Implicit Closeness to Blacks, Support for Affirmative Action, Slavery Reparations, and Vote Intentions for Barack Obama in the 2008 Elections

Abstract

Does pro-Black policy support require an individual to be unbiased? I distinguish two types of implicit attitudes based on whether the attitude-target is evaluated as an object (evaluative associations) or as an independent social agent (relational associations). In a series of studies (N = 3,073), a significant anti-Black evaluative association bias emerges. In contrast, relational associations are significantly pro-Black and are unrelated to evaluative associations. Relational associations predict opinions regarding affirmative action, government help for Blacks, slavery reparations, and intentions to vote for Barack Obama. Thus, minority representation based on relational associations may not require absence of anti-minority evaluative bias.

Notes

¹Five positive and five negative race-unrelated words were chosen from this list of 1,036 normed words with a frequency of at least F = 25 and valence ratings greater than 8 for positive words and less than 2 for negative ones.

²The question at the top of the screen read, “Does this word have a GOOD or a BAD meaning?” followed by the word and the response options “1 = Good word” and “0 = Bad Word”.

³To reduce skewness associated with raw reaction time measures, all responses shorter than 250 ms and exceeding 2,500 ms were eliminated (truncation), log-transformed, and converted to facilitation scores by subtracting the reaction times associated with racial prime-target pairs from neutral prime-target pairs. To prevent retinal after-images from becoming consciously perceptible, subliminal primes were masked with random character strings.

⁴To keep the underlying psychological mechanism comparable to the IAT, I used word rather than picture primes. Fazio and Olson (2003) argued that the mechanisms underlying sequential priming with pictures and the IAT may differ because the IAT forces respondents to “correctly” classify pictures as representing either a Black or White individual. They argued that, in contrast to priming, “the IAT seems to assess associations to the category labels, not automatically activated responses to the individual exemplars” (Fazio & Olson, 2003, p. 315). They argued that this distinction does not apply when racial labels rather than pictures are used as primes (footnote 5; Fazio & Olson, 2003, p. 315).

⁵Racial primes were displayed for merely three frames. In a computer with a 60 Hz monitor this represents 50 ms, and in one with a 100 Hz monitor merely 30 ms. In Study 1 (the college student study; n = 555) only 12% of the respondents claimed seeing racial group names displayed on the computer screen, an estimate likely to be inflated by guessing as 3% falsely claimed seeing pictures of animals displayed on the screen. The interpretation of results does not substantially change when the 12% prime-recognizing respondents are removed from analyses in Study 1 (results not shown). It is important to note that the IAT would have required 100% prime recognition.

⁶Because there are many individuals or groups an individual may feel close to, each characterized by a unique combination of “matching” and “mismatching” traits, a large number of traits must be used. Whether implicit closeness toward two different racial groups can be distinguished with the “self–other overlap” measure is addressed in the Results section.

⁷To investigate possible confounds that could arise if self-Black matching traits are evaluated more positively than self-Black mismatching traits, I compared average likability ratings (Anderson, 1968, in percentage of the likability scale) for self-Black matching and self-Black mismatching traits using a paired-samples t test. The average likability score for self-Black matching traits was 47.50% (SE = 0.14) and for self-Black mismatching ones 47.67% (SE = 0.18, n = 2,716). The difference (−0.17%) failed to reach significance at conventional levels (paired t = − 0.59, p = .56).

*p < .05. **p < .01. Correlation coefficients in the bottom panel represent Spearman's rho. The coefficient d represents effect size according to Cohen's d (Cohen, 1988) for bias (difference from zero) with d = m/s, whereby m is the mean and s the standard deviation; effect size for correlation coefficients: . B-W = pro-Black versus pro-White.

⁸Pearson's r requires that data in each variable must be normally distributed, whereas Spearman's rho relaxes the normality assumption. I used Spearman's rho because feeling thermometer ratings toward African Americans were significantly skewed (Skewness = −0.61, SE = 0.05), thus violating the normality assumption. However, the interpretation of the results does not change when Pearson's r is used (results not shown).

^a Favorability of traits rated in the explicit trait survey to be characteristic of Blacks based on Anderson's (1968) likability norms.

^b Neutral Implicit Black Closeness: Implicit Closeness measure computed only from 30 neutral traits based on Anderson's (1968) trait ratings: worrier, self-righteous, sarcastic, unpredictable, submissive, shy, spendthrift, aggressive, timid, forward, argumentative, nonchalant, materialistic, methodical, wordy, shrewd, skeptical, persistent, naïve, proud, opportunist, systematic, choosy, objective, restless, sophisticated, emotional, persuasive, blunt, serious.

^a Favorability of traits rated in the explicit trait survey to be characteristic of Blacks based on Anderson's (1968) likability norms.

^b Neutral Implicit Black Closeness: Implicit Closeness measure computed only from 30 neutral traits based on Anderson's (1968) trait ratings: worrier, self-righteous, sarcastic, unpredictable, submissive, shy, spendthrift, aggressive, timid, forward, argumentative, nonchalant, materialistic, methodical, wordy, shrewd, skeptical, persistent, naïve, proud, opportunist, systematic, choosy, objective, restless, sophisticated, emotional, persuasive, blunt, serious.

⁹To investigate potential effects of other relevant demographic variables, I compared the regressions in the third row of panels in Table 2 and in the left and center panels of Table 3 to separate regressions for Black non-Hispanic and White non-Hispanic respondents, respondents above and below median educational achievement, above and below median income, as well as for male and female respondents. Comparing 362 pairs of coefficients yielded only 11 significant differences (3.04%). This is well below the level expected by pure chance at the p < .05 level.

¹⁰I replicated the analyses in the third row of panels in Tables 2 and 3 with the neutral implicit closeness score. Because the number of trials for the neutral score is only one third of the number of trials for the overall score, measurement error is expected to be larger, and the chance of observing the effect with control variables entered is diminished. The effect of the neutral implicit closeness variable emerges as significant at p = .02 with respect to slavery reparations but fails to reach significance at conventional levels with respect to the other dependent variables.

¹¹Even some postelection analyses raise serious questions. For example, Böhm, Funke, and Harth (2010) found a consistent same-race voting preference among voters in the 2008 Democratic primaries and caucuses. With 60% of the Democratic electorate being White and only 22% Black as recently as 2012 (Newport, 2013), Barack Obama's nomination seems difficult to explain.