Explicit and implicit racial attitudes a test of their convergent and predictive validity

The number of respondents classified as prejudiced on the implicit measure substantially exceeds the corresponding number based on explicit indicators, suggesting that survey respondents

Explicit and Implicit Racial Attitudes:

A Test of their Convergent and Predictive Validity

Shanto Iyengar and Solomon Messing, Stanford University

Kyu Hahn, Seoul National University Mahzarin Banaji and Christopher Dial, Harvard University

Abstract

Using data from national samples, we examine the convergent and predictive validity of explicit and implicit measures of racial prejudice First, we show that explicit measures diverge from a measure of implicit racial bias The number of respondents classified as prejudiced on the implicit measure substantially exceeds the corresponding number based on explicit indicators, suggesting that survey respondents may be masking their racial attitudes Second, in three

different experimental contexts, we demonstrate that implicit racial bias predicts a preference for individuals with lighter complexions People classified as prejudiced on the basis of explicit measures, however, do not discriminate on the basis of complexion Our findings suggest that future efforts to assess prejudice should incorporate both implicit and explicit racial attitudes

The measurement of Americans‟ racial attitudes has become especially challenging in the post-civil rights era because survey respondents are motivated to answer questions in a manner that suggests the absence of racial bias (see McConahay, Hardee, and Batts 1981) For instance, the percentage of white Americans who use stereotypic and derogatory terms (e.g “lazy”) to describe African-Americans has declined sharply since the 1960s (Gaertner and Dovidio 2005; Virtanen and Huddy 1998; Taylor, Sheatsley, and Greeley 1978), and by the end of the twentieth century, whites evaluated blacks just as favorably as their own group But survey questions that disguise the racial cue elicit higher levels of prejudice (Kuklinski and Cobb, 1998; Crosby et al., 1980) In one study with unobtrusive measures, the authors concluded that “blatantly prejudiced attitudes still pervade the white population” (Kuklinski et al 1997, p 403) Thus, when people are unaware that they are violating egalitarian norms, they feel free to express preferences and stereotyped judgments that are hostile to minorities

Evidence of lingering racial bias in Americans‟ policy preferences raises further doubts about the decline of prejudice (see Fording 2003; Quillian 2006) In the case of crime, support for punitive policies increases significantly when whites learn that the criminal perpetrator is non-white (Gilliam and Iyengar 2000; Peffley and Hurwitz 2007; Eberhardt et al 2004) Race bias also characterizes a variety of economic markets (Ayres, 2001); job applicants with

European-sounding first names are preferred (by 50 percent) over applicants with identical resumes but African American-sounding names (Bertrand and Mullainathan 2004) In short, Americans say they are free of racial bias, but their attitudes and behaviors frequently indicate otherwise

Partly in response to normative and impression management pressures, survey

researchers have shifted the definition of prejudice away from explicit racial animus in the

direction of more indirect and diffuse indicators of “symbolic racism” or “racial resentment.” In this revisionist view, prejudice in the contemporary era is more appropriately defined as a blend

of racial animosity and mainstream cultural values (Kinder and Sears, 1981) Operationally, the blend is captured by survey questions that focus on beliefs about minorities‟ adherence to the American way (Kinder and Sanders 1996; Feldman and Huddy 2005)

Because the new measures of racism include questions tapping both racial sentiment and support for mainstream cultural values, there is understandable concern over their measurement validity Do they, in fact, measure racial prejudice or some other construct related to prejudice? Proponents cite the close relationship between these measures and a variety of race-related attitudes and policy preferences, e.g racial stereotypes, support for affirmative action, etc (see Kinder and Sanders 1996; Sears and Henry 2005; Tesler and Sears 2010) as evidence of their predictive validity Critics challenge this interpretation on the ground that the manifest content of the racial resentment questions which encompasses both prejudice and ideological

predispositions make them definitionally intertwined with questions of public policy (see Sniderman and Tetlock 1986; Sniderman and Carmines 1997) In the most recent critique along these lines, Carmines et al argue that the criterion measures used to assess the predictive validity

of new racism (e.g support for affirmative action) may in fact be alternative indicators of the same underlying construct Their analysis demonstrates that the survey items making up the racial resentment scale consistently load on the same underlying factor as a variety of racial policy questions Carmines et al therefore conclude that “both the racial resentment and racial

policy measures measure a single underlying dimension, not two different concepts” (Carmines, Sniderman and Easter, 2011, p 106).1

Our goal in this paper is to compare the relative validity of survey and implicit measures

of racial bias We used two validation tests First, we correlated survey measures of prejudice with an implicit measure that is not subject to conscious control We found that the survey measures correlated only weakly with implicit bias Second, we assessed the validity of explicit and implicit measures based on their ability to predict a participant‟s choice between two non-white individuals who were identical in all respects except for their degree of Afrocentric

physical features (primarily skin complexion) Our logic was that white respondents classified as racially prejudiced, should reveal a preference for the less Afrocentric individual We found that the measure of implicit bias predicted preferences based on Afrocentric features, but the survey measures had no bearing on responsiveness to the visual racial cue; indeed, the participants classified as more prejudiced on the survey measures were “color blind” in their preferences

The paper is organized as follows We begin with a brief discussion of the distinction between implicit and explicit racial attitudes We then describe our indicators of racial prejudice, both implicit and explicit Next, we present results of a convergent validity analysis, which compares racial attitudes with candidate evaluations While explicit and implicit measures of candidate preference converge, explicit measures of prejudice were quite distinct from the most commonly used measure of implicit racial bias These findings suggest that the divergence cannot be attributed simply to differences in measurement modality, but must represent an underlying behavioral reality Last, we report the results from three predictive validity tests in

1 Using the American National Election Studies data, Carmines et al replicate their analysis on seven different election years with consistent results.

which respondents indicated their preference as between candidates for elective office or

prospective immigrants In each case, we used morphing software to vary the level of

Afrocentrism in the target individuals‟ faces Our results show that individuals classified as relatively prejudiced on the implicit measure were significantly less likely to prefer the more Afrocentric individual, while those with higher levels of prejudice according to the survey

measures were unresponsive to the racial cue, i.e they treated the more and less Afrocentric targets uniformly In closing, we discuss the implications of our findings for the study of race and politics

Implicit Versus Explicit Racial Attitudes

The measurement of racial attitudes has long interested social psychologists, but not because of concerns that people may deliberately misrepresent their attitudes Instead, it is

largely assumed that conscious aspects of attitudes and beliefs represent but a thin sliver of the mind‟s work Experiments on the most fundamental aspects of the human mind, such as vision and memory, have shown not only that the human brain operates outside conscious awareness, but also that unintentional thought and feeling may be the dominant mode of operation (Bargh 1999) Based on this evidence, psychologists now believe that the mind‟s architecture precludes introspective access for the most part and have therefore sought to develop measures of attitudes and beliefs that exist independent of conscious attitudes (see Banaji and Heiphetz 2010, for a review) Explicit racial attitudes may or may not reflect genuine conscious racial preferences, but

in either case they shed no light on less conscious or implicit preferences

There is a rapidly growing literature on the relationship between implicit and explicit attitudes and the effect of each on behavioral outcomes (Nosek 2005; Nock and Banaji 2007; McConnell and Liebold 2001; Greenwald et al., 2009) In the area of race, the evidence suggests

that implicit attitudes predict race-related behaviors For example, Dovidio et al (2002) found that whites‟ implicit attitudes predicted their non-verbal behavior toward blacks in a classroom task setting, while survey measures only predicted their verbal behavior Towles-Schwen and Fazio (2006) found that anti-black implicit attitudes of white freshmen who had been randomly assigned a black roommate, predicted the stability and duration of the roommate relationships Rooth (2010) found that implicit measures of anti-Muslim stereotypes among Swedish hiring managers predicted the decision to favor Swedes over Arab and Muslim job applicants

A recent meta-analysis of attitude-behavior linkages (Greenwald et al., 2009) found not only that implicit racial attitudes reliably predicted relevant behavioral outcomes, but also that the predictive validity of explicit attitudes was compromised in socially sensitive attitude domains In fact, race was the only domain in which the predictive validity of implicit attitudes surpassed explicit attitudes by a significant margin

The nature of implicit attitudes necessitates measurement approaches that bypass the standard method posing of questions altogether Researchers rely instead on reaction time to concepts (such as “black” and “white”) and attributes (such as “good” and “bad”) Based on the theory that people respond faster to category-attribute pairs for which they have learned

automatic associations, these measures focus on the time taken to respond to pairings of white + good and black + bad and the opposite (black + good and white + bad) to generate an indirect measure of racial preference.2 There are several such latency-based methods, the most common being the Implicit Association Test (IAT; Greenwald, McGhee, and Schwarz, 1998) and

2 Automatic associations should exhibit lower error rates, but error rates have been shown to be a noisier signal than latencies alone (see Greenwald et al., 1998, p 1467)

evaluative priming (see Banaji and Heiphetz 2010) As described below, our analysis relies on the former

Study 1

This study focused on the convergent and predictive validity of implicit and explicit measures of racial attitudes We assess the former by comparing the distributions of several indicators of explicit racial prejudice with the IAT We find that the level of prejudice captured

by the implicit measure is substantially higher than that revealed by the explicit measures Unlike the case of racial attitudes, indicators of explicit and implicit candidate preference have the same distribution and are highly correlated Our criterion of predictive validity is reduced support for Democratic presidential candidate Barack Obama when his complexion is darkened The

expected aversion for the darker-skinned version of Obama occurs among study participants with high levels of anti-black implicit bias High scorers on the measures of explicit bias, although less likely to support Obama, are unaffected by the visual cue

Indicators

The Race Implicit Association Test (IAT)

The IAT (Greenwald, McGhee, and Schwartz 1998) is a computer-based task that

requires participants to rapidly sort items into categories Based on the time it takes to sort these items and the errors made in sorting, the IAT measures the strength of association between any set of categories (say animals vs plants, Hispanics vs Africans) and attributes (good vs bad, strong vs weak) Most IATs contain four distinct categories consisting of a pair of targets (e.g., African American and European American) and a pair of attributes (e.g., good and bad) These category labels are displayed on either the left or right side of the screen while words or pictures representing the categories appear one by one in the center of the screen

Participants sort each item as it appears into its corresponding category using only two computer keys For example, „E‟ might be the key assigned for items representing category A (say African American), which then appears on the left side of the screen, and „I‟ might be assigned to items representing category B, which then appears on the right side of the screen In the case of the race IAT, the target categories African American and European American are represented by images of black and white faces, while the attribute categories good and bad are represented by words conveying positive and negative concepts (e.g., wonderful, joy, laughter and terrible, hurt, failure)

Implicit racial attitudes are assessed by subtracting the response times during blocks with hypothesized compatible pairings (e.g., African American paired with bad and European

American paired with good) from the response times during blocks with hypothesized

incompatible pairings (e.g., African American paired with good and European American paired with bad) Positive values represent faster sorting when African American is paired with bad and European American is paired with good (compared to the opposite); negative values represent faster sorting when African American is paired with good and European American is paired with bad (compared to the opposite).3 An “IAT score” (also referred to as a “D-score”) ranging from -2 to 2 is calculated for each participant based on the difference in response times between the

„white-good, black-bad‟ and „white-bad, black-good‟ pairings (full details on scoring an IAT are

3 We used the full version of the race IAT with five blocks in our first study of presidential vote choice In the second study, featuring a pair of prospective immigrants, and in the third study, featuring an unknown black candidate for an Illinois state office, we substituted the brief version of the IAT, which consisted of four trial blocks The brief version of the IAT has been found to yield results that are consistent with those based on the full IAT (see Sriram et al., 2010) For the two studies based on the brief IAT, we removed respondents who incorrectly classified more than 15 percent of all IAT trials.

presented in Greenwald et al, 2003) Thus, positive IAT scores represent a racial bias in favor of whites over blacks.4

Explicit Racial Attitudes

We relied on two widely utilized survey indices of explicit racial attitudes overt racism and racial resentment The former is based on a set of four trait ratings that respondents apply to African-Americans and whites.5 The latter is based on a set of four agree-disagree items that tap beliefs about minorities, individualist cultural values, and support for racial equality.6 In addition

to the indices of overt racism and racial resentment, we also compare respondents‟ thermometer ratings (on a 1-10 scale) of self-reported warm or cold feelings towards African-Americans and European-Americans

4 Since it was developed in the 1990s, the race IAT has been used in dozens of papers as a measure of implicit race bias and in studies of intergroup variation in race attitudes (for a review see, Nosek et al 2002; for critical commentary on the IAT and responses, see Blanton and Jaccard 2006; Greenwald, Nosek, and Sriram 2006).

5 The first item in the set was worded as follows: “We‟re interested in your opinions about different groups in our society Using the scale shown below, where a score of 1 would mean that you think most

of the people in the group tend to be “hard working,” while a score of 7 would mean that most of the people are “lazy,” where would you place African-Americans.” This was followed by scales with end points of “violent” and “peaceful,” “self-reliant” and “prefer to be on welfare,” and “interact with people

of different backgrounds” and “stick to themselves.” We converted each item to a 0-1 metric, summed the four responses aimed at each group and divided by four The final indicator was the difference between the ratings of whites and blacks The Alpha values for the African-American and White indices were 77 and 67 respectively.

6 The items, taken from Kinder and Sanders (1996) were as follows (1) “Over the past few years, blacks have got less than they deserve.” (2) “The Irish, Italians, Jews, Vietnamese and other minorities

overcame prejudice and worked their way up Blacks should do the same without any special favors.” (3)

“It‟s really a matter of some people not trying hard enough; if blacks would only try harder they could be just as well off as whites.” (4) “Generations of slavery and discrimination have created conditions that make it difficult for blacks to work their way out of the lower class.” Respondents answered each item along a four-point scale that ranged from “strongly agree” to “strongly disagree.” Items 2 and 3 were reflected, the items were converted to a 0-1 metric and an index score was computed as the average of the four items Coefficient Alpha was 89.

Implicit Candidate Preference

Since the development of the race IAT, the methodology has been extended to several other attitude domains including politics The candidate IAT is based on the same procedures and measurement as the race IAT In this case, the categories „European American‟ and „African American‟ are replaced by the categories „John McCain‟ and „Barack Obama.‟ Multiple images

of each candidate constituted the stimuli for the Obama and McCain categories and were

matched along obvious dimensions such as clarity, pose, facial expression and background Positive candidate IAT scores represent faster sorting of Barack Obama paired with bad and John McCain paired with good (compared to the inverse); negative values represent the opposite, i.e.,

a relatively more positive implicit attitude toward Obama over McCain Thus, higher scores indicate a stronger implicit preference for McCain over Obama

Explicit Candidate Preference

Our study, conducted just before the 2008 election, included an extensive set of questions measuring respondents‟ evaluations of John McCain and Barack Obama Respondents indicated their feelings (warm or cold) towards each candidate on a 100-point thermometer scale They also indicated whether they intended to vote for Obama or McCain

The Sample

All three studies reported in this paper utilize matched online samples recruited from the Polimetrix National Panel Polimetrix (PMX) maintains a large online panel of American adults (N in excess of one million) who agree to participate in surveys in exchange for accumulating credit points applicable towards acquiring various consumer products (e.g an Ipod) PMX has developed a matching-based methodology for sampling from their pools of opt-in respondents (details of the sampling methodology are available at www.polimetrix.com.) First, PMX

constructs a sampling frame from the American Community Study with additional data from the Current Population Survey voter supplement and the Pew Religious Life study.7 From this frame, PMX draws a stratified random sample (the target sample) of people similar in size to the desired sample from their opt-in panel Next, PMX searches their opt-in online panel for

respondents who most closely match the individuals in the target sample on the variables of race, gender, age, education, and imputed party identification On average, 2-3 matches are drawn for every person in the target sample all of whom are invited to complete the study From this set of completed interviews, PMX draws the final matched-sample taking the panelists who most closely match the target sample counterparts The end result is a sample of opt-in respondents with equivalent characteristics as the target sample on the matched characteristics listed above (see Rivers 2005).8

The panelists for all three studies were recruited to participate in a survey of related or political attitudes For Study 1, PMX fielded the online survey during the second week

election-in October and recruited 1100 registered voters On completelection-ing the survey, respondents were directed to the Project Implicit website where they were given a “warm-up” IAT designed to

7 The 2006 American Community Survey (ACS), conducted by the U.S Bureau of the Census, is based

on a probability sample of size 1,194,354 with a response rate of 93.1 percent.

8 The fact that PMX matches according to a set of demographic characteristics does not imply that their samples are unbiased All sampling modes are characterized by different forms of bias and opt-in Internet panels are no exception Systematic comparisons of PMX matched samples with RDD (telephone) samples and face-to-face interviews indicate trivial differences between the telephone and online modes, but substantial divergences from the face-to-face mode (Hill, Vavreck, and Zaller 2007) In general, the online samples appear biased in the direction of politically attentive voters For instance, in comparison with National Election Study respondents (interviewed face-to-face), PMX respondents were more likely

by eight percentage points to correctly identify the Vice-President of the US Because attentiveness is likely to be associated with recognition of cultural norms, it is possible that the level of under-reporting of racial bias may be somewhat higher in online samples in comparison with RDD samples.

acclimatize them to the reaction time protocol followed by the race and candidate IATs Finally, the IAT data were merged with the survey data

Results Convergent Validity

Convergent validity requires consistency across indicators of the same underlying

concept We examine the consistency of implicit and explicit attitudes within the race and

candidate preference domains, presenting the percentage of the national sample that favors

whites and Obama (see Table 1) Where appropriate, we compute Cohen‟s d as an approximate

measure of effect size.9 We also present the simple correlations (r) between the implicit and

explicit indicators

(Table 1 here) There is an unmistakable pattern to the data implicit and explicit attitudes diverged for race, but converged in the case of candidates for elective office The significantly lower level of

preference for whites (and the corresponding smaller values of Cohen‟s d) associated with the

explicit indicators suggests a considerable mismatch of explicit with implicit racial attitudes As generally documented in previous studies based on less representative samples (Nosek et al., 2002), explicit measures significantly understate the level of race bias in American society The estimate of racial preference based on the feeling thermometers, for instance, was 41 points

9 Cohen‟s d requires comparability of stimuli across the implicit and explicit domains (Cohen 1982) In

the case of race, we have full comparability between the IAT, the survey measure of overt racism, and the race thermometers In all these cases, the responses indicate positive or negative affect for blacks/whites The index of racial resentment, however, mixes items about race with items about political values Accordingly, it is not possible to calculate any measure of effect size attributable to race per se

Comparing effect size across indicators assumes equivalent midpoints (and endpoints) The items we compare here have very different metrics; the d values are thus presented as rough approximations of effect size

lower than the estimate based on the IAT: while 81 percent of the sample had at least some preference for whites on the IAT, only 40 percent were above the midpoint of the difference between the two feeling thermometers The average correlation of the three explicit measures with the race IAT was 25, suggesting only weak correspondence between explicit and implicit anti-black attitudes

Explicit and implicit evaluations of the 2008 presidential candidates, on the other hand, proved much more consistent on both mean levels of preference and implicit-explicit correlation

Cohen‟s d showed a relatively modest and uniform effect size associated with the race of the

candidate and the average spread in support for Obama between the three explicit measures and the IAT was less than five points Nonetheless, there was some evidence of a “Bradley effect” higher levels of explicit than implicit support for Obama Thus Obama “loses” the election on the basis of the candidate IAT (where McCain obtains 53 percent of the “vote”) The overall

correspondence of implicit and explicit evaluations was substantial the average correlation between the implicit indicator and the survey measures was 67, significantly higher than the same correlation for black and white social groups of 25.10

Predictive Validity

Our general objective in all three studies was to predict an aversive response to cues suggesting an individual is African-American or non-white Skin complexion is among the most visible of racial attributes and we incorporated a visual manipulation of Barack Obama‟s

10 We also carried out a confirmatory factor analysis of the implicit and explicit measures of racial bias and candidate evaluations In the case of the former, implicit and explicit measures loaded on two distinct factors; for the latter, however, the underlying factor structure was unidimensional (for further details, see Iyengar et al., 2009) The factor analysis thus reinforces the results presented in Table 1; implicit and explicit measures of prejudice diverge, but implicit and explicit measures of candidate evaluation

converge.

complexion into the 2008 election study Respondents were provided a photograph of Obama and McCain before they answered questions concerning their intended vote choice and feelings about the candidates We altered Obama‟s complexion by either lightening or darkening the facial image.11 Depending on the condition to which they were assigned, participants were either exposed to the actual photographs of the two candidates, the dark or light version of the Obama photograph, or no photographs at all.12 The full set of Obama images are shown in Appendix 1

We initially set the no picture condition as the baseline and estimated the effects of the dark, actual and light conditions separately The mean evaluations of Obama were no different across the three levels of complexion We then combined the light and no picture conditions on the grounds that in relation to the dark and actual complexion conditions, both represent a

weaker racial cue As the V measure for the facial pixels indicates, the actual level of darkness in the light condition differs from both the dark and actual conditions (mean V for four light images

= 72, mean for eight dark and actual images = 61; t = 3.68, p < 01, two-tailed) We then

estimated separate interactions between the three indicators of racial prejudice and exposure to the dark and actual picture conditions In the case of both survey indicators, the interactions were

11 We calibrated the darkness manipulation according to the brightness (V) component of the HSV color space using the ImageMetrics R package (see Messing et al., 2009) Lower values of V for facial pixels indicate a darker complexion In the dark conditions, the mean V score for the facial pixels was 54, in the actual image conditions the measure was 69, and in the light image conditions it was 72.

12 The study includes multiple images of light, dark, and actual complexion conditions because we also manipulated the presence of Afrocentric facial features by morphing Obama‟s face with two different prototypical Afrocentric or Eurocentric faces in the ratio of 70:30 The complexion of the four Afrocentric and Eurocentric source faces was first matched with the complexion level of the Obama images in the dark and light conditions In effect, as shown in Appendix 1, this procedure produces relatively Afro- and Eurocentric images with light and dark complexion respectively In this analysis, we pool across the four sets of morphs and focus on differences attributable to complexion because the morphing produced no effects – direct or indirect on evaluations of Obama Because face morphing tends to enhance the attractiveness of the morphed face, we also morphed McCain‟s face with the face of an unknown white male in the same 70:30 ratio The photographs of McCain and Obama remained on the screen while respondents provided their responses to the vote choice and thermometer questions

consistently non-significant In the case of the IAT, we obtained significant interactions on the thermometer score in both the dark and actual complexion conditions and a marginally

significant interaction on vote intention in both conditions Based on the pattern of these results,

we pooled the dark and actual complexion conditions In effect, our specification compares respondents given no image of Obama or a light image with those given a darker image

Our dependent measures are intended vote choice (scored 0 for a McCain vote, 5 for undecided or “can‟t say” responses, and 1 for an Obama vote) and the net difference in

thermometer ratings of each candidate (Obama feeling thermometer – McCain feeling

thermometer) As the survey included measures of overt racism and racial resentment as well as the race IAT, we are in a position to compare explicit and implicit racial attitudes as moderators

of the complexion effect

As shown in Table 2, both survey-based measures of prejudice (but not the IAT) proved significant predictors of support for Obama The more resentful and more racist respondents were more pro-McCain, but their preference was stable across the complexion manipulation Thus, the effects of resentment and overt racism on candidate preference were homogeneous in the two sets of complexion conditions Although the IAT score had no direct effect on the mean evaluations of the candidates, the IAT x complexion interaction was robust (p < 05) in the case

of the feeling thermometer and marginally significant (p < 10) for vote intention.13 In the dark conditions, the effects of the D-score proved non-significant In the darkened and actual

13 Table 2 presents the results of an ordinal logit model of the vote choice variable and an ordinary least squares (OLS) model of the net thermometer measure In each case we present Huber-White standard errors All estimates were produced with the “Design” package in R; all tables produced with the

“apsrtable” package.

complexion conditions, however, support for Obama dropped at higher levels of anti-black implicit bias and increased among those with an implicit preference for blacks

(Table 2 here) The effect size of the IAT x complexion interactions was non-trivial We simulated predicted probabilities based on random draws (N=10,000) from the asymptotic distribution of model parameters, while varying complexion and our measures of prejudice (racial resentment, overt racism and IAT score) and holding other variables at their mean values (see King, Tomz, and Wittenberg, 2000) The left panels of Figure 1 plot the simulated predicted probabilities of voting for Obama and Obama‟s net thermometer rating (with 95 percent confidence intervals) for respondents with IAT scores ranging from two standard deviations below and above the mean The middle and right panels repeat this exercise, but with the racial resentment and overt racism indices in place of the IAT

(Figure 1 here) Among respondents with IAT scores two standard deviations above the mean, the

predicted probability of voting for Obama decreased by 114 (24 percent) in the dark condition compared to the non-dark conditions Conversely, among respondents with IAT scores two standard deviations below the mean (i.e those with pro-black implicit bias), exposure to the dark condition increased the likelihood of voting for Obama by 164 (46 percent) When we use a one standard deviation departure from the mean as the basis for comparison, the probability of voting for Obama fell by nearly 045 (10percent) in the dark condition at higher levels of anti-black implicit bias and increased by 095 (25 percent) among respondents with pro-black implicit bias

In the case of the feeling thermometer (which ranges from -100 to 100), the effect size of the complexion manipulation increased by 15 points among respondents with extremely high IAT scores and by nine points among those with moderately high levels of implicit bias

The middle panel of Figure 1 shows that racial resentment exerted very strong effects on support for Obama across both levels of the complexion manipulation Among respondents with resentment scores one standard deviation above the mean, the probability of voting for the darkened version of Obama decreased by 199 (45 percent) compared to the average respondent;

in the case of the lightened condition, their support fell by the same margin (.196 or 47 percent; for similar evidence concerning the importance of resentment as a predictor of support for Obama, see Tesler and Sears, 2010; Jackman and Vavreck, 2010) Clearly, the effects of

resentment on candidate preference are substantial, but unlike the case of the race IAT, the effects are invariant to differing levels of the racial cue The more resentful are no more likely to support a light-skinned African-American candidate

Finally, the bottom panel of the figure presents the predicted mean scores based on different levels of the overt racism index As in the case of resentment, overt racism had a significant, but uniform effect on the indicators of candidate support; respondents with relatively high racism scores were equally averse to Obama in the lighter and darker conditions

We also tested for an interaction between respondent race and skin complexion American respondents were more likely than whites to evaluate the darker images of Obama favorably with respect to the thermometer ratings (p < 05), and voting intent (p < 05).14 This

African-14 Covariates in these models included respondent education, race, party identification, an index of economic voting the complexion dummy, and the race x complexion interaction These models excluded Hispanic participants.

result serves as a useful manipulation check, and also supports the argument that blacks are more likely to respond to the in-group race cue provided by darker complexion than the status cue conveyed by lighter complexion (see Hochschild and Weaver, 2007 for a review of the literature

on the social advantages accruing to blacks with lighter complexion)

The absence of interaction effects between the two indicators of explicit prejudice and skin complexion may reflect the fact that by the time of this study (mid-October) most people had already learned about Obama‟s ethnicity Since media coverage of Obama‟s racial heritage was extensive, his ethnicity was common knowledge Alterations to his complexion, therefore, may have done little to prime explicit racial attitudes In effect, the overt measures may have failed the predictive validity test because, by the end of the 2008 campaign, there was nothing we could do to impact voters‟ conscious categorization of Obama‟s race

The presence of the expected interactions between the complexion manipulation and the IAT, however, suggests that the complexion cue however subtle was sufficient to strengthen the effects of implicit bias on candidate preference As an implicit cue, the manipulation would

be more likely to activate implicit over explicit attitudes Whatever the underlying mechanism, the pattern of interactions suggests that it is implicit but not explicit attitudes that predict

responsiveness to race cues

Study 2

In this study we replicated the complexion manipulation, but in the context of public opinion toward prospective immigrants Unlike Study 1, the target individual was completely unknown and not African-American, but either Hispanic or Middle Eastern We presented a national sample of 1250 Polimetrix respondents with two brief vignettes each describing a

potential male immigrant Subjects were told that the vignettes represented brief excerpts from

an actual application for a work permit, and that their task was to consider each applicant‟s

background to decide whether (1) the applicant should be awarded a work permit, (2) in the event the temporary worker later decided to remain in the country, whether he should be awarded citizenship.15

The text of the vignette manipulated the immigrant‟s occupational background and

nationality Half the participants encountered a manual laborer with only a high school education, the remaining half were told that the immigrant was a professional (either an engineer or a

computer scientist) with advanced college degrees Respondents also learned that the applicant was either a citizen of Kuwait or Mexico As shown in the Appendix, the accompanying

photograph representing the Kuwaiti/Mexican was morphed with either a prototypically

Afrocentric or Eurocentric face (the faces in the extreme left and right columns respectively).16The treatment images of the two candidates (shown in the middle columns) represent a 60:40 blend in which each immigrant‟s face includes 60 percent of the original features and 40 percent

of either Eurocentric or Afrocentric features.17

We computed an index score representing respondents‟ overall treatment of the

immigrant‟s application A score of zero represents the extreme negative posture (denial of the

15 In the case of the work permit and citizenship questions, respondents could select “approve,”

“disapprove,” or “can‟t say” as the response options We scored these as 1, 0 and 5, respectively

Responses to the length of stay question were made on a four-point time scales that ranged from six months to three years We scored the maximum length as 1 and the minimum as 0.

16 We selected the prototypical faces from Dr Jennifer Eberhardt‟s face database (Stanford University, Psychology Dept) This database includes 100 Afrocentric and Eurocentric faces that were rated by

student judges for stereotypicality, attractiveness, and age We selected two exemplars of each category with high stereotypicality ratings that were matched for attractiveness and age.

17 As in Study 1, we calibrated our indicator of Afrocentrism according to the brightness (V) component

of the HSV color space In the Hispanic-Eurocentric condition, the average V measure was 71; in the Afrocentric condition it was 54 The corresponding entries for the Middle Eastern condition were 73 and 56 respectively The scale of the manipulation is thus nearly identical to the difference between the light and dark conditions in Study 1.

work permit, denial of citizenship), while a score of 1 would indicate the opposite – granting of both work permit and citizenship

Results

We present both separate and pooled (across the two candidates) results showing the effects of the Afrocentrism manipulation on willingness to admit individual immigrants (see Table 3) In this study our measure of explicit prejudice is limited to the index of racial

resentment and our measure of implicit bias is based on the brief version of the IAT,18 which generates the same D-score ranging from -2 to 2 Once again, our key test of predictive validity

is the interaction between the visual manipulation and the measures of racial prejudice The results are consistent with Study 1, but stronger After adjusting for the effects of the status and nationality manipulations (both of which were robust) and other covariates including a measure

of support for open immigration policy, respondents with higher D-scores (more prejudiced) were significantly more likely to alter their evaluations of the immigrant based on the

complexion cue As shown in Table 3, the IAT x Afrocentric complexion interaction was

significant for both immigrant candidates presented (p < 05 separately, p < 01 pooled)

Respondents with higher resentment scores, however, evaluated the immigrants negatively irrespective of their appearance, as indicated by the lack of a significant interaction between the racial resentment index and the Afrocentric complexion manipulation

(Table 3 here)

18 The D-score based on the brief IAT has the same properties as the score based on the full version of the test The difference is that the brief IAT tests only category-positive pairings and thus only entails two blocks and as few as 40 trials, compared to the seven blocks and 180 or more trials that comprise the full IAT (For evidence concerning the predictive utility of the brief IAT, see Sriram et al, 2010.)

To assess the effect size of the IAT x complexion interaction, we again simulated

predicted probabilities based on the model coefficients for the pooled model, allowing

complexion and racial resentment to vary while holding other varaibles at their mean values (see King, Tomz, and Wittenberg, 2000) The left-hand panel of Figure 2 plots the predicted values of the approval index for work permits and citizenship (with 95 percent confidence intervals) for respondents with IAT scores one or two standard deviations below and above the mean The right-hand panel repeates this excercise with the racial resentment index in place of the IAT

Among respondents with IAT scores two standard deviations above the mean, the

predicted approval index score declined by 156 (22 percent) in the dark conditions compared to the light conditions Conversely, among respondents with IAT scores two standard deviations below the mean (i.e., those with pro-black implicit bias), exposure to the Afrocentric

manipulation increased the predicted approval index score by 114 (18 percent) When we used a one-standard deviation departure from the mean as the basis for comparison, the predicted work permit-citizenship approval index score fell by 086 (12 percent) in the dark condition at higher levels of anti-black implicit bias, and increased by 05 (8 percent) among respondents with pro-black implicit bias

(Figure 2 here) Unlike the case of support for a black candidate for political office, racial resentment did not have a substantial independent effect on willingness to admit immigrants, nor did the impact

of racial resentment vary according to whether the immigration candidate had a more Afrocentric appearance or Eurocentric appearance The right panel of Figure 2 shows that among

respondents with resentment scores one standard deviation above the mean, the predicted

immigration index for the Afrocentric version of the immigration candidate fell by 004 (2