantly stronger than the correlations

47 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 48

Group Processes & Intergroup Relations 8(1)

Table 3. Strength of implicit–explicit relationship as a function of Information condition

Correlation between MRS and IAT effect ———————————————————————————— Composite IAT effect for IAT effect for Experimental condition IAT effect latency errors

Accurate .51* .39* .46* Inaccurate .18* .25* .04* No Information .14* .12* .10*

Note: Asterisks denote statistically signiﬁcant correlations, p < .05.

observed in the Inaccurate (z = 1.65, p < .05; potentially explain the increased implicit– one-tailed) and No Information (z = 1.75, p < explicit relationship is that people simply .05; one-tailed). Thus the pattern of correla- reported more negative attitudes in the tions mirrors the relationships observed in the Accurate condition, regardless of their level of multiple regression analysis. implicit prejudice. However, the data are not consistent with this interpretation. If all partici- Were there any other plausible interpretations pants in the Accurate condition simply elevated of the results? their MRS scores to a similar degree, the Although the predicted pattern of results was relationship between post-test MRS scores and observed, there remained two possible alterna- IAT scores would not have been any stronger in tive interpretations of the results. One possible the Accurate condition. An increase in the alternative interpretation of the results revolves implicit–explicit relationship requires, by defi- around the incidental feedback that partici- nition, that those who had more negative IAT pants receive when completing the IAT. When scores reported negative explicit attitudes to a completing the IAT, participants are often greater degree than those who had more aware that they fare poorly on the test. As a positive IAT scores. Therefore the stronger result, one possible interpretation of the implicit–explicit relationship observed in the strengthened implicit–explicit relationship is Accurate condition is not consistent with the that participants adjust their explicit responses notion that everyone simply reported more to ‘fit’ with their performance on the IAT. negative MRS scores in the Accurate condition. According to this interpretation, the explicit attitudes expressed in the Accurate condition Discussion would reflect their perceived performance on the IAT, rather than their actual racial atti- The current study was designed to demonstrate tudes. If this alternative interpretation was that the apparent dissociation between implicit correct, then the crucial IAT effect Infor- and explicit measures of racial attitudes is exag- mation condition interaction should have been gerated in most research, due to the conditions qualified by the three-way interaction effect under which explicit attitudes are assessed. The involving the Order variable. However, the results indicated that when participants three-way interaction term that tested this believed their ‘true attitudes’ were being accu- effect did not approach significance (see Table rately assessed, there was a significant relation- 2). Also, the correlations between MRS scores ship between an implicit measure of racial and the IAT effect among participants in the attitudes (the IAT) and an explicit measure of Accurate condition were very similar in magni- racial attitudes (the MRS). When participants tude regardless of the order of completion did not believe that their self-reported explicit (IAT first r = .46, IAT last r = .53). attitudes could be accurately corroborated with Another alternative interpretation that could an implicit measure, there was no association

48 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 49

Nier implicit and explicit racial attitudes

between implicit and explicit attitudes. Thus measure. In the current analysis, the original the results suggest that as the motivation to value of the implicit–explicit correlation in the report explicit attitudes that are consistent with Accurate condition (r = .51) was divided by the implicit attitudes increases, the implicit– square root of the product of the reliabilities of explicit relationship strengthens due to each measure (MRS alpha = .86; IAT alpha = changes in self-reported explicit attitudes .50). This calculation yielded a strong (implicit attitudes were not inﬂuenced by the implicit–explicit correlation (r = .79). While the information that participants received about results of this analysis must certainly be inter- the IAT). Furthermore, the results of the preted with caution, it nonetheless suggests that current study cannot be explained as an order employing a measurement procedure that both effect, since the magnitude of implicit–explicit minimizes the impact of error variance (such as link did not vary as a function of the order in structural equation modeling) and yields more which the IAT and MRS were completed. forthright explicit responses (such as the bogus These results have implications for the repre- pipeline) may result in a strong implicit– sentational explanation of the poor correspon- explicit relationship that shows much less dence between implicit and explicit racial evidence of dissociation than has been attitudes. Some theorists have argued that the commonly observed. dissociation between implicit and explicit atti- Furthermore, the results of the current study tudes is evidence that implicit and explicit provide additional evidence for the utility of attitudes are independent cognitive represen- implicit measures. Despite the obvious value tations that are the result of different psycho- of implicit measures as being more resistant to logical processes (e.g. Devine, 1989; Wilson et social desirability forces (e.g. Banse, Seise, & al., 2000). However, the results of the current Zerbes, 2001), some have questioned the value study indicate that this dissociation, evident in of measures like the IAT (Cameron et al., so many previous studies, is at least partially due 2000). One criticism of implicit measures has to social desirability concerns. To conclusively been that they are, at best, only weakly related demonstrate the true extent to which implicit to explicit measures. Ironically, the current and explicit racial attitudes are dissociated, results suggest that the apparent poor conver- social desirability motivations and other factors gent validity of implicit measures is partially that are known to underestimate the magni- due to a shortcoming of explicit measures. tude of the implicit–explicit link, such as When the impact of social desirability concerns measurement error, must be minimized. on explicit attitudes is minimized, the conver- How strong might the implicit–explicit link gent validity of implicit measures markedly be under ideal conditions that minimize improves. the impact of factors that cause the Future research could examine the behav- implicit–explicit link to be underestimated? ioral correlates of the implicit–explicit relation- While the current study cannot definitively ship under varying conditions. Research thus address this question, a supplementary analysis far indicates that explicit attitudes are generally was performed in which the magnitude of the correlated with deliberative responses and that implicit–explicit relationship was corrected for explicit measures are typically not correlated the effect of measurement error, which is some- with subtle, spontaneous behaviors (Dovidio et times referred to as correction for attenuation al., 1997). Research could examine whether (Cohen & Cohen, 1983). This analysis allows explicit attitudes that are assessed while partici- one to estimate the magnitude of a correlation pants are ‘on the pipeline’ are more predictive between any two variables if the measurement of subtle interracial behaviors. Furthermore, error due to unreliability could be removed. Plant, Devine, and Brazy (2003) demonstrated This analysis is performed by dividing the that there are important individual differences original correlation coefﬁcient by the square in the nature of the attitudes that Whites’ root of the product of the reliabilities of each report while on the bogus pipeline. Additional

49 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 50

Group Processes & Intergroup Relations 8(1)

research could examine whether these individ- asked ‘what the computer would predict’ their ual differences may moderate the magnitude of responses to be on the MRS, they may have the implicit–explicit relationship under bogus guessed the computer would inaccurately label pipeline conditions. them as having relatively negative attitudes toward African-Americans. 6. These significant differences among post-test Notes MRS scores remained significant even after statistically controlling for pre-test MRS scores 1. The time between completion of the pre-test and using a covariance analysis, ps < .05. participation in the laboratory portion of the study 7. The reliability of this composite measure (alpha = ranged from 3 to 11 weeks. The time elapsed .50) was similar to the reliability of other implicit between the pretest and the laboratory session did measures (Dovidio et al., 2001), which generally not appear to influence participants’ responses on have lower reliabilities than traditional self-report the MRS; the time elapsed (in weeks) was not measures of attitudes. correlated with MRS scores reported in the 8. While no effects involving the Order variable laboratory session, or the change in MRS scores were significant at the .05 level, the main effect from pre-test to the laboratory session. for the order condition was marginally significant, 2. In their meta-analysis of the validity of the bogus indicating the MRS scores were slightly higher pipeline procedure, Roese and Jamieson (1993) when participants completed the IAT first. Also explicitly recommended that participants should the interaction between Order and the second be asked to predict the output of the pipeline Information condition contrast was marginally device, rather than just being instructed to significant. This effect indicated that the answer accurately. This recommendation was for elevation of MRS scores in the Inaccurate and several reasons. One, this response format yields Accurate condition was greater when the IAT was larger effects than does asking participants to completed first. While these effects suggest that only respond accurately. Two, a substantial Order may have had some impact on MRS scores, majority of past research with the bogus pipeline neither of these two marginal effects qualified the has employed this type of response format. major hypothesized effects. Three, this response format does not appear to 9. Two separate regression analyses were performed simply induce a demand to respond more to examine whether the latency effect and the extremely, but rather it yields responses that are error rate effect could predict MRS scores when more accurate (Arkin & Lake, 1983). For each IAT effect was treated independently, example, Quigly-Fernandez and Tedeschi (1978) instead of being combined into a composite demonstrated that participants who were asked to measure. In each of these analyses the critical guess the output of a bogus EMG apparatus gave interaction effect was marginally significant ps < responses that more accurately reflected their .10, one-tailed). actual experiences. 3. IAT scores were also calculated using alternative analytic procedures including Greenwald et al.’s (2003) revised scoring procedure as well as a Acknowledgements natural logarithmic transformation. Our results I would like to thank Beth Chase, Kim Chula, Raya were somewhat weaker, although still very Gorcheva, Christina Johnson, Amy Kasiski, Allison similar, using these alternative scoring Kuhbach, Lee Rowsey, and Mike Vergnani for procedures. serving as experimenters. 4. We also performed two ANOVAs in which each IAT effect was analyzed separately. Nearly identical results were obtained using this References approach. 5. This unexpected increase in MRS scores in the Aiken, L. S., & West, S. G. (1991). Multiple regression: Inaccurate condition may have been due to the Testing and interpreting interaction. Newbury Park, fact that participants were aware that their CA: Sage. responses on the MRS would be compared to Arkin, R. M., & Lake, E. A. (1983). Plumbing the their IAT scores. It is also possible that when depths of the bogus pipeline: A reprise. Journal of participants in the Inaccurate condition were Research in Personality, 17, 81–88.

50 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 51

Nier implicit and explicit racial attitudes

Banaji, M. R., & Greenwald, A. G. (2000, March 19). Dovidio, J. F., Kawakami, K., Johnson, C., Johnson, Pride and prejudice. In G. Moriba-Meadows B., & Howard, A. (1997). The nature of prejudice: (Producer), Dateline NBC. New York: National Automatic and controlled processes. Journal of Broadcasting Corporation. Experimental Social Psychology, 33, 510–540. Banse, R., Seise, J., & Zerbes, N. (2001). Implicit Dunton, B. C., & Fazio, R. H. (1997). An individual attitudes towards homosexuality: Reliability, difference measure of motivation to control validity, and controllability of the IAT. Zeitschrift prejudiced reactions. Personality and Social für Experimentelle Psychologie, 48, 145–160. Psychology Bulletin, 23, 316–326. Blair, I. V. (2001). Implicit stereotypes and Fazio, R. H., Jackson, J. R., Dunton, B. C., & prejudice. In G. B. Moskowitz (Ed.), Cognitive Williams, C. J. (1995). Variability in automatic social psychology: The Princeton symposium on the legacy activation as an unobstrusive measure of racial and future of social cognition (pp. 359–374). attitudes: A bona fide pipeline? Journal of Mahwah, NJ: Erlbaum. Personality and Social Psychology, 69, 1013–1027. Boniecki, K. A., & Jacks, J. Z. (2002). The elusive Fazio, R. H., & Olson, M. A. (2003). Implicit relationship between measures of implicit and measures in social cognition research: Their explicit prejudice, Representative Research in Social meaning and use. Annual Review of Psychology, 54, Psychology, 26, 1–14. 297–327. Cameron, J. A., Alvarez, J. M., & Bargh, J. A. (2000). Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. Examining the validity of implicit and explicit measures K. (1998). Measuring individual differences in of prejudice: Is there really a bonafide pipeline? Paper implicit cognition: The implicit association test. presented at the first annual meeting of the Journal of Personality and Social Psychology. 74, Society for Personality and Social Psychology, 1464–1480. Nashville, TN. Greenwald, A. G., Nosek, B. A., & Banaji, M. R. Cohen, J., & Cohen, P. (1983). Applied multiple (2003). Understanding and using the implicit regression/correlation analysis for the behavioral association test: An improved scoring system. sciences. New York: Erlbaum. Journal of Personality and Social Psychology, 85, Cunningham, W. A., Preacher, K. J., & Banaji, M. R. 197–216. (2001). Implicit attitude measures: Consistency, Jones, J. M. (1997). Prejudice and racism. New York: stability, and change. Psychological Science, 12, McGraw-Hill. 163–170. Kawakami, K., & Dovidio, J. F. (2001). The reliability Dasgupta, N., McGhee, D. E., Greenwald, A. G., & of implicit stereotyping. Personality and Social Banaji, M. R. (2000). Automatic preference for Psychology Bulletin, 27, 212–225. White Americans: Eliminating the familiarity McConahay, J. B. (1986). Modern racism, explanation. Journal of Experimental Social ambivalence, and the modern racism scale. In J. F. Psychology, 36, 316–328. Dovidio & S. L. Gaertner (Eds.), Prejudice, Devine, P. G. (1989). Stereotypes and prejudice: discrimination, and racism (pp. 91–126). Orlando, Their automatic and controlled components. FL: Academic Press. Journal of Personality and Social Psychology, 56, 5–18. Monteith, M. J., Voils, C. I., & Ashburn-Nardo, L. Dovidio, J. F., & Gaertner, S. L. (1998). On the (2001). Taking a look underground: Detecting, nature of contemporary prejudice: The causes, interpreting, and reacting to implicit racial biases. consequences, and challenges of aversive racism. Social Cognition, 19, 395–417. In J. Eberhardt & S. T. Fiske (Eds.), Confronting Plant, E. A., Devine, P. G., & Brazy, P. C. (2003). The racism: The problem and the response (pp. 3–32). bogus pipeline and motivations to respond Newbury Park, CA: Sage. without prejudice: Revisiting the fading and faking Dovidio, J. F., Kawakami, K., & Beach, K. R. (2001). of racial prejudice. Group Processes and Intergroup Implicit and explicit attitudes: Examination of the Relations, 6, 187–200. relationship between measures of intergroup bias. Quigley-Fernandez, B., & Tedeschi, J. T. (1978). The In R. Brown & S. L. Gaertner (Eds.), Blackwell bogus pipeline as a lie detector: Two validity handbook of social psychology, Volume 4: Intergroup studies. Journal of Personality and Social Psychology, processes (pp.175–197). Oxford, UK: Blackwell. 36, 247–256. Dovidio, J. F., Kawakami, K., & Gaertner S. L. Roese N. J., & Jamieson, D. W. (1993). Twenty (2002). Implicit and explicit racial attitudes and years of bogus pipeline research: A critical review interracial interaction. Journal of Personality and and meta-analysis. Psychological Bulletin, 114, Social Psychology, 82, 62–68. 363–375.

51 03 GPI 048615 (to/d) 8/12/04 12:19 pm Page 52

Group Processes & Intergroup Relations 8(1)

Schlenker, B. R., & Weingold, M. F. (1992). Wittenbrink, B., Judd, C. M., & Park, B. (1997). Interpersonal processes involving impression Evidence for racial prejudice at the implicit level regulation and management. Annual Review of and its relationship with questionnaire measures. Psychology, 43, 133–168. Journal of Personality and Social Psychology, 72, Sigall, H., & Page, R. (1971). Current stereotypes: A 262–274. little fading, a little faking. Journal of Personality and Social Psychology, 18, 247–255. Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). Paper received 12 March 2004; revised version accepted A model of dual attitudes. Psychological Review, 107, 26 June 2004. 101–126.